Multi-water-source analysis method and system for cross-basin water transfer project based on scheduling model

Abstract

The invention discloses a multi-water-source analysis method and system for a cross-basin water transfer project based on a scheduling model, and the method comprises the steps: building a multi-mode collaborative incoming water and required water prediction model based on error compensation, and obtaining incoming water and required water prediction data through the prediction model; determining a priority weight between the water supply benefit and the risk, and establishing a water supply benefit maximization and risk collaboration large-scale complex cross-basin water transfer model; obtaining an optimal water transfer scheme by using the water transfer model according to the priority weight and forecast data; historical water supply data are collected, current water supply rule prediction data are obtained based on a bidirectional recurrent neural network, and a branch water source model is constructed; and obtaining the water source composition of each water receiving area by using the branch water source model. According to the invention, the intellectualization level and decision-making efficiency of cross-basin water transfer project scheduling management are obviously improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of multi-source optimization analysis, and particularly relates to a multi-source analysis method considering multiple uncertainties. BACKGROUND

[0002] As an important means to solve water resource shortage problems, cross-regional water transfer projects have been widely used in water resource scarce areas. However, existing cross-basin water transfer systems mostly focus on deterministic modeling to meet water demand, and cannot effectively cope with uncertainties such as climate change and changes in water demand, making it difficult to meet the real-time scheduling needs of water transfer systems under the constraints of supply and demand information uncertainty analysis.

[0003] The multi-objective optimization scheduling model of large-scale cross-basin water transfer system is a huge multi-dimensional nonlinear complex constrained optimization problem. The contradictions between each sub-objective, the incommensurability, and the competitive relationship between them make them unable to achieve optimal simultaneously, and the decision-making problem of "when to supply water, how much water to supply" and "when to transfer water, how much water to transfer" needs to be optimized to maximize the overall efficiency of the water resource system and ensure the sustainable management of regional water resources. In addition, the cross-basin water transfer system involves rivers, lakes and other multiple water sources. In the actual water transfer process, each water supply source is mixed with each other, and it is difficult to clarify the water source proportion of each water receiving area.

[0004] The joint operation of large-scale cross-basin water transfer system includes a series of processes such as inflow and demand forecasting, optimization scheduling and decision-making. However, most of the existing researches construct optimization scheduling models based on deterministic inflow forecasting, usually ignoring the impact of supply and demand information uncertainty on the water transfer system. The uncertainty of water supply and demand will make it difficult for the water transfer system to achieve accurate water allocation, increasing the risk and uncertainty of operation. In addition, in the actual water transfer process, each water supply source is mixed with each other, and it is difficult to clarify the water source composition of each water receiving area, affecting the management efficiency and water resource utilization effect of the water transfer system.

[0005] Accurate inflow and demand forecasting can improve the efficient use and safe and stable operation of cross-basin water transfer system. However, compared with short-term real-time inflow process prediction, due to the inability to effectively predict weather systems and precipitation processes at medium and long-term scales, it is difficult for runoff prediction models based on physical process driving to provide stable and reliable prediction results, showing high randomness and chaotic characteristics. For water demand prediction, due to the complexity of the water supply system, it is impossible to establish a deterministic model to describe it, which is a technical problem to be solved by the present application. SUMMARY

[0006] In order to solve the problems of being difficult to deal with uncertainty, being unable to effectively carry out multi-objective collaborative optimization and unclear water source composition in the prior art, the present application aims to provide a multi-water source analysis method and system for a cross-basin water diversion project based on a scheduling model, which on the one hand constructs a multi-objective large cross-basin water diversion system considering the uncertainty of inflow and demand, and the collaborative risk of water supply benefit, and on the other hand, from the actual problem, uses an improved IMODE algorithm with priority weight to solve the model, and combines the water supply law predicted by the bidirectional recurrent neural network, and uses the deep learning method to construct a large complex cross-basin water source model, so as to solve the problem that the water source of the current complex large cross-basin water diversion system is difficult to divide.

