LightGBM-Transform-LSTM-based hydropower cluster output prediction method

Abstract

The invention relates to a LightGBM-Transform-LSTM-based hydropower cluster output prediction method. The method comprises the following steps: collecting historical power generation data and historical meteorological data of a target hydropower cluster, and carrying out the preprocessing of the historical power generation data and historical meteorological data; time delay characteristics are determined through historical power generation data, and basic characteristic data are constructed; constructing an output prediction model which comprises three base models, namely a Light GBM base model, a Transform base model and an LSTM (Long Short Term Memory) base model; respectively constructing three types of input feature data corresponding to the three base models, and training an output prediction model; and performing power generation data prediction of the target hydropower cluster at the future moment through the trained output prediction model.

Description

Technical Field

[0001] The present invention relates to a hydropower cluster output prediction method based on LightGBM-Transformer-LSTM, belonging to the technical field of hydropower output prediction. Background Art

[0002] As an important part of renewable energy, hydropower generation is widely used around the world, especially playing an important role in power system dispatch and energy management. However, hydropower generation has strong seasonality and uncertainty, and is affected by many factors such as precipitation, temperature, and river basin conditions. Therefore, accurate hydropower generation forecasting is of vital importance for the stable operation of the power system, energy optimization dispatching, and unit start-up and shutdown arrangements.

[0003] Existing hydropower generation forecasting methods can be roughly divided into two categories: traditional statistical models and modern machine learning models. Traditional statistical methods, such as the autoregressive integrated moving average model (ARIMA), rely on the regularity of historical data and predict future power generation through time series modeling. However, these methods are weak in dealing with high-frequency noise and nonlinear relationships, and cannot effectively capture the complex dynamic changes in hydropower generation data. For example, under the influence of factors such as precipitation and temperature, hydropower generation may fluctuate violently, and models such as ARIMA often perform poorly in such situations. Summary of the invention

[0004] In order to solve the above-mentioned problems existing in the prior art, the present invention proposes a hydropower cluster output prediction method based on LightGBM-Transformer-LSTM.

[0005] The technical solution of the present invention is as follows:

[0006] On the one hand, the present invention proposes a hydropower cluster output prediction method based on LightGBM-Transformer-LSTM, comprising the following steps:

[0007] Collect historical power generation data and historical meteorological data of the target hydropower cluster and perform preprocessing;

[0008] Through historical power generation data, determine the time delay feature with the greatest correlation between the power generation data at a given moment and the power generation data after the delay; and construct basic feature data based on the pre-processed historical power generation data, historical meteorological data and time delay features;

[0009] Construct an output prediction model, including three base models, namely LightGBM base model, Transformer base model and LSTM base model; based on the basic feature data, construct three types of input feature data corresponding to the three base models respectively, and input the three types of input feature data into the three base models respectively. In the training process of the output prediction model, the prediction results of each base model are fused by the weighted integration method. The output prediction model is iteratively trained according to the loss of the fused prediction results and the real power generation data, and the weights assigned to each base model in the weighted integration process are adjusted;

[0010] The trained output prediction model is used to predict the power generation data of the target hydropower cluster at future times.

[0011] As a preferred implementation, the method for determining the time delay feature with the greatest correlation between the power generation data at a given moment and the power generation data after the delay through the historical power generation data is specifically as follows:

[0012] The Pearson correlation calculation method is used to calculate the Pearson correlation coefficients of the power generation data under different time delays and the power generation data at a given moment, and the time delay with the largest Pearson correlation coefficient is taken as the time delay feature.

[0013] As a preferred implementation, the method of constructing three types of input feature data corresponding to three base models based on the basic feature data is specifically as follows:

[0014] For the LightGBM base model, on the basis of basic feature data, the environmental data related to the hydropower cluster are introduced to form the first input feature data;

[0015] For the Transformer base model, on the basis of the basic feature data, the meteorological series data within the first given time window is introduced to form the second input feature data;

[0016] For the LSTM base model, on the basis of the basic feature data, the power generation sequence data and the meteorological sequence data within the second given time window are introduced to form the third input feature;

[0017] The first given time window is twice as large as the second given time window.

[0018] As a preferred implementation, in the output prediction model, the fused prediction results are specifically:

[0019]

[0020] in, To integrate the prediction results, is the prediction result of the i-th base model, w i is the weight parameter of the ith base model,m is the number of base models;

[0021] The loss function of the output prediction model training process is set as:

[0022]

[0023] in, is the loss function, n is the number of samples, y i represents the true value of the i-th sample, Represents the predicted value of the j-th base model for the i-th sample;

[0024] According to the calculated loss, the partial derivative of the weight of each base model is obtained to obtain the weight gradient of each base model, and the weight of the corresponding base model is updated according to the weight gradient of each base model.

