New energy power distribution strategy decision-making method based on deep learning

Abstract

The invention discloses a new energy power distribution strategy decision-making method based on deep learning, and belongs to the technical field of energy management and distribution, and the method comprises the steps: collecting and obtaining a new energy power data set through a big data technology; performing standardized preprocessing on the new energy power data set through a data preprocessing program; constructing a new energy power transaction prediction model by using a deep learning network; determining a new energy power stage prediction result based on the new energy power transaction prediction model; obtaining a power distribution target, and establishing a new energy power period distribution function based on the new energy power phase prediction result; and obtaining a new energy power distribution scheme cluster according to the new energy power stage prediction result, and performing evaluation iterative optimization on the new energy power distribution scheme cluster based on the new energy power period distribution function to serve as a target new energy power period distribution strategy. The technical problem of insufficient accuracy and optimization of a new energy power distribution strategy in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of energy management and distribution, and in particular to a new energy power distribution strategy decision-making method based on deep learning. Background Art

[0002] In recent years, with the increase in global energy demand and the enhancement of environmental protection awareness, new energy technologies have been rapidly developed and widely used. Renewable energy such as wind energy, solar energy, and hydropower have gradually occupied an important position in the power system. However, due to the intermittent and volatile characteristics of these new energy sources, they have brought new challenges to the stable operation of the power system.

[0003] In order to better meet user needs, promote economic development and improve energy efficiency, the power market needs to achieve optimal allocation and efficient use of resources. However, traditional power distribution strategies are often based on static data and empirical judgments, lacking intelligent and dynamic adjustment and optimization capabilities.

[0004] In summary, this application aims to solve the technical problems of insufficient accuracy and optimization of new energy power distribution strategies in the prior art. Summary of the invention

[0005] This application provides a new energy power distribution strategy decision-making method based on deep learning, aiming to solve the technical problems of insufficient accuracy and optimization of new energy power distribution strategies in the prior art.

[0006] In view of the above problems, the present application provides a new energy power distribution strategy decision-making method based on deep learning.

[0007] In one aspect disclosed in the present application, a new energy power distribution strategy decision-making method based on deep learning is provided, the method comprising acquiring a new energy power data set through big data technology, the new energy power data set comprising new energy time series power generation information, power market basic information, power generation side transaction information, medium- and long-term and spot transaction time series data, and grid-side power shortage data; acquiring a data preprocessing program, performing standardized preprocessing on the new energy power data set through the data preprocessing program to obtain a new energy standard power data set; training and optimizing the new energy standard power data set using a deep learning network to construct a new energy power transaction prediction model; performing power prediction for a preset time window based on the new energy power transaction prediction model to determine a new energy power stage prediction result; acquiring a power distribution target, optimizing the weight distribution of the power distribution target based on the new energy power stage prediction result, and establishing a new energy power time period allocation function; obtaining a new energy power distribution scheme cluster according to the new energy power stage prediction result, evaluating and iteratively optimizing the new energy power distribution scheme cluster based on the new energy power time period allocation function, and obtaining an optimal power distribution scheme as a target new energy power time period allocation strategy.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0009] The new energy power data set including new energy time series power generation information, power market basic information, power generation side transaction information, medium- and long-term and spot transaction time series data and grid-side power shortage data is collected through big data technology. Then, the data set is standardized and preprocessed using a data preprocessing program to obtain a new energy standard power data set. On this basis, the new energy standard power data set is trained and optimized using a deep learning network to construct a new energy power transaction prediction model. Based on the prediction model, the power demand and supply situation within the preset time window is predicted to determine the new energy power stage prediction result. Next, the power allocation target is obtained, and the power allocation target is optimized based on the new energy power stage prediction result to establish a new energy power time period allocation function. According to the prediction results, multiple new energy power allocation scheme clusters are generated, and these schemes are evaluated and iteratively optimized using the new energy power time period allocation function, and finally the optimal power allocation scheme is obtained as the target new energy power time period allocation strategy. The technical problems of the accuracy and optimization of the new energy power allocation strategy in the prior art are solved, the accuracy and optimization efficiency of the new energy power allocation are effectively improved, and the stable operation and economic benefits of the power system are ensured.

[0010] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 A flowchart of a new energy power distribution strategy decision-making method based on deep learning is provided for an embodiment of the present application.

[0012] Figure 2 A flowchart of constructing a new energy power transaction prediction model in a new energy power distribution strategy decision-making method based on deep learning is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0013] The overall idea of ​​the technical solution provided by this application is as follows:

[0014] The embodiment of the present application provides a new energy power distribution strategy decision-making method based on deep learning. First, a new energy power data set including new energy time series power generation, power market basic information, power generation side transaction information, medium- and long-term and spot transaction time series data, and power shortage data on the grid side is collected through big data technology. Then, the data set is standardized through a data preprocessing program to obtain a new energy standard power data set. The data set is trained and optimized using a deep learning network to construct a new energy power trading prediction model. Based on the prediction model, the power demand and supply situation within a preset time window is predicted to determine the new energy power stage prediction results. The power distribution target is obtained, and the weight distribution is optimized based on the prediction results to establish a new energy power time period allocation function. Finally, the optimal power distribution plan is obtained by evaluating and iteratively optimizing the generated multiple power distribution plans as the target new energy power time period allocation strategy.

