Hydropower production simulation data generation method and system based on overseas new energy consumption

Abstract

The invention discloses a hydroelectric production simulation data generation method and system based on overseas new energy consumption, and the method comprises the steps: 1, collecting the weather, load and hydroelectric output coefficient of a certain place, obtaining a data set A, training and testing a plurality of models, and obtaining a selected model; 2, collecting meteorological data of an overseas prediction place to obtain a data set B except hydropower output coefficient characteristics, and predicting a hydropower output coefficient in the data set B by using a selected model; 3, constructing a data set C; 4, data identical to the data set C is removed from the data set A to obtain an alternative data set, and then a specified number of data is selected from the alternative data set to construct a data set D; and 5, performing secondary training on the selected model in the step 1 by using the data set C, predicting the data set D by using the selected model after the secondary training, calculating a prediction error, and then determining whether the prediction error is correct or not through a nonparametric hypothesis testing method. The method provides decision support for development and investment of an overseas new energy market.

Description

Technical Field

[0001] The present invention relates to the technical field of data generation, and in particular to a method and system for generating hydropower production simulation data based on overseas new energy consumption. Background Art

[0002] If domestic companies want to develop new energy projects overseas, they must analyze the country's absorption capacity based on the existing power system and planned power system. The data of each power source and load at 8760 o'clock throughout the year will be the key data for calculating the absorption level and power abundance. Among them, the data generation of wind power is highly correlated with wind speed, and photovoltaic is highly correlated with radiation. However, there are many influencing factors for the generation of hydropower data, and the uncertainty is relatively large.

[0003] Domestic project developers who go abroad to develop projects need to have a rough idea of ​​the country's absorption capacity, and then calculate the installed capacity of the country's developable renewable energy. However, in many cases, the degree of disclosure of foreign power system data is insufficient, so it is impossible to conduct absorption level analysis. Summary of the invention

[0004] The present invention provides a method and system for generating hydropower production simulation data based on overseas renewable energy consumption, so as to solve the technical problems mentioned in the background technology.

[0005] To achieve the above object, the technical solution of the present invention is achieved as follows: The present invention provides a method for generating hydropower production simulation data based on overseas new energy consumption, comprising the following steps: S1. Collect the weather, load and hydropower output coefficient of a certain place to obtain a data set A, then divide the data set A into two parts, and name them as a training set and a test set respectively, use the training set and the test set to train and test the selected multiple models respectively, obtain multiple test results, evaluate the multiple test results, and select the model with the smallest prediction error as the selected model; S2. Calculate the marginal contribution of each feature in data set A according to the predicted value of the selected model, and then calculate the Shapley value of each feature in data set A according to the marginal contribution of each feature; collect meteorological data of overseas prediction sites, obtain data set B except the hydropower output coefficient feature, and calculate the Shapley value of each feature in data set B; and use the selected model to predict the hydropower output coefficient in data set B, and calculate the distribution difference value of each feature in data set B and data set A; S3, selecting a set amount of data from data set A, and constructing data set C based on the Shapley value of each feature in data set A and data set B, and the distribution difference value of each feature in data set A and data set B; S4, removing the data identical to the data set C from the data set A to obtain an alternative data set, then selecting a specified number of data from the alternative data set, and constructing the data set D based on the distribution difference value of each feature in the data set A and the Shapley value of each feature in the data set A; S5. Use the data set C to perform secondary training on the selected model in S1 to obtain the selected model after secondary training; use the selected model after secondary training to calculate the prediction of data set D and calculate the prediction error, and then confirm whether the prediction error is correct through the non-parametric hypothesis testing method, so as to quantitatively characterize the prediction accuracy of data set A for predicting data set B according to the recursive relationship of distance.