[0025] On the other hand, the present invention also proposes a hydropower cluster output prediction system based on LightGBM-Transformer-LSTM, comprising:

[0026] Data acquisition module, used to collect historical power generation data and historical meteorological data of the target hydropower cluster and perform preprocessing;

[0027] A basic feature construction module is used to determine the time delay feature with the greatest correlation between the power generation data at a given moment and the power generation data after the time delay through the historical power generation data; and to construct basic feature data based on the pre-processed historical power generation data, historical meteorological data and time delay features;

[0028] The integrated model training module is used to construct an output prediction model, including three base models, namely the LightGBM base model, the Transformer base model and the LSTM base model. Based on the basic feature data, three types of input feature data corresponding to the three base models are constructed respectively, and the three types of input feature data are input into the three base models respectively. In the training process of the output prediction model, the prediction results of each base model are fused by the weighted integration method. The output prediction model is iteratively trained according to the loss of the fused prediction results and the real power generation data, and the weights assigned to each base model in the weighted integration process are adjusted.

[0029] The hydropower cluster output prediction module is used to predict the power generation data of the target hydropower cluster at a future moment through the trained output prediction model.

[0030] As a preferred implementation, the method for determining the time delay feature with the greatest correlation between the power generation data at a given moment and the power generation data after the delay through the historical power generation data is specifically as follows:

[0031] The Pearson correlation calculation method is used to calculate the Pearson correlation coefficients of the power generation data under different time delays and the power generation data at a given moment, and the time delay with the largest Pearson correlation coefficient is taken as the time delay feature.

[0032] As a preferred implementation, the method of constructing three types of input feature data corresponding to three base models based on the basic feature data is specifically as follows:

[0033] For the LightGBM base model, on the basis of basic feature data, the environmental data related to the hydropower cluster are introduced to form the first input feature data;

[0034] For the Transformer base model, on the basis of the basic feature data, the meteorological series data within the first given time window is introduced to form the second input feature data;

[0035] For the LSTM base model, on the basis of the basic feature data, the power generation sequence data and the meteorological sequence data within the second given time window are introduced to form the third input feature;

[0036] The first given time window is twice as large as the second given time window.

[0037] As a preferred implementation, in the output prediction model, the fused prediction results are specifically:

[0038]

[0039] in, To integrate the prediction results, is the prediction result of the i-th base model, w i is the weight parameter of the ith base model, m is the number of base models;

[0040] The loss function of the output prediction model training process is set as:

[0041]

[0042] in, is the loss function, n is the number of samples, y i represents the true value of the i-th sample, Represents the predicted value of the j-th base model for the i-th sample;

[0043] According to the calculated loss, the partial derivative of the weight of each base model is obtained to obtain the weight gradient of each base model, and the weight of the corresponding base model is updated according to the weight gradient of each base model.

[0044] On the other hand, the present invention also proposes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for predicting the output of a hydropower cluster based on LightGBM-Transformer-LSTM as described in any embodiment of the present invention is implemented.

[0045] On the other hand, the present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the hydropower cluster output prediction method based on LightGBM-Transformer-LSTM as described in any embodiment of the present invention.

[0046] Additional aspects and advantages of the present invention will be set forth in the following description, and some of them will be apparent from the description, or may be understood by practicing the present invention. In addition, the various aspects and advantages of the present invention may be realized and obtained by the method steps and combinations particularly pointed out in the appended claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a flow chart of a method according to Embodiment 1 of the present invention;

[0048] Figure 2 This is a schematic diagram of the workflow of Embodiment 1 of the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] It should be understood that the step numbers used in this document are only for convenience of description and are not intended to limit the order in which the steps are executed.

[0051] It should be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0052] The terms “include” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0053] The term "and / or" means and includes any and all possible combinations of one or more of the associated listed items.

[0054] Embodiment 1:

[0055] With the rapid development of machine learning technology, more and more nonlinear modeling methods such as support vector machine (SVM), random forest (RF), gradient boosting tree (GBT) are applied to hydropower generation forecasting. These methods can handle complex nonlinear relationships, especially when there are high interactions and complex patterns between data features, and show better modeling capabilities. However, although machine learning models can fit nonlinear data well, there are still some problems: first, many machine learning models rely on a large amount of training data, and their prediction effects may be unstable when the sample size is insufficient or the data is unbalanced; second, machine learning methods are usually "black box" models, lacking the interpretability of data, and it is difficult to provide detailed decision-making basis for power dispatching departments.