[0015] After introducing the basic principles of the present application, various non-limiting implementation methods of the present application will be specifically described below in conjunction with the drawings in the specification.

[0016] Example

[0017] like Figure 1 As shown, the embodiment of the present application provides a new energy power distribution strategy decision method based on deep learning, the method comprising:

[0018] Step S100: acquiring a new energy power data set through big data technology, wherein the new energy power data set includes new energy time series power generation information, power market basic information, power generation side transaction information, medium- and long-term and spot transaction time series data, and grid side power shortage data.

[0019] Specifically, big data technology is used to collect new energy power data sets from various sources (such as power monitoring systems, market trading platforms, etc.).

[0020] The renewable energy time-series power generation information refers to the time series data recording renewable energy power generation, such as the time changes of wind power and photovoltaic power generation. The data sources include monitoring systems and sensors of wind farms, photovoltaic power stations, hydropower stations and other power generation facilities. With the characteristics of time series, the data changes over time, reflecting the dynamic changes of renewable energy power generation.

[0021] The basic information of the power market includes the price, supply and demand of the power market and other basic information. The data sources include the power trading center and the power market operation platform. It can provide the market environment background and provide reference for power distribution decision-making.

[0022] The power generation side transaction information includes the transaction data of the power generation side in different transaction markets. It comes from the transaction records of power generation enterprises and power transaction platforms. It reflects the market behavior and transaction status of the power generation side and is important data for formulating power generation strategies.

[0023] The medium- and long-term and spot transaction time series data include the transaction time series data of medium- and long-term contracts and spot markets. They are derived from the power trading market and contract recording system. They can provide long-term and short-term transaction information to help predict future market behavior and price fluctuations.

[0024] The power shortage data on the grid side refers to the data recording the power shortage on the power user side. It comes from the power company's customer service system, smart meter data, and user feedback. It reflects the demand and emergency status on the power user side and is an important basis for adjusting power distribution.

[0025] Based on this step, a comprehensive, accurate, and real-time updated new energy power data set can be constructed, providing a solid data foundation for subsequent power distribution strategy decisions.

[0026] Step S200: obtaining a data preprocessing program, and performing standardization preprocessing on the new energy power data set through the data preprocessing program to obtain a new energy standard power data set.

[0027] Specifically, data preprocessing is a key step in data analysis and machine learning, and its purpose is to convert raw data into a format suitable for model training and analysis. For the new energy power data set, the preprocessing steps include data cleaning, normalization, feature extraction and other operations.

[0028] The new energy standard power dataset processed based on this step can effectively support the training and optimization of deep learning models and improve the model's predictive ability and decision-making effect.

[0029] Step S300: Use a deep learning network to train and optimize the new energy standard power data set to build a new energy power transaction prediction model.

[0030] Specifically, after completing data preprocessing, the standardized new energy power data set is trained and optimized using a deep learning network to build a model that can effectively predict new energy power transactions.

[0031] First, select a suitable deep learning model to build a prediction model. Commonly used deep learning models include: Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Long Short-Term Memory Network (LSTM), Transformer model. This application uses Long Short-Term Memory Network (LSTM), which is an improved version of RNN and can effectively handle long-term dependency problems.

[0032] Then, a deep learning model architecture suitable for the characteristics of renewable energy power data is designed. The following is an example of a typical LSTM model architecture: Input layer, which accepts the standardized renewable energy power data set. The LSTM layer is used to capture the long-term dependencies of time series data. The fully connected layer (Dense Layer) connects the output of the LSTM layer to the fully connected layer for feature combination and mapping. The output layer predicts the power trading volume in the future time window.

[0033] Then, the new energy standard electricity dataset is divided into training set, validation set and test set for model training and evaluation. Select appropriate loss function (such as mean square error) and optimizer (such as Adam) for model compilation. Train the model using training data and perform hyperparameter tuning on the validation set to prevent overfitting. Evaluate model performance on the test set to ensure that the model has good generalization ability.

[0034] Finally, adjust the model's hyperparameters (such as learning rate, batch size, number of layers, and number of units) to optimize model performance. Use regularization techniques (such as L2 regularization, Dropout) to prevent overfitting and improve the generalization ability of the model. Combine the prediction results of multiple models (such as model integration, voting method) to improve the accuracy and stability of the prediction.

[0035] Based on this step, the deep learning network can be used to train and optimize the new energy standard power data set to build an effective new energy power trading prediction model. This model can accurately predict the future power trading volume, provide a scientific basis for the dispatch and resource optimization of the power system, and improve the efficiency and stability of the system.

[0036] Step S400: Perform power forecasting for a preset time window based on the new energy power transaction forecasting model to determine a new energy power stage forecasting result.

