researchonindustrialwasterecoverynetworkoptimization ...conv feature map 256-d figure...

Research ArticleResearch on Industrial Waste Recovery Network Optimization:Opportunities Brought by Artificial Intelligence

Bin Liao and Ting Wang

School of Management, University of Guizhou, Guiyang, China

Correspondence should be addressed to Ting Wang; [email protected]

Received 3 November 2019; Revised 26 February 2020; Accepted 23 March 2020; Published 25 April 2020

Guest Editor: Davide Castellano

Copyright © 2020 Bin Liao and Ting Wang. /is is an open access article distributed under the Creative Commons AttributionLicense, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work isproperly cited.

With the acceleration of industrialization and urbanization in China, a large amount of waste in industrial parks has become themain cause of regional environmental pollution. In order to solve this problem, this paper relied on artificial intelligence’sprediction technology and image recognition technology to intelligently upgrade the traditional industrial waste planningmanagement system and designed a waste intelligent classification center with intelligent prediction and intelligent classificationcapabilities. So, as to realize this new intelligent classification center and explain its value, this paper explains the key imple-mentation technology of this intelligent classification center and validates it by constructing a multitarget location model thatconsiders both economic and environmental benefits.

1. Introduction

With the rapid development of industry and economy, theproblem of environmental pollution is becoming more andmore serious. In the face of the “three great mountains” ofcompetition proliferation, resource depletion, and envi-ronmental degradation, governments have begun to attachgreat importance to the coordinated balance between eco-nomic development and environmental protection issues.As of 2017, China produces about 3.3 billion tons of in-dustrial waste each year, with a total stockpiled over 60billion tons over the years. Industrial waste that has not beendisposed of in a timely manner has not only caused a hugewaste of land resources but also caused regional ecosystemdisturbances [1–3]. How to safely, economically, environ-mentally, and efficiently treat these wastes has become one ofthe important research topics for the development of asustainable economy in China.

/e social demand for recovering and reusing discardedproducts has stimulated a research boom in product-recy-cling networks [4]. After years of accumulation, the researchon the waste product-recycling network has achieved results,but it mainly focuses on the recovery and reuse of mobile

phones, automobiles, washing machines, and other elec-tronic products. Few studies have discussed the issue ofindustrial waste recycling [5–9]. In the existing research onindustrial waste management, most scholars treat industrialwaste and domestic waste as urban waste for discussion.Huang and Li discussed the problems of Wuhan municipalwaste recycling, treatment, and resource reuse. Finally, theyproposed the scientific classification of municipal waste anddiversified garbage collection, charging, and treatmentmanagement methods to promote Wuhan’s waste man-agement related development of the industry [10]; Li et al.proposed an urban waste recycling system based on radiofrequency wireless network, including system function de-scription, hardware design, and software design. /e systemuses sensor technology and radio frequency wireless net-work technology to realize the information collection ofurban garbage bins and the positioning of urban garbagecollection systems. /is research has realized the dynamictracking and monitoring of garbage from the source to thedisposal terminal and has made some contributions to theintelligent development of garbage recycling management[11]; Wang et al. established a game relationship model forresidents, receiving and transporting enterprises, and waste

HindawiMathematical Problems in EngineeringVolume 2020, Article ID 3618424, 11 pageshttps://doi.org/10.1155/2020/3618424

mailto:[email protected]://orcid.org/0000-0002-7663-6406https://orcid.org/0000-0002-4335-9078https://creativecommons.org/licenses/by/4.0/https://creativecommons.org/licenses/by/4.0/https://doi.org/10.1155/2020/3618424

processing enterprises under different classification ratiosbased on the Stackelberg game theory. /e proportion of theresponsibility of the waste management of the subject andthe simulation analysis of the model show that the wasteclassification capacity of industrial waste service enterprisesis the main factor affecting the interests of the subjects in therecycling network [12].

Importantly, as the process of industrialization andurbanization continues to accelerate, the large concentrationof industries and populations has saturated the load of theprevious urban waste transportation network, which has ledto a sudden decline in the efficiency of the traditional in-dustrial waste. As a result, severe waste accumulation andenvironmental pollution have occurred in many regions ofChina. /erefore, the separation of traditional waste recy-cling systems and the design of a recycling managementnetwork dedicated to the treatment of industrial waste toimprove the efficiency of waste treatment are becomingresearch hotspots in the field of industrial waste manage-ment and reuse.