[0056] With the rise of deep learning, long short-term memory network (LSTM) as a classic time series modeling method has gradually been applied to hydropower generation forecasting. LSTM can process long-term dependencies in time series data through its memory units, so it has obvious advantages in capturing seasonal changes and periodic fluctuations in hydropower generation. However, LSTM may face computational efficiency issues when processing long time series, and is also easily affected by data noise, resulting in reduced prediction accuracy.

[0057] The Transformer model was first used in the field of natural language processing and has attracted widespread attention for its high efficiency in processing long sequence data. Unlike LSTM, the Transformer captures long-term dependencies through the self-attention mechanism, and exhibits high parallelism and computational efficiency when modeling complex time series. However, although the Transformer has advantages in processing the global dependencies of time series, its ability to model short-term dependencies and local dynamic changes is relatively weak.

[0058] In view of the limitations of the prior art, this embodiment provides a hydropower cluster output prediction method based on LightGBM-Transformer-LSTM, which comprehensively improves the accuracy and robustness of hydropower generation prediction by combining the global modeling capability of LightGBM, the long-term dependency capture capability of Transformer, and the short-term dependency modeling capability of LSTM. By weighted integration of the prediction results of multiple base models, the present invention can simultaneously handle global characteristics, long-term and short-term dependencies, and nonlinear dynamic changes, thereby providing a new solution for the accurate prediction of hydropower generation.

[0059] See Figure 1 and Figure 2 , the method provided in this embodiment comprises the following steps:

[0060] S100: Collect historical power generation data and historical meteorological data of the target hydropower cluster and perform preprocessing.

[0061] In step S100, for the historical power generation data, the missing values ​​in the data caused by sensor failure or recording error are filled by spline interpolation method to ensure data integrity and continuity; the historical meteorological data are obtained through the meteorological data of the European Center for Medium-Term Weather Forecasts (ECMWF), including relative humidity, specific rain water content, specific snow water content and air temperature; the historical power generation data and meteorological data are normalized, and the selected historical power generation data and meteorological data will be normalized to ensure that the input feature data are within the same scale range, avoiding the influence of different dimensions on model training. The normalization formula is as follows:

[0062]

[0063] Among them, x is the original data, x min and x max are the minimum and maximum values ​​of the data set, respectively.

[0064] The normalized data not only includes the historical power generation data of the target power station, but also integrates the target power station and its related meteorological data (such as rainfall, temperature, humidity, etc.) to fully reflect the impact of environmental factors on power generation. Through this step, we can comprehensively consider historical power generation characteristics and environmental changes in the input data, providing richer input information for the model.

[0065] S200. Determine the time delay feature with the greatest correlation between the power generation data at a given moment and the power generation data after the delay through historical power generation data; the goal is to find the best position of the data delay in the time dimension, so as to capture the time dependence of the power generation change to the greatest extent.

[0066] Then, basic feature data are constructed based on the preprocessed historical power generation data, historical meteorological data and time delay characteristics.

[0067] S300, build an output prediction model, including three base models, namely LightGBM base model, Transformer base model and LSTM base model; among them:

[0068] The LightGBM base model learns the relationship between input features and the power generation of the target hydropower station by training a decision tree model, which is mainly used to capture the global characteristics and nonlinear characteristics in the data.

[0069] The Transformer base model captures the long-term dependencies in the target power station power generation data through a multi-head self-attention mechanism and generates prediction results with global characteristics. The present invention captures the long-term dependencies in the target power station power generation data through the Transformer model. Transformer adopts a multi-head self-attention mechanism, which can consider the relationship between different positions in the sequence when processing the input sequence and capture long-distance dependency characteristics. In this way, the model can not only pay attention to the power generation situation at the current time point, but also combine the influencing factors at more distant time points to generate prediction results with global information. The advantage of Transformer is that it can perform more efficiently than traditional recursive neural networks (RNNs) and LSTMs when processing long time series data.

[0070] The LSTM base model models time series data through forget gates, input gates, and output gates. The memory unit can efficiently capture the short-term dependencies of data by retaining important historical information and discarding irrelevant information. The local prediction results generated by the LSTM model can accurately reflect the short-term fluctuations of the target power station and provide accurate local information for the final prediction.