[0006] Furthermore, the S1 specifically includes the following steps: S11, collecting the weather, load and hydropower output coefficient of a certain place to obtain a data set A; S12, then divide the data set A into a training set and a test set according to a set ratio; S13, on the basis of considering working days and holidays, using the training set to train the selected multiple models, the selected model is a linear regression model or a nonlinear regression model, and after the training, using the test set to test, to obtain multiple test results; S14. Evaluate multiple test results based on the root mean square error (RMSE) and the mean absolute percentage error (MAPE), and then select the model with the smallest prediction error as the selected model. ; Among them, the calculation formulas of root mean square error RMSE and mean absolute percentage error MAPE are as follows: ; ; in, is the sample size, It is Observations, It is predicted value.

[0007] Furthermore, the S2 specifically includes the following steps: S21. Input dataset A into the selected model Among them Indicates input to the selected model A set of independent variable data in the data set A; and , Represents a set of data containing k features, and the k features are , , ,..., ; S22, for Group Data , calculate the marginal contribution , S represents a subset of the data set A; where the marginal contribution The details are as follows: ; in, Indicates adding features to subset S The selected model The predicted value of Indicates that the model is selected when only features in subset S are used The predicted value of S23. Obtain the Shapley value of each feature in data set A by taking the weighted average of the marginal contribution of each feature in data set A in all possible subsets. ; S24, collecting meteorological data of overseas prediction sites, obtaining other data of data set B except the hydropower output coefficient feature, and using the selected model to predict the hydropower output coefficient in data set B, obtaining data set B containing all features; S25, using the same method as S21 to S23, obtain the Shapley value of each feature in the data set B; S26, binning data set A and data set B by using equal interval binning or equal frequency binning; then calculating the proportion of each bin in data set A and data set B, so as to obtain the proportion of each bin in data set A And the proportion in dataset B ; S27. Calculate the distribution difference value of each feature in data set A and data set B based on the proportion of each box in different data sets obtained in the previous calculation.

[0008] Furthermore, the Shapley value of the i-th feature in S23 is The calculation formula is as follows: ; in, It means S is N except the Group Data The subset after that, N represents the set of all features, and ; Representation Subset The number of features, yes The total number of permutations of elements.

[0009] Furthermore, the calculation formula of the distribution difference value of each feature in S26 is as follows: ; Among them, PSI is the distribution difference value of the feature.

[0010] Furthermore, the S3 specifically includes the following steps: Selecting all feature data input to the selected model from the data set A to obtain a plurality of selected data; S32, constructing constraint condition 1 and objective function 1; constraint condition 1 includes two, one of which is that the data selected in S31 are all data in data set A, and the other is that the data selected in S31 must be greater than a set number; Then, the distance between dataset B and dataset C is solved by using the Shapley value of each feature in dataset A and dataset B and the distribution difference value of each feature in dataset A and dataset B; S33. Use a heuristic algorithm to solve constraint condition 1 and objective function 1, find multiple data corresponding to approximate optimal solutions, and construct data set C.

[0011] Furthermore, the objective function 1 in S32 is specifically as follows: ; in, Represents the first data, Indicates the first data, Represents the distance between dataset B and dataset C.

[0012] Furthermore, the S4 specifically includes the following steps: S41, removing the data identical to the data set C from the data set A to obtain a candidate data set, and then selecting a specified number of multiple data from the candidate data set to obtain multiple selected data; S42, constructing constraint condition 2 and objective function 2, where constraint condition 2 includes two conditions, one of which is that the number of selected data must be greater than a specified number; the other is that the distance between a data set composed of multiple selected data and data set C must be greater than or equal to the distance between data set C and data set B; the data set composed of multiple selected data is the pre-constructed data set D; Then use the Shapley value of each feature of dataset A And the distribution difference value of each feature in dataset A solves the distance between dataset C and pre-constructed dataset D; The specific objective function 2 is as follows: ; ; in, Represents the first data, Represents the distance between dataset C and pre-constructed dataset D; S43. Use a heuristic algorithm to solve constraint condition 2 and objective function 2, find multiple data corresponding to approximate optimal solutions, and construct data set D.