[0007] In order to achieve the above-mentioned application purposes, the present application provides the following technical solutions:

[0008] In a first aspect, the present application realizes a multi-water source analysis method for a cross-basin water diversion project based on a scheduling model, which comprises the following steps:

[0009] S1, collecting regional historical lake inflow data and water receiving area data, and establishing a multi-modal collaborative inflow and demand prediction model based on error compensation;

[0010] S2, obtaining inflow and demand forecast data by using the prediction model;

[0011] S3, determining the priority weight between water supply benefit and risk, and establishing a large complex cross-basin water diversion model for maximizing water supply benefit and risk collaboration;

[0012] S4, obtaining an optimal water diversion scheme by using the water diversion model according to the priority weight and the forecast data;

[0013] S5, collecting historical water supply data, obtaining current water supply law prediction data based on a bidirectional recurrent neural network, and constructing a water source model;

[0014] S6, obtaining the water source composition of each water receiving area by using the water source model.

[0015] In some embodiments, the step S1 comprises:

[0016] Step S11: collecting regional historical lake inflow data, and water demand data, precipitation, evaporation, groundwater level, land use, population, economic development index and industry index of the water receiving area;

[0017] Step S12: determining the lake inflow and demand characteristic factors, and dividing them into high-frequency factors and low-frequency factors;

[0018] Step S13: The divided factors are subjected to multi-modal coordination training, high-frequency features are predicted using a DTW-GRU model, and low-frequency features are predicted using a DTW-MLP model;

[0019] Step S14: The high-frequency feature and low-frequency feature errors are learned through KNN regression to compensate for the errors caused by different features, and the final prediction result is obtained.

[0020] In some embodiments, the step S12 comprises: step S12a, taking the precipitation factor, the evaporation factor, the groundwater factor, and the land use factor as the lake inflow characteristic factors; step S12b, taking the population index factor, the economic index factor, and the industry index factor as the water demand characteristic factors of the water receiving area; and step S12c, using Fourier transform to divide the lake inflow characteristic factors and the water demand characteristic factors of the water receiving area into high and low frequencies.

[0021] In some embodiments, the step S13 comprises: step S13a, using GRU to predict the high-frequency feature sequence, using dynamic programming to solve the DTW distance between the prediction result and the true value as the DTW loss, combining the DTW loss and the mean square error to form the total loss, using the total loss to optimize the parameters of the GRU model, and obtaining the DTW-GRU model; step S13b, using MLP to predict the low-frequency feature sequence, using dynamic programming to solve the DTW distance between the prediction result and the true value as the DTW loss, combining the DTW loss and the mean square error to form the total loss, using the total loss to optimize the parameters of the MLP model, and obtaining the DTW-MLP model; and step S13c, dividing the data set into a training set and a test set, using the training set to train the DTW-GRU model and the DTW-MLP model respectively, and using the trained models to predict the test set to obtain the prediction results of the high-frequency features and the low-frequency features respectively.

[0022] In some embodiments, the large complex inter-basin water transfer model established in the step S3 has a target function including: the minimum total water shortage rate of the water receiving area of the inter-basin water transfer system and the minimum power consumption of the water pumping station; and the priority weight between the water supply benefit and the risk is determined in the step S3, specifically: the minimum total water shortage rate of the water receiving area is taken as the first target, the minimum power consumption of the water pumping station is taken as the second target, and the priority of the first target is higher than that of the second target; the improved IMODE algorithm with the priority weight is used to solve the model, and in the iteration process, if the current optimal individual does not meet the second target, the remaining water supply capacity of the replaceable Jiangjiang water quantity between the lakes along the line is combined as a parameter, and the current optimal individual is corrected in the direction of the second target without weakening its optimal characteristics.

[0023] In some embodiments, the step S5 uses a bidirectional RNN to calculate the correlation coefficient of the water supply of each lake and the water demand of each water receiving area, and to formulate a water supply rule and construct a water source model according to the correlation coefficient.

[0024] In some embodiments, the step S5 includes: a step S51 of collecting historical water supply of each lake and water demand of each section of water receiving area; a step S52 of inputting the data into a bidirectional recurrent neural network to calculate the correlation coefficient of the water supply of each lake and the water demand of each water receiving area; and a step S53 of formulating a water supply rule and establishing a water source model according to the correlation characteristics of the water supply of each lake and the water demand of each water receiving area.