[0071] Afterwards, taking the basic feature data as the shared feature, new features are added to the shared features according to the characteristics of the three base models, and three types of input feature data corresponding to the three base models are constructed. The three types of input feature data are input into the three base models respectively. In the training process of the output prediction model, the prediction results of each base model are fused through the weighted integration method. The output prediction model is iteratively trained according to the loss of the fused prediction results and the actual power generation data, and the weight assigned to each base model in the weighted integration process is adjusted.

[0072] S400, predicting the power generation data of the target hydropower cluster at a future time by using the trained output prediction model.

[0073] As a preferred implementation of this embodiment, in step S200, the method for determining the time delay feature with the greatest correlation between the power generation data at a given moment and the power generation data after the delay through the historical power generation data is specifically:

[0074] The Pearson correlation calculation method is used to calculate the Pearson correlation coefficient between the power generation data under different time delays and the power generation data at a given time, and the time delay with the largest Pearson correlation coefficient is taken as the time delay feature. The specific Pearson correlation calculation formula is:

[0075]

[0076] Among them, r XYis the Pearson correlation coefficient, cov(X,Y) is the covariance of the power generation data at a given moment and the power generation data under time delay Y, σ X and σ Y are the standard deviation of the power generation data at a given moment and the power generation data under time delay Y, respectively.

[0077] As a preferred implementation of this embodiment, in step S300, the method of constructing three types of input feature data corresponding to three base models based on the basic feature data is specifically as follows:

[0078] For the LightGBM base model, considering that LightGBM focuses on the nonlinear relationship between global characteristics and features, on the basis of basic feature data, the environmental data of the hydropower cluster is introduced to form the first input feature data, such as reservoir capacity, altitude, etc.

[0079] For the Transformer base model, as a model for processing long-term dependencies, on the basis of basic feature data, we can focus on historical data over a longer time range, especially when dealing with seasonal and cyclical fluctuations, and add meteorological series data within a first given time window for a longer time to form the second input feature data; the meteorological series data within the first given time window, for example, daily meteorological data for the past 30 days.

[0080] For the LSTM base model, more emphasis is placed on the temporal nature of time series data. Therefore, its input features, in addition to the basic feature data, also include historical series data such as power generation and meteorological data within the time window. The power generation series data and meteorological series data within the second given time window are introduced to form the third input feature. The meteorological series data within the second given time window, for example, the daily power generation and meteorological data for the past 10 days.

[0081] As a preferred implementation of this embodiment, in the output prediction model,

[0082] For each base model, the loss function is:

[0083]

[0084] in, is the loss function of the ith base model, y i is the true value of the i-th sample, is the predicted value of the jth base model for the ith sample, and n is the number of samples.

[0085] The prediction results after fusion are as follows:

[0086]

[0087] in, To integrate the prediction results, is the prediction result of the i-th base model, w i is the weight parameter of the i-th base model, and m is the number of base models;

[0088] The loss function of the entire output prediction model training process is set as:

[0089]

[0090] in, is the loss function, n is the number of samples, y i represents the true value of the i-th sample, Represents the predicted value of the j-th base model for the i-th sample;

[0091] In the processing prediction model, the weight wi of each base model is determined by optimizing the objective function of the entire integrated model, that is, minimizing the loss. The optimization of weights is performed by the gradient descent method. The gradient of the loss function to each weight is calculated by the back propagation algorithm, that is, the contribution of each weight to the total loss function. The weight gradient calculation formula for each base model is:

[0092]

[0093] in, is the gradient of the loss function with respect to the jth basis model weight,

[0094] It is the weighted fusion prediction result.

[0095] After calculating the gradient of each base model, the weights are updated using the gradient descent method. The formula for weight update is as follows:

[0096]

[0097] Among them, η is the learning rate, which controls the step size of each update.