[0013] Furthermore, the S5 specifically includes the following steps: S51, using the data set C to train the selected model in S1 to obtain a trained selected model; S52, then under the constraint that the distance between data set C and data set D is greater than or equal to the distance between data set C and data set B, the prediction error of data set D is calculated using the trained selected model; S53, performing a non-parametric hypothesis test on the prediction error in the data set D, and confirming whether the prediction error in S52 is correct through the non-parametric hypothesis test, so as to quantitatively characterize the prediction accuracy of the data set A for predicting the data set B according to the recursive relationship of the distance; The recursive relationship of the distance is the second constraint in S42 and the constraint in S52.

[0014] Another aspect of the present invention provides a hydropower production simulation data generation system, including a computer device, wherein the computer device is programmed or configured to execute the above hydropower production simulation data generation method.

[0015] Beneficial effects of the present invention: The present invention discloses a method for generating hydropower production simulation data based on overseas renewable energy consumption. In view of the fact that overseas power data is not disclosed to a sufficient degree and data acquisition is difficult, the present invention can provide project developers with basic data for consumption production simulation and provide an accuracy evaluation of the basic data, thereby providing decision-making support for the development and investment in overseas renewable energy markets. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The present invention is a flowchart of the method for generating hydropower production simulation data. DETAILED DESCRIPTION

[0017] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Preferred embodiments of the present invention are provided in the drawings. However, the present invention can be implemented in many other different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present invention more thorough and comprehensive.

[0018] Reference Figure 1 The embodiment of the present application provides a method for generating hydropower production simulation data based on overseas new energy consumption, comprising the following steps: S1. Collect the weather, load and hydropower output coefficient of a certain place to obtain a data set A, then divide the data set A into two parts, and name them as a training set and a test set respectively, use the training set and the test set to train and test the selected multiple models respectively, obtain multiple test results, evaluate the multiple test results, and select the model with the smallest prediction error as the selected model; Preferably, data set A is the meteorological data of a province in central China, including rainfall, runoff, evaporation, 2m air temperature, dew point temperature, load and hydropower output 8760 time point data; S2. Calculate the marginal contribution of each feature in data set A according to the predicted value of the selected model, and then calculate the Shapley value of each feature in data set A according to the marginal contribution of each feature; collect meteorological data of overseas prediction sites, obtain data set B except the hydropower output coefficient feature, and calculate the Shapley value of each feature in data set B; and use the selected model to predict the hydropower output coefficient in data set B, and calculate the distribution difference value of each feature in data set B and data set A; Preferably, data set B is meteorological data of the overseas prediction location, including rainfall, runoff, evaporation, 2m air temperature, dew point temperature and load 8760 time point data and predicted hydropower output coefficient; S3, selecting a set amount of data from data set A, and constructing data set C based on the Shapley value of each feature in data set A and data set B, and the distribution difference value of each feature in data set A and data set B; S4, removing the data identical to the data set C from the data set A to obtain an alternative data set, then selecting a specified number of data from the alternative data set, and constructing the data set D based on the distribution difference value of each feature in the data set A and the Shapley value of each feature in the data set A; S5. Use the data set C to perform secondary training on the selected model in S1 to obtain the selected model after secondary training; use the selected model after secondary training to calculate the prediction of data set D and calculate the prediction error, and then confirm whether the prediction error is correct through the non-parametric hypothesis testing method, so as to quantitatively characterize the prediction accuracy of data set A for predicting data set B according to the recursive relationship of distance.

[0019] The present invention discloses a method for generating hydropower production simulation data based on overseas renewable energy consumption. In view of the fact that overseas power data is not disclosed to a sufficient degree and data acquisition is difficult, the present invention can provide project developers with basic data for consumption production simulation and provide an accuracy evaluation of the basic data, thereby providing decision-making support for the development and investment in overseas renewable energy markets.