[0037] Specifically, after constructing and optimizing the new energy power trading prediction model, the model is used to predict future power demand and supply. The preset time window is the time window that needs to be predicted according to specific needs, such as the next hour, day, week, etc. The selection of the time window depends on the specific application scenario and decision-making needs.

[0038] Get real-time data from the power system, including the latest renewable energy generation, meteorological data, power market data, etc. Integrate real-time data with historical data to form a complete input data set for the model to make predictions.

[0039] Standardize the real-time data to make it meet the model input requirements. Use the same standardization method (such as mean and standard deviation) as when training the model. Extract the required features from the real-time data, such as time features (such as hours, weeks, seasons, etc.) and statistical features.

[0040] The preprocessed data is imported into the trained new energy power trading prediction model. The model calculates based on the input data and outputs the prediction results. These results include the power demand and supply within the preset time window.

[0041] Based on the forecast results output by the model, determine the phased change trend and specific values ​​of renewable energy power. For example, predict the power generation and demand per hour in the next day. Evaluate the uncertainty of the forecast results, such as using confidence intervals or prediction intervals to indicate the reliability of the forecast results.

[0042] Use the forecast results to guide power dispatch decisions. For example, adjust power generation plans and load management strategies based on the forecasted power demand and supply. Optimize the configuration and use of new energy resources to improve the operating efficiency and economic benefits of the power system. Identify potential power shortages or surpluses in advance based on the forecast results and take corresponding preventive measures.

[0043] Based on this step, it is possible to make power forecasts for the preset time window based on the new energy power trading forecasting model, determine the new energy power stage forecast results, and provide a scientific basis for the scheduling and optimization of the power system.

[0044] Step S500: Acquire a power allocation target, optimize weight allocation of the power allocation target based on the new energy power stage prediction result, and establish a new energy power time period allocation function.

[0045] Specifically, after completing the power forecast and obtaining the new energy power stage forecast results, the power allocation target is obtained, and the weight allocation is optimized based on the forecast results to establish a new energy power time period allocation function.

[0046] Among them, the power distribution target includes power demand target, power generation target, system stability target, and economic target. The power demand target refers to determining the demand target of the power system in each time period, including basic demand, peak demand, and peak-shaving demand. The power generation target refers to determining the power generation target of each new energy power generation unit, including power generation, output rate, power generation cost, etc. The system stability target refers to ensuring the stable operation of the power system, including frequency stability, voltage stability, and supply and demand balance. The economic target refers to economic targets such as optimizing power generation costs, reducing wind and solar power abandonment, and improving the utilization rate of new energy.

[0047] Furthermore, based on the predicted power demand data, analyze the power demand in each time period, including peak time period, valley time period, etc. Based on the predicted new energy power generation data, analyze the power generation capacity in each time period, including power generation volatility, intermittency, etc.

[0048] Optimization weight allocation requires clear optimization targets, such as improving the utilization rate of new energy, reducing power generation costs, and ensuring system stability. Different weights are assigned according to the importance of different optimization targets. For example, different weights can be assigned to economic targets, stability targets, environmental protection targets, etc. to reflect their priority in the overall optimization. For example, the weight of the system stability target is 0.4, the utilization rate of new energy is 0.3, the economy is 0.2, and the environmental protection target is 0.1. According to the real-time operation status and forecast results, the weights of each target are dynamically adjusted to adapt to the changing power demand and power generation situation.

[0049] Finally, based on the optimized weight allocation results, a new energy power time allocation function is established. This function takes the predicted value of new energy power as input and calculates the power allocation amount for each time period according to the weight allocation results. The allocation function can consider a variety of factors, such as the forecast of power demand, the transmission capacity of the power network, the price mechanism of the power market, etc., to ensure the rationality and effectiveness of power allocation.

[0050] The new energy power time allocation function can be expressed as:

[0051] Objective function = ∑ i w i ·f i (x);

[0052] Among them, w i is the weight of the i-th target, f i (x) is the function of the i-th objective, and x is the decision variable. Determine the constraints that need to be met in the allocation process, such as power balance constraints, generation capacity constraints, line transmission capacity constraints, etc. Use optimization algorithms (such as linear programming, nonlinear programming, genetic algorithms, etc.) to solve the objective function and obtain the optimal power allocation plan.

[0053] Based on this step, the prediction results can be effectively used to optimize the weight allocation of power distribution targets, establish a new energy power time allocation function, and achieve the optimal power distribution strategy.

[0054] Step S600: Obtain a new energy power allocation scheme cluster according to the new energy power stage prediction result, evaluate and iteratively optimize the new energy power allocation scheme cluster based on the new energy power period allocation function, and obtain the optimal power allocation scheme as the target new energy power period allocation strategy.

[0055] Specifically, based on the prediction results of the renewable energy power stage, a set of power distribution schemes are initially generated. These schemes consider different power generation strategies and load management strategies to meet the predicted power demand and supply. Among them, a scheme cluster is constructed, that is, a collection of multiple candidate power distribution schemes. Each scheme has different power distribution strategies in different time periods to maximize the utilization of renewable energy, minimize costs, and ensure system stability.

