New energy power generation prediction system based on big data and AI

Abstract

The invention provides a new energy power generation prediction method and system based on big data and AI. The system firstly collects historical data, meteorological data, equipment operation data and the like related to new energy power generation from a plurality of data sources, and performs data cleaning and preprocessing, including processing missing values in the data and standardizing the data. Next, the processed data is trained and predicted based on a deep learning model LSTM, the model can capture time dependence and nonlinear relation in the data, and accurate prediction of future power generation is provided. In addition, the system also supports adaptive optimization and dynamic updating of the model so as to cope with possible environment change and data distribution drift in the new energy power generation process.

Description

Technical Field

[0001] The present invention relates to data analysis, in particular to a new energy power generation prediction system based on big data and AI Background Art

[0002] With the continuous growth of global energy demand and the increasingly severe environmental problems, renewable energy power generation (such as wind power, solar energy, etc.) has become an important part of the modern power system. However, since these renewable energy power generation methods are greatly affected by factors such as weather and climate, and have high volatility and uncertainty, how to accurately predict renewable energy power generation has become a key issue in grid dispatching and energy management.

[0003] Traditional power generation forecasting methods usually rely on simple statistical models or empirical formulas, which are difficult to handle complex nonlinear relationships and large-scale multidimensional data. With the rapid development of big data and artificial intelligence (AI) technology, the use of big data analysis and deep learning models to forecast renewable energy power generation has gradually become a research hotspot. In many cases, the integration between meteorological models and wind power generation forecasting models is not high enough. The input data provided by the meteorological model (such as wind speed) may not fully take into account the physical characteristics of wind power generation, such as the specific response characteristics of wind turbines. This disconnect affects the prediction accuracy of the model in extreme weather. Summary of the invention

[0004] The present invention aims to at least solve the technical problems existing in the prior art, and particularly innovatively proposes a renewable energy power generation prediction system based on big data and AI. The system combines data cleaning, feature engineering, deep learning models and optimization strategies, and can handle the uncertainty and volatility in renewable energy power generation data, provide more accurate short-term and long-term power generation forecasts, help optimize power grid scheduling, and reduce the risks caused by power generation fluctuations.

[0005] 1. Data preprocessing module

[0006] The complete wind speed dataset x is collected from multiple wind speed sensors at the wind farm, which contains seasonal variations and diurnal fluctuations.

[0007] All the acquired wind speed data sets x are processed, and the missing values ​​in all the wind speed data sets x are interpolated by mixed probability.

[0008]

[0009] in, is the missing value of wind speed data calculated by probability, t is the current time point, t-1 and t+1 are the time points before and after t respectively, and x t-1 and x t+1are the values ​​of the front and back items of the missing value of the wind speed data, respectively. D is the observed wind speed data set. The model infers the missing value of wind speed based on the observed wind speed data set D. is a probability term, which is used to estimate reasonable interpolation values ​​based on the probability prediction of missing values ​​of wind speed data;

[0010] The probability term The calculation method is:

[0011]

[0012] where θ represents the parameter vector of the missing value probability model, Represents the missing value of wind speed data at the parameter vector θ and the current time point t The conditional probability distribution of , P(θ|D) represents the posterior distribution of parameter θ after observing the wind speed data set D;

[0013] This formula represents the predicted value of the wind speed data at the current time point t under the condition of observing the wind speed data set D. The probability distribution of This probability distribution is obtained by weighted integration of the posterior distribution P(θ|D) of the parameter θ;

[0014] After the mixed probability interpolation is completed, the wind speed data x t Standardization

[0015]

[0016] Among them, μ x is the mean of the wind speed data set x, σ x is the standard deviation of the wind speed data set x. The standardized wind speed data The mean is 0 and the standard deviation is 1.

[0017] The mixed probability interpolation method was used to handle the missing values ​​in the wind speed data to ensure the continuity and reliability of the wind speed data. The standardization step used the mean and standard deviation of the historical data of the wind farm.