[0020] In some embodiments, the S1 specifically includes the following steps: S11, collecting the weather, load and hydropower output coefficient of a certain place to obtain a data set A; S12, then divide the data set A into a training set and a test set according to a set ratio; S13. Based on the consideration of working days and holidays, the selected multiple models are trained using the training set. The selected models are linear regression models or nonlinear regression models. The linear regression models include linear neural networks or others. The nonlinear regression models include decision trees, random forests, nonlinear neural networks or others. During the training process, based on the characteristics of temperature, precipitation, evaporation, runoff, humidity and load, the previous moment, the average value of the previous three moments, the average value of the previous six moments and the average value of the previous twelve moments of each feature, the value of the moment on the same day in the past, the average value of the moment on the same day in the past, the value of the moment 7 days ago and the average value of the moment 7 days ago are used as extended features for research; After training, the test set is used for testing to obtain multiple test results; S14. Evaluate multiple test results based on the root mean square error (RMSE) and the mean absolute percentage error (MAPE), and then select the model with the smallest prediction error as the selected model. ; Among them, the calculation formulas of root mean square error RMSE and mean absolute percentage error MAPE are as follows: ; ; in, is the sample size, It is Observations, It is predicted value.

[0021] In some embodiments, S2 specifically includes the following steps: S21. Input dataset A into the selected model Among them Indicates input to the selected model A set of independent variable data in the data set A; and , Represents a set of data containing k features, and the k features are , , ,..., ; S22, for Group Data , calculate the marginal contribution , S represents a subset of the data set A; where the marginal contribution The details are as follows: ; in, Indicates adding features to subset S The selected model The predicted value of Indicates that the model is selected when only features in subset S are used The predicted value of S23. Obtain the Shapley value of each feature in data set A by taking a weighted average of the marginal contribution of each feature in data set A in all possible subsets. ; S24, collecting meteorological data of overseas prediction sites, obtaining other data of data set B except the hydropower output coefficient feature, and using the selected model to predict the hydropower output coefficient in data set B, obtaining data set B containing all features; S25, using the same method as S21 to S23, obtain the Shapley value of each feature in the data set B; S26, binning data set A and data set B by using equal interval binning or equal frequency binning; then calculating the proportion of each bin in data set A and data set B, so as to obtain the proportion of each bin in data set A And the proportion in dataset B ; in, ; is the first The number of samples in a box; N A is the total number of samples in data set A; and, , is the first The number of samples in a box; is the total number of samples in data set B; S27, according to the proportion of each box in different data sets obtained by the previous calculation, calculate the distribution difference value of each feature in data set A and data set B; the distribution difference value can be a population stability index (PSI value), Kulibec divergence (KL divergence) and bulldozing distance, etc.; The distribution difference value of the feature needs to specify the expected data set (i.e., data set B) as a benchmark, because the calculation of the shap value is based on the expected data set, and most calculations of distribution difference values ​​are asymmetric.

[0022] In some embodiments, the Shapley value of each feature in the data set A in S23 is The calculation formula is as follows: ; in, express is N except the Group Data The subset after that, N represents the set of all features, and ; Representation Subset The number of features, yes The total number of permutations of elements.

[0023] In some embodiments, the calculation formula of the distribution difference value of each feature in S26 is as follows: ; Among them, PSI is the distribution difference value of the feature.

[0024] In some embodiments, S3 specifically includes the following steps: S31, selecting from data set A all feature data input into the selected model, to obtain a plurality of selected data; S32, construct constraint condition 1 and objective function 1; constraint condition 1 includes two, one of which is that the data selected in S31 are all data in data set A, and the other is that the data selected in S31 needs to be greater than a set number, that is, the data selected in S31 needs to be large enough; Then, the distance between dataset B and dataset C is solved by using the Shapley value of each feature in dataset A and dataset B and the distribution difference value of each feature in dataset A and dataset B; S33, using a heuristic algorithm to solve constraint condition 1 and objective function 1, optimizing the solving speed, finding multiple data corresponding to the approximate optimal solution, and constructing a data set C. The heuristic algorithm here includes but is not limited to simulated annealing, Pareto, and particle swarm optimization.