[0056] Then, based on the renewable energy power time allocation function, the fitness of each allocation scheme is calculated. The fitness function comprehensively considers the weights of various allocation targets and evaluates the performance of each scheme on different targets. The evaluation criteria include system stability, renewable energy utilization, economic and environmental benefits, etc. The evaluation criteria are used to judge the pros and cons of each scheme.

[0057] Evaluate the initially generated solution cluster and calculate the fitness value of each solution. Preliminarily select the better solutions according to the fitness value. Optimize and adjust the selected better solutions. The adjustment direction includes improving the utilization rate of new energy, reducing the cost of power generation, and enhancing system stability. Use optimization algorithms (such as genetic algorithms, particle swarm optimization algorithms, etc.) to improve the solutions. Re-evaluate and calculate the improved solutions to obtain new fitness values. Continuously iterate optimization and evaluation to gradually approach the optimal solution. Set the termination conditions of iterative optimization, such as reaching the preset fitness threshold or converging after a certain number of iterations. According to the final evaluation results, select the solution with the highest fitness value as the optimal power distribution solution. This solution can achieve the best balance in all aspects after comprehensively considering various goals. Take the optimal power distribution solution as the target new energy power time allocation strategy. This strategy will be used for power distribution decisions in different time windows in actual operation to guide the dispatch and optimization of the power system.

[0058] Based on this step, the optimal power distribution scheme can be evaluated and optimized from the generated multiple power distribution schemes, realizing an efficient, stable, economical and environmentally friendly new energy power distribution strategy.

[0059] Furthermore, the method described in the present application also includes:

[0060] Abnormal data identification is performed on the new energy power data set to obtain an abnormal power data set, wherein the abnormal power data set includes inconsistent data, missing data and out-of-bounds data; the abnormal power data set is mapped and matched with the data preprocessing program to determine an abnormal data preprocessing matching program; based on the abnormal data preprocessing matching program, the inconsistent data, missing data and out-of-bounds data are preprocessed respectively to obtain consistent power data, complete power data and normative power data; based on the consistent power data, complete power data and normative power data, the new energy standard power data set is obtained.

[0061] Specifically, identify inconsistent data in the dataset, which may be caused by different formats or recording standards of different data sources. For example, data timestamp formats are not uniform, value units are inconsistent, etc. Identify missing data in the dataset. Missing data may be missing records due to sensor failure, data transmission errors, or other reasons. Identify data points that are out of reasonable range, which may be caused by collection errors or abnormal events. For example, wind speed sensor readings are abnormally high or low.

[0062] Map and match the identified abnormal data with the preprocessing program to determine the preprocessing methods for different types of abnormal data. Design specific preprocessing programs for different types of abnormal data to ensure the consistency, integrity and standardization of the preprocessed data.

[0063] Preprocess inconsistent data. For example, unify the timestamp format and convert data in different units into a unified unit to ensure data consistency. Fill in missing data. Common methods include: fill missing values ​​with the mean of the data; fill missing values ​​with the median of the data; fill missing data with linear interpolation or other interpolation methods; use regression models or other machine learning methods to predict and fill missing values. Process out-of-bounds data. Common methods include: directly delete out-of-bounds data and correct out-of-bounds data to values ​​within a reasonable range. For example, correct abnormally high wind speed readings to reasonable maximum wind speed values. Replace out-of-bounds data with normal data at adjacent time points.

[0064] Finally, the consistent power data, complete power data and standardized power data are integrated to form a standardized new energy power data set. This can effectively process abnormal data in the new energy power data set, obtain high-quality standardized new energy power data sets, and provide a reliable data foundation for subsequent deep learning model training and optimization.

[0065] Further, such as Figure 2 As shown, the method described in the present application also includes:

[0066] The new energy standard power data set is labeled with power supply and demand conditions, power generation and transaction electricity prices to obtain a power supply and demand sample set, a power generation sample set and a power transaction electricity price sample set; based on a deep learning network, the power supply and demand sample set, the power generation sample set and the power transaction electricity price sample set are subjected to cyclic supervised training to generate a power supply and demand prediction model, a power generation prediction model and a power transaction electricity price prediction model; the power supply and demand prediction model, the power generation prediction model and the power transaction electricity price prediction model are merged to generate an initial power transaction prediction model; the initial power transaction prediction model is verified and optimized to construct the new energy power transaction prediction model.

[0067] Specifically, the electricity supply and demand situation is identified based on the historical data and real-time data in the new energy standard power data set. Including electricity demand, load changes, etc. Generate a sample set of electricity supply and demand situation, which contains electricity demand and supply data in different time periods. Identify the historical power generation data of each new energy power generation unit, including wind power, photovoltaics, etc. Generate a power generation sample set, which contains the power generation data of each new energy power generation unit in different time periods. Identify the transaction electricity price information in the power market, including spot market electricity prices, medium- and long-term transaction electricity prices, etc. Generate a power transaction price sample set, which contains power transaction price data in different time periods.