[0098] Embodiment 2:

[0099] This embodiment proposes a hydropower cluster output prediction system based on LightGBM-Transformer-LSTM, including:

[0100] A data acquisition module, used to collect historical power generation data and historical meteorological data of the target hydropower cluster and perform preprocessing; this module is used to implement the function of step S100 in the first embodiment, and will not be described in detail here;

[0101] A basic feature construction module is used to determine the time delay feature with the greatest correlation between the power generation data at a given moment and the power generation data after the delay through the historical power generation data; and to construct basic feature data based on the pre-processed historical power generation data, historical meteorological data and time delay features; this module is used to implement the function of step S200 in the first embodiment, and will not be described in detail here;

[0102] The integrated model training module is used to construct an output prediction model, including three base models, namely, the LightGBM base model, the Transformer base model and the LSTM base model; based on the basic feature data, three types of input feature data corresponding to the three base models are respectively constructed, and the three types of input feature data are respectively input into the three base models. In the training process of the output prediction model, the prediction results of each base model are fused by the weighted integration method, and the output prediction model is iteratively trained according to the loss of the fused prediction results and the real power generation data, and the weight assigned to each base model in the weighted integration process is adjusted; this module is used to implement the function of step S300 in Example 1, which will not be repeated here;

[0103] The hydropower cluster output prediction module is used to predict the power generation data of the target hydropower cluster at a future moment through the trained output prediction model; this module is used to implement the function of step S400 in the first embodiment, which will not be described in detail here.

[0104] As a preferred implementation of this embodiment, the method for determining the time delay feature with the greatest correlation between the power generation data at a given moment and the power generation data after the delay through the historical power generation data is specifically:

[0105] The Pearson correlation calculation method is used to calculate the Pearson correlation coefficients of the power generation data under different time delays and the power generation data at a given moment, and the time delay with the largest Pearson correlation coefficient is taken as the time delay feature.

[0106] As a preferred implementation of this embodiment, the method of constructing three types of input feature data corresponding to three base models based on the basic feature data is specifically as follows:

[0107] For the LightGBM base model, on the basis of basic feature data, the environmental data related to the hydropower cluster are introduced to form the first input feature data;

[0108] For the Transformer base model, on the basis of the basic feature data, the meteorological series data within the first given time window is introduced to form the second input feature data;

[0109] For the LSTM base model, on the basis of the basic feature data, the power generation sequence data and the meteorological sequence data within the second given time window are introduced to form the third input feature;

[0110] The first given time window is twice as large as the second given time window.

[0111] As a preferred implementation of this embodiment, in the output prediction model, the fused prediction result is specifically:

[0112]

[0113] in, To integrate the prediction results, is the prediction result of the i-th base model, w i

[0114] is the weight parameter of the ith base model, m is the number of base models;

[0115] The loss function of the output prediction model training process is set as:

[0116]

[0117] in, is the loss function, n is the number of samples, y i represents the true value of the i-th sample, Represents the predicted value of the j-th base model for the i-th sample;

[0118] According to the calculated loss, the partial derivative of the weight of each base model is obtained to obtain the weight gradient of each base model, and the weight of the corresponding base model is updated according to the weight gradient of each base model.

[0119] Embodiment three:

[0120] This embodiment proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the method for predicting the output of a hydropower cluster based on LightGBM-Transformer-LSTM as described in any embodiment of the present invention is implemented.

[0121] Embodiment 4:

[0122] This embodiment proposes a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the method for predicting the output of a hydropower cluster based on LightGBM-Transformer-LSTM as described in any embodiment of the present invention is implemented.

[0123] In the embodiments of the present application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, c can be single or multiple.

[0124] Those of ordinary skill in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented in a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0125] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0126] In several embodiments provided in the present application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), disk or optical disk, and other media that can store program codes.

[0127] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A hydropower cluster output prediction method based on LightGBM-Transformer-LSTM, characterized in that: The following steps are involved: Collect historical power generation data and historical meteorological data of the target hydropower cluster and perform preprocessing; Using historical power generation data, determine the time delay feature with the greatest correlation between power generation data at a given moment and power generation data after a time delay; And construct basic characteristic data based on pre-processed historical power generation data, historical meteorological data and time delay characteristics; Construct an output prediction model, including three base models, namely LightGBM base model, Transformer base model and LSTM base model; based on the basic feature data, construct three types of input feature data corresponding to the three base models respectively, and input the three types of input feature data into the three base models respectively. In the training process of the output prediction model, the prediction results of each base model are fused by the weighted integration method. The output prediction model is iteratively trained according to the loss of the fused prediction results and the real power generation data, and the weights assigned to each base model in the weighted integration process are adjusted; The trained output prediction model is used to predict the power generation data of the target hydropower cluster at future times.

2. According to claim 1, a method for predicting hydropower cluster output based on LightGBM-Transformer-LSTM is characterized in that: The method for determining the time delay feature with the greatest correlation between the power generation data at a given moment and the power generation data after the time delay through the historical power generation data is specifically: The Pearson correlation calculation method is used to calculate the Pearson correlation coefficients of the power generation data under different time delays and the power generation data at a given moment, and the time delay with the largest Pearson correlation coefficient is taken as the time delay feature.