[0018] 2. Model training module

[0019] The preprocessed wind speed data was used as input and trained using the LSTM model. The model architecture was optimized several times, including increasing the number of hidden layers to improve the model's ability to capture complex wind speed patterns;

[0020] The LSTM model of this system consists of three parts: forget gate, input gate and output gate;

[0021] The forget gate of the LSTM model determines which information in the previous time step will be retained and which will be discarded. The calculation formula of the forget gate is:

[0022]

[0023] f t is the output of the current time step of the forget gate of the LSTM model, ranging between [0,1], W f (t) is the weight matrix of the forget gate of the LSTM model at the current time step, h t-1 is the hidden state at the previous time step, is the input of the current time step, b f is the bias term of the forget gate of the LSTM model, and σ is the activation function;

[0024] The calculation formula of the LSTM model input gate is:

[0025]

[0026] i t is the output of the current time step of the LSTM model input gate, ranging between [0,1], W i (t) is the weight matrix of the LSTM model input gate at the current time step, b i is the bias term of the input gate of the LSTM model;

[0027] The calculation formula of the LSTM model output gate is:

[0028]

[0029] o t The output of the LSTM model output gate at the current time step, W o (t) is the weight matrix of the LSTM model output gate at the current time step, b o is the bias term of the output gate;

[0030] Among them, the LSTM model forget gate weight matrix W f (t), LSTM model input gate weight matrix W i (t) and the LSTM model output gate weight matrix W o The update methods of (t) are:

[0031]

[0032] Among them, W f (t+1), W i (t+1) and W o (t+1) are the weight matrices for the next time step;

[0033] and It is the gradient of the loss function to the weight matrix, reflecting the sensitivity of the prediction error of the model to the weight matrix, and is used to know the direction and magnitude of the weight update;

[0034] Represents the wind speed data predicted by the LSTM model for the observed data D The probability distribution of , which reflects the uncertainty estimate of the current time step t for the future time step t+1;

[0035] Overall, these update formulas are designed to ensure that the LSTM model can adjust its weights appropriately based on past data and model uncertainty, so that the model can better predict future outputs. By combining the gradient information of the loss function and the uncertainty of the prediction, the LSTM model can gradually optimize its weights during the training process, making the final prediction results more accurate and reliable.

[0036] Repeat the above steps until the model converges.

[0037] 3. Power generation prediction module

[0038] Use the trained model data for power generation prediction;

[0039] And according to the model evaluation results and actual application feedback, the model structure and data preprocessing process are continuously optimized;

[0040] Feed new data and information about new environmental changes back into the model for periodic retraining to adapt to changes in the environment and demand.

[0041] In the daily operation of the wind farm, the system continuously predicts the power generation in the next 24 hours and transmits the prediction results to the grid dispatch center. The dispatch center optimizes the distribution of power load based on the prediction results, ensuring the stability of the grid and the power generation efficiency of the wind farm.

[0042] During the forecasting process, the system automatically identified and processed outliers in the weather data, ensuring the reliability of the forecast results.

[0043] Based on the forecast data provided by the system, the dispatch center adjusted the backup power plan in advance when wind power generation fluctuated significantly, significantly reducing the power dispatch costs caused by wind power fluctuations.

[0044] The system also provides credibility indicators for prediction results to help dispatchers assess prediction risks when making decisions, thereby further improving the accuracy and reliability of scheduling.

[0045] When applying the trained model to power generation forecasting, you first need to load the trained deep learning model and ensure that the model parameters and architecture are consistent with those during training. In addition, you also need to load the statistical parameters used for standardization or normalization so that the input data can be preprocessed in the same way.

