Multi-energy complementary system optimization scheduling rule extraction method simultaneously considering new energy consumption and river ecological protection

Abstract

A multi-energy complementary system optimization scheduling rule extraction method considering new energy consumption and river ecological protection at the same time comprises the steps that basic conditions of multi-energy complementation are arranged, a sample test is carried out, and a scheduling optimal solution and reservoir feature data such as runoff situation and water level are obtained through a multi-objective optimization algorithm. The result structure is layered, and a series of key factors including the cycle number, the reservoir front water level, the reservoir inflow and the inflow of the current time period are selected as the input of the model by selecting significant features. The correlation among other factors is evaluated through a mutual information method (MIR) and an F test (FR), and a decision tree algorithm model is constructed by using an optimization result of a multi-objective optimization algorithm. And the step tree assigner optimizes the number of input factors and finds an optimal hyper-parameter combination of the model. The method can be suitable for multi-energy complementary systems in different areas and different modes, and is good in adaptability and high in practicability.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy regulation and control, and in particular relates to a method for extracting optimal dispatching rules for a multi-energy complementary system that simultaneously considers new energy consumption and river ecological protection. Background Art

[0002] Over the past decade, my country's wind-hydro-photovoltaic multi-energy complementary system has achieved unprecedented development, and the installed capacity has continued to grow. In particular, the installed capacity of wind power and solar power generation is in a leading position worldwide. However, affected by the instability, randomness and variability of wind and solar power generation, the consumption of new energy has always been a prominent challenge. The optimization dispatching rules of the multi-energy complementary system are centered on hydropower stations, comprehensively consider the benefits of wind and solar power generation, fully tap the flexibility and adjustment capabilities of the multi-energy complementary system, promote the full consumption of new energy, and improve the effective method of hydropower for intermittent energy such as wind power and photovoltaics. However, compared with the independent dispatching of cascade reservoirs, the multi-energy complementary operation has changed the dispatching and operation boundaries of the reservoir group, resulting in the reservoir group dispatching rules no longer being applicable. Therefore, it is necessary to analyze the impact of the strong volatility, randomness and intermittent characteristics of wind and solar output on the dispatching mode of the multi-energy complementary system based on the actual situation of the multi-energy complementary system, and explore the formulation method of the multi-energy complementary dispatching rules.

[0003] Existing research uses scheduling rule formulation methods such as machine learning and linear programming. Most of these methods use linear rules with fewer parameters and a single structure. Although simple and intuitive, it is difficult to accurately describe complex scheduling rule decisions. Although the machine learning scheduling rule formulation method is better than the traditional linear scheduling rule, this method is a black box model that lacks interpretability and is difficult to directly apply to engineering practice.

[0004] The wind-water-light multi-energy complementary operation optimization dispatching model in the upper reaches of the Yellow River, with the main goals of new energy consumption and river ecological protection, combined with a decision tree algorithm based on interpretable rules, has formulated a comprehensive dispatching rule for the multi-energy complementary system that takes into account the maximum stability of the multi-energy complementary system operation and the minimum change in river runoff. The research results are of great significance for realizing the optimal operation of the multi-energy complementary system and the application of the optimization dispatching rules in practice.

[0005] To sum up, the technical problem to be solved by the present invention is: how to achieve the target balance between the output stability of the multi-energy complementary system and the maximization of power generation, ensure the maximum output of the reservoir group after absorbing the wind and solar energy within the scheduling cycle, and at the same time improve the output stability of the multi-energy complementary system. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a method for extracting optimization scheduling rules for a multi-energy complementary system that takes into account both new energy consumption and river ecological protection, taking into account multiple objectives: simultaneously taking into account new energy consumption, output stability of the multi-energy complementary system, maximizing power generation, and river ecological protection, thereby achieving comprehensive optimization of the multi-energy complementary system and improving the overall performance of the system.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0008] The method for extracting optimal dispatching rules for a multi-energy complementary system that considers both new energy consumption and river ecological protection is as follows:

[0009] Step 1: Establish a multi-energy complementary model, with the goal of ensuring the output stability of the multi-energy complementary system and maximizing the power generation, while reducing the impact of the operation and scheduling of the multi-energy complementary system on the river runoff situation, and construct a multi-objective optimization scheduling model for the multi-energy complementary system.

[0010] Step 1.1, construct a multi-energy complementary system with the maximum power generation objective function. The maximum power generation objective is to ensure that the output of the reservoir group after absorbing wind and solar energy is the maximum within the dispatch period T. It is expressed in terms of total power generation P, and the calculation formula is as follows:

[0011]

[0012] Where: P max is the maximum amount of total power generation, kw; is the average output of the time period, kw; Δt is the length of the time period, 24h; P i,t , P j,t , P k,t They are the power generation of the i-th wind power station, the j-th hydropower station and the k-th photovoltaic field in the t-th time period, kw; T is the total number of time periods in the scheduling period.