In the research of industrial waste recyclingmanagementnetwork, scholars mostly discuss the industrial waste net-work planning and design based on the perspectives oflogistics cost, service level, and equipment utilization effi-ciency [13–15]. Wang et al. constructed a two-objectiveoptimization model when designing the logistics network toanalyze the impact of different input costs on the envi-ronment in the design of the supply chain while consideringthe two goals of cost and environmental protection level toanalyze the project’s return on investment. /e rate has apositive enlightening effect on the future green managementof the supply chain [16]. Since then, Elhedhli and Ryanfurther explore the relationship between distance and car-bon emissions [17]; Rui et al. considered a logistics networkplanning model with a hard time window and analyzed theimpact of different time windows on the economy and theenvironment [18]; Zhang and Li-rong others have consid-ered the uncertainty of the quality of the recycled products inthe market, set up a multiobjective stochastic programmingmodel with the goal of minimizing logistics network costsand environmental pollution, and solved the stochasticprogramming model with scenario analysis [19].

/is thesis considers the intelligent upgrading of in-dustrial waste recycling management network in an in-dustrial park with multiple factories. First, the artificialintelligence prediction technology and image recognitiontechnology are used to intelligently upgrade the traditionalindustrial waste planning management system. An intelli-gent classification center for industrial waste with intelligentprediction and intelligent classification capabilities isdesigned, and its key implementation technology is dis-cussed. Finally, the value and significance of the intelligentclassification center designed in this paper are illustrated byconstructing a multiobjective site selection model consid-ering both economic and environmental benefits.

/e rest of this article is organized as follows. /e nextsection puts forward the key technologies for the realization ofthe industrial waste intelligent classification center systemarchitecture, including the industrial waste production and

component predictionmodels based onBP learning algorithmsand deep learning-based waste identification and detectionimplementation technologies. In the fourth section, the siteselection model, implementation algorithm, and numericalsimulation of the industrial waste intelligent recycling centerconsidering economic and environmental benefits aredesigned. /e last section discusses and summarizes the thesiswork.

2. Design of Artificial Intelligent WasteSorting Center

2.1. System Framework of Artificial Intelligence RecyclingClassification Center. /e optimization of an industrial wasterecycling network system is essentially a multilevel, multi-objective management decision problem. To upgrade thetraditional model, it is necessary to redesign a decision modelwith corresponding functions. In this research, artificial in-telligence technology is integrated into the traditional wastelogistics recycling system. /e system framework of the arti-ficial intelligence recycling classification center that integratesthe two core technologies of waste production predictiontechnology and image recognition is designed, as well as itsfunctional decisionmodel. It can be seen fromFigure 1 that thisintelligent recycling classification center has basic decision-making functions for industrial waste production forecasting,component forecasting, image recognition, and intelligentlocation selection. During the operation of the classificationcenter, the quantity and category prediction of the waste to beprocessed is first based on the preset related parameters. On theone hand, these prediction results can be used as training sets totrain and detect the image recognition model. On the otherhand, it can be combinedwith processing power as a parametersource for the location model of the intelligent classificationcenter to verify the economic and environmental benefits ofsuch intelligent classification centers replacing traditional ar-tificial classification centers. It can be seen that the data flow isprogressively advanced between different models, forming anorganic whole with complex relationships.

2.2. Key Technologies Implemented by Artificial IntelligenceRecycling Classification Center. As can be seen from theabove, the basis for realizing the artificial intelligence clas-sification center designed in this paper is the design of in-dustrial waste prediction technology and image recognitiontechnology. In this section, we focus on the design ideas andalgorithms of these two technologies.

2.2.1. Prediction Model of Industrial Waste Products andComponents Based on BP Learning Algorithm. /e neuralnetwork is a kind of multilayer feedforward neural networktrained according to the error backpropagation algorithm. Itis the most widely used neural network. /e artificial neuralnetwork does not need to determine the mathematicalequation of the mapping relationship between input andoutput in advance. It only learns certain rules through itsown training and obtains the result closest to the expectedoutput value when given the input value. With the

2 Mathematical Problems in Engineering

advantages of high prediction accuracy and robust results,this method is widely used in intelligent prediction, machinelearning, and other fields [20]. /is paper designs the BPnetwork prediction model of industrial waste products andcomposition, as shown in Figure 2.