[0025] In the second aspect, a multi-water source analysis system of a cross-basin water diversion project based on a scheduling model includes: a data collection and prediction module for performing steps S1 and S2 to establish a water inflow and water demand prediction model and obtain prediction data; an optimal scheduling module for performing steps S3 and S4 to establish a water diversion model and solve an optimal water diversion scheme; and a water source analysis module for performing steps S5 and S6 to construct a water source model and analyze the water source composition of each water receiving area.

[0026] In the third aspect, an electronic device includes at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the processor, and the processor implements any of the methods when executing the computer program.

[0027] In the fourth aspect, a computer readable storage medium stores a computer program executable by a processor to implement any of the methods.

[0028] Compared with the prior art, the beneficial effects and advantages of the present application are as follows:

[0029] 1. By constructing a multi-modal collaborative prediction model based on error compensation (DTW-GRU for high-frequency features, DTW-MLP for low-frequency features, and KNN for error compensation), the accuracy of medium and long-term water inflow and water demand prediction is effectively improved, and the risk caused by uncertainty is reduced.

[0030] 2. A multi-objective optimization model considering water supply benefits (minimum water shortage rate) and operation risks (minimum power consumption) is established, and an improved IMODE algorithm with priority weight is used to solve it. Under the premise of ensuring water supply safety, economic operation is realized, and a more practical optimization scheduling scheme is obtained.

[0031] 3. Innovatively, a bidirectional recurrent neural network (Bi-RNN) is introduced to analyze historical water supply patterns and construct a water source division model. This model can clearly quantify the water source composition ratio of each water receiving area and solve the problem of difficult water source division in complex water transfer systems through water source analysis, thereby achieving refined management and traceability of water resources.

[0032] 4. By integrating the prediction, optimization, and analysis processes into a single system, the entire process from data input to scheme output and water source analysis is automated, significantly improving the intelligence level and decision-making efficiency of inter-basin water transfer project scheduling and management. Attached Figure Description

[0033] Figure 1 This is a flowchart of the multi-source analysis method for inter-basin water transfer projects based on a scheduling model, as described in this invention.

[0034] Figure 2 This is a flowchart of step S1 of the present invention.

[0035] Figure 3 This is a flowchart of step S12 of the present invention.

[0036] Figure 4 This is a flowchart of step S13 of the present invention.

[0037] Figure 5 This is a flowchart of step S14 of the present invention.

[0038] Figure 6 This is a flowchart of step S3 of the present invention.

[0039] Figure 7 This is a flowchart of step S4 of the present invention.

[0040] Figure 8 This is a flowchart of step S5 of the present invention.

[0041] Figure 9 This is a flowchart of step S52 of the present invention.

[0042] Figure 10 This is a flowchart of step S53 of the present invention.

[0043] Figure 11 This is a block diagram of the multi-source water transfer system for inter-basin water transfer projects based on the scheduling model of the present invention. Detailed Implementation

[0044] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0045] Example 1: As Figure 1 The diagram shows the overall process of the multi-source water transfer project analysis method based on the scheduling model proposed in this invention, which specifically includes the following steps:

[0046] Step S1, collect regional historical lake inflow data and water receiving area data, and establish a multi-modal coordinated error compensation based inflow and demand prediction model;

[0047] Step S2, obtain inflow and demand prediction data using the prediction model;

[0048] Step S3, determine the priority weight between water supply benefit and risk, and establish a large-scale complex inter-basin water transfer model for maximizing water supply benefit and risk coordination;

[0049] Step S4, obtain the optimal water transfer scheme using the water transfer model according to the priority weight and prediction data;

[0050] Step S5, collect historical water supply data, obtain current water supply rule prediction data based on a bidirectional recurrent neural network, and construct a water source model;

[0051] Step S6, obtain the water source composition of each water receiving area using the water source model.