[0013] Step 1.2, construct a multi-energy complementary system with the highest objective function of total output stability. The output stability of the multi-energy complementary system can be measured by the coefficient of variation of the power generation output. The smaller the coefficient of variation, the more stable the output. The calculation formula is as follows:

[0014]

[0015] Where: CV min 、std min (P) and They are the minimum coefficient of variation, minimum standard deviation and mean value of the total output of the multi-energy complementary system respectively.

[0016] Step 1.3, construct the objective function of the minimum change of river runoff pattern in the multi-energy complementary system. The minimum change of river runoff pattern is the difference between the downstream runoff Q of the hydropower station during the fish spawning period from April to June and the natural runoff pattern q t The difference between them is as small as possible, and the calculation formula is as follows:

[0017]

[0018] Where: Q is the natural runoff situation of downstream runoff, m / s, q is the downstream runoff, m / s; a and b are the end time and the start time of the spawning period (days), respectively.

[0019] Step 2: Use a multi-objective optimization algorithm to comprehensively consider multiple objectives, such as rationality, environmental impact, and system stability, to find the optimal solution set that balances various indicators, thereby improving the overall performance of the multi-energy complementary system.

[0020] Step 2.1, generate a new population. First, randomly generate a parent population N. Then, operate the parent generation through selection, crossover and mutation to generate a child population N. The father and the child are combined into 2N to form a new parent generation. Then the population is added to the next generation of the child generation through fast non-dominated sorting and elite strategy selection to form a new parent generation D. When the number of new parents D is greater than N, T selects N individuals and generates a new population.

[0021] Step 2.2: The number of reference points is mainly related to the number of objective functions, the location of the reference points, and the number of shares each objective is divided into. The specific calculation formula is as follows:

[0022]

[0023] Where: R is the number of reference points; U is the number of objective functions; T is the number of parts for each objective.

[0024] Step 2.3, for the processing of individuals in the population, first perform a translation operation on the population so that the ideal point becomes the origin. The calculation formula is as follows:

[0025]

[0026] Where: f'i(x) and fi(x) are the values ​​of the individuals of the i-th target dimension in the U target functions in the population after and before the translation operation, respectively; To calculate the individual minimum value of each target dimension in the U target functions in the population.

[0027] Secondly, calculate the extreme points of each objective function. The calculation formula is as follows:

[0028]

[0029] In the formula: AFF(x i ,σ) is the extreme point of the i-th objective function, and σ is the unit direction vector as the coordinate axis. When σ is 0, take 10 -6 By forming a straight line between the extreme point of each coordinate axis and its ideal point, a U-dimensional linear hyperplane is established, and the intercept αi is obtained by intersecting with the focus of the coordinate axis. Using the general equation of the linear hyperplane, the specific function value is substituted to obtain the intercept, and then the intercept is normalized according to formula (4).

[0030] Finally, the population individuals are normalized, and the calculation formula is as follows:

[0031]

[0032] Step 2.4, output the configuration plan. The reference line takes the origin and the reference point connecting line, calculates the distance from each individual in the population to each reference line, and establishes the relationship between the individual and the reference line. The reference point elite strategy selects the offspring individual with a scale of N as the reference point to join the next generation.

[0033] Step 3, by stratifying the optimal solution of the multi-objective optimization algorithm, analyzing the main influencing factors, and selecting the significant characteristics of wind, water and photovoltaic power generation, a series of key factors were selected as the input of the model, including the number of cycles, the water level before the reservoir, the reservoir inflow and the inflow in the current period, due to the difficulty in collecting reservoir output, load and forecast flow data. Considering their close relationship with the current storage period and decision-making, they are determined as mandatory input factors.

[0034] Step 4: By building an appropriate decision tree model, you can quickly predict and make decisions, classify and evaluate different renewable energy power generation plans, solve classification and regression problems, accurately select the best power generation plan, and improve the efficiency of renewable energy utilization.

[0035] Step 4.1, divide the training data set according to the input variables and output variables. The calculation formula is as follows:

[0036] E={(A1,B1),(A2,B2),...,(A n ,B n );

[0037] Where: E is the partitioned training data set; A is the input variable; B is the continuous output variable.

[0038] Step 4.2, by checking each eigenvector and its corresponding eigenvalue one by one. The two regions are divided and defined by iterating each eigenvariable and its corresponding eigenvalue in turn. Through the above steps, the input space is divided into N subspaces, each of which has a part of sample data and corresponding output values. The solution of the model is generated. The calculation formula is as follows:

[0039]

[0040] Where: k is the current segmentation variable; t is the segmentation feature value corresponding to the k segmentation variable; δ n For each subspace, there is a part of sample data and the corresponding output value; a1, a2, ..., a N Subspace.

[0041] Step 4.3, use the square error minimization criterion to determine the optimal split point and prediction value, and compare the loss function error. According to the least squares principle, determine the optimal value of the mean of all outputs in the subspace. The calculation formula is as follows:

[0042]

[0043] Where: S is the square error; y i Subspace a N All output values ​​on; f(x i ) Subspace a N Solutions of all models on ; is δ n The optimal value of y i Subspace a N All output values ​​on .

