Water resource management method coupled with supply chain analysis and neural network prediction

Abstract

The invention discloses a water resource management method coupled with supply chain analysis and neural network prediction, and the method specifically comprises the steps: analyzing a supply chain distribution rule of regional water in combination with a supply chain water consumption accounting technology and a structure decomposition technology, and recognizing key water consumption links and main driving variables thereof; constructing a key driving variable and water consumption data set to train and verify a convolutional neural network-long short-term memory network (CNN-LSTM) model for high-precision prediction of the dynamic change trend of regional water consumption; a Monte Carlo method is adopted, uncertainty of main driving variables is represented through random sampling, robustness of model output is verified, a minimum water consumption path is extracted to generate an optimization scheme, and a dynamic change rule in a system is revealed. Through coupling of supply chain resource analysis, an intelligent prediction technology and uncertainty analysis, the method faces a future development trend, and provides decision support for supply and demand management and efficiency optimization of water resources.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of water resources management, and particularly relates to a water resources management method coupling supply chain analysis and neural network prediction. BACKGROUND

[0002] Water scarcity is a major global challenge, and the competition for water among multiple users such as cities, industries, and agriculture is intensifying, which seriously affects the sustainable use of water resources. The United Nations Sustainable Development Goals in 2030 lists water resources management as a core issue, emphasizing its key role in environmental sustainability and social development. Existing research explores water resources management strategies through various methods, but the complexity of water resources systems increases the difficulty of accurate management. This complexity is due to changes in water demand, diversity of industrial activities, and interactions of environmental factors. Analyzing the dynamic changes of these driving factors is the key to achieving efficient use and sustainable management of water resources.

[0003] The supply chain analysis method in the prior art is often used to evaluate the distribution characteristics of water resources, but it is mainly limited to static analysis of historical data, lacking dynamic prediction ability for future water trends. Neural network technology has shown certain potential in the field of water resources prediction, but existing models rely heavily on data-driven methods, making it difficult to clearly understand the dynamic changes of driving factors in the water resources system through technical means, and lacking sufficient quantification of the uncertainty of these factors. Existing research lacks a systematic technical solution to reveal water usage patterns, generate confidence intervals, and optimize management strategies by integrating supply chain analysis, neural network prediction, and uncertainty analysis through computer equipment. Therefore, there is an urgent need for a water resources management method that combines supply chain analysis and neural network prediction, which integrates multi-source data, intelligent prediction technology, and uncertainty analysis to solve the practical technical problems of dynamic prediction and optimized management in water resources systems, and provides reliable technical support for efficient water resources management. SUMMARY

[0004] The purpose of the present application is to provide a water resources management method coupling supply chain analysis and neural network prediction, comprising the following steps:

[0005] Step A: Construct an integrated water resources efficiency optimization model combining supply chain analysis and neural network prediction to represent the multi-department effect and multi-factor uncertainty of water resources system management;

[0006] Step B: Based on the water consumption obtained by the acquisition device, statistical data is formed; based on the statistical data, the water consumption distribution characteristics of multiple departments are quantified using supply chain water accounting technology, and key water consumption departments are identified; key driving variables affecting regional water consumption changes are identified using structural decomposition technology as input parameters of the water resources efficiency optimization model;

[0007] Step C: Construct key inputs according to the input parameters determined in step B, combine the time series data of regional water consumption to construct the supply chain water resource dataset, and train and verify the convolutional neural network-long short-term memory network model with the supply chain water resource dataset to realize the prediction of regional water consumption; the key inputs include: water resource utilization intensity, industrial structure, and population consumption;

[0008] Step D: Based on the intermediate path scenario of social development, extract the dataset of key driving variables, represent the uncertainty of variable value by performing Monte Carlo random sampling on the computing device, and verify the robustness of the model output; combine the regional water consumption predicted by the convolutional neural network-long short-term memory network model in step C to generate water resource optimization control instructions; the control instructions include: regional water supply adjustment, industrial structure optimization, and water resource utilization efficiency improvement scheme.

[0009] The integrated water resource efficiency optimization model of supply chain analysis and neural network prediction in the step A is represented as:

[0010] X input =f MRIO,SDA (X w )

[0011] S opt =f CNN-LSTM,FA (X input )

[0012] In the formula, S opt is the output of the optimization control instruction; MRIO represents the supply chain water accounting technology; SDA represents the structure decomposition technology; CNN-LSTM represents the neural network prediction module; FA represents the variable effect analysis technology; X w is the input dataset of MRIO and SDA; X input is the key driving variable identified by MRIO and SDA.