[0052] As shown in Figure 2 , the step S1 further comprises:

[0053] Step S11, collect regional historical lake inflow data, and collect water receiving area data, including water receiving area demand data, precipitation data, evaporation data, groundwater level data, land use data, population, economic development index, and industrial index;

[0054] Step S12, determine the lake inflow and water receiving area demand characteristic factors, and divide the factors into high frequency and low frequency. High frequency factors reflect short-term fluctuations and help reduce prediction errors. Low frequency factors reveal long-term trends and provide a basis for long-term planning;

[0055] Step S13, perform multi-modal coordination training on the divided factors, use DTW-GRU model for high frequency characteristics prediction, and use DTW-MLP for low frequency characteristics prediction;

[0056] Step S14, learn the high and low frequency characteristic errors through KNN regression, compensate for the errors caused by different characteristics, and obtain the final prediction result;

[0057] In this embodiment, a multi-modal collaborative water supply and demand prediction model based on error compensation is constructed. First, the lake inflow and the water demand characteristics of the water receiving area are determined. Since the high-frequency factor mainly reflects short-term fluctuations, it can reduce prediction errors, and the low-frequency factor can reveal long-term trends, which can provide a basis for long-term planning and decision-making. Therefore, the factors are divided into high-frequency and low-frequency. Then, the divided factors are trained in multi-modal coordination. In hydrological calculation, the mean square error only calculates the difference between the predicted sequence and the actual sequence at the corresponding time step, assuming that the two are strictly aligned. However, in actual hydrological data, the predicted sequence and the actual sequence may not be synchronized in time, making it difficult to effectively measure the true prediction accuracy. To solve this problem, the model introduces a dynamic time warping (DTW) loss to capture more flexible time relationships, allowing the prediction results to better match the dynamic changes of actual data. For different frequency characteristics, the model uses different prediction methods: the DTW-GRU model is used to capture short-term volatility in high-frequency characteristics, improving the ability to adapt to short-term dynamics; the DTW-MLP model is used to extract the overall trend of low-frequency characteristics, providing support for long-term trend prediction. Different frequency characteristics can be processed specifically, improving the accuracy and adaptability of the model in hydrological time series prediction; finally, KNN regression is used to learn the error of high and low frequency characteristics, and the error brought by different characteristics is compensated to obtain the final prediction result.

[0058] As shown in Figure 3 , the step S12 further comprises:

[0059] Step S12a, the actual situation of lake inflow depends on precipitation, evaporation, groundwater level, and land use, so the selected characteristic factors are precipitation factor, evaporation factor, groundwater factor, and land use factor;

[0060] Step S12b, the water demand of the water receiving area depends on the population, per capita water consumption, economic development level, industrial index, and water consumption quota, so the selected characteristic factors are population index factor, economic index factor, and industrial index factor;

[0061] Step S12c, use Fourier transform to divide the lake inflow characteristic factors and the water demand characteristic factors of the water receiving area into high and low frequencies;

[0062] The characteristic factors are decomposed into different frequency components, which can further explore the information in the data and understand which factors have greater impact on the predicted variables. By weighting or selecting different frequency characteristics, the generalization ability and prediction ability of the model can be improved, thereby better constructing the model. The actual situation of the lake inflow depends on precipitation, evaporation, groundwater level, and land use, so the selected characteristic factors are precipitation factors, evaporation factors, groundwater factors, and land use factors. The water demand of the receiving area depends on the population, per capita water consumption, economic development level, industrial index, and water consumption quota, so the selected characteristic factors are population index factor, economic index factor, and industrial index factor. Fourier transform can decompose a function or signal into a superposition of sinusoidal waves of different frequencies, which is commonly used in hydrological analysis. In this embodiment, Fourier transform is used to divide the lake inflow characteristic factors and the water demand characteristic factors of the receiving area into high and low frequencies.