/e core of an artificial neural network to realize itsfunction is the algorithm. BP neural network is a multilayerfeedforward network trained by error backpropagation. /eBP learning algorithm is an error correction method basedon this principle. /e learning process of the algorithmconsists of forwarding and backpropagation (partial com-position). When an input mode of a network is given, it ispassed from the input layer unit to the hidden layer unit,processed by the hidden layer unit, and then sent to theoutput layer unit, and an output mode is generated by theoutput layer unit. Because this is a layer-by-layer effect in theprocess, each layer of neurons can only affect the state of thenext layer of neurons, which is called forward propagation; ifthe output response is in error with the expected outputmode and does not meet the requirements, then the error istransferred to the backpropagation and then transmitted./e value is transmitted layer by layer along the connectionpath, and the connection weight of each layer is modified.For a set of samples, learning is performed by using differenttraining modes, and the model is continuously repeated inthe forward and backward propagation processes. Onlywhen each training mode meets the requirements, the BPnetwork training is completed. Network learning is a processof minimizing the objective function to complete the input-to-output mapping. Generally, the objective function isdefined as the sum of the squared error of the output layerunit’s desired output and the actual output on all inputmodes. /e prediction algorithm of industrial waste pro-duction and composition in this paper is as follows:

Step 1: suppose the input layer has nneurons, the inputvector is X(x1, x2, . . . , xn); the hidden layer has mneurons, the hidden layer vector is H(h1, h2, . . . , hm);the output layer has k neurons, the output vector is

Y(y1, y2, . . . , y)k. /e weight between the input layerand the output layer is and the threshold is ; the weightbetween the hidden layer and the output layer is and thethreshold is.Step 2: assign initial values to all connection weightsand node thresholdsStep 3: do the following calculations for each inputsample:

(a) Forward calculation:

yk � m

j�0ωjk·hj, hj �

n

j�0ωij·xi. (1)

Let the transfer function be a sigmoid function, that is,f(x) � (1/1 + exp(−ax))(a> 0).For the Pi sample in the training set, the network inputvector is X(x1, x2, . . . , xn), the actual output isY(y1, y2, . . . , yk), the expected output isT(t1, t2, . . . , tn), and the error function is defined asfollows:

E �12

i

k�1tk − yk(

2. (2)

Let the total number of samples is N, then the mean ofthe squared error is

EAV �1N

N

N�1E. (3)

EAV is the mean of the squared error. /e purpose ofthe algorithm learning is to minimize EAV.

(b) Inverse calculationWhen the algorithm fails to meet the target expec-tations, it will calculate backward. /e first is tocalculate the local gradient δk. /e first is to calculate

Plant number forecast Indicators of industrial economicdevelopment

Industrial waste production

Garbage components

Optimal waste treatment scheme

Industrialwaste

productionforecasting

model

Industrialwaste

componentprediction

model

Imagerecognition

classificationmodel

Consider total system cost

Consider system efficiency

Best siteselection scheme

Optimal logisticscost model

Best wastedelivery solution Industrial layout andgeographic information

Intelligentindustrial waste

recyclingnetwork and

systemoptimization

Industrial waste

Functional framework of artificial intelligence recyclingclassification center Optimization model of recycling network system

Figure 1: System framework of artificial intelligence recycling classification center and its functional decision-making model.

Mathematical Problems in Engineering 3

the local gradient. It is worth noting that for theoutput and hidden layers, the algorithms of the localgradient are different, as shown in formulas (4) and(5):

Output layer δk � tk − yk( f xk( , (4)

Hidden layer δk �δEδyk

f xk( . (5)

(c) Correction weight:

Using the minimum value of the gradient descentmethod, the updated amount Δωij of ωij can beexpressed by the following formula:

Δωij � −ηzE

zωij. (6)

η represents the learning rate of the neural network,and its value is greater than zero.Step3: enter a new sample until EAV meets the pre-determined requirements. Eventually, the model canoutput the annual industrial waste output and com-position of the region. /ese data will be used as atraining sample for image data and site selection fortraining to implement waste identification and moni-toring technology data source for the model.

2.2.2. Realization Technology of Waste Recognition andDetection Based on Deep Learning. Deep learning is anemerging research field of machine learning. Its researchcontent is to automatically extract multilayer features fromthe data and represent them. In the process, a series ofnonlinear transformations can be used to extract features

from the original data and finally realize machine vision andintelligent operation such as machine sensing. /is researchuses the idea of deep learning to design the target detectiontechnology of industrial waste images and uses the type,position, size, and confidence of the target object as thedetermination parameters to achieve intelligent detectionand recognition of the predetermined target object. /ebasic flow of the target detection technology is shown inFigure 3.

/e target detection algorithm is the key to achieveintelligent detection. Considering the complexity of in-dustrial waste detection tasks in different environments,this study improves the Faster RCNN [21] target detectionmodel, which combines the RPN network and Fast-RCNNUnified to identify the location and angle of industrialwaste at the same time. Faster RCNN model, which iscomposed of two subnetworks: Region Proposal Gener-ation (RPN) network and VGG-16 classification networkas shown in Figure 4.

