Electrical load prediction method and system
By integrating multiple prediction models and cross-verification methods, the power load prediction model is optimized, and the problem of inaccurate prediction of power load in the existing technology is solved, and the prediction accuracy and stability is achieved, supporting the safe and economic operation of the power system.
Patent Information
- Application Number
- CN202410198045.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-02-22
- Publication Date
- 2025-04-22
AI Technical Summary
The prior art is difficult to provide accurate and stable power load prediction, resulting in insufficient grid capacity or over-utilization of equipment, affecting the safety and economics of the power system.
Various prediction models (such as linear regression, support vector machines, neural networks, time series models) are used for integrated learning, the prediction results of multiple models are weighted and averaged, and the model parameters are optimized through cross-validation and online learning to improve the accuracy and stability of prediction.
It significantly improves the accuracy and stability of power load prediction, can more accurately reflect power load changes, help power system scheduling and production planning, reduce power generation costs, and improve system safety and economic benefits.
Smart Images
Figure CN119864790A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of electric power systems, and in particular relates to a method and system for predicting electric load. Background Art
[0002] Electricity load forecasting refers to the process of estimating the demand on the power system and studying the impact of related factors on the power load by analyzing historical data of power load based on changes in power load and external factors, using specific mathematical methods or establishing mathematical models. The purpose of load forecasting is to provide load development status and levels, and to determine the power supply and consumption in each power supply area and each planned year, the maximum load of power supply and consumption, and the total load development level in the planned area, and to determine the load composition of each planned year. Power system load forecasting is related to the dispatching operation and production plan of the power system. Accurate load forecasting helps to improve the safety and stability of the system and can reduce the cost of power generation. Power load forecasting is an important task for power companies. Based on accurate load forecasting, the internal operations of the power system can be arranged economically and reasonably. The start and stop of the generating set, maintain the safety and stability of the power grid operation, reduce unnecessary rotating reserve capacity, reasonably arrange the unit maintenance plan, maintain the normal production and life of the society, reduce the power generation cost to a limited extent, and improve the economic and social benefits; the accuracy of the prediction directly affects the rationality of investment and operation, and is directly related to the safe, economical and reliable operation of the power grid; if the load forecast is lower than the actual, it will lead to insufficient power grid capacity and fail to meet the normal power supply requirements of users, and may even cause power shortages; on the other hand, if the load forecast is higher than the actual, it will lead to the installation of too many equipment that cannot be fully utilized, thereby causing a waste of investment; therefore, it is very necessary to provide a power load prediction method and system that integrates the prediction results of multiple models, has high prediction accuracy and good prediction stability. Summary of the invention
[0003] The purpose of the present invention is to overcome the deficiencies of the prior art and to provide a method and system for predicting power load which integrates prediction results of multiple models, has high prediction accuracy and good prediction stability.
[0004] The object of the present invention is achieved by: a method for predicting power load, the method comprising the following steps:
[0005] Step 1: Collect multi-dimensional big data parameters of electricity consumption data to obtain a big data set of electricity consumption;
[0006] Step 2: Preprocess the collected electricity consumption big data set to obtain a data processing set;
[0007] Step 3: Extracting features related to power load from the data processing set to obtain a data feature set;
[0008] Step 4: Based on the data feature set, multiple prediction models are used to integrate and learn multiple different prediction models, and the prediction results of multiple models are integrated to obtain the optimal prediction model set;
[0009] Step 5: Use the optimal prediction model set, input prediction parameters, and obtain a prediction result set;
[0010] Step 6: Evaluate the prediction results, set evaluation conditions, and determine whether the prediction structure set meets the evaluation condition set;
[0011] Step 7: If the prediction result set satisfies the evaluation condition set, the prediction result is output; if the prediction result set does not satisfy the evaluation condition set, the prediction model set is optimized and adjusted based on the evaluation condition set to obtain an optimization result set and an optimization analysis result set;
[0012] Step 8: Output the optimization result set and the optimization analysis result set;
[0013] Step 9: Use online learning and incremental learning methods to receive new data and update model parameters in real time.
[0014] The multi-dimensional big data parameters of the electricity consumption data in step 1 include:
[0015] Collect daily, weekly and monthly historical electricity load data through the power company's database to obtain the first electricity big data parameter;
[0016] Collect meteorological data, including temperature, humidity, and rainfall, to obtain the second electricity consumption big data parameters;
[0017] Collect holiday data, including statutory holidays and weekends, to obtain third