[0046] In the data preparation stage, the latest input data, such as weather data, historical power generation data, and equipment status data, are collected to ensure the timeliness and completeness of the data. Next, the new data is preprocessed, including handling missing values ​​(such as using interpolation or other methods), outliers (such as deletion or replacement), and standardization or normalization (using the statistical parameters of the training set). The preprocessed new data is then input into the model for prediction to generate a forecast of future power generation. A preliminary quality check is performed on the forecast results to ensure that there are no obvious abnormal forecast values. Finally, the model's forecast results are passed to the power grid dispatch system or other decision-making system for actual power generation management and dispatch optimization, and the forecast results and actual power generation data are recorded for subsequent model evaluation and optimization.

[0047] According to the prediction performance of the model, the prediction results are compared with the actual power generation data by evaluating the performance of the model, analyzing the causes of prediction errors, and identifying the weaknesses of the model, such as prediction deviations under specific weather conditions. Through the feedback information collected in actual applications, such as the dependence of the dispatch system on the prediction and the user's satisfaction with the prediction results, it is determined whether the model structure or data preprocessing process needs to be adjusted.

[0048] When optimizing the model structure, consider adjusting the model architecture based on the evaluation results and feedback, such as increasing or decreasing the number of neural network layers, changing the activation function, trying different model types, or introducing new features such as finer time features or more complex meteorological data. In addition, adjust the existing feature engineering process to better capture the key patterns in the data and add regularization techniques such as Dropout or L2 regularization to prevent overfitting.

[0049] When optimizing data preprocessing, it is necessary to adjust the data preprocessing process according to the actual situation. For example, improve the missing value processing strategy and fill it with more context information; optimize the outlier detection and processing methods and use more complex detection algorithms; re-evaluate the standardization or normalization method, especially after introducing new data features, to ensure that the preprocessing method is still applicable.

[0050] In order to adapt to changes in the environment and demand, the latest power generation data, weather data, equipment status data, etc. are continuously collected and organized to ensure the real-time and accuracy of the data. After marking important events or environmental changes in the data (such as extreme weather and equipment maintenance), the model is retrained using the latest data to capture new patterns brought about by changes in the environment and demand. When retraining, the time window of the training data needs to be adjusted to ensure that the model can learn the latest trends and changes.

[0051] After retraining, evaluate the performance of the new model to ensure that its prediction accuracy is better than the old model or that it performs better in key application scenarios. Based on the evaluation results, decide whether to deploy the retrained model to the production environment. After the new model is deployed, it is necessary to continuously monitor its performance in the production environment and promptly identify and respond to new problems to ensure the stability and accuracy of power generation forecasts.

[0052] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0053] The present invention aims to propose a system and method for forecasting renewable energy power generation based on big data and AI technology. The system combines data cleaning, feature engineering, deep learning models and their optimization strategies, and can handle the uncertainty and volatility in renewable energy power generation data, provide more accurate short-term and long-term power generation forecasts, help optimize power grid scheduling, and reduce the risks caused by power generation fluctuations.

[0054] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0056] Figure 1 It is a system schematic diagram of the present invention. DETAILED DESCRIPTION

[0057] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0058] like Figure 1 As shown, the present invention discloses a new energy power generation prediction system based on big data and AI, including the following contents:

[0059] 1. Data preprocessing module

[0060] The entire wind speed dataset x collected from multiple wind speed sensors at the wind farm is processed, which contains seasonal variations and diurnal fluctuations.

[0061] All the acquired wind speed data sets x are processed, and the missing values ​​in all the wind speed data sets x are interpolated by mixed probability.