[0025] In some embodiments, the objective function 1 in S32 is specifically as follows: ; in, Represents the first data, Indicates the first data, Represents the distance between dataset B and dataset C.

[0026] In some embodiments, the S4 specifically includes the following steps: S41, removing the data identical to the data set C from the data set A to obtain a candidate data set, and then selecting a specified number of multiple data from the candidate data set to obtain multiple selected data; S42, constructing constraint condition 2 and objective function 2, constraint condition 2 includes two, one of which is that the number of selected data needs to be greater than the specified number; that is, the number of selected data needs to be sufficient; the other is that the distance between the data set composed of multiple selected data and data set C must be greater than or equal to the distance between data set C and data set B; the data set composed of multiple selected data is the pre-constructed data set D; Then use the Shapley value of each feature of dataset A And the distribution difference value of each feature in dataset A solves the distance between dataset C and pre-constructed dataset D; The specific objective function 2 is as follows: ; ; in, Represents the first data, Represents the distance between dataset C and pre-constructed dataset D; S43, using a heuristic algorithm to solve the second constraint condition and the second objective function, optimizing the solving speed, finding multiple data corresponding to the approximate optimal solution, and constructing a data set D. The heuristic algorithm here includes but is not limited to simulated annealing, Pareto, particle swarm optimization, and genetic algorithm.

[0027] In some embodiments, S5 specifically includes the following steps: S51, using the data set C to train the selected model in S1 to obtain a trained selected model; S52, then under the constraint that the distance between data set C and data set D is greater than or equal to the distance between data set C and data set B, the prediction error of data set D is calculated using the trained selected model; S53, performing a non-parametric hypothesis test on the prediction error in the data set D, and confirming whether the prediction error in S52 is correct through the non-parametric hypothesis test, so as to quantitatively characterize the prediction accuracy of the data set A for predicting the data set B according to the recursive relationship of the distance; The recursive relationship of the distance is the second constraint in S42 and the constraint in S52.

[0028] The specific process of nonparametric hypothesis testing is as follows: First, assume that the results obtained by S52 are correct and reliable, assume the significance level (α=0.05), and then calculate the P value (probability value). The P value is calculated in the form of a one-sided test: One-sided test : ; ; One-sided test : ; ; in, is the population standard deviation, is the sample size, To test the sample mean, is the overall mean, is the assumed population mean.

[0029] The calculated P value is compared with the pre-set significance level to determine whether to accept the null hypothesis and finally draw a conclusion.

[0030] For ease of understanding, the following examples are used to illustrate: Step S1, in this example, data set A uses public data from the global climate atmospheric reanalysis data set produced by the European Centre for Medium-Range Weather Forecasts (ECMWF), such as precipitation, evaporation, runoff, 2m maximum temperature, dew point temperature, 2m temperature, etc. In addition to the above public data, some derived features need to be calculated, such as firstly, the water vapor pressure and saturated water vapor pressure can be calculated based on the above dew point temperature and 2m temperature using the Antoine equation or the Wexler formula, thereby obtaining the relative humidity; secondly, combined with the load generation data, the above features and variables such as the known hydropower output coefficient of a province are used as basic features to simulate the generation of hydropower.

[0031] Relative humidity is calculated based on water vapor pressure and saturated water vapor pressure. The calculation formula is as follows: Water vapor pressure (e): ; Saturated water vapor pressure ( ): ; Relative humidity (RH) (%): ; in, is the saturated water vapor pressure at 0°C, usually 6.1078; T is the temperature at 2m; is the dew point temperature; the values ​​of a and b are related to the temperature: when or When the water surface: ; when or When the ice surface: .

[0032] The dependent variable and independent variable data of the known data set A are shown in the following table.

[0033] Table 1: Simulation data of dataset A;

[0034] In the process of model training and testing, the output coefficient is selected as the dependent variable. All feature data are divided into training set and test set in a ratio of 0.7:0.3, and decision tree, random forest, neural network, Lasso and other models are used for training.