[0068] Then, the long short-term memory network (LSTM) is used to train the power supply and demand sample set to obtain the power supply and demand prediction model. During the training process, the power demand in the future time period is predicted by inputting the historical power demand data. A similar method is used to train the power generation sample set to obtain the power generation prediction model. By inputting the historical power generation data and meteorological data, the new energy power generation in the future time period is predicted. The power trading price sample set is trained to obtain the power trading price prediction model. By inputting the historical power price data and market supply and demand, the power trading price in the future time period is predicted.

[0069] The output results of the three models are concatenated to form a comprehensive input feature matrix. For example, assuming that the output of each model is the predicted value for the next 24 hours, the final feature matrix will contain three 24-dimensional vectors (power demand, power generation, and power trading price). Design a new deep learning model to process the concatenated feature matrix. Common fusion model structures include: fully connected neural networks (DNNs) to further nonlinearly transform and combine the concatenated features. Time series models, such as LSTM or Transformer, can capture the temporal dependencies between features. Use historical data to train the fusion model with the goal of minimizing the prediction error. During the training process, the input is the concatenated feature matrix and the output is the actual power trading data (such as actual supply and demand balance, power price, etc.).

[0070] The initial power trading prediction model is verified using the validation set to evaluate the prediction accuracy and generalization ability of the model. The model's error and accuracy are calculated by comparing the model's prediction results with the actual data. Based on the validation results, the initial power trading prediction model is optimized. Optimization methods include adjusting model parameters, improving training algorithms, and increasing the amount of training data. The optimized model is used to re-predict and verify to ensure that the model has higher prediction accuracy and reliability.

[0071] Based on this step, the electricity supply and demand forecasting model, the electricity generation forecasting model and the electricity trading price forecasting model can be effectively integrated to generate a high-precision new energy power trading forecasting model, providing strong support for the optimization and scheduling of the power system.

[0072] Furthermore, the method described in the present application also includes:

[0073] A recurrent neural network structure is selected according to the deep learning network, and an initial hidden layer value and a model weight matrix of the recurrent neural network structure are obtained; based on the initial hidden layer value and the model weight matrix, hidden layer calculations are performed on the electricity supply and demand sample set, the electricity generation sample set, and the electricity trading price sample set, respectively, to obtain current hidden layer values; based on the current hidden layer value and the model weight matrix, iterative calculation training is performed on the electricity supply and demand sample set, the electricity generation sample set, and the electricity trading price sample set, respectively, to generate the electricity supply and demand prediction model, the electricity generation prediction model, and the electricity trading price prediction model.

[0074] Specifically, select a suitable RNN structure, such as a long short-term memory network (LSTM), and initialize the hidden layer values ​​of the RNN (usually zero) and the model weight matrix before training. The weight matrix can be initialized using a random initialization method (such as Xavier initialization). The power supply and demand sample set, power generation sample set, and power trading price sample set are used as input data and input into the RNN model in sequence. At each time step, input the data xt of the current time step and combine it with the hidden layer value h of the previous time step. t-1 , calculate the hidden layer value ht and output value of the current time step. The update formula is:

[0075] ht=σ(W h ·h t-1 +W x ·X t +b h );

[0076] Among them, W h is the hidden layer weight matrix, W x is the input weight matrix, b h is the bias term, and σ is the activation function (such as tanh or ReLU). Calculate the data at each time step to obtain the hidden layer value h at the current time step t .

[0077] Iterative training is performed using the back propagation through time (BPTT) algorithm, forward calculation and backward propagation are performed on the data of each time step, and the weight matrix of the model is updated. In each iteration, the loss function (such as mean square error MSE) is calculated and the gradient is calculated. The weight matrix and bias terms are updated so that the model gradually approaches the optimal prediction effect. After multiple iterative training, the power supply and demand prediction model, the power generation prediction model and the power transaction price prediction model are generated respectively. Each model can accurately predict the relevant data of the future time period based on the input time series data.

[0078] This step uses deep learning networks to generate power supply and demand forecasting models, power generation forecasting models, and power trading price forecasting models. These models can accurately predict relevant data for future time periods and provide a scientific basis for new energy power distribution strategy decision-making.

[0079] Furthermore, the method described in the present application also includes:

[0080] The initial power trading prediction model is evaluated and verified by using a model loss function to determine model performance loss parameters; based on the model performance loss parameters, gradient calculation is performed on the cyclic model parameters of the initial power trading prediction model to obtain prediction model parameter gradient information; based on the prediction model parameter gradient information, the cyclic model parameters are iteratively updated through a back propagation algorithm to obtain cyclic model update parameters; based on the cyclic model update parameters, the initial power trading prediction model is optimized and configured to obtain the new energy power trading prediction model.

[0081] Specifically, the initial power trading prediction model is evaluated using a loss function. Commonly used loss functions include mean square error (MSE) and mean absolute error (MAE). The loss parameter of the model performance is determined by calculating the prediction error of the model on the validation set.