[0013] The combination of structure decomposition technology in step B to identify the key driving variables affecting regional water consumption change includes:

[0014] Based on the official statistics of industry supply and demand and industrial water data, the supply chain water accounting technology is adopted to clarify the supply chain water characteristics of multiple regions and departments; combined with the structure decomposition technology, the key driving variables affecting water consumption change are identified:

[0015] W=QL(YF+E) (1)

[0016] VW=W / VA (2)

[0017] W t-0 =ΔQ+ΔL+ΔSc+Δfc+ΔSo+Δfo+ΔP+ΔE (3)

[0018]

[0019] where W represents water use based on supply chain accounting; Q represents water resource use efficiency of the whole region; L is the Leontief inverse matrix, L = (I-A) -1 ; I represents the unit matrix; YF = [yf i rs ] nk×1 , yf i rs represents the final use product contributed by department i of region r; represents the export of department i of region r; VW represents the water resource use efficiency of different industries based on supply chain accounting; represents the added value of department j of region s; t and 0 represent different periods; W t-0 represents the change of water use between two periods; the final use product YF is divided into resident use Yc and other use Yo; Sc and So represent the use structure of Yc and Yo, respectively; fc and fo represent the per capita use of Yc and Yo, respectively; Δ represents the influence of each variable on the change of water use, ΔQ reflects the technical effect of water resource use; ΔL embodies the industrial structure effect; ΔSc represents the resident use structure effect; Δfc represents the resident per capita use effect; ΔSo represents the other use structure effect; Δfo represents the other per capita use effect; ΔP represents the population effect; ΔE represents the export effect.

[0020] The convolutional neural network-long short-term memory network model in step C comprises:

[0021] The CNN performs convolution operation on the input data to extract features, and then passes the features to the LSTM; the LSTM layer learns the features and remembers long-term dependencies, and the full connection layer converts the output of the LSTM into the final prediction result; 80% of the samples in the supply chain water resource data set are used for model training, and 20% of the samples are used for testing; the root mean square error RMSE, the mean absolute error MAE and R 2 evaluate the quality of the model;

[0022] h l = ELU (ω l ⊙X input + bs l ) (12)

[0023]

[0024] i t = σ (ω i · [h t-1 , xt ]+bs i ) (14)

[0025] f t =σ(ω f ·[h t-1 ,x t ]+bs f ) (15)

[0026]

[0027] o t =σ(ω o ·[h t-1 ,x t ]+bs o ) (18)

[0028] h t =o t tanh(c t ) (19)

[0029] In the formula, the CNN architecture contains one convolutional layer, the convolutional kernel size of the convolutional layer is [10, 1], including 32 filters, h l represents the output feature map of the lth filter; ELU is an activation function; ω l is a convolution kernel weight matrix; ⊙ represents a convolution operation; X input is input features into the convolutional layer; bs is a bias vector; H(i, j) represents the value of the feature map at position (i, j) after convolution; IS(i+m, j+n) represents the value of the input feature map at position (i+m, j+n); K(m, n) represents the weight value of the convolution kernel at index (m, n); the LSTM architecture includes a first LSTM layer and a second LSTM layer, the first LSTM layer contains 50 hidden units, and the second LSTM layer contains 32 hidden units, i t , f t and o t represent the input gate, the forget gate and the output gate, respectively; is a candidate cell state used to update the cell state; c t is a cell state used to store long-term memory; h t is a hidden state, i.e., the output of the current time step; σ represents a Sigmoid activation function; tanh represents a hyperbolic tangent activation function; ω i , ω f , ω o and ω c represent weight matrices; x t is the input of time step t; bs i , bs f , bs oAnd bs c The bias vector is represented.

[0030] The specific process of generating the water resource optimization control instruction in the step D is: based on the data set of the SSP2 scenario, using the Monte Carlo method, performing random sampling on the computing device, characterizing the uncertainty of the main driving variable, and verifying the robustness of the model output; in combination with the water consumption predicted by the convolutional neural network-long short-term memory network model in the step C, the water resource optimization control instruction is generated.

[0031] The Monte Carlo method characterizes the uncertainty of the main driving variable by performing random sampling on the computing device, and verifies the robustness of the model output, and the main formula is:

[0032]

[0033] In the formula, represents the sampling value of the i-th input variable in the j-th simulation; represents the mean value of the input variable x i represents the standard deviation; COV represents the coefficient of variation; n sim represents the number of Monte Carlo simulations.

[0034] Another object of the present application is to provide a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program stored in the memory, the processor executes the water resource management method coupled with supply chain analysis and neural network prediction according to the present application.