[0035] By predicting the above model, we can get the model training and test results for a certain year. The selected model is Lasso, and the model Lasso is the selected model. The model evaluation value is calculated, and the results are shown in the following table: Table 2: Evaluation indicators of selected model training and test models in dataset A;

[0036] From the above table, we can see that the evaluation errors of the training and test models calculated by the selected model are low, which meets the prediction accuracy requirements. The above model can be used to migrate the prediction data set.

[0037] Step S2, calculating the feature contribution, distribution difference value and hydropower output coefficient of the prediction data set B; The Shapley value additive explanatory model (SHAP model) is used to calculate the Shapley value (Shapley value) of each feature in the expected data set of the selected model obtained in step S1. The Shapley value can quantify the contribution of each feature to the model prediction.

[0038] Specifically, for each feature and each subset, calculate the marginal contribution The Shapley value of the feature It is obtained by taking a weighted average of its marginal contributions in all possible subsets. The weights are determined based on the size of the subsets and the number of permutations.

[0039] ; in, , Representation Subset The number of elements of yes The total number of permutations of elements.

[0040] Through calculation, the importance of each feature is ranked as temperature, precipitation, evaporation, load, runoff, and humidity. The Shapley contribution weights are 'precipitation': 0.0089, 'maximum temperature': 0.0626, 'load': 0.000195, 'evaporation': 0.0257, 'runoff': 0.01288, 'humidity': 0.0369.

[0041] Using the equal frequency binning method, each box of each feature contains 200 data. The calculated distribution difference values ​​(PSI) of each feature of data set A and data set B are precipitation': 13.86, maximum temperature': 1.97, load': 633, evaporation': 4.8, runoff': 9.58, and humidity': 3.34.

[0042] The S1 selected model is used to predict the hydropower output coefficient in data set B.

[0043] The distance between data sets is the sum of the distribution difference value of each feature multiplied by the Shapley value.

[0044] Step S3, constructing a data set C;

[0045] The style of data set B is shown in the following table, which only contains independent variables and lacks dependent variables that need to be predicted.

[0046] Table 3: Simulation data of dataset B;

[0047] The simulated annealing algorithm is used to optimize the solution speed and find the data set C corresponding to the approximate optimal solution. The data in the above data set can be considered as a collection of multiple particles.

[0048] ; in, ; ; ; Among them, Fitness(X) is the fitness value of each particle after each expansion optimization. The objective function value of the distance-optimized target model (i.e., the selected model), is the penalty coefficient, T is the current annealing temperature, is the penalty function, is a piecewise function, is the conversion function of multiple constraint functions, express Penalty exponent when the threshold is exceeded.

[0049] The simulated annealing criterion is used to set the extended optimization end condition, and the extended optimization end condition is specifically: ; in, For the The probability that a particle accepts the value after the nth iteration is is the current annealing temperature, is the value after the nth iteration, is the value after the n-1th iteration, is the fitness value corresponding to the n-1th round of iteration, is the fitness value corresponding to the nth round of iteration.

[0050] The calculation result is that 1035 data are selected from data set A to form data set C. The obtained data set C is as follows: Table 4: Dataset C;

[0051] Step S4, constructing a data set D; According to the optimization goal of minimizing the distance between data set C and data set D, the simulated annealing algorithm is used to obtain data set D. The optimization solution method is the same as that of data set C. According to the calculation results, a total of 100 data are selected, as shown in the following table.

[0052] Table 5: Dataset D;

[0053] Step S5, model inference error, and prediction of data set B; Step S51: Use the selected model obtained in step 1 to train the data set C, and then predict the data set D. Under the constraint that the distance between data sets C and D is greater than or equal to the distance between data sets C and B, calculate the prediction error of data set D. The RMSE and MAPE are 11.93% and 32.9%, respectively. Since the distance between CD is greater than the distance between CB, it can be obtained that the error of C predicting B is lower than the error of C predicting D, that is, the error of A predicting B is lower than the model evaluation value of C predicting D.