[0062]

[0063] in, is the missing value of wind speed data calculated by probability, t is the current time point, t-1 and t+1 are the time points before and after t respectively, and x t-1 and x t+1 are the values ​​of the front and back items of the missing value of the wind speed data, respectively. D is the observed wind speed data set. The model infers the missing value of wind speed based on the observed wind speed data set D. is a probability term, which is used to estimate reasonable interpolation values ​​based on the probability prediction of missing values ​​of wind speed data;

[0064] The probability term The calculation method is:

[0065]

[0066] where θ represents the parameter vector of the missing value probability model, Represents the missing value of wind speed data at the parameter vector θ and the current time point t The conditional probability distribution of , P(θ|D) represents the posterior distribution of parameter θ after observing the wind speed data set D;

[0067] This formula represents the predicted value of the wind speed data at the current time point t under the condition of observing the wind speed data set D. The probability distribution of This probability distribution is obtained by weighted integration of the posterior distribution P(θ|D) of the parameter θ;

[0068] After the mixed probability interpolation is completed, the wind speed data x t Standardization

[0069]

[0070] Among them, μ x is the mean of the wind speed data set x, σ x is the standard deviation of the wind speed data set x. The standardized wind speed data The mean is 0 and the standard deviation is 1.

[0071] The mixed probability interpolation method was used to handle the missing values ​​in the wind speed data to ensure the continuity and reliability of the wind speed data. The standardization step used the mean and standard deviation of the historical data of the wind farm.

[0072] 2. Model training module

[0073] The preprocessed wind speed data was used as input and trained using the LSTM model. The model architecture was optimized several times, including increasing the number of hidden layers to improve the model's ability to capture complex wind speed patterns;

[0074] The LSTM model of this system consists of three parts: forget gate, input gate and output gate;

[0075] The forget gate of the LSTM model determines which information in the previous time step will be retained and which will be discarded. The calculation formula of the forget gate is:

[0076]

[0077] f t is the output of the current time step of the forget gate of the LSTM model, ranging between [0,1], W f (t) is the weight matrix of the forget gate of the LSTM model at the current time step, h t-1 is the hidden state at the previous time step, is the input of the current time step, b f is the bias term of the forget gate of the LSTM model, and σ is the activation function;

[0078] The calculation formula of the LSTM model input gate is:

[0079]

[0080] i t is the output of the current time step of the LSTM model input gate, ranging between [0,1], W i (t) is the weight matrix of the LSTM model input gate at the current time step, b i is the bias term of the input gate of the LSTM model;

[0081] The calculation formula of the LSTM model output gate is:

[0082]

[0083] o t The output of the LSTM model output gate at the current time step, W o (t) is the weight matrix of the LSTM model output gate at the current time step, b o is the bias term of the output gate;

[0084] Among them, the LSTM model forget gate weight matrix W f (t), LSTM model input gate weight matrix W i (t) and the LSTM model output gate weight matrix W o The update methods of (t) are:

[0085]

[0086] Among them, W f (t+1), W i (t+1) and W o (t+1) are the weight matrices for the next time step;

[0087] and It is the gradient of the loss function to the weight matrix, reflecting the sensitivity of the model's prediction error to the weight matrix, and is used to know the direction and magnitude of the weight update;

[0088] Represents the wind speed data predicted by the LSTM model for the observed data D The probability distribution of , which reflects the uncertainty estimate of the current time step t for the future time step t+1;

[0089] Overall, these update formulas are designed to ensure that the LSTM model can adjust its weights appropriately based on past data and model uncertainty, so that the model can better predict future outputs. By combining the gradient information of the loss function and the uncertainty of the prediction, the LSTM model can gradually optimize its weights during the training process, making the final prediction results more accurate and reliable.

[0090] Repeat the above steps until the model converges.

[0091] 3. Power generation prediction module

[0092] Use the trained model data for power generation prediction;

[0093] And according to the model evaluation results and actual application feedback, the model structure and data preprocessing process are continuously optimized;

[0094] Feed new data and information about new environmental changes back into the model for periodic retraining to adapt to changes in the environment and demand.

[0095] In the daily operation of the wind farm, the system continuously predicts the power generation in the next 24 hours and transmits the prediction results to the grid dispatch center. The dispatch center optimizes the distribution of power load based on the prediction results, ensuring the stability of the grid and the power generation efficiency of the wind farm.

[0096] During the forecasting process, the system automatically identified and processed outliers in the weather data, ensuring the reliability of the forecast results.