[0082] The gradient information of the initial power trading prediction model is calculated based on the performance loss parameters of the model. The gradient information is used to guide the update of model parameters to ensure that the model is optimized in the direction of reducing losses.

[0083] The back-propagation algorithm is used to iteratively update the model parameters based on the calculated gradient information. The back-propagation algorithm calculates the contribution of each parameter to the loss through the chain rule and updates the parameters based on the learning rate.

[0084] The hidden layer parameters and weight matrix of the recurrent neural network are updated to obtain new recurrent model parameters. In each iteration, the updated parameters gradually approach the optimal solution, which continuously improves the prediction accuracy of the model.

[0085] The updated cycle model parameters are used to optimize the initial power trading prediction model. Through multiple iterations of updates and optimizations, the performance of the model on the validation set is gradually improved. After sufficient training and optimization, the final new energy power trading prediction model is constructed. The model can comprehensively predict the power supply and demand, power generation and trading electricity prices in the future time period, providing a scientific basis for power distribution strategy decision-making.

[0086] This step uses the model loss function to evaluate and verify the initial power trading prediction model, calculates the gradient information and iteratively updates the model parameters through the back propagation algorithm, and finally optimizes the configuration to obtain a high-precision new energy power trading prediction model. It can accurately predict future power supply and demand, power generation and transaction prices, and provide reliable decision support for new energy power allocation strategies.

[0087] Furthermore, the method described in the present application also includes:

[0088] The power distribution target is parsed as sub-targets to determine a set of power distribution sub-targets; a new energy power distribution strategy space is constructed, and functions are fitted to the new energy power distribution strategy space based on the power distribution sub-target set to obtain a set of sub-target fitness functions; a decision-making evaluation is performed on the power distribution sub-target set based on the new energy power stage prediction results to obtain a set of sub-target weight factors; the sub-target fitness function set is weighted and fused based on the sub-target weight factor set to establish the new energy power time period allocation function.

[0089] Specifically, the overall power distribution target is analyzed in detail and decomposed into multiple specific sub-targets. Sub-target examples: Maximize the utilization rate of new energy: increase the proportion of new energy such as wind energy and solar energy. Minimize power generation costs: reduce overall power generation costs and improve economic benefits. Ensure the stability of the power system: ensure the stability of the grid frequency and voltage to meet load requirements. Reduce carbon emissions: optimize the power generation structure and reduce the emission of harmful substances such as carbon dioxide. The analyzed sub-targets are summarized to form a set of power distribution sub-targets.

[0090] Define the renewable energy power distribution strategy space, including all possible power distribution schemes. The strategy space considers factors such as power supply and demand, power generation capacity, and market price in different time periods. Based on the set of power distribution sub-goals, perform function fitting on the renewable energy power distribution strategy space. The fitting process uses historical data and forecast data to establish the sub-goal fitness function. The fitness function represents the effect of each power distribution scheme in achieving each sub-goal.

[0091] Use the output results of the power trading forecasting model to evaluate the performance of each sub-goal in different time periods. The evaluation process includes analyzing the changing trends of power demand, power generation and market prices. Determine the importance and priority of each sub-goal based on the decision-making evaluation results. Assign weight factors to form a set of sub-goal weight factors. The decision-making evaluation refers to a systematic evaluation of the realization of each power distribution sub-goal in different time periods during the formulation of the new energy power distribution strategy, and determine the importance and priority of each sub-goal based on the evaluation results. The purpose of this evaluation process is to provide a scientific basis for the final power distribution decision, to ensure that the optimized power distribution plan can balance the requirements of multiple sub-goals and achieve the best overall effect.

[0092] Based on the sub-goal weight factor set, the fitness functions of each sub-goal are weighted and fused. The fused fitness function constitutes the final new energy power time allocation function. This function can comprehensively consider the realization of each sub-goal and provide an optimization plan for power allocation in different time periods.

[0093] Through the above steps, a new energy power time allocation function that comprehensively considers multiple sub-goals can be established to optimize the power allocation strategy in different time periods, improve the utilization rate of new energy, reduce power generation costs, ensure system stability, and reduce carbon emissions.

[0094] Furthermore, the method described in the present application also includes:

[0095] Parameter optimization is performed within the new energy power distribution scheme cluster by using a preset parameter step size to obtain multiple power distribution schemes; the multiple power distribution schemes are evaluated and calculated based on the new energy power time period allocation function to obtain the fitness of multiple power distribution schemes; based on the fitness of the multiple power distribution schemes, the new energy power distribution scheme cluster is compared and optimized to determine the optimal power distribution scheme.

[0096] Specifically, a preset parameter step size, i.e., the increment of the parameter adjustment in each iteration, is set for optimization within the cluster of renewable energy power distribution schemes. According to the preset parameter step size, parameter optimization is performed within the cluster of renewable energy power distribution schemes. Multiple different power distribution schemes are generated by systematically adjusting the power distribution parameters.