[0054] Step S52: Perform a non-parametric hypothesis test on the prediction error of data set D, assuming that the conclusion obtained in point 1 is correct and credible. Without presetting the distribution type of the features in data sets B and C, it can be calculated that the p-values ​​are all greater than 0.1. Therefore, whether the significance level is assumed to be 0.05 or 0.01, the p-value is greater than the significance level. Therefore, it can be obtained that the assumption of the non-parametric hypothesis test is credible, that is, the regression model trained and optimized based on data set A is trained on data set C, and the error of the prediction of data set B is close to the error of data set D.

[0055] Another aspect of the present invention provides a hydropower production simulation data generation system, including a computer device, wherein the computer device is programmed or configured to execute the above hydropower production simulation data generation method.

[0056] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered within the protection scope of the present invention. In addition, the technical solutions between the various embodiments of the present invention can be combined with each other, but it must be based on the ability of ordinary technicians in the field to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for generating hydropower production simulation data based on overseas renewable energy consumption, characterized in that: The steps include: S1. Collect the weather, load and hydropower output coefficient of a certain place to obtain a data set A, then divide the data set A into two parts, and name them as a training set and a test set respectively, use the training set and the test set to train and test the selected multiple models respectively, obtain multiple test results, evaluate the multiple test results, and select the model with the smallest prediction error as the selected model; S2. Calculate the marginal contribution of each feature in data set A according to the predicted value of the selected model, and then calculate the Shapley value of each feature in data set A according to the marginal contribution of each feature; collect meteorological data of overseas prediction sites, obtain data set B except the hydropower output coefficient feature, and calculate the Shapley value of each feature in data set B; and use the selected model to predict the hydropower output coefficient in data set B, and calculate the distribution difference value of each feature in data set B and data set A; S3, selecting a set amount of data from data set A, and constructing data set C based on the Shapley value of each feature in data set A and data set B, and the distribution difference value of each feature in data set A and data set B; S4, removing the data identical to the data set C from the data set A to obtain an alternative data set, then selecting a specified number of data from the alternative data set, and constructing the data set D based on the distribution difference value of each feature in the data set A and the Shapley value of each feature in the data set A; S5. Use the data set C to perform secondary training on the selected model in S1 to obtain the selected model after secondary training; use the selected model after secondary training to calculate the prediction of data set D and calculate the prediction error, and then confirm whether the prediction error is correct through the non-parametric hypothesis testing method, so as to quantitatively characterize the prediction accuracy of data set A for predicting data set B according to the recursive relationship of distance.

2. The method for generating hydropower production simulation data according to claim 1, characterized in that: The S1 specifically includes the following steps: S11, collecting the weather, load and hydropower output coefficient of a certain place to obtain a data set A; S12, then divide the data set A into a training set and a test set according to a set ratio; S13. Based on the consideration of working days and holidays, the selected multiple models are trained using the training set, where the selected model is a linear regression model or a nonlinear regression model; After training, the test set is used for testing to obtain multiple test results; S14. Evaluate multiple test results based on the root mean square error (RMSE) and the mean absolute percentage error (MAPE), and then select the model with the smallest prediction error as the selected model. ; Among them, the calculation formulas of root mean square error RMSE and mean absolute percentage error MAPE are as follows: ； ； in, is the sample size, It is Observations, It is predicted value.