[0044] Step 4.4, select the best segmentation attribute and value to divide the input space, traverse all input features and values, find the best split point, and divide the space into two sub-areas. If the sub-area cannot be divided any further, the best output value is obtained. The calculation formula is as follows:

[0045]

[0046] Where: e and d are the optimal segmentation attribute and attribute value respectively.

[0047] According to steps 4.1-4.4, as long as one area can still be divided, continue to repeat the above steps until the entire decision tree is no longer divisible.

[0048] Step 4.5: Accuracy evaluation of multi-energy complementary scheduling rule extraction

[0049] Step 4.5.1, evaluation of the importance of dispatch rule prediction; In order to study the comprehensive power generation benefits of new energy and multi-energy complementary systems, the decision tree model is used to predict the water level of the reservoir based on the typical year optimization dispatch results, with the reservoir inflow and wind and solar output as input prediction factors. The total number of days of typical year optimization dispatch results is obtained, in which the training set and the test set are divided. After the model is run multiple times, the average wind and solar power generation and reservoir flow of the multi-energy complementary model are used as prediction factors. The importance of the prediction factor is calculated by summing up the changes caused by the splitting of each prediction factor, and then dividing the sum by the number of branch nodes.

[0050] Step 4.5.2, automatic optimization of hyperparameter evaluation of scheduling rules; by automatically optimizing hyperparameters, the robustness of the extraction of scheduling rules for the multi-energy complementary model can be improved, and errors and uncertainties that may occur when manually adjusting parameters can be avoided. At the same time, the risk of overfitting or underfitting can be reduced, and better generalization to new data sets can be achieved. In multiple simulation tests, when the highest accuracy is achieved, the maximum number of target evaluations, the total number of function calculations, the total duration, the total objective function calculation time, the observed best feasible point (minimum number of leaf nodes), the observed objective function value, the estimated objective function value, the function calculation time, the best feasible point estimated according to the model, the estimated objective function value, the estimated function calculation time and other optimization parameter results are obtained.

[0051] Step 4.5.3, parameter evaluation of the number of leaf nodes in the scheduling rule; adjusting the minimum number of samples for the number of leaf nodes can help control the complexity of the model and improve the generalization ability of the model, so that the model can be better applied to unseen data. The influence of leaf nodes on the minimum number of samples in multiple simulation tests was obtained, from which it can be seen that the resampling error, cross-validation error, prediction accuracy and error rate all change with the number of leaf nodes.

[0052] Step 4.5.4, evaluation of the scheduling rule pruning depth parameter; Since the depth of the optimal decision tree model obtained by the above training process is too complex, further pruning operations are needed to further improve the adaptability of the decision tree model. Pruning is used to reduce the complexity of the model and prevent overfitting in model training, which can help the decision tree model achieve better performance and generalization ability. The impact of pruning depth on model performance in multiple simulation tests is obtained, and the performance of the model (revalidation error, cross-validation error, prediction accuracy and prediction error rate, etc.) is affected when the number of pruning changes. The optimal value and average model performance of the comprehensive evaluation model are obtained.

[0053] Step 4.5.5, dispatching rule balance accuracy evaluation and decision analysis: The balance relationship between the number of leaf nodes and the pruning depth depends on the complexity of the model and the requirements of generalization ability, which can be used to dispatch the water-wind-solar multi-energy complementary system in the upper reaches of the Yellow River. Therefore, according to the inflow of the reservoir, the output of wind and solar energy is balanced, and the comprehensive dispatching rules of hydropower stations are formulated to take into account the maximum operation stability of the multi-energy complementary system and the minimum change in river runoff. The reservoir level of the reservoir is predicted to help the multi-energy complementary system achieve efficient task allocation and resource scheduling, and improve overall performance and efficiency.

[0054] The present invention can achieve the following beneficial effects:

[0055] The present invention provides a method for extracting optimized dispatching rules for a multi-energy complementary system that takes into account both new energy consumption and river ecological protection. The method analyzes the power generation dispatching rules of the multi-energy complementary system through a multi-objective optimization algorithm, introduces a decision tree algorithm, extracts the power generation dispatching rules of the multi-energy complementary system, and proposes a comprehensive dispatching rule for a multi-energy complementary system that takes into account both ecology and power generation. The method is applicable to a variety of multi-energy complementary systems and different river ecological protection requirements and targets, has good adaptability and strong practicality, can effectively reduce the adverse effects of the operation of the multi-energy complementary system on the river ecology, and promote the sustainable and high-quality development of new energy.

[0056] The present invention comprehensively considers multiple goals: simultaneously considering multiple goals such as new energy consumption, output stability of the multi-energy complementary system, maximizing power generation and river ecological protection, and can achieve comprehensive optimization of the multi-energy complementary system and improve the overall performance of the system.