/e RPN network adopts the design method of thefully convolutional network, in which the convolutionallayer uses the VGG-16 model and shares convolutionswith subsequent classification networks to reduce doublecounting. Figure 5 shows the structure of the RPN net-work. A small network is used to slide the feature imageoutput by the last shared convolution layer to generate theregion of interest. /e input of this small network comesfrom the n × n size window input on the convolutionalfeature image, and then the features of each slidingwindow will be mapped to the lower dimensional featurevector. After the mapping, Re LU will perform nonlinearprocessing. /e features, thus, obtained are used as inputto two parallel fully connected layers—the bounding boxreturn layer (REG) and the classification layer (CLS). /eregression layer is responsible for readjusting the

Number of factories

Regional industrialproduct

Input node Hidden layer Output node

Industrial waste production

.

.

Input node Hidden layer Output node

Waste type 1

Waste type 2

Waste type 3

Waste type 4

Waste type 5

Waste type 6

Waste type N

. .

.

.

Figure 2: BP network prediction model for industrial waste products and components.


Industrial wasteimages Candidate area Feature extraction Classification

Border regressionanalysis

Training samples

Positive sample

Negative sample

Feature extraction

Training samples

Figure 3: Basic flowchart of industrial waste monitoring technology.

2k scores 4k coordinates

Cls layerReg layer

Intermediate layer

Sliding window

Conv feature map

256-d

Figure 4: RPN network structure.

Waste generation Artificial intelligence recycling classification center

Resale marketResale market

Remanufacturing center

Waste landfill center

Clouddistribution

platformWaste type2

Waste type1

Waste type3

Figure 5: Industrial waste smart recycling model concept illustration.


inaccurate area of interest, and the function of the clas-sification layer is responsible for determining whether thearea belongs to a candidate area. /is processing methodensures that all features are associated with the REG andCLS layers. At the position of each sliding window, de-termine whether multiple regions of interest containobjects at the same time. /e number of regions predictedsimultaneously at each position is recorded as k. /us, thebounding box regression layer outputs a total of 4k co-ordinates for k regions, and the classification layer outputsa total of 2k scores to evaluate whether each regioncontains an object. In each sliding window position, 3scales and 3 aspect ratios are used by default, so there arek � 9 anchor points in total. So, for a featured image of sizew × h, there are “w∗ h ∗ k anchor points. After obtainingcandidate regions of interest through the RPN network,the VGG-16 network pretrained on the ImageNet datasetis used for classification.

/e VGG-16 network is a convolutional neural net-work structure proposed by the VGG group in 2014. /edataset is ImageNet. It uses 16 network layers with pa-rameters, including 13 convolution kernels with a size of3 × 3. It consists of 3 fully connected layers [22]. In ad-dition, the same as Fast R-CNN, on the last convolutionallayer, the spatial pyramid pooling (SPP) layer is also usedto pool the feature layers to the same size. In the originalRPN scheme, a decoupling scheme different from thetraditional model is adopted, that is, the angle predictionlayer is separately connected behind the copied FC layer ofthe VGG-16 network, so that it is not updated at the sametime as the position of the industrial waste and the cat-egory layer. /e scheme can further reduce the error of thebottle object angle prediction. As with Fast R-CNN, ourtraining model uses a minimizing multitask loss function./e loss function of the RPN and VGG-16 networks is thesame. /e loss function consists of two terms in thefollowing formula:

L Pi , ti ( �1

Ncls

i

Lcls pi, p∗i( + λ

1Nreg

i

pi∗

Lreg ti, ti∗

( .

(7)

Among them, i represents the pulling of anchor points inmini-batch and Pi represents the probability that the ianchor point is predicted to be an object. If the anchor pointbelongs to a positive sample, then p∗i is 1, otherwise, it is 0. tiis a 5-dimensional vector and belongs to the parameterizedpredicted object’s bounding box coordinates, and t∗i is the tilabel associated with the positive anchor. /e classificationloss function Lcls is calculated using a softmax loss strategyand the regression loss calculation is performed using arobust regression loss function (Smooth L1). When p∗i � 1,it means that the regression loss is only activated at theanchor point for processing positive samples. At this time,the outputs of the corresponding CLS and REG layers arePi and ti .

Among them, for the graphs bounding box regression,we adopt the R-CNN parameterization method to unify theregression parameters:

tx �x − xa(

wa ,

ty �y − ya(

ha ,

tx∗

�x∗ − xa(

wa ,

ty∗

�y∗ − ya(

ha ,

tw � logw

wa ,

th � logh

ha ,

tw∗

� logw∗

wa ,

th∗

� logh∗

ha ,

tp �t

2π ,

tp∗

�t∗

2π .

(8)

Among them, x, y, andw represent the center coordi-nates of the frame, h represents the length, xa represents theposition of the anchor point, and x∗ and y∗ represent theposition of the callout frame.