[0097] Using the established renewable energy power time allocation function, the generated multiple power allocation schemes are evaluated and calculated. The evaluation process includes calculating the performance of each scheme in achieving each sub-goal and obtaining the fitness value. The fitness of each power allocation scheme is calculated, and the fitness function reflects the comprehensive effect of the scheme in achieving each sub-goal. The fitness value is used to compare the advantages and disadvantages of different schemes.

[0098] Compare the fitness of all generated power distribution plans to find the plan with the highest fitness. The plan with the highest fitness is the plan with the best comprehensive performance in all sub-goals. By comparing the fitness, the optimal power distribution plan is determined. This plan achieves the best balance in maximizing the utilization of new energy, minimizing the cost of power generation, ensuring system stability and reducing carbon emissions.

[0099] This step uses the method of parameter optimization, evaluation calculation and fitness comparison to find the optimal power distribution plan within the cluster of renewable energy power distribution plans. This plan can achieve the best balance among multiple sub-goals and provide reliable decision support for the optimal operation of renewable energy power systems.

[0100] Furthermore, the method described in the present application also includes:

[0101] Based on the adaptability of the multiple power distribution schemes, the multiple power distribution schemes are compared and analyzed to determine the parameter optimization direction; according to the parameter optimization direction, the new energy power distribution scheme cluster is matched and divided to obtain a local power distribution scheme set; a local search step is set, and based on the local search step, the local power distribution scheme set is optimized and analyzed to determine the optimal power distribution scheme.

[0102] Specifically, the fitness of all generated power distribution schemes is calculated and the fitness values ​​of each scheme are compared. By comparing the fitness values, the effect of each scheme in achieving each sub-goal is evaluated. According to the fitness comparison results, the direction of further optimization is determined. The parameter optimization direction indicates how to adjust the power distribution parameters to improve the overall fitness of the scheme.

[0103] According to the determined parameter optimization direction, the new energy power distribution plan clusters are matched and divided. A local power distribution plan set is generated, focusing on the most potential parameter area for further optimization.

[0104] Set the local search step size, that is, the increment of adjusting the parameters in each iteration when optimizing within the local power allocation solution set. The local search step size is usually smaller than the initial parameter step size to perform fine optimization. In the local power allocation solution set, use the set local search step size for optimization analysis. Iterate and adjust the power allocation parameters to gradually approach the optimal solution. Through fine optimization within the local solution set, determine the power allocation solution with the highest fitness. This solution performs best in comprehensively achieving multiple sub-goals and becomes the final optimal power allocation solution.

[0105] This step is based on the methods of comparison analysis, matching division and local search optimization, which can finely optimize within the cluster of renewable energy power distribution schemes and find the optimal power distribution scheme. This scheme can achieve the best balance in multiple sub-goals and provide reliable decision support for the optimized operation of renewable energy power systems.

[0106] In summary, the new energy power distribution strategy decision method based on deep learning provided by the embodiment of the present application has the following technical effects:

[0107] 1. We collected and standardized the preprocessed new energy power data set through big data technology, and used deep learning networks for training and optimization to build a new energy power trading prediction model. Based on the prediction results, we optimized the power distribution target and established the power time allocation function, and finally obtained the optimal power distribution plan through evaluation and iterative optimization. This method significantly improved the accuracy and optimization efficiency of new energy power distribution.

[0108] 2. Detailed identification of new energy standard power data sets, generation of power supply and demand sample sets, power generation sample sets and power transaction price sample sets, and use of deep learning networks for cyclic supervised training to generate multiple prediction models. Through model fusion and verification optimization, a comprehensive new energy power transaction prediction model was constructed, which improved the prediction accuracy of future power supply and demand, power generation and transaction prices.

[0109] 3. Use the model loss function to evaluate and verify the initial power trading prediction model, calculate the gradient information, and iteratively update the model parameters through the back propagation algorithm, and finally optimize the configuration to obtain a high-precision new energy power trading prediction model. This process improves the prediction accuracy and generalization ability of the model.

[0110] 4. Generate multiple power distribution plans through parameter optimization, evaluate and calculate the plans based on the new energy power time allocation function, and determine the optimal power distribution plan with the highest adaptability. This step improves the scientificity and rationality of the power distribution plan and ensures the comprehensive performance of the optimal plan on multiple sub-goals.

[0111] Any step of the method described above can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor to implement any method in the embodiments of the present application, without any unnecessary restrictions.

[0112] Furthermore, the first or second mentioned above may not only represent an order relationship, but may also represent a specific concept, and / or refer to multiple elements that can be selected individually or in full. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A new energy power distribution strategy decision-making method based on deep learning, characterized in that: The method comprises: The new energy power data set is acquired through big data technology, and the new energy power data set includes new energy time-series power generation information, power market basic information, power generation side transaction information, medium- and long-term and spot transaction time-series data, and grid-side power shortage data; Acquire a data preprocessing program, and perform standardized preprocessing on the new energy power data set through the data preprocessing program to obtain a new energy standard power data set; Using a deep learning network to train and optimize the new energy standard power data set, and construct a new energy power transaction prediction model; Based on the new energy power transaction prediction model, power prediction is performed for a preset time window to determine the new energy power stage prediction result; Acquire a power distribution target, optimize weight allocation for the power distribution target based on the new energy power stage prediction result, and establish a new energy power time period allocation function; A new energy power allocation scheme cluster is obtained according to the new energy power stage prediction result, and the new energy power allocation scheme cluster is evaluated and iteratively optimized based on the new energy power period allocation function to obtain the optimal power allocation scheme as the target new energy power period allocation strategy.