[0057] The present invention uses a variety of accuracy evaluation methods, including scheduling rule prediction importance evaluation, automatic optimization hyperparameter evaluation, leaf node number parameter evaluation, pruning depth parameter evaluation, and balanced accuracy evaluation and decision analysis, to comprehensively evaluate the scheduling rules and improve the reliability and generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0059] Figure 1 A flow chart of a method for extracting optimal dispatching rules for a multi-energy complementary system that takes into account both new energy consumption and river ecological protection provided by the present invention;

[0060] Figure 2 A schematic diagram of the solution results of the multi-energy complementary model provided by the present invention;

[0061] Figure 3 A schematic diagram of the importance ranking of the multi-energy complementation model provided by the present invention;

[0062] Figure 4A schematic diagram of the automatic optimization results of hyperparameters of the multi-energy complementary model provided by the present invention;

[0063] Figure 5 A schematic diagram of the influence of leaf nodes on decision-making performance provided by the present invention;

[0064] Figure 6 A schematic diagram of the influence of pruning depth on decision parameters provided by the present invention;

[0065] Figure 7 A schematic diagram of the number of leaf nodes and the frequency of occurrence of pruning depth when the model provided by the present invention has the maximum accuracy;

[0066] Figure 8 A schematic diagram of the decision tree dispatching rules for the multi-energy complementary system in the upper Yellow River provided by the present invention. DETAILED DESCRIPTION

[0067] like Figure 1 As shown, the present invention provides a method for extracting optimal dispatching rules for a multi-energy complementary system for new energy consumption and river ecological protection. Taking the Maerdang-Banduo-Yangqu Hydropower Station in the upper reaches of the Yellow River (the basic data of the reservoir are shown in Table 1), two wind power stations (Qieji and Ertala) and two photovoltaic power stations (Yitala and Santala) forming a complementary system for supplying power to the power grid as an example, a wind, water and light multi-energy complementary model is established, including the following steps:

[0068] Table 1 Basic data of reservoirs

[0069]

[0070] Step 1: Establish a wind-water-solar multi-energy complementary model, with the goal of stabilizing the output of the multi-energy complementary system and maximizing the power generation, while reducing the impact of the operation and scheduling of the multi-energy complementary system on the river runoff situation, and construct a multi-objective optimization scheduling model for the wind-water-solar multi-energy complementary system. The calculation results are as follows: Figure 2 Shown

[0071] Step 1.1, construct the wind, water and light multi-energy complementary system with the maximum power generation objective function. The maximum power generation objective is to ensure that the output of the Yangqu-Banduo-Maerdang reservoir group after absorbing the wind and light new energy is the maximum within the dispatch period T. It is expressed in terms of total power generation P, and the calculation formula is as follows:

[0072]

[0073] Where: P max is the maximum amount of total power generation, kw; is the average output of the time period, kw; Δt is the length of the time period, 24h; P i,t , P j,t , P k,tThey are the power generation of the i-th wind power station, the j-th hydropower station and the k-th photovoltaic field in the t-th time period, kw; T is the total number of time periods in the scheduling period.

[0074] Step 1.2, construct the wind, water and light multi-energy complementary system with the highest objective function of total output stability. The output stability of the wind, water and light multi-energy complementary system can be measured by the coefficient of variation of the power output. The smaller the coefficient of variation, the more stable the output. The calculation formula is as follows:

[0075]

[0076] Where: CV min 、std min (P) and P are the minimum coefficient of variation, minimum standard deviation and mean value of the total output of the multi-energy complementary system, respectively.

[0077] Step 1.3, construct the objective function of the minimum change of river runoff pattern in the wind, water and light multi-energy complementary system. The minimum change of river runoff pattern is the difference between the downstream runoff Q of the hydropower station during the fish spawning period from April to June and the natural runoff pattern q after the hydropower station is dispatched and operated. t The difference between them is as small as possible, and the calculation formula is as follows:

[0078]

[0079] Where: Q is the natural runoff situation of downstream runoff, m / s, q is the downstream runoff, m / s; a and b are the end time and the start time of the spawning period (days), respectively.

[0080] Step 2: Use a multi-objective optimization algorithm to comprehensively consider multiple objectives, such as rationality, environmental impact, and system stability, to find the optimal solution set that balances various indicators, thereby improving the overall performance of the multi-energy complementary system.

[0081] Step 2.1, generate a new population. First, randomly generate a parent population N. Then, operate the parent generation through selection, crossover and mutation to generate a child population N. The father and the child are combined into 2N to form a new parent generation. Then the population is added to the next generation of the child generation through fast non-dominated sorting and elite strategy selection to form a new parent generation D. When the number of new parents D is greater than N, T selects N individuals and generates a new population.

[0082] Step 2.2: The number of reference points is mainly related to the number of objective functions, the location of the reference points, and the number of shares each objective is divided into. The specific calculation formula is as follows:

[0083]

[0084] Where: R is the number of reference points; U is the number of objective functions; T is the number of parts for each objective.