3. Study on Site Selection of IntelligentIndustrial Waste Recycling Center

3.1. Problem Statement. /e key technologies of the artificialintelligence recycling classification center have been explainedabove. It is not difficult to find that the artificial intelligencerecycling classification center with image recognition technologyand prediction technology can save a lot of labor costs. In orderto verify the economic benefits of the designed artificial intel-ligence classification and processing center, from the perspectiveof system integrity, this paper designs a two-level goal of systemplanning research with system cost minimization and systemefficiencymaximization./e intelligent processing center will beintegrated and sent to the waste data according to the operationprocess. At the same time, the optimal number and location ofintelligent processing centers in the industrial waste processingnetwork will be determined based on the geographical locationof the various end-of-waste processing nodes. Finally, theeconomic benefits are judged by comparing the differencebetween the construction cost and the saved human resources.

Before this, it is necessary to briefly explain the operationmode of the entire waste system. /is research integratesartificial intelligence technology into the traditional wastelogistics recycling network and connects the various


recycling nodes of industrial waste through the artificialintelligence classification center. /e goal is to achieve theintelligence of the full recycling network, minimize costs,and maximize efficiency. /e waste in the city is classifiedand processed (e.g., landfill, remanufacturing, and resale).Simplifying the problem can be summarized as the mainpoint of industrial waste generation, artificial intelligenceprocessing center, remanufacturing center, and landfillcenter, and for the small-cap market, the relationship isshown in Figure 5.

3.2. Mathematical Formula

3.2.1. Model Assumptions. In order to facilitate the analysisand description of the problem, this article makes someassumptions:

(1) All activities included in the recycling model pro-posed in this article are carried out in this model

(2) /e system relies on existing secondary market,remanufacturing plant, no longer considers theirconstruction costs, and only considers their oper-ating costs

(3) /e transportation costs of industrial waste arelinearly related to the distance

(4) /e waste output predicted by the trained neuralnetwork is equal to the real output

3.2.2. Model Parameter Setting and Description. /e pa-rameters involved in the model built in this paper mainlyinclude node parameters, other parameters, general pa-rameters, and decision variables, as shown in Table 1.

3.2.3. Objective Function and Constraints. /e objectivefunction considered in this paper is mainly the economicand environmental benefits of building an artificial intelli-gence classification center, so the following two objectivefunctions are constructed:

maxF1 �kZ

ERmL + kZ

WRkl(

jZQWjk

⎛⎝ ⎞⎠, (9)

minF2 � k

bnk + λ Z

WRkl + Z

ETml

⎡⎣ ⎤⎦

− k

pWk + α

k

ZQWjk + β

k

ZWEkm + c Z

WTmn + Z

ETmn

⎡⎣ ⎤⎦

− k

dZQWjk g +

k

dZWEkm g +

k

dZWTmn g +

k

dZETmng

⎛⎝ ⎞⎠⎡⎢⎢⎣

+ k

dZWRkl g +

k

dZERml g

⎛⎝ ⎞⎠⎤⎥⎥⎦.

(10)

Among them, equation (9) indicates that the maximumreuse rate of waste is the reflection of the environmentalprotection benefits of the recycling network, and equation (10)indicates that the overall revenue of the entire recycling networksystem is maximized, which is a reflection of economic benefits

s.t.

j

ZQWjk �

k

ZWEkm +

k

ZWRkl +

k

ZWTkn , ∀k, j, (11)

k

ZWEkm � Z

ETml + Z

ERmn, ∀k, (12)

XWEk ·

j

ZQWjk �

k

ZWEkm , ∀k, (13)

XWRk ·

j

ZQWjk �

k

ZWRkl , ∀k, (14)

XWTk ·

j

ZQWjk �

k

ZWTkn , ∀k, (15)

ZQWjk � Yc, (16)

j

ZQWjk ≤

k

Wk, (17)

D ∈ (0, 1). (18)

Constraints (11) and (12) indicate the conservation ofmaterial quality at each logistics node, that is, the amount ofindustrial waste input at each waste generation point, artificialintelligence recycling classification center, remanufacturingcenter, resales center, and waste landfill center equal to theoutput amount; constraints (13) to (15) restrict the amount ofwaste transported by the classification center to each secondarynode through a proportional limit; constraint (16) indicatesthat the predicted value in the artificial intelligence center is thewaste input entered in the model; constraint (17) indicates theprocessing limit of each waste treatment center; and constraint(17) indicates the range of decision variables.