[0097] Based on the forecast data provided by the system, the dispatch center adjusted the backup power plan in advance when wind power generation fluctuated significantly, significantly reducing the power dispatch costs caused by wind power fluctuations.

[0098] The system also provides credibility indicators for prediction results to help dispatchers assess prediction risks when making decisions, thereby further improving the accuracy and reliability of scheduling.

[0099] When applying the trained model to power generation forecasting, you first need to load the trained deep learning model and ensure that the model parameters and architecture are consistent with those during training. In addition, you also need to load the statistical parameters used for standardization or normalization so that the input data can be preprocessed in the same way.

[0100] In the data preparation stage, the latest input data, such as weather data, historical power generation data, and equipment status data, are collected to ensure the timeliness and completeness of the data. Next, the new data is preprocessed, including handling missing values ​​(such as using interpolation or other methods), outliers (such as deletion or replacement), and standardization or normalization (using the statistical parameters of the training set). The preprocessed new data is then input into the model for prediction to generate a forecast of future power generation. A preliminary quality check is performed on the forecast results to ensure that there are no obvious abnormal forecast values. Finally, the model's forecast results are passed to the power grid dispatch system or other decision-making system for actual power generation management and dispatch optimization, and the forecast results and actual power generation data are recorded for subsequent model evaluation and optimization.

[0101] According to the prediction performance of the model, the prediction results are compared with the actual power generation data by evaluating the performance of the model, analyzing the causes of prediction errors, and identifying the weaknesses of the model, such as prediction deviations under specific weather conditions. Through the feedback information collected in actual applications, such as the dependence of the dispatch system on the prediction and the user's satisfaction with the prediction results, it is determined whether the model structure or data preprocessing process needs to be adjusted.

[0102] When optimizing the model structure, consider adjusting the model architecture based on the evaluation results and feedback, such as increasing or decreasing the number of neural network layers, changing the activation function, trying different model types, or introducing new features such as finer time features or more complex meteorological data. In addition, adjust the existing feature engineering process to better capture the key patterns in the data and add regularization techniques such as Dropout or L2 regularization to prevent overfitting.

[0103] When optimizing data preprocessing, you need to adjust the data preprocessing process according to the actual situation. For example, you can improve the missing value processing strategy and fill it with more context information; optimize the outlier detection and processing methods and use more complex detection algorithms; re-evaluate the standardization or normalization method, especially after introducing new data features, to ensure that the preprocessing method is still applicable.

[0104] In order to adapt to changes in the environment and demand, the latest power generation data, weather data, equipment status data, etc. are continuously collected and organized to ensure the real-time and accuracy of the data. After marking important events or environmental changes in the data (such as extreme weather and equipment maintenance), the model is retrained using the latest data to capture new patterns brought about by changes in the environment and demand. When retraining, the time window of the training data needs to be adjusted to ensure that the model can learn the latest trends and changes.

[0105] After retraining, evaluate the performance of the new model to ensure that its prediction accuracy is better than the old model or that it performs better in key application scenarios. Based on the evaluation results, decide whether to deploy the retrained model to the production environment. After the new model is deployed, it is necessary to continuously monitor its performance in the production environment and promptly identify and respond to new problems to ensure the stability and accuracy of power generation forecasts.

[0106] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A new energy power generation prediction system based on big data and AI, characterized in that: include: Data preprocessing module: preprocess the acquired power generation data, perform mixed probability interpolation on missing values, and standardize the data; Model training module: input the preprocessed data into the LSTM model for training until the model converges; Power generation prediction module: The trained model data is used for power generation prediction, and the model is continuously optimized based on the new data obtained.