2. A new energy power distribution strategy decision method based on deep learning as claimed in claim 1, characterized in that: The obtaining of the new energy standard power data set includes: Performing abnormal data identification on the new energy power data set to obtain an abnormal power data set, wherein the abnormal power data set includes inconsistent data, missing data, and out-of-bounds data; Mapping and matching the power abnormality data set with the data preprocessing program to determine an abnormal data preprocessing matching program; Preprocessing the inconsistent data, missing data and out-of-bounds data based on the abnormal data preprocessing matching program to obtain consistent power data, complete power data and standardized power data; The new energy standard power data set is obtained according to the consistent power data, the complete power data and the normative power data.

3. A new energy power distribution strategy decision method based on deep learning as claimed in claim 1, characterized in that: The constructing of a new energy power transaction prediction model comprises: Marking the power supply and demand situation, power generation and transaction price of the new energy standard power data set to obtain a power supply and demand situation sample set, a power generation sample set and a power transaction price sample set; Based on the deep learning network, the power supply and demand sample set, the power generation sample set and the power transaction price sample set are subjected to cyclic supervised training to generate a power supply and demand prediction model, a power generation prediction model and a power transaction price prediction model; The power supply and demand prediction model, the power generation prediction model and the power transaction price prediction model are combined to generate an initial power transaction prediction model; The initial power transaction prediction model is verified and optimized to construct the new energy power transaction prediction model.

4. A new energy power distribution strategy decision method based on deep learning as claimed in claim 3, characterized in that: The generating of the power supply and demand prediction model, the power generation prediction model and the power transaction price prediction model comprises: Selecting a recurrent neural network structure according to a deep learning network, and obtaining an initial hidden layer value and a model weight matrix of the recurrent neural network structure; Based on the initial hidden layer value and the model weight matrix, performing hidden layer calculations on the power supply and demand sample set, the power generation sample set, and the power transaction price sample set to obtain current hidden layer values; Based on the current hidden layer value and the model weight matrix, the electricity supply and demand sample set, the electricity generation sample set and the electricity trading price sample set are respectively subjected to iterative calculation training to generate the electricity supply and demand prediction model, the electricity generation prediction model and the electricity trading price prediction model.

5. A new energy power distribution strategy decision method based on deep learning as claimed in claim 3, characterized in that: The constructing of the new energy power transaction prediction model comprises: Using the model loss function to evaluate and verify the initial power trading prediction model, and determine the model performance loss parameter; Performing gradient calculation on the cyclic model parameters of the initial power trading prediction model based on the model performance loss parameters to obtain prediction model parameter gradient information; Iteratively updating the cycle model parameters based on the prediction model parameter gradient information through a back propagation algorithm to obtain cycle model update parameters; The initial power trading prediction model is optimized and configured based on the cycle model update parameters to obtain the new energy power trading prediction model.

6. A new energy power distribution strategy decision method based on deep learning as claimed in claim 1, characterized in that: The step of establishing a new energy power time period allocation function comprises: Performing sub-goal analysis on the power distribution goal to determine a power distribution sub-goal set; Constructing a new energy power allocation strategy space, performing function fitting on the new energy power allocation strategy space based on the power allocation sub-goal set, and obtaining a sub-goal fitness function set; Based on the prediction result of the new energy power stage, a decision-making evaluation is performed on the power allocation sub-goal set to obtain a sub-goal weight factor set; The sub-objective fitness function set is weighted and fused based on the sub-objective weight factor set to establish the new energy power time period allocation function.

7. A new energy power distribution strategy decision method based on deep learning as claimed in claim 1, characterized in that: The obtaining of the optimal power distribution scheme comprises: Performing parameter optimization within the new energy power distribution scheme cluster by using a preset parameter step size to obtain multiple power distribution schemes; Based on the new energy power time period allocation function, the plurality of power allocation schemes are evaluated and calculated to obtain the fitness of the plurality of power allocation schemes; The new energy power distribution scheme cluster is compared and optimized based on the adaptability of the multiple power distribution schemes to determine the optimal power distribution scheme.

8. A new energy power distribution strategy decision method based on deep learning as claimed in claim 7, characterized in that: Determining the optimal power distribution scheme includes: Comparing and analyzing the multiple power distribution schemes based on the adaptability of the multiple power distribution schemes, and determining a parameter optimization direction; Matching and dividing the new energy power distribution scheme cluster according to the parameter optimization direction to obtain a local power distribution scheme set; A local search step is set, and optimization analysis is performed on the local power allocation scheme set based on the local search step to determine the optimal power allocation scheme.

Images