3. The method for generating hydropower production simulation data according to claim 1, characterized in that: The S2 specifically includes the following steps: S21. Input dataset A into the selected model Among them Indicates input to the selected model A set of independent variable data in the data set A; and , Represents a set of data containing k features, and the k features are , , ,..., ; S22, for Group Data , calculate the marginal contribution , S represents a subset of the data set A; where the marginal contribution The details are as follows: ； in, Indicates adding features to subset S The selected model The predicted value of Indicates that the model is selected when only features in subset S are used The predicted value of S23. Obtain the Shapley value of each feature in data set A by taking the weighted average of the marginal contribution of each feature in data set A in all possible subsets. ; S24, collecting meteorological data of overseas prediction sites, obtaining other data of data set B except the hydropower output coefficient feature, and using the selected model to predict the hydropower output coefficient in data set B, obtaining data set B containing all features; S25, using the same method as S21 to S23, obtain the Shapley value of each feature in the data set B; S26, binning data set A and data set B by using equal interval binning or equal frequency binning; then calculating the proportion of each bin in data set A and data set B, so as to obtain the proportion of each bin in data set A And the proportion in dataset B ; S27. Calculate the distribution difference value of each feature in data set A and data set B based on the proportion of each box in different data sets obtained in the previous calculation.

4. The method for generating hydropower production simulation data according to claim 3, characterized in that: The Shapley value of each feature in the data set A in S23 The calculation formula is as follows: ； express is N except the Group Data The subset after that, N represents the set of all features, and ; Representation Subset The number of features, yes The total number of permutations of elements.

5. The method for generating hydropower production simulation data according to claim 4, characterized in that: The calculation formula of the distribution difference value of each feature in S26 is as follows: ; Among them, PSI is the distribution difference value of the feature.

6. The method for generating hydropower production simulation data according to claim 5, characterized in that: The S3 specifically includes the following steps: S31, selecting from data set A all feature data input into the selected model, to obtain a plurality of selected data; S32, constructing constraint condition 1 and objective function 1; constraint condition 1 includes two, one of which is that the data selected in S31 are all data in data set A, and the other is that the data selected in S31 must be greater than a set number; Then, the distance between dataset B and dataset C is solved by using the Shapley value of each feature in dataset A and dataset B and the distribution difference value of each feature in dataset A and dataset B; S33. Use a heuristic algorithm to solve constraint condition 1 and objective function 1, find multiple data corresponding to approximate optimal solutions, and construct data set C.

7. The method for generating hydropower production simulation data according to claim 6, characterized in that: The objective function 1 in S32 is specifically as follows: ； in, Represents the first data, Indicates the first data, Represents the distance between dataset B and dataset C.

8. The method for generating hydropower production simulation data according to claim 7, characterized in that: The S4 specifically includes the following steps: S41, removing the data identical to the data set C from the data set A to obtain a candidate data set, and then selecting a specified number of multiple data from the candidate data set to obtain multiple selected data; S42, constructing constraint condition 2 and objective function 2, where constraint condition 2 includes two conditions, one of which is that the number of selected data must be greater than a specified number; the other is that the distance between a data set composed of multiple selected data and data set C must be greater than or equal to the distance between data set C and data set B; the data set composed of multiple selected data is the pre-constructed data set D; Then use the Shapley value of each feature of dataset A And the distribution difference value of each feature in dataset A solves the distance between dataset C and pre-constructed dataset D; The specific objective function 2 is as follows: ； ； in, Represents the first data, Represents the distance between dataset C and pre-constructed dataset D; S43. Use a heuristic algorithm to solve constraint condition 2 and objective function 2, find multiple data corresponding to approximate optimal solutions, and construct data set D.

9. The method for generating hydropower production simulation data according to claim 1, characterized in that: The S5 specifically includes the following steps: S51, using the data set C to train the selected model in S1 to obtain a trained selected model; S52, then under the constraint that the distance between data set C and data set D is greater than or equal to the distance between data set C and data set B, the prediction error of data set D is calculated using the trained selected model; S53, performing a non-parametric hypothesis test on the prediction error in the data set D, and confirming whether the prediction error in S52 is correct through the non-parametric hypothesis test, so as to quantitatively characterize the prediction accuracy of the data set A for predicting the data set B according to the recursive relationship of the distance; The recursive relationship of the distance is the second constraint in S42 and the constraint in S52.

10. A hydropower production simulation data generation system, comprising a computer device, characterized in that: The computer device is programmed or configured to execute the method for generating hydropower production simulation data as described in any one of claims 1 to 9.

Images