3. According to claim 1, a method for predicting hydropower cluster output based on LightGBM-Transformer-LSTM is characterized in that: The method of constructing three types of input feature data corresponding to three base models based on the basic feature data is specifically as follows: For the LightGBM base model, on the basis of basic feature data, the environmental data related to the hydropower cluster are introduced to form the first input feature data; For the Transformer base model, on the basis of the basic feature data, the meteorological series data within the first given time window is introduced to form the second input feature data; For the LSTM base model, on the basis of the basic feature data, the power generation sequence data and the meteorological sequence data within the second given time window are introduced to form the third input feature; The first given time window is twice as large as the second given time window.

4. According to claim 1, a method for predicting hydropower cluster output based on LightGBM-Transformer-LSTM is characterized in that: In the output prediction model, the fused prediction results are specifically: in, To integrate the prediction results, is the prediction result of the i-th base model, w i is the weight parameter of the ith base model, m is the number of base models; The loss function of the output prediction model training process is set as: in, is the loss function, n is the number of samples, y i represents the true value of the i-th sample, Represents the predicted value of the j-th base model for the i-th sample; According to the calculated loss, the partial derivative of the weight of each base model is obtained to obtain the weight gradient of each base model, and the weight of the corresponding base model is updated according to the weight gradient of each base model.

5. A hydropower cluster output prediction system based on LightGBM-Transformer-LSTM, characterized in that: include: Data acquisition module, used to collect historical power generation data and historical meteorological data of the target hydropower cluster and perform preprocessing; A basic feature building module, used to determine the time delay feature with the greatest correlation between the power generation data at a given moment and the power generation data after the time delay through the historical power generation data; And construct basic characteristic data based on pre-processed historical power generation data, historical meteorological data and time delay characteristics; The integrated model training module is used to construct an output prediction model, including three base models, namely the LightGBM base model, the Transformer base model and the LSTM base model. Based on the basic feature data, three types of input feature data corresponding to the three base models are constructed respectively, and the three types of input feature data are input into the three base models respectively. In the training process of the output prediction model, the prediction results of each base model are fused by the weighted integration method. The output prediction model is iteratively trained according to the loss of the fused prediction results and the real power generation data, and the weights assigned to each base model in the weighted integration process are adjusted. The hydropower cluster output prediction module is used to predict the power generation data of the target hydropower cluster at a future moment through the trained output prediction model.

6. According to claim 5, a hydropower cluster output prediction system based on LightGBM-Transformer-LSTM is characterized in that: The method for determining the time delay feature with the greatest correlation between the power generation data at a given moment and the power generation data after the time delay through the historical power generation data is specifically: The Pearson correlation calculation method is used to calculate the Pearson correlation coefficients of the power generation data under different time delays and the power generation data at a given moment, and the time delay with the largest Pearson correlation coefficient is taken as the time delay feature.

7. The hydropower cluster output prediction system based on LightGBM-Transformer-LSTM according to claim 5 is characterized in that: The method of constructing three types of input feature data corresponding to three base models based on the basic feature data is specifically as follows: For the LightGBM base model, on the basis of basic feature data, the environmental data related to the hydropower cluster are introduced to form the first input feature data; For the Transformer base model, on the basis of the basic feature data, the meteorological series data within the first given time window is introduced to form the second input feature data; For the LSTM base model, on the basis of the basic feature data, the power generation sequence data and the meteorological sequence data within the second given time window are introduced to form the third input feature; The first given time window is twice as large as the second given time window.

8. The hydropower cluster output prediction system based on LightGBM-Transformer-LSTM according to claim 5 is characterized in that: In the output prediction model, the fused prediction results are as follows: in, To integrate the prediction results, is the prediction result of the i-th base model, w i is the weight parameter of the ith base model, m is the number of base models; The loss function of the output prediction model training process is set as: in, is the loss function, n is the number of samples, y i represents the true value of the i-th sample, Represents the predicted value of the j-th base model for the i-th sample; According to the calculated loss, the partial derivative of the weight of each base model is obtained to obtain the weight gradient of each base model, and the weight of the corresponding base model is updated according to the weight gradient of each base model.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the hydropower cluster output prediction method based on LightGBM-Transformer-LSTM as described in any one of claims 1 to 4 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the hydropower cluster output prediction method based on LightGBM-Transformer-LSTM as described in any one of claims 1 to 4 is implemented.

Images