3.2.4. Multiobjective Function Solution Design. /e abovemodel belongs to amultiobjective mixed-integer programmingmodel. Multiobjective programming generally belongs to thePareto optimization problem. Because of the mathematicaloptimization problem involvingmultiple objective functions, itcan only be as close as possible to the ideal solution based onthe coordination of the objective functions. Based on the lit-erature [23], this paper uses the fuzzy membership function infuzzy theory to fuzzify the objective function and converts themultiobjective problem into a single-objective problem forsolving. Specific steps are as follows:

Step 1: First, disassemble the dual objective function intotwo single-objective functions and use Lingo software toobtain the value interval of each objective function. First,the value range [F1min, F1max] of the first objectivefunction is calculated by programming when only the


environmental benefit target is considered; similarly, whenonly the economic benefits are considered, the value rangeof the second objective function can be obtained byprogramming calculation as [F2min, F2max].Step 2: the membership degree fuzzy function method isused to assign a fuzzy wish value to each objectivefunction, and the membership degree of the fuzzy wish isused to indicate the degree of satisfaction of the decision-maker team in responding to the target level. /is modelbelongs to the maximization optimization problem, andthe membership function of its fuzzy desire is as follows:

μ1 �

1, F1 >F1min,

F1 − F1minF1max − F1min

, F1min ≤F1 ≤F1max,

0, F1 Fmin,

F2 − F2min

F2max − F2min, F2min ≤F2 ≤F2max,

0, F2

Table 4 shows the data use formulas (1)–(6) to predict the wasteproducts in the next cycle of G city. Table 5 shows thetransportation ratio of each node. Table 6 shows the operatingcosts of processing waste per ton of waste at each node. Inaddition, according to the basic local conditions, the trans-portation cost is 4 yuan/ton/km, the cost of traditional manualwaste treatment is 200 yuan/ton, and the reuse rate of tradi-tional industrial waste is 74.6%.

Using Lingo software, the objective functions underdifferent weights are solved, and the logistics network fa-cilities location schemes under three weights and the cor-responding objective function values of environmental andeconomic benefits of the network system are obtained (asshown in Table 7).

It can be known from the example that when only envi-ronmental protection benefits are used as the objective func-tion, alternative points 1, 2, 4, 5, 6, 7, and 8 need to be selectedintelligence to build an artificial classification center. /eeconomic benefits generated at this time are 174,206 yuan perunit cycle. /e environmental protection benefit is to achieve84.6% of the reuse rate of waste; when only economic benefitsare considered and when only economic benefits are used asthe objective function, alternative points 2, 3, 4, 5, 6, and 8 needto be selected. During the construction of artificial intelligenceclassification center, the economic benefit generated at thistime is 288,014 yuan per cycle, and the environmental pro-tection benefit is to achieve 82.7% of the reuse rate of waste;when considering both economic and environmental benefits,at this time, alternative points 2, 4, 5, 6, 7, and 8 are selected. Atthis time, the economic benefit is 246,137 yuan per cycle, andthe environmental protection benefit is to achieve 83.5% of thereuse rate of the waste. /is plan takes into account botheconomic and environmental benefits. /erefore, the solutionof the model in this paper is shown in Figure 6 at this time.

Table 2: Remanufacturing center, resale market, and landfill center.

X-axis Y-axisRemanufacturing center 35 85Resale market 70 38Landfill center 20 37

Table 3: Coordinates of alternative points, construction cost, and cycle capacity.

Recycling classification center X-axis Y-axis Construction cost (yuan) Cycle capacity (tons)1 76 12 200000 12002 71 16 200000 12003 68 25 200000 12004 65 66 200000 12005 55 47 200000 12006 45 68 200000 12007 35 41 200000 12008 49 27 200000 1200

Table 4: Coordinates of waste generation points, predicted value,and original classification costs.

Waste generation point X-axisY-axis Cycle production (tons)

1 92 19 3282 22 47 2353 51 59 2244 89 92 1665 55 76 2056 14 76 3627 26 57 4008 84 7 2809 82 53 19110 93 94 16111 24 35 34612 57 45 26313 42 22 46014 1 79 13515 67 15 26516 98 48 15817 65 76 41018 37 85 26019 88 54 37220 29 67 340

Table 5: Transportation proportion of each node afterclassification.

Transportation ratio (%)XWEk 50XWRk 30XWTk 20

Table 6: Operating costs of each node.

Unit processing cost (yuan/ton/km)Z

QWjk 4

ZWEkm 3ZWRkl 3ZWTkn 3ZERml 3ZETmn 3


4. Conclusion and Discussion

With the acceleration of China’s industrialization and ur-banization process, the traditional urban waste treatmentand logistics network has been unable to meet the rapidlyexpanding population and production capacity, and thelarge accumulation of waste in industrial parks has become amajor cause of environmental pollution. One of the mainchallenges of the sustainable development strategy is to solvethe problem of upgrading and designing waste networksystems in industrial parks.