[0085] Step 2.3, for the processing of individuals in the population, first perform a translation operation on the population so that the ideal point becomes the origin. The calculation formula is as follows:

[0086]

[0087] Where: f'i(x) and fi(x) are the values ​​of the individuals of the i-th target dimension in the U target functions in the population after and before the translation operation, respectively; To calculate the individual minimum value of each target dimension in the U target functions in the population.

[0088] Secondly, calculate the extreme points of each objective function. The calculation formula is as follows:

[0089]

[0090] In the formula: AFF(x i ,σ) is the extreme point of the i-th objective function, and σ is the unit direction vector as the coordinate axis. When σ is 0, take 10 -6 By forming a straight line between the extreme point of each coordinate axis and its ideal point, a U-dimensional linear hyperplane is established, and the intercept αi is obtained by intersecting with the focus of the coordinate axis. Using the general equation of the linear hyperplane, the specific function value is substituted to obtain the intercept, and then the intercept is normalized according to formula (4).

[0091] Finally, the population individuals are normalized, and the calculation formula is as follows:

[0092]

[0093] Step 2.4, output the configuration plan. The reference line takes the origin and the reference point connecting line, calculates the distance from each individual in the population to each reference line, and establishes the relationship between the individual and the reference line. The reference point elite strategy selects the offspring individual with a scale of N as the reference point to join the next generation.

[0094] Step 3, by stratifying the optimal solution of the multi-objective optimization algorithm, analyzing the main influencing factors, and selecting the significant characteristics of wind, water and photovoltaic power generation, a series of key factors were selected as the input of the model, including the number of cycles, the water level before the reservoir, the reservoir inflow and the inflow in the current period, due to the difficulty in collecting reservoir output, load and forecast flow data. Considering their close relationship with the current storage period and decision-making, they are determined as mandatory input factors.

[0095] Step 4, by constructing an appropriate decision tree model, rapid prediction and decision-making can be achieved, and different renewable energy power generation schemes can be classified and evaluated. Solve the classification and regression problems to accurately select the best power generation scheme and improve the efficiency of renewable energy utilization. Given that the decision tree algorithm is suitable for processing conditional attributes of continuous values, but not for processing decision attributes of continuous values, consider grading the reservoir water level to meet the requirements of the decision tree algorithm. The grading results are shown in Table 2. First, the multi-energy complementary system power generation of the wind-water-light new energy base composed of the Erdang-Banduo-Yangqu reservoir group in normal water years, dry years, and Fengshui years is divided into training samples and test samples, of which 900 are training samples and the remaining 145 are verification samples.

[0096] Table 2 Water level classification of reservoirs in the upper Yellow River multi-energy complementary system

[0097]

[0098] Step 4.1, divide the training data set according to the input variables and output variables. The calculation formula is as follows:

[0099] E={(A1,B1),(A2,B2),...,(A n ,B n );

[0100] Where: E is the partitioned training data set; A is the input variable; B is the continuous output variable.

[0101] Step 4.2, by checking each eigenvector and its corresponding eigenvalue one by one. The two regions are divided and defined by iterating each eigenvariable and its corresponding eigenvalue in turn. Through the above steps, the input space is divided into N subspaces, each of which has a part of sample data and corresponding output values. The solution of the model is generated. The calculation formula is as follows:

[0102]

[0103] Where: k is the current segmentation variable; t is the segmentation feature value corresponding to the k segmentation variable; δ n For each subspace, there is a part of sample data and the corresponding output value; a1, a2, ..., a N Subspace.

[0104] Step 4.3, use the square error minimization criterion to determine the optimal split point and prediction value, and compare the loss function error. According to the least squares principle, determine the optimal value of the mean of all outputs in the subspace. The calculation formula is as follows:

[0105]

[0106] Where: S is the square error; yi Subspace a N All output values ​​on; f(x i ) Subspace a N Solutions of all models on ; is δ n The optimal value of y i Subspace a N All output values ​​on .

[0107] Step 4.4, select the best segmentation attribute and value to divide the input space, traverse all input features and values, find the best split point, and divide the space into two sub-areas. If the sub-area cannot be divided any further, the best output value is obtained. The calculation formula is as follows:

[0108]

[0109] Where: e and d are the optimal segmentation attribute and attribute value respectively.

[0110] According to these steps, as long as an area can still be divided, the process is repeated until the entire decision tree is no longer divisible.

[0111] Step 4.5: Accuracy evaluation of multi-energy complementary scheduling rule extraction

[0112] Step 4.5.1, evaluation of the importance of dispatch rule prediction; in order to study the comprehensive power generation benefits of new energy and wind-hydro-solar multi-energy complementary systems, the decision tree model is used to predict the water level of the reservoir based on the typical year optimization dispatch results, with the reservoir inflow and wind and solar output as input prediction factors. The total number of days of typical year optimization dispatch results is obtained, which is divided into training sets and test sets. After the model is run multiple times, the average wind and solar power generation and reservoir flow of the multi-energy complementary model are used as prediction factors. The importance of the prediction factor is calculated by summing up the changes in each prediction factor caused by the split, and then dividing the sum by the number of branch nodes. The calculation results are as follows Figure 3 Shown