[0035] Another object of the present application is to provide a computer readable storage medium having a computer program stored thereon, wherein when the processor executes the computer program, the processor executes the water resource management method coupled with supply chain analysis and neural network prediction according to the present application.

[0036] The present application has the following advantages:

[0037] The application of the water resource management method coupled with supply chain analysis and neural network prediction disclosed by the present application can realize fine identification of multi-regional and multi-department water distribution characteristics and prediction of dynamic change trend, thereby supporting scientific optimization of water resource management scheme.

[0038] ​The application is based on official statistics of industry supply and demand and industrial water data, combines supply chain water accounting and structural decomposition method, identifies key water features and main driving factors of multi-region and multi-department, adopts convolutional neural network-long short-term memory network (CNN-LSTM) model to train and learn multi-source water data, so as to realize high-precision prediction of regional water dynamic change trend. Meanwhile, random sampling is performed on a computer device, the main driving variable uncertainty is quantified through a Monte Carlo method, a confidence interval of water consumption and water resource utilization efficiency is generated, the robustness of model output is verified, the driving variable path with minimum water consumption is extracted, a water resource optimization scheme is generated, and the dynamic change law of the water resource system is revealed. Through the above technical means, the application realizes the fusion application of multi-source data and intelligent prediction model, has obvious advantages in dynamic prediction, uncertainty quantification and optimization management strategy generation compared with the existing method, and can provide reliable technical support for efficient allocation and supply and demand regulation of water resources. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 It is a water resource management method process schematic diagram of coupling supply chain analysis and neural network prediction of the application;

[0040] Figure 2 It is a multi-department water feature identification schematic diagram in the embodiment of the application;

[0041] Figure 3 It is a driving factor influence analysis schematic diagram on water change in the embodiment of the application; wherein, (a) is a variable contribution degree statistical chart from 2012 to 2015; (b) is a variable contribution degree statistical chart from 2015 to 2017;

[0042] Figure 4 It is a water resource efficiency optimization path schematic diagram in the embodiment of the application; wherein, (a) is a water consumption statistical chart of the baseline analysis; (b) is a water consumption statistical chart of the Monte Carlo simulation; (c) is a water resource utilization efficiency statistical chart of the baseline analysis; (d) is a water resource utilization efficiency statistical chart of the Monte Carlo simulation; (e) is a statistical chart of annual average change rate of key driving variables under the optimal path; DETAILED DESCRIPTION

[0043] The application provides a water resource management method coupling supply chain analysis and neural network prediction, which is further described in detail below in combination with the drawings.

[0044] As shown in Figure 1 , the embodiment of the application discloses a water resource management method coupling supply chain analysis and neural network prediction, which comprises the following steps:

[0045] Step A: build an integrated water resource efficiency optimization model of supply chain analysis and neural network prediction, representing the multi-department effect and multi-factor uncertainty of water resource system management;

[0046] Step B: based on the water consumption obtained by the acquisition device, statistical data is formed; based on the statistical data, the water consumption distribution characteristics of multiple departments are quantified by using supply chain water accounting technology, and key water consumption departments are identified; key driving variables affecting regional water consumption changes are identified by combining structural decomposition technology as input parameters of the water resource efficiency optimization model;

[0047] Step C: according to the input parameters determined in step B, key inputs are constructed, a supply chain water resource data set is constructed combining time series data of regional water consumption, and a convolutional neural network-long short-term memory network model is trained and verified with the supply chain water resource data set to realize prediction of regional water consumption; the key inputs include: water resource utilization intensity, industrial structure, and population consumption;

[0048] Step D: based on the intermediate path scenario of social development, the data set of key driving variables is extracted, the robustness of the model output is verified by performing Monte Carlo random sampling on the computing device to represent the uncertainty of variable values; combining the regional water consumption predicted by the convolutional neural network-long short-term memory network model in step C, water resource optimization control instructions are generated; the control instructions include: regional water supply adjustment, industrial structure optimization, and water resource utilization efficiency improvement scheme.

[0049] The integrated water resource efficiency optimization model of supply chain analysis and neural network prediction in the step A is represented as:

[0050] X input =f MRIO,SDA (X w )

[0051] S opt =f CNN-LSTM,FA (X input )

[0052] In the formula, S opt is the output of the optimization control instruction; MRIO represents the supply chain water accounting technology; SDA represents the structural decomposition technology; CNN-LSTM represents the neural network prediction module; FA represents the variable effect analysis technology; X w is the input data set of MRIO and SDA; X input is the key driving variable identified by MRIO and SDA.