[0063] As shown in Figure 4 , wherein the step S13 further comprises:

[0064] Step S13a, using GRU to predict the high-frequency characteristic sequence, using dynamic programming to solve the DTW distance between the prediction result and the true value as the DTW loss. Combine the DTW loss and the mean square error to form the total loss, use the total loss to optimize the parameters of the GRU model, and get the DTW-GRU model;

[0065] Step S13b, using MLP to predict the low-frequency characteristic sequence, using dynamic programming to solve the DTW distance between the prediction result and the true value as the DTW loss. Combine the DTW loss and the mean square error to form the total loss, use the total loss to optimize the parameters of the MLP model, and get the DTW-MLP model;

[0066] Step S13c, divide the data set into training set (first 60%) and test set (last 40%), use the training set to train the DTW-GRU model and the DTW-MLP model respectively, use the trained model to predict the test set, and get the prediction results of the high-frequency characteristics and the low-frequency characteristics respectively;

[0067] In the prediction of incoming water and water demand, the traditional model calculates the difference between the predicted sequence and the actual sequence at the corresponding time step using the mean square error. However, in actual hydrological data, the predicted sequence and the actual sequence may not be synchronized in time, making it difficult for the mean square error to effectively measure the true prediction accuracy. Therefore, in this embodiment, the DTW loss is introduced, and different prediction methods are used for different frequency characteristic factors: the DTW-GRU model is used for high-frequency characteristics to capture short-term volatility in high-frequency characteristics and improve the adaptability to short-term dynamics; the DTW-MLP model is used for low-frequency characteristics to extract the overall trend of low-frequency characteristics and provide support for long-term trend prediction; then, the data set is divided into a training set (the first 60%) and a test set (the last 40%), and the DTW-GRU model and the DTW-MLP model are trained using the training set, and the trained models are used to predict the test set to obtain the prediction results of high-frequency characteristics and low-frequency characteristics, respectively.

[0068] Specifically, the design principles of the DTW loss are as follows:

[0069] Given the true sequence: , the predicted sequence: , the DTW loss formula is defined as the minimum cumulative cost on all legal paths .

[0070] ; (1)

[0071] wherein, is the number of time steps of the predicted sequence; is an alignment path; indicates that the true sequence at the th time point is aligned with the predicted sequence at the th time point; is the length of the alignment path; is a local distance measure, commonly the squared difference or the absolute difference .

[0072] As shown in Figure 5 , the step S14 further includes:

[0073] Step S14a, subtract the prediction result from the measured data to obtain error data, and learn the high and low frequency error data through KNN regression to compensate for the errors caused by different characteristics.

[0074] Step S14b, convert the high and low frequency prediction characteristics and error compensation characteristics into vector form, perform weighted calculation on each characteristic vector through the back propagation algorithm, perform nonlinear processing on the weighted combination result through the activation function, introduce the full connection layer, reconstruct the combined characteristics, and obtain the final prediction result.

[0075] The error data is obtained by subtracting the prediction result from the measured data. The high and low frequency error data is learned by KNN regression to compensate for the error caused by different features. The prediction error is transmitted back to the fully connected layer layer by layer through the back propagation algorithm model, and the contribution of each weight to the error is calculated. In this way, it can be identified which weights cause the error to be too large and which weights help more accurate prediction. Using the gradient descent algorithm, the weight matrix is updated according to the calculated gradient. Each update will adjust the weight in the direction of reducing the error, thereby optimizing the contribution proportion of the feature in the combination. Each iteration gradually improves the weight, so that the expression ability of the combined feature is enhanced. As the training progresses, the weight matrix is gradually optimized, and the contribution proportion of each feature is dynamically adjusted. The higher the weight, the greater the impact on the prediction result, and the model will also be biased towards more valuable features. The optimized combined feature can better reflect the pattern in the data. After multiple rounds of training, the weight matrix optimizes the relationship between features to the best state, thereby generating a more useful combined feature representation for the prediction task. The weighted combination result is processed by the activation function for non-linear processing to capture the complex interaction between features and enhance the expression ability of the combined feature. Through the fully connected layer, the mutual relationship between high and low frequency features is learned, and the error compensation feature is fused into the combination to correct the potential prediction deviation. The combined feature is reconstructed through the fully connected layer to form a comprehensive composite feature containing high and low frequency information and error compensation information, providing complete support for the final prediction. Using the reconstructed feature combination, a more accurate prediction result is output, realizing the full integration and optimization of high and low frequency features and error compensation features.