[0113] Step 4.5.2, automatic optimization of hyperparameters of scheduling rules. By automatically optimizing hyperparameters, the robustness of the extraction of scheduling rules for the multi-energy complementary model can be improved, and the errors and uncertainties that may occur when manually adjusting parameters can be avoided. At the same time, the risk of overfitting or underfitting can be reduced, and it can be better generalized to new data sets. Figure 4The maximum number of target evaluations is 30 when the highest accuracy is achieved in 500 simulation tests. The total number of function calculations is 30, and the total duration is 22.75s. The total objective function calculation time is 1.86s, the observed best feasible point (minimum number of leaf nodes) is 54, the observed objective function value is 0.21579, the estimated objective function value = 0.2158, the function calculation time = 0.057723s, and the best feasible point estimated by the model is 69. The estimated objective function value is 0.21572, and the estimated function calculation time is 0.059427s.

[0114] Step 4.5.3, evaluate the parameter of the number of leaf nodes of the scheduling rule; adjusting the minimum number of samples of the number of leaf nodes can help control the complexity of the model and improve the generalization ability of the model, so that the model can be better applied to unseen data. Figure 5 The figure shows the influence of leaf nodes on the minimum number of samples in 500 simulation tests. It can be seen that the resampling error gradually increases and finally stabilizes, the cross-validation error first decreases and then stabilizes, and the prediction accuracy and error rate vary with the number of leaf nodes. When the minimum number of samples contained in the leaf node is less than 14, the resampling error is low and the cross-validation error is high, which is caused by underfitting of the model. As the minimum number of samples contained in the leaf node increases to 38, the cross-validation error gradually decreases because the model can better fit the training data. However, when the minimum number of samples contained in the leaf node is greater than 108, the cross-validation error may begin to increase, which is caused by overfitting of the model; the multi-energy complementary scheduling rule extraction model performs best on the training data when the minimum number of samples contained in the leaf node is 54, with a resampling error of 0.2211, a cross-validation error of 0.2253, and a prediction accuracy of 80.00%, so the model under the parameters is selected as the final model.

[0115] Step 4.5.4, evaluation of scheduling rule pruning depth parameters; Since the depth of the optimal decision tree model obtained by the above training process is too complex, further pruning operations are required to further improve the adaptability of the decision tree model. Pruning is used to reduce the complexity of the model and prevent overfitting during model training, which can help the decision tree model achieve better performance and generalization ability. Figure 6The effect of pruning depth on model performance in 500 simulation tests shows that when the number of pruning changes, the performance of the model (revalidation error, cross-validation error, prediction accuracy and prediction error rate, etc.) will be affected to varying degrees. In the comprehensive evaluation of the optimal value of the model and the average model performance, when the number of pruning is 11, the model performance is the best. At this time, the maximum accuracy of the model in 500 simulation tests is 85.52%, the average resampling error is 0.2185, the average cross-validation error is 0.3327, and the average prediction accuracy is 75.82%. The model is selected as the final model.

[0116] Step 4.5.5, evaluation of scheduling rule balance accuracy and decision analysis: The balance relationship between the number of leaf nodes and the pruning depth depends on the complexity of the model and the requirements of generalization ability, which can be used to schedule the water-wind-solar multi-energy complementary system in the upper reaches of the Yellow River.

[0117] Figure 7 (a) shows the number and frequency of leaf nodes when the accuracy rate reaches the maximum in 5000 simulation tests. It can be seen that the number and frequency of leaf nodes when the accuracy rate reaches the maximum generally show a trend of first increasing and then decreasing. The overall performance parameters are concentrated in the leaf nodes with the highest frequency between 10 and 20. The best performance point of the model is at the leaf node number 14, accounting for 4.38%. Figure 7 (b) shows the number and frequency of pruning depths corresponding to the maximum accuracy in 500 simulation tests. It can be seen that the number and frequency of leaf nodes when the accuracy reaches the maximum generally show a trend of first increasing and then decreasing. The best point of model performance is when the pruning depth is 11, accounting for 18.80%, which can control the complexity and generalization ability of the decision tree model.

[0118] Table 3 compares the average results of each evaluation index of the mining scheduling rules before and after 500 model optimizations on the test set. As can be seen from Table 3, the simulation accuracy of different types of optimizations is slightly different. Among them, the model with better performance is automatic hyperparameter optimization, which performs well in cross-validation error and calculation prediction accuracy. The maximum prediction accuracy is 87.59%, and the average prediction accuracy can reach 77.49%. The cross-validation error is 0.2235, and the resampling error is 0.2181.