[0053] The combination of structural decomposition technology to identify key driving variables affecting regional water consumption changes in the step B includes:

[0054] Based on official statistics on industry supply and demand and industrial water use data, supply chain water accounting technology is used to clarify the water use characteristics of multiple sectors in the supply chain. Combined with structural decomposition technology, the key driving variables that affect changes in water use are identified:

[0055] W=QL(YF+E) (1)

[0056] VW=W / VA (2)

[0057] W t-0 =ΔQ+ΔL+ΔSc+Δfc+ΔSo+Δfo+ΔP+ΔE (3)

[0058]

[0059] Where W represents the water consumption based on supply chain accounting; Q represents the overall water resource utilization efficiency of the region; L is the Leontief inverse matrix, L = (IA) -1 ; I represents the identity matrix; YF = [yf i rs ] nk×1 , yf i rs represents the final use product contributed by sector i in region r; represents the export of sector i in r; VW represents the water resource utilization efficiency of each industry under supply chain accounting; represents the added value of sector j in region s; t and 0 represent different periods; W t-0 represents the change in water consumption between two periods; the final use product YF is decomposed into resident use Yc and other use Yo; Sc and So represent the use structure of Yc and Yo respectively; fc and fo represent the per capita use of Yc and Yo respectively; Δ represents the impact of each variable on the change in water consumption, ΔQ reflects the effect of water resource utilization technology; ΔL reflects the industrial structure effect; ΔSc represents the resident use structure effect; Δfc represents the per capita resident use effect; ΔSo represents the other use structure effect; Δfo represents the other per capita use effect; ΔP represents the population effect; ΔE represents the export effect.

[0060] The convolutional neural network-long short-term memory network model in step C includes:

[0061] The advantages of CNN and LSTM architecture are fused to build a CNN-LSTM model. In the model, CNN performs convolution operation on the input data to extract features, and then passes the features to LSTM. The LSTM layer learns the features and remembers long-term dependencies, and the fully connected layer converts the output of LSTM into the final prediction result. The dataset is constructed based on the identified key variables and long sequence water consumption. 80% of the samples are used for model training, and 20% are used for testing to evaluate the model quality. The root mean square error (RMSE), mean absolute error (MAE) and R 2 The quality of the model is evaluated. The main formula of the CNN-LSTM model is:

[0062] h l =ELU(ω l ⊙X input +bs l ) (12)

[0063]

[0064] i t =σ(ω i ·[h t-1 ,x t ]+bs i ) (14)

[0065] f t =σ(ω f ·[h t-1 ,x t ]+bs f ) (15)

[0066]

[0067] o t =σ(ω o ·[h t-1 ,x t ]+bs o ) (18)

[0068] h t =o t tanh(c t ) (19)

[0069] In the formula, the CNN architecture contains one convolutional layer, the convolutional kernel size of the convolutional layer is [10, 1], including 32 filters, h l represents the output feature map of the lth filter; ELU is an activation function; ω l is a convolution kernel weight matrix; ⊙ represents convolution operation; X inputIt is passed into the convolution layer as input features; bs is the bias vector; H(i,j) represents the value of the feature map at position (i,j) after convolution; IS(i+m,j+n) represents the value of the input feature map at position (i+m,j+n); K(m,n) represents the weight value of the convolution kernel at index (m,n); the LSTM architecture includes the first LSTM layer and the second LSTM layer. The first LSTM layer contains 50 hidden units and the second LSTM layer contains 32 hidden units. t 、f t and o t Represent the input gate, forget gate and output gate respectively; is the candidate cell state, used to update the cell state; c t is the cell state, used to store long-term memory; h t is the hidden state, that is, the output of the current time step; σ represents the Sigmoid activation function; tanh represents the hyperbolic tangent activation function; ω i 、ω f 、ω o and ω c represents the weight matrix; x t is the input at time step t; bs i ,bs f ,bs o and bs c Represents the bias vector.

[0070] The specific process of generating water resource optimization control instructions in step D is as follows: based on the SSP2 scenario data set, Monte Carlo random sampling is performed on a computing device to simulate the fluctuation of key driving variables using random sampling rules to verify the robustness of the model output; the convolutional neural network-long short-term memory network model in step C is used to predict water consumption, quantify the probability distribution of water consumption and water resource utilization efficiency, and generate water resource optimization control instructions (i.e., regional water supply, industrial structure, and water resource utilization efficiency);

[0071] The main formula for Monte Carlo random sampling is:

[0072]

[0073] Where, represents the sample value of the i-th input variable in the j-th simulation; Represents the input variable x i The mean of represents standard deviation; COV represents coefficient of variation; n sim Indicates the number of Monte Carlo simulations.