[0076] As shown in Figure 6 , the step S3 further comprises:

[0077] Step S31, constructing a target function, including: minimizing the total water shortage rate of the water receiving area and minimizing the power consumption of the pumping station in the inter-basin water transfer system;

[0078] Step S32, constructing a constraint condition, including: lake water balance constraint, lake water level constraint, and pumping station water lifting constraint;

[0079] Target function 1: minimizing the total water shortage rate of the water receiving area

[0080] (2)

[0081] In the formula: is the time period number, is the partition number, is the water source number, is the total number of time periods in the scheduling period, is the total number of water receiving areas, Total number of water sources, Total number of water sources, Total number of water sources, Total number of water sources, Water quantity coefficient obtained from water sources, Total number of water sources, Total number of water sources Total number of water sources Water supply and demand of water sources, unit / m3 Total number of water sources, Total number of water sources, Total number of water sources Actual water supply of subareas, unit / m3 Total number of water sources, Total number of water sources,

[0082] Objective function 2: the minimum total power consumption of pump stations as the objective function:

[0083] (3)

[0084] In the formula: Period number, Pump station number, Water density, Total number of water sources, Total number of water sources, Total number of water sources Pumping flow of pump stations, unit / m3 Total number of water sources, Total number of water sources, Total number of water sources Pumping head of pump stations, unit / m Total number of water sources, Working efficiency of pump stations, Total number of water sources, Total number of water sources Power consumption of pump stations.

[0085] As shown in Figure 7 The step S4 further comprises:

[0086] Step S41, the primary objective based on the actual large-scale inter-basin water transfer project is to ensure water use in the water receiving area, so the minimum total water shortage rate of the project water receiving area is considered as the first objective;

[0087] Step S42, on the basis of meeting the water supply demand to the greatest extent, the minimum power consumption of the system is pursued to save the water supply cost, so the minimum power consumption of the pump station is considered as the second objective;

[0088] Step S43, according to the priority of the first target being higher than the requirement of the second target, the IMODE algorithm is first used to solve the first target, and in the iterative evolutionary search process, the optimal individual in each generation population is checked to see whether the decision state represented by the optimal individual meets the second target, if not, it indicates that there is a residual water supply capacity of replaceable water quantity of the water source along the line between the lakes, and the optimal individual is modified in the direction of the second target without weakening the optimal characteristics of the optimal individual of the current generation;

[0089] The multi-objective optimization problem is characterized in that there is usually a contradiction and competition relationship between each sub-target, and it is not easy to achieve the optimal state at the same time. Therefore, the multi-objective optimization problem often adopts a coordinated solution, that is, each sub-target is assigned a weight, and is integrated into a single objective function. In the embodiment, the primary target is to guarantee the water use of the water receiving area, and therefore the minimum water shortage of the water receiving area of the project is considered as the first target. At the same time, on the basis of meeting the water supply demand to the greatest extent, the minimum power consumption of the water pumping system is pursued to save the water supply cost, and therefore the minimum power consumption of the water pumping station is considered as the second target. According to the requirement that the priority of the first target is higher than the requirement of the second target, the IMODE algorithm is first used to solve the first target, and in the iterative evolutionary search process, the optimal individual in each generation population is checked to see whether the decision state represented by the optimal individual meets the second target, if not, it indicates that there is a residual water supply capacity of replaceable water quantity of the water source along the line between the lakes, and the optimal individual is modified in the direction of the second target without weakening the optimal characteristics of the optimal individual of the current generation.

[0090] As shown in Figure 8 , the step S5 further comprises:

[0091] Step S51, collect the historical water supply quantity of each lake and the water demand of each section of the water receiving area;

[0092] Step S52, input the data into the bidirectional RNN (including the forward RNN and the backward RNN), and calculate the correlation coefficient of the water supply quantity of each lake and the water demand of each water receiving area;

[0093] Step S53, according to the correlation characteristics of the water supply quantity of each lake and the water demand of each water receiving area, formulate a water supply rule, and establish a water source model;

[0094] The bidirectional RNN considers not only the information in front of the sequence but also the information behind the sequence when processing the sequence data. It is composed of two independent RNN layers (forward RNN and backward RNN): one runs from the beginning of the sequence to the end, and the other runs from the end of the sequence to the beginning. The output at each time is the concatenation of the hidden states of the two directions, and the final output is calculated based on the merged hidden state. In this embodiment, the data is input into the bidirectional RNN (including the forward RNN and the backward RNN), the correlation coefficients of the water supply of each lake and the water demand of each water receiving area are calculated, the water supply rules are formulated according to the correlation characteristics of the water supply of each lake and the water demand of each water receiving area, and the water source allocation model is established.