[0119] Table 3 Comparison of decision performance parameters before and after optimization

[0120]

[0121] According to the above scheduling scheme formulation process, the decision tree is used to adjust the normal water year, dry year, and Fengshui year. The scheduling results of the multi-energy complementary system in the upper reaches of the Yellow River are as follows: Figure 8It can be seen that the inflow of the reservoir has the greatest impact on the power generation of the water-wind-light multi-energy complementary system in the upper reaches of the Yellow River, and the wind power station (Qieji) has little impact. 3 / s, after the optimization and regulation of wind and solar complementation through the multi-energy complementary system, the water level of Yangqu Hydropower Station fluctuates within the specified range of 2711m to 2713m; when the inflow flow is 177.50m 3 / s to 182.56m 3 / s, after the optimization and regulation of wind and solar complementation through the multi-energy complementary system, the water level of Yangqu Hydropower Station fluctuates within the specified range of 2712m~2714m; when the inflow flow is less than 177.50m 3 / s, the water level of Yangqu Hydropower Station fluctuates within the prescribed range of 2712m to 2714m; through calculation and analysis, the average proportion of the measured level of the training samples and the calculated level of the decision tree is 77.49%, and the maximum proportion is 87.59%, both reaching more than 75%, with high accuracy, which can be used to dispatch the water-wind-light multi-energy complementary system in the upper reaches of the Yellow River. Therefore, according to the inflow of the reservoir, the output of wind and solar energy is balanced, and the comprehensive dispatching rules for hydropower stations that take into account the maximum stability of the multi-energy complementary system and the minimum change in river runoff are formulated. This can predict the reservoir level of the reservoir, help the multi-energy complementary system to achieve efficient task allocation and resource dispatch, and improve overall performance and efficiency.

[0122] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for extracting optimal dispatching rules for a multi-energy complementary system that considers both new energy consumption and river ecological protection, characterized by The following steps are involved: Step 1: Establish a multi-energy complementary model, with the goal of ensuring the output stability of the multi-energy complementary system and maximizing the power generation, as well as reducing the impact of the operation and scheduling of the multi-energy complementary system on the river runoff situation, to construct a multi-objective optimization scheduling model for the multi-energy complementary system; Step 2: Using a multi-objective optimization algorithm to comprehensively consider multiple objectives and find the optimal solution set that balances various indicators, thereby improving the overall performance of the multi-energy complementary system; the multiple objectives include rationality, environmental impact and system stability; Step 3: By stratifying the optimal solution structure of the multi-objective optimization algorithm, analyzing the main influencing factors, selecting the significant characteristics of wind, water and photovoltaic power generation, and determining them as mandatory input factors based on the close relationship between the significant characteristics of wind, water and photovoltaic power generation and the current storage period and decision-making; Step 4: By building an appropriate decision tree model, quickly predict and make decisions, classify and evaluate different renewable energy power generation plans; solve classification and regression problems to accurately select the best power generation plan and improve the efficiency of renewable energy utilization.

2. The method for extracting optimal dispatching rules for a multi-energy complementary system that simultaneously considers new energy consumption and river ecological protection according to claim 1 is characterized in that: In step 1, the following method is used to construct a multi-objective optimization scheduling model of the multi-energy complementary system to obtain reservoir characteristic data of the new energy and multi-energy complementary system, wherein the reservoir characteristic data includes the optimal scheduling solution, runoff situation and water level: Step 1.1, construct a multi-energy complementary system with the maximum power generation objective function. The maximum power generation objective is to ensure that the output of the reservoir group after absorbing wind and solar energy is the maximum within the dispatch period T; it is expressed by the total power generation P, and the calculation formula is as follows: Where: P max is the maximum amount of total power generation, kw; is the average output during the period, kw; Δt is the time period length, 24h; P i,t , P j,t , P k,t are the power generation of the i-th wind power station, the j-th hydropower station and the k-th photovoltaic power station in the t-th period, kw; T is the total number of time periods in the scheduling period; Step 1.2, construct a multi-energy complementary system with the highest objective function of total output stability. The output stability of the multi-energy complementary system is measured by the coefficient of variation of the power generation output. The smaller the coefficient of variation, the more stable the output. The calculation formula is as follows: Where: CV min 、std min (P) and They are the minimum coefficient of variation, minimum standard deviation and average value of the total output of the multi-energy complementary system; Step 1.3, construct the objective function of the minimum change of river runoff in the multi-energy complementary system. The minimum change of river runoff is the difference between the downstream runoff Q of the hydropower station during the fish spawning period from April to June and the natural runoff situation q t The difference between them is as small as possible, and the calculation formula is as follows: Where: Q is the natural runoff situation of downstream runoff, m / s, q is the downstream runoff, m / s; a and b are the end time and the start time of the spawning period respectively.