[0074] In a specific embodiment, a water resources management method coupling supply chain analysis with neural network prediction is proposed, including:

[0075] Step 1: Build an integrated water resource efficiency optimization model combining supply chain analysis and neural network prediction to quantify the multi-sector effects and multi-factor uncertainties of water resource system management;

[0076] Step 2: Based on official statistics of industry supply and demand and industrial water data, reveal the multi-sector and full-chain supply chain water distribution structure through supply chain water accounting technology; introduce structural decomposition technology to identify key driving variables affecting regional water use changes;

[0077] Step 3: Based on the time series data of key driving variables and water consumption, build a supply chain water resource dataset, and train and validate the convolutional neural network-long short-term memory network model with this dataset to predict future regional water consumption;

[0078] Step 4: Based on the intermediate path scenario of social development, perform random sampling on the computer device to simulate the uncertainty of key driving variables and verify the robustness of the model output; combine the convolutional neural network-long short-term memory network model to predict the water consumption of the corresponding scenario, quantify the probability distribution of water consumption and water resource utilization efficiency, and form control instructions for regional water supply adjustment, industrial structure optimization, and water resource utilization efficiency improvement.

[0079] In this embodiment, the computing device is: Intel Core i5-12500H processor, NVIDIA GeForce RTX 2050 graphics card, 16GB RAM and 512GB SSD memory and storage, Windows 11 operating system, RStudio 2024 development environment, R 4.4.1 R language version.

[0080] The specific process of building an integrated water resource efficiency optimization model combining supply chain analysis and neural network prediction in step 1 is to obtain key driving variables by combining supply chain water accounting technology and structural decomposition technology; based on the dataset of key driving variables, use neural network model and variable effect analysis technology to predict water consumption and generate water resource optimization control instructions:

[0081] X input =f MRIO,SDA (X w )

[0082] S opt =f CNN-LSTM,FA (X input )

[0083] In the formula, S optfor optimizing the output of control instructions; MRIO stands for supply chain water accounting technology; SDA stands for structural decomposition technique; CNN-LSTM stands for neural network prediction module; FA stands for variable effect analysis technology; X w is the input data set of MRIO and SDA; X input is the key driving variable identified by MRIO and SDA.

[0084] The specific process of accounting for supply chain water use and identifying key driving variables in step 2 is to quantify the water distribution structure of multiple departments based on official statistics of industry supply and demand and industrial water data using supply chain water accounting technology; combined with structural decomposition technology, identify the key driving variables affecting the change of water consumption:

[0085] W = QL(YF + E) (1)

[0086] VW = W / VA (2)

[0087] W t-0 = ΔQ + ΔL + ΔSc + Δfc + ΔSo + Δfo + ΔP + ΔE (3)

[0088]

[0089]

[0090] In the formula, W represents the water consumption based on supply chain accounting; Q represents the water resource utilization efficiency of the whole region; L is the Leontief inverse matrix, L = (I-A) -1 ; I represents the unit matrix; YF = [yf i rs ] nk×1 , yf i rs represents the final use product contributed by department i in region r; represents the export of department i in region r; VW represents the water resource utilization efficiency of different industries under supply chain accounting; represents the added value of department j in region s; t and 0 represent different periods; W t-0 represents the change of water consumption between two periods; the final use product YF is divided into resident use Yc and other use Yo; Sc and So represent the use structure of Yc and Yo respectively; fc and fo represent the per capita use of Yc and Yo respectively; Δ represents the influence of each variable on the change of water consumption, ΔQ reflects the water resource utilization technology effect; ΔL reflects the industrial structure effect; ΔSc represents the resident use structure effect; Δfc represents the resident per capita use effect; ΔSo represents the other use structure effect; Δfo represents the other per capita use effect; ΔP represents the population effect; ΔE represents the export effect.

[0091] The main water use sectors and key driving variables identified in step 2 are that, according to the water resource utilization efficiency of each sector, agriculture, manufacturing, construction, and accommodation and catering industry are established as key water use sectors; direct water intensity, industrial structure, and per capita resident consumption are established as key driving variables; the effect of industrial structure can be refined as the effect of key water use sectors.