[0095] As shown in Figure 9 , wherein the step S52 further comprises:

[0096] Step S52a, the historical data is input into the forward RNN from front to back one by one. At each time step, the forward RNN calculates a new hidden state and output according to the current input and the hidden state of the last time step, which represents the prediction of the future based on the past information at this time point;

[0097] Step S52b, the historical data is input into the backward RNN from back to front one by one. At each time step, the backward RNN calculates a new hidden state and output according to the current input and the hidden state of the next time step, which represents the prediction of the past based on the future information at this time point;

[0098] Step S52c, the outputs of the forward and backward RNNs at the last time step are connected and input into the fully connected layer for linear combination to obtain the final prediction values of the water supply of each lake and the water demand of each water receiving area, and the Spearman rank correlation coefficients are calculated;

[0099] As shown in Figure 10 , wherein the step S53 further comprises:

[0100] Step S53a, the correlation coefficients of the water supply of each lake and the water demand of each water receiving area are converted into the proportion of the water supply of each lake to each water receiving area;

[0101] Step S53b, the water supply rules are formulated according to the proportion of the water supply of each lake to each water receiving area, and the water source allocation model is established.

[0102] Embodiment two: as Figure 11As shown, it comprises: a scheduling model-based multi-source analysis system for cross-basin water diversion projects, comprising: a data collection and prediction module for performing steps S1 and S2, establishing a water inflow and water demand prediction model and obtaining prediction data; an optimal scheduling module for performing steps S3 and S4, establishing a water diversion model and solving an optimal water diversion scheme; and a water source analysis module for performing steps S5 and S6, constructing a water source model and analyzing the water source composition of each water receiving area.

[0103] Embodiment three: an electronic device, comprising: at least one processor; and a memory connected in communication with the at least one processor; wherein the memory stores instructions executable by the processor, and the processor implements the scheduling model-based multi-source analysis method for cross-basin water diversion projects of the application in embodiment one when executing the computer program.

[0104] Embodiment four: a computer-readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the scheduling model-based multi-source analysis method for cross-basin water diversion projects of the application in embodiment one.

[0105] To sum up, the scheduling model-based multi-source analysis method for cross-basin water diversion projects can meet the actual needs of multi-objective large cross-basin water diversion systems with water supply benefit coordination risks and water inflow and water demand uncertainty disturbances, and solve the problem of difficult water source division in current complex large cross-basin water diversion systems.

[0106] The above only describes the preferred embodiments of the application, and is not intended to limit the application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the application shall be included in the protection scope of the application.

[0107] The preferred embodiments of the application are described in detail above in combination with the drawings and technical solutions, and only describe the preferred embodiments of the application, and do not limit the concept and scope of the application. In order to avoid unnecessary repetition, various possible combinations are not described again in the application. Each specific technical feature described in the above specific embodiments can be combined, modified or improved in any appropriate manner without departing from the concept of the application, and shall fall within the protection scope of the application.

Claims

1. A method for multi-source analysis of inter-basin water transfer projects based on a scheduling model, characterized in that, Includes the following steps: S1. Collect historical lake inflow data and water-receiving area data in the region, and establish a multimodal collaborative water inflow and demand prediction model based on error compensation; S2. Obtain water inflow and water demand forecast data using the aforementioned prediction model; S3. Determine the priority weight between water supply benefits and risks, and establish a large-scale complex inter-basin water transfer model that maximizes water supply benefits and coordinates risks. S4. Obtain the optimal water diversion scheme using the water diversion model based on the priority weights and forecast data; S5. Collect historical water supply data, obtain current water supply pattern prediction data based on bidirectional recurrent neural network, and construct a water source differentiation model; S6. Use the water source model to obtain the water source composition of each water receiving area.