2. The new energy power generation prediction system based on big data and AI according to claim 1 is characterized in that: The data preprocessing module comprises: The entire wind speed dataset x collected from multiple wind speed sensors in the wind farm is processed, and the missing values ​​in the entire wind speed dataset x are interpolated using mixed probability. in, is the missing value of wind speed data calculated by probability, t is the current time point, t-1 and t+1 are the time points before and after t respectively, and x t-1 and x t+1 are the values ​​of the front and back items of the missing value of the wind speed data, respectively. D is the observed wind speed data set. The model infers the missing value of wind speed based on the observed wind speed data set D. is a probability term, which is used to estimate reasonable interpolation values ​​based on the probability prediction of missing values ​​of wind speed data; The probability term The calculation method is: where θ represents the parameter vector of the missing value probability model, Represents the missing value of wind speed data at the parameter vector θ and the current time point t The conditional probability distribution of , P(θ|D) represents the posterior distribution of parameter θ after observing the wind speed data set D; This formula represents the predicted value of the wind speed data at the current time point t under the condition of observing the wind speed data set D. The probability distribution of This probability distribution is obtained by weighted integration of the posterior distribution P(θ|D) of the parameter θ; After the mixed probability interpolation is completed, the wind speed data x t Standardization Among them, μ x is the mean of the wind speed data set x, σ x is the standard deviation of the wind speed data set x, the standardized wind speed data The mean is 0 and the standard deviation is 1.

3. The new energy power generation prediction system based on big data and AI according to claim 1 is characterized in that: The model training module includes: The LSTM model of this system consists of three parts: forget gate, input gate and output gate; The forget gate of the LSTM model determines which information in the previous time step will be retained and which will be discarded. The calculation formula of the forget gate is: f t is the output of the current time step of the forget gate of the LSTM model, ranging between [0,1], W f (t) is the weight matrix of the forget gate of the LSTM model at the current time step, h t-1 is the hidden state at the previous time step, is the standardized wind speed data, b f is the bias term of the forget gate of the LSTM model, and σ is the activation function; The calculation formula of the LSTM model input gate is: i t is the output of the current time step of the LSTM model input gate, ranging between [0,1], W i (t) is the weight matrix of the LSTM model input gate at the current time step, b i is the bias term of the input gate of the LSTM model; The calculation formula of the LSTM model output gate is: o t The output of the LSTM model output gate at the current time step, W o (t) is the weight matrix of the LSTM model output gate at the current time step, b o is the bias term of the output gate; Among them, the LSTM model forget gate weight matrix W f (t), LSTM model input gate weight matrix W i (t) and the LSTM model output gate weight matrix W o The update methods of (t) are: Among them, W f (t+1), W i (t+1) and W o (t+1) are the weight matrices for the next time step; and It is the gradient of the loss function to the weight matrix, reflecting the sensitivity of the model's prediction error to the weight matrix, and is used to know the direction and magnitude of the weight update; Represents the wind speed data predicted by the LSTM model for the observed data D The probability distribution of , which reflects the uncertainty estimate of the current time step t for the future time step t+1; Repeat the above steps until the model converges.

4. The new energy power generation prediction system based on big data and AI according to claim 1 is characterized in that: The power generation prediction module comprises: The power generation prediction unit uses the trained model data for power generation prediction; A model structure adjustment unit, used to dynamically adjust the structure of the deep learning model based on evaluation results and feedback information; Data preprocessing optimization unit, used to adjust the data preprocessing process according to actual application feedback and evaluation results, including improving missing value processing strategies, optimizing outlier detection and processing methods, and re-evaluating standardization or normalization methods; Data update unit, used to collect and integrate the latest power generation data, weather data and equipment status data in real time to ensure the timeliness and integrity of the data; An environmental change capture unit, which is used to identify and mark important environmental changes or events and incorporate this information into the retraining process; A periodic retraining unit that periodically retrains the model based on the latest data and environment changes, adjusting the time window of the training data to learn new trends and patterns; The transfer learning unit is used to apply transfer learning techniques to use previously trained model parameters as the initial parameters for new training to reduce training time and improve the adaptability of the model.

Images