[0092] The specific process of constructing the convolutional neural network-long short-term memory network model in step 3 is to fuse the advantages of CNN and LSTM architecture to construct a CNN-LSTM model; in this model, CNN performs convolution operation on input data to extract features, and then passes these features to LSTM; the LSTM layer learns these features and remembers long-term dependencies, and the fully connected layer converts the output of LSTM into the final prediction result; based on the identified key variables and long sequence water consumption, a data set (1998-2022) is constructed, 80% of the samples are used for model training, and 20% are used for testing, to evaluate the quality of the model with root mean square error (RMSE), mean absolute error (MAE), and R2; the main formula of the CNN-LSTM model is:

[0093] h l =ELU(ω l ⊙X input +bs l ) (12)

[0094]

[0095] i t =σ(ω i ·[h t-1 ,x t ]+bs i ) (14)

[0096] f t =σ(ω f ·[h t-1 ,x t ]+bs f ) (15)

[0097]

[0098] o t =σ(ω o ·[h t-1 ,x t ]+bs o ) (18)

[0099] h t =o t tanh(c t) (19)

[0100] In the formula, the CNN architecture contains one convolutional layer, the convolution kernel size of the convolutional layer is [10, 1], including 32 filters, h l Represents the output feature map of the lth filter; ELU is the activation function; ω l is the convolution kernel weight matrix; ⊙ represents the convolution operation; X input It is passed into the convolution layer as input features; bs is the bias vector; H(i,j) represents the value of the feature map at position (i,j) after convolution; IS(i+m,j+n) represents the value of the input feature map at position (i+m,j+n); K(m,n) represents the weight value of the convolution kernel at index (m,n); the LSTM architecture includes the first LSTM layer and the second LSTM layer. The first LSTM layer contains 50 hidden units and the second LSTM layer contains 32 hidden units. t 、f t and o t Represent the input gate, forget gate and output gate respectively; is the candidate cell state, used to update the cell state; c t is the cell state, used to store long-term memory; h t is the hidden state, that is, the output of the current time step; σ represents the Sigmoid activation function; tanh represents the hyperbolic tangent activation function; ω i 、ω f 、ω o and ω c represents the weight matrix; x t is the input at time step t; bs i ,bs f ,bs o and bs c Represents the bias vector.

[0101] The specific process of generating water resource optimization control instructions in step 4 is to perform Monte Carlo random sampling on a computer device based on a dataset of the intermediate path of social development to simulate the fluctuations of key driving variables and verify the robustness of the model; combine the convolutional neural network-long short-term memory network model to predict water consumption from 2026 to 2050, quantify the probability distribution of water consumption and water resource utilization efficiency, and form water resource optimization control instructions (i.e., regional water supply, industrial structure, and water resource utilization efficiency) to provide a reliable basis for risk management;

[0102] The main formula for Monte Carlo random sampling is:

[0103]

[0104] Where, the sample value of the i-th input variable in the j-th simulation; the mean value of the input variable x i ; and the standard deviation; COV represents the coefficient of variation; n sim represents the number of Monte Carlo simulations.

[0105] Figure 2 is the water consumption characteristics of the department's supply chain. It can be seen from Figure 2 that the water resource utilization intensity of agriculture (Agr), manufacturing (Man), construction (Con), and accommodation and catering (Acc) is outstanding; therefore, these four departments are established as the key water consumption departments.

[0106] Figure 3 is the impact of driving variables on regional water consumption. Among them, (a) is the variable contribution degree statistics from 2012 to 2015; (b) is the variable contribution degree statistics from 2015 to 2017; as shown in Figure 3 , the direct water intensity (Q), the industrial structure (L), and the per capita resident consumption (fc) are the key driving variables affecting water consumption changes. These variables are crucial to the regional water resource utilization mode and should be given priority in water resource management.

[0107] Figure 4 is the water resource efficiency optimization path of the region. Among them, (a) is the water consumption statistics of the benchmark analysis; (b) is the water consumption statistics of the Monte Carlo simulation; (c) is the water resource utilization efficiency statistics of the benchmark analysis; (d) is the water resource utilization efficiency statistics of the Monte Carlo simulation; (e) is the annual average change rate statistics of the key driving variables under the optimal path; as shown in Figure 4 , compared with the benchmark and the Monte Carlo simulation, the average values of water consumption and water resource utilization efficiency are very close, which shows that the model has high stability and reliability in the central tendency. And the Monte Carlo simulation further reveals the sensitivity and uncertainty range of the indicators to the input disturbance: the 95% confidence interval of water consumption is [159.52, 167.24] m 3 , with small fluctuations, indicating that the sensitivity of water consumption to driving variables is low. The 95% confidence interval of water resource utilization efficiency is [179.78, 239.34] m 3 / ￥, with large fluctuations, indicating that it is more sensitive to changes in policies or external conditions. To achieve optimal water resource management, the annual average change rate of the industrial structure of the key departments (agriculture, manufacturing, construction, and accommodation and catering) is controlled as follows: -1.50%, -3.20%, -7.13%, and +0.05% from 2026 to 2050; the annual average growth rate of per capita resident consumption is controlled at 0.19%. By 2050, the minimum water consumption is 161.87 m 3, the water resource utilization efficiency is highest 220.73m 3 The optimal path shows that the key to realizing the sustainable development of water resources lies in the adjustment of industrial structure to different degrees and the effective management of resident consumption and water resource utilization efficiency.