Based on artificial intelligence’s prediction technologyand image recognition technology, this article intelligentlyupgrades the traditional industrial waste planning andmanagement system, designs an industrial waste intelligentclassification center with intelligent prediction and intelli-gent classification capabilities, and implements its keyimplementation. /e technology was explored. Finally, thevalue and significance of the intelligent classification centerdesigned in this paper are illustrated by constructing amultiobjective site selection model considering both eco-nomic and environmental benefits.

It can be seen from the different weighted results ofeconomic and environmental benefits of fuzzy mathematicalmethods that regardless of focusing on economic benefits orenvironmental benefits, huge economic benefits can beobtained by reducing the dependence on human resourcesafter building a new intelligent classification center./erefore, relevant departments and responding companiescan be advised to fully apply such artificial intelligencetechnology to the existing industrial waste recycling net-work, thereby enhancing the global benefits of the existing

industrial waste recycling network, improving system effi-ciency and robustness, and ultimately achieve green sus-tainable development.

In previous research, [11] proposed an urban wasterecycling system based on radio frequency wireless networks.Using sensor technology and radio frequency wireless networktechnology, the information collection of urban waste bins andthe apex and positioning of urban trash cans were achieved./is has a certain enlightening effect on us. Big data technologyand artificial intelligence technology are gradually changingour way of life. /erefore, this article uses new artificial in-telligence technology to optimize the existing industrial wasterecycling network and realize the idea of intelligent, eco-nomical, and sustainable recycling systems. It can promoteacademic innovation and disciplinary intersection in the fieldof waste management. In [16], while considering the cost, andtaking into account the benefits of investment in environ-mental protection, a green supply chain network design modelwas proposed, and the optimization method was used to studythe supply network construction cost and the minimum en-vironmental pollution of the entire supply chain system. /eirresearch found that the main way to reduce carbon dioxideemissions and total costs is to increase network capacity andenhance the service capabilities of facilities, which coincideswith our research conclusions. To some extent, this shows thatthe only way for the supply chain network to achieve greeningand maximize economic benefits is to improve service capa-bilities and response capabilities through technological im-provements. References [17, 19], etc., proposed or appliedclassic multiobjective function tools such as particle swarmoptimization and genetic algorithms when solving multi-objective function problems. /e multiobjective function al-gorithm designed in this paper uses the fuzzy membershipfunction in the fuzzy theory to obfuscate the objective functionand converts the multiobjective problem into a single-objectiveproblem for solving. On the one hand, the solution process andresults can be more clearly explained and analyzed. For ex-ample, the effect of the objective function on the overall systemoptimization under different preferences can be further ana-lyzed. On the other hand, only 0 and 1 were the two extremeweights. Researchers can also design new solving algorithmsbased on different weight preferences.

Our further research direction is to consider morefactors in the supply chain, such as transportation mode,demand uncertainty, artificial intelligence learning ability,image recognition fault tolerance rate, etc., in order toenhance its applicability to real-world scenarios. On thecontrary, we can also extend our research by designing newsolutions to solve this multiobjective model.

Data Availability

/e data used to support the findings of this study areavailable from the corresponding author upon request.

Conflicts of Interest

/e authors declare that there are no conflicts of interestregarding the publication of this paper.

Table 7: Logistics network facility location plan.

Weightcoefficient(ω1,ω2)

Totalnetwork cost

(yuan)

Landfillrate (%)

AI recyclingclassification centeralternative point

(1, 0) 174206 84.6 1, 2, 4, 5, 6, 7, 8(0, 1) 288014 82.7 2, 3, 4, 5, 6, 8(1, 1) 246137 83.5 2.4, 5, 6, 7, 8

0102030405060708090

100

0 10 20 30 40 50 60 70 80 90 100

Waste generation point Remanufacturing centerResale market Landfill centerAI Recycling classificationcenter

Figure 6: Distribution of site selection schemes.


Acknowledgments

/is work was supported by the Construction Project forDomestic 2017 First-Class-Discipline in Guizhou Province(GNYL[2017]005), Guizhou Provincial Philosophy andSocial Science Planning Joint Fund (18GZLH03), and Majorproject fund for social science & humanities of GuizhouUniversity (GDZT201702).

References

[1] Z. Lu, J. Zhang, S. Lu et al., “Pollution characteristics andevaluation of heavy metal elements in the soil around themunicipal solid waste incineration power plant and in theplant area,” Environmental Science, vol. 40, no. 5, pp. 483–2492, 2019.

[2] Y. Peng, L. Lei, X. Peng, P. Yang, X. Zhao, and W. Ma, “/edevelopment of domestic waste classification in China? Ob-stacles and countermeasures,” China Environmental Science,vol. 38, no. 10, pp. 3874–3879, 2018.