2. The method according to claim 1, characterized in that, Step S1 includes: Step S11: Collect historical lake inflow data for the region, as well as water demand data, precipitation, evaporation, groundwater level, land use, population, economic development index, and industry index for the water-receiving area; Step S12: Determine the characteristic factors of lake inflow and demand, and classify them into high-frequency factors and low-frequency factors; Step S13: Perform multimodal coordination training on the divided factors. High-frequency features are predicted using the DTW-GRU model, and low-frequency features are predicted using the DTW-MLP model. Step S14: Learn the errors of high-frequency and low-frequency features through KNN regression, compensate for the errors caused by different features, and obtain the final prediction result.

3. The method according to claim 2, characterized in that, Step S12 includes: Step S12a, using precipitation factor, evaporation factor, groundwater factor, and land use factor as lake water inflow characteristic factors; Step S12b, using population index factor, economic index factor, and industrial index factor as water demand characteristic factors of the water-receiving area; Step S12c, using Fourier transform to classify the lake water inflow characteristic factors and the water demand characteristic factors of the water-receiving area into high and low frequencies.

4. The method according to claim 2, characterized in that, Step S13 includes: Step S13a: Predicting high-frequency feature sequences using GRU, solving for the DTW distance between the predicted result and the true value using dynamic programming as the DTW loss, combining the DTW loss and mean squared error to form the total loss, and using the total loss to optimize the parameters of the GRU model to obtain the DTW-GRU model; Step S13b: Predicting low-frequency feature sequences using MLP, solving for the DTW distance between the predicted result and the true value using dynamic programming as the DTW loss, combining the DTW loss and mean squared error to form the total loss, and using the total loss to optimize the parameters of the MLP model to obtain the DTW-MLP model; Step S13c: Dividing the dataset into a training set and a test set, training the DTW-GRU model and the DTW-MLP model using the training set respectively, and using the trained model to predict the test set to obtain the prediction results for high-frequency features and low-frequency features respectively.

5. The method according to claim 1, characterized in that, The large-scale complex inter-basin water transfer model established in step S3 has objective functions including: minimizing the total water shortage rate in the water-receiving area of ​​the inter-basin water transfer system and minimizing the power consumption of pumping stations. In step S3, the priority weight between water supply benefits and risks is determined as follows: minimizing the total water shortage rate in the water-receiving area is taken as the first objective, and minimizing the power consumption of pumping stations is taken as the second objective, with the first objective having higher priority than the second objective. The improved IMODE algorithm with priority weights is used to solve the model. During the iteration process, if the current optimal individual does not meet the second objective, it is corrected towards the second objective without weakening its optimal characteristics, taking into account the parameter that there is remaining water supply capacity among the lakes along the route that can replace the diverted water volume.

6. The method according to claim 1, characterized in that, In step S5, a bidirectional RNN is used to calculate the correlation coefficient between the water supply of each lake and the water demand of each water-receiving area, and water supply rules are formulated accordingly to construct a water source separation model.

7. The method according to claim 1, characterized in that, Step S5 includes: Step S51, collecting historical water supply data for each lake and water demand data for each water-receiving area; Step S52, inputting the data into a bidirectional recurrent neural network to calculate the correlation coefficient between the water supply data of each lake and the water demand data of each water-receiving area; Step S53, formulating water supply rules and establishing a water source separation model based on the correlation characteristics between the water supply data of each lake and the water demand data of each water-receiving area.

8. A multi-source water transfer project analysis system based on a scheduling model, the system comprising: The data collection and forecasting module is used to execute steps S1 and S2, establish water inflow and water demand forecasting models and obtain forecast data; The optimization scheduling module is used to execute steps S3 and S4, establish a water transfer model, and solve for the optimal water transfer scheme; the water source analysis module is used to execute steps S5 and S6, construct a sub-water source model, and analyze the water source composition of each water receiving area.

9. An electronic device, comprising: At least one processor; And a memory communicatively connected to at least one of the processors; wherein the memory stores instructions executable by the processor, which, when executing the computer program, implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 7.

Images