[0108] The water resource management method coupling supply chain analysis and neural network prediction in the embodiment of the application is different from the prior art, and the application of the supply chain water analysis in the prior art is limited to analyzing the relationship between industrial association and water distribution based on historical data, evaluating the cascading influence of water management policy, and lacking simulation and prediction of future development trend. Although the neural network model is applied to the prediction module in water resource management, the dynamic change law of the internal system of the water resource system is not explained, and the uncertainty of the driving variable and its influence on water change are not quantified. Under the future possible development mode, the supply chain water characteristics of the region, the dynamic influence of various driving variables on water change, and the optimization path of water resource efficiency cannot be fully analyzed by the prior art. The multi-model coupling framework in the embodiment of the application evaluates the supply chain water structure of multiple regions and multiple departments, identifies the influence of the main driving variable on water, simulates the dynamic change of water under the condition of driving variable uncertainty, generates the optimization path of water resources, and reveals the dynamic change law of the internal system, thereby providing scientific guidance for the optimization management of the water resource system.

[0109] In the embodiment of the application, the water resource system is focused, the internal relationship and development trend of the system are analyzed in depth by using the multi-model coupling framework, the key water consumption department and the key driving variable influencing water consumption are identified, the water consumption change under the future uncertain condition is predicted, the dynamic change law of the water resource system is revealed, and the optimization control instruction of water resources is generated.

[0110] Another embodiment of the application discloses a computer device comprising a memory and a processor, and the memory stores a computer program, and when the processor executes the computer program stored in the memory, the processor executes the water resource management method coupling supply chain analysis and neural network prediction according to the application.

[0111] Another embodiment of the application discloses a computer readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the processor executes the water resource management method coupling supply chain analysis and neural network prediction according to the application.

Claims

1. A water resources management method that couples supply chain analysis and neural network prediction, characterized in that: The steps include: Step A: Build an integrated water resource efficiency optimization model combining supply chain analysis and neural network prediction to characterize the multi-sector effects and multi-factor uncertainties of water resource system management; Step B: generating statistical data based on the water consumption obtained by the collection device; Based on the statistical data, supply chain water accounting technology is used to quantify the water use distribution characteristics of multiple sectors and identify key water-using sectors; Combining structural decomposition techniques to identify key driving variables that affect regional water use changes, which serve as input parameters for the water resource efficiency optimization model; Step C: Construct key inputs based on the input parameters determined in Step B, construct a supply chain water resource dataset based on the time series data of regional water consumption, and use the supply chain water resource dataset to train and validate a convolutional neural network-long short-term memory network model to predict regional water consumption; the key inputs include: water resource utilization intensity, industrial structure, and population consumption; Step D: Based on the intermediate path scenario of social development, extract a data set of key driving variables, characterize the uncertainty of variable values ​​by performing Monte Carlo random sampling on a computing device, and verify the robustness of the model output; combine the regional water consumption predicted by the convolutional neural network-long short-term memory network model in step C to generate water resource optimization control instructions; the control instructions include: plans for adjusting regional water supply, optimizing industrial structure, and improving water resource utilization efficiency.

2. The water resource management method of coupling supply chain analysis and neural network prediction according to claim 1 is characterized in that: The integrated water resource efficiency optimization model of supply chain analysis and neural network prediction in step A is expressed as: X input =f MRIO,SDA (X w ) S opt =f CNN-LSTM,FA (X input ) Where S opt To optimize the output of control instructions; MRIO stands for supply chain water accounting technology; SDA stands for structured decomposition technology; CNN-LSTM stands for neural network prediction module; FA stands for variable effect analysis technique; X w is the input dataset of MRIO and SDA; X input are the key driving variables identified by MRIO and SDA.