3. The method for extracting optimal dispatching rules for a multi-energy complementary system that simultaneously considers new energy consumption and river ecological protection according to claim 1 is characterized in that: The sub-steps of step 2 are as follows: Step 2.1, generate a new population: First, randomly generate the parent group N; Then, the parent generation is operated through selection, crossover and mutation to generate a child population N; the father and the child are combined into 2N to form a new parent generation; then the population is added to the next generation of the child generation through fast non-dominated sorting and elite strategy selection to form a new parent generation D; when the number of new parent generations D is greater than N, T selects N individuals and generates a new population; Step 2.2: The number of reference points is mainly related to the number of objective functions, the location of the reference points, and the number of shares each objective is divided into. The calculation formula is as follows: Where: R is the number of reference points; U is the number of objective functions; T is the number of parts for each objective; Step 2.3, for the processing of individuals in the population, first perform a translation operation on the population so that the ideal point becomes the origin. The calculation formula is as follows: Where: f'i(x) and fi(x) are the values ​​of the individuals of the i-th target dimension in the U target functions in the population after and before the translation operation, respectively; To calculate the individual minimum value of each target dimension in the U target functions in the population; Secondly, calculate the extreme points of each objective function. The calculation formula is as follows: In the formula: AFF(x i ,σ) is the extreme point of the i-th objective function, σ is the unit direction vector as the coordinate axis; when σ is 0, take 10 -6 ; By forming a straight line between the extreme points of each coordinate axis and its ideal point, a U-dimensional linear hyperplane is established, and the intercept αi is obtained by intersecting with the focus of the coordinate axis; Using the general equation of the linear hyperplane, substitute the function value to find the intercept, and then normalize the intercept; Finally, the population individuals are normalized, and the calculation formula is as follows: Step 2.4, output the configuration plan, the reference line takes the line connecting the origin and the reference point, calculates the distance from each individual in the population to each reference line, and establishes the relationship between the individual and the reference line: through the reference point elite strategy, select the offspring individual with a scale of N as the reference point to join the next generation.

4. The method for extracting optimal dispatching rules for a multi-energy complementary system that simultaneously considers new energy consumption and river ecological protection according to claim 1 is characterized in that: The mandatory input factors include the number of cycles, the reservoir head water level, the reservoir inflow and the inflow during the current period.

5. The method for extracting optimal dispatching rules for a multi-energy complementary system that simultaneously considers new energy consumption and river ecological protection according to claim 1 is characterized in that: In step 4, the following methods are used to build an appropriate decision tree model for rapid prediction and decision-making to classify and evaluate different renewable energy generation options: Step 4.1, divide the training data set according to the input variables and output variables; The calculation formula is as follows: <h2 style=";text-align:left;direction:ltr">E = {(A1,B1),(A2,B2),...,(A<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> B<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> ); Where: E is the partitioned training data set; A is the input variable; B is the continuous output variable; Step 4.2, by checking each eigenvector and the corresponding eigenvalue one by one; The two regions are divided and defined by iterating each feature variable and its corresponding feature value in turn; the input space is divided into N subspaces, each of which has a portion of sample data and corresponding output values; the solution of the model is generated; the calculation formula is as follows: Where: k is the current segmentation variable; t is the segmentation feature value corresponding to the k segmentation variable; δ n For each subspace, there is a part of sample data and the corresponding output value; a1, a2, ..., a N subspaces; Step 4.3, use the square error minimization criterion to determine the optimal split point and prediction value, and compare the loss function error; according to the least squares principle, determine the optimal value of the mean of all outputs in the subspace; the calculation formula is as follows: Where: S is the square error; y i Subspace a N All output values ​​on; f(x i ) Subspace a N Solutions of all models on ; is δ n The optimal value of y i Subspace a N All output values ​​on ; Step 4.4, select the best segmentation attribute and value to divide the input space, traverse all input features and values, find the best split point, and divide the space into two sub-regions; if the sub-region cannot be divided any further, obtain the best output value; The calculation formula is as follows: Where: e and d are the optimal segmentation attribute and attribute value respectively; According to steps 4.1-4.4, as long as one region can still be divided, continue to repeat the above process until the entire decision tree is no longer divisible; Step 4.5, predict the importance of multi-energy complementary scheduling rules extraction, automatically optimize hyperparameters, leaf node number, pruning depth parameter accuracy evaluation, and obtain the scheduling results of the multi-energy complementary system in the upper reaches of the Yellow River.

6. The method for extracting optimal dispatching rules for a multi-energy complementary system that simultaneously considers new energy consumption and river ecological protection according to claim 5 is characterized in that: The sub-steps of step 4.5 are: Step 4.5.1, evaluate the importance of scheduling rule prediction: based on the optimization scheduling results of a typical year, use the reservoir inflow and wind and solar output as input prediction factors to predict the reservoir water level; The total number of days of the typical year optimization scheduling results is obtained, in which the training set and the test set are divided. After the model is run multiple times, the importance of the average wind and solar power generation and reservoir flow of the multi-energy complementary model as prediction factors is obtained; Step 4.5.2, evaluate the hyperparameters of automatic optimization of scheduling rules; Step 4.5.3, evaluate the number of leaf nodes of the scheduling rule; Step 4.5.4, evaluate the scheduling rule pruning depth parameters; Step 4.5.5, conduct dispatch rule balancing accuracy evaluation and decision analysis.

7. A device for extracting optimized dispatching rules for a multi-energy complementary system, characterized in that: The program instructions executed by the device are a method for extracting optimal scheduling rules for a multi-energy complementary system that simultaneously considers new energy consumption and river ecological protection according to any one of claims 1 to 6.

Images