[3] L. Dan, G. Chen, W. Ma, and N. Duan, “Characteristics andtreatment status of domestic garbage in villages and towns inChina,” China Environmental Science, vol. 38, no. 11,pp. 4187–4197, 2018.

[4] T. Lindhqvist, Extended Producer in Cleaner Production:Policy Principle to Promote Environmental Improvements ofProduct Systems, IIIEE, Lund, Sweden, 2000.

[5] J. Li, D. Li, and M. Lu, “Research on the design of C2B wastemobile phone reverse logistics recycling network based oncarbon emission,” Manufacturing Automation, vol. 40, no. 9,pp. 95–101, 2018.

[6] Y. Chen, Y. Yang, and Q. Zhang, “Design of reverse logisticsnetwork for waste household appliances based on third partyrecycling,”Mathematics in Practice and@eory, vol. 46, no. 17,pp. 81–89, 2016.

[7] D. Wang, H. Du, and W. Yang, “A two-stage fourth-partyreverse logistics network design model based on CVaR,”Industrial Engineering and Management, vol. 20, no. 1,pp. 22–27, 2015.

[8] H. Prajapati, R. Kant, and R. Shankar, “Bequeath life to death:state-of-art review on reverse logistics,” Journal of CleanerProduction, vol. 211, pp. 503–520, 2019.

[9] S. Agrawal, R. K. Singh, and Q. Murtaza, “A literature reviewand perspectives in reverse logistics,” Resources, Conservationand Recycling, vol. 97, pp. 76–92, 2015.

[10] D. Huang and Y. Li, “Problems in urban domestic wastetreatment in Wuhan and suggestions for improvement,”Resources and Industries, vol. 14, no. 3, pp. 167–173, 2012.

[11] L. I. Yan, G. Yang, and A. Ren, “Research on urban garbagerecycling system based on radio frequency wireless network,”Modern Electronic Technology, vol. 38, no. 16, pp. 48–50, 2015.

[12] W. Wang, X. Ge, and Y. Xu, “Analysis of multi-agent interestgame and differential responsibility in urban waste separationand recycling,” China Population, Resources and Environ-ment, vol. 27, no. S2, pp. 41–44, 2017.

[13] Q. Wang, Li Xia, and C. Tong, “Selection and evaluation ofreverse logistics suppliers based on program satisfaction,”Logistics Technology, vol. 37, no. 11, pp. 50–55, 2018.

[14] Q. Zhang and X. Lu, “Research on e-commerce return vehiclerouting problem and ant colony algorithm,” Computer En-gineering and Applications, vol. 54, no. 22, pp. 239–245, 2018.

[15] M. Lu, Z. Ye, J. Tian, and F. Wang, “Optimization design ofreverse logistics network considering random reproducibilityratio,” System Engineering, vol. 35, no. 6, pp. 113–120, 2017.

[16] F. Wang, X. Lai, and N. Shi, “A multi-objective optimizationfor green supply chain network design,” Decision SupportSystems, vol. 51, no. 2, pp. 262–269, 2011.

[17] S. Elhedhli and M. Ryan, “Green supply chain network designto reduce carbon emissions,” Transportation Research Part D:Transport and Environment, vol. 17, no. 5, pp. 370–379, 2012.

[18] Li Rui, M. Huang, and F. Sun, “Multi-to-multiple networkdesign of fourth-party logistics with hard time windows,”Control Engineering, vol. 25, no. 10, pp. 1927–1933, 2018.

[19] Q. Zhang and W. Li-rong, “Multi-objective stochastic pro-gramming model for reverse logistics network design,” SoftScience, vol. 29, no. 10, pp. 120–124, 2015.

[20] S. Yang and L. Huang, “Financial early warning model oflisted companies based on BP neural network,” Systems En-gineering @eory & Practice, vol. 25, no. 1, pp. 12–18, 2005.

[21] W. Yang, H. Wang, J. Zhang et al., “An improved vehicle real-time detection algorithm based on Faster-RCNN,” Journal ofNanjing University (Natural Science Edition), vol. 55, no. 2,2019.

[22] K. Simonyan and A. Zisserman, “Very deep convolutionalnetworks for large-scale image recognition,” 2014, https://arxiv.org/abs/1409.1556.

[23] X. Xu and J. Li, @eory and Method of Multi-Objective De-cision Making, Tsinghua University Press, Beijing, China,2005.

https://arxiv.org/abs/1409.1556https://arxiv.org/abs/1409.1556

researchonindustrialwasterecoverynetworkoptimization ...conv feature map 256-d figure...

Documents