3. The water resource management method of coupling supply chain analysis and neural network prediction according to claim 1 is characterized in that: The key driving variables that affect regional water use changes identified by combining structural decomposition techniques in step B include: Based on official statistics on industry supply and demand and industrial water use data, supply chain water accounting technology is used to clarify the water use characteristics of multi-regional and multi-sector supply chains. Combined with structural decomposition technology, the key driving variables that affect changes in water use are identified: W=QL(YF+E) (1) VW=W / VA (2) W t-0 =ΔQ+ΔL+ΔSc+Δfc+ΔSo+Δfo+ΔP+ΔE (3) Where W represents the water consumption based on supply chain accounting; Q represents the overall water resource utilization efficiency of the region; L is the Leontief inverse matrix, L = (IA) -1 ; I represents the identity matrix; YF = [yf i rs ] nk×1 , yf i rs represents the final use product contributed by sector i in region r; represents the export of sector i in r; VW represents the water resource utilization efficiency of each industry under supply chain accounting; represents the added value of sector j in region s; t and 0 represent different periods; W t-0 represents the change in water consumption between two periods; the final use product YF is decomposed into resident use Yc and other use Yo; Sc and So represent the use structure of Yc and Yo respectively; fc and fo represent the per capita use of Yc and Yo respectively; Δ represents the impact of each variable on the change in water consumption, ΔQ reflects the effect of water resource utilization technology; ΔL reflects the industrial structure effect; ΔSc represents the resident use structure effect; Δfc represents the per capita resident use effect; ΔSo represents the other use structure effect; Δfo represents the other per capita use effect; ΔP represents the population effect; ΔE represents the export effect.

4. The water resource management method of coupling supply chain analysis and neural network prediction according to claim 1 is characterized in that: The convolutional neural network-long short-term memory network model in step C includes: CNN performs convolution operations on the input data to extract features, and then passes these features to LSTM; the LSTM layer learns these features and remembers long-term dependencies, and the fully connected layer converts the output of LSTM into the final prediction result; 80% of the samples in the supply chain water resources dataset are used for model training, and 20% of the samples are used for testing; the root mean square error (RMSE), mean absolute error (MAE), and R 2 Evaluate model quality; h l =ELU(ω l ⊙X input +bs l ) (12) i t =σ(ω i ·[h t-1 ,x t ]+bs i ) (14) f t =σ(ω f ·[h t-1 ,x t ]+bs f ) (15) o t =σ(ω o ·[h t-1 ,x t ]+bs o ) (18) h t =o t fishy t ) (19) In the formula, the CNN architecture contains one convolutional layer, the convolution kernel size of the convolutional layer is [10, 1], including 32 filters, h l Represents the output feature map of the lth filter; ELU is the activation function; ω l is the convolution kernel weight matrix; ⊙ represents the convolution operation; X input It is passed into the convolution layer as input features; bs is the bias vector; H(i,j) represents the value of the feature map at position (i,j) after convolution; IS(i+m,j+n) represents the value of the input feature map at position (i+m,j+n); K(m,n) represents the weight value of the convolution kernel at index (m,n); the LSTM architecture includes the first LSTM layer and the second LSTM layer. The first LSTM layer contains 50 hidden units and the second LSTM layer contains 32 hidden units. t 、f t and o t Represent the input gate, forget gate and output gate respectively; is the candidate cell state, used to update the cell state; c t is the cell state, used to store long-term memory; h t is the hidden state, that is, the output of the current time step; σ represents the Sigmoid activation function; tanh represents the hyperbolic tangent activation function; ω i 、ω f 、ω o and ω c represents the weight matrix; x t is the input at time step t; bs i ,bs f ,bs o and bs c Represents the bias vector.

5. The water resource management method of coupling supply chain analysis and neural network prediction according to claim 1 is characterized in that: The specific process of generating water resource optimization control instructions in step D is: based on the data set of the SSP2 scenario, using the Monte Carlo method, performing random sampling on the computing device to characterize the uncertainty of the main driving variables and verify the robustness of the model output; combining the convolutional neural network-long short-term memory network model in step C to predict water consumption and generate water resource optimization control instructions.

6. The water resource management method of coupling supply chain analysis and neural network prediction according to claim 5 is characterized in that: The Monte Carlo method characterizes the uncertainty of the main driving variables and verifies the robustness of the model output by performing random sampling on the computing device. The main formula is: Where, represents the sample value of the i-th input variable in the j-th simulation; Represents the input variable x i The mean of represents standard deviation; COV represents coefficient of variation; n sim Indicates the number of Monte Carlo simulations.

7. A computer device, characterized in that: The method comprises a memory and a processor, wherein a computer program is stored in the memory. When the processor runs the computer program stored in the memory, the processor executes the water resource management method of coupling supply chain analysis and neural network prediction according to any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor performs the water resource management method coupled with supply chain analysis and neural network prediction according to any one of claims 1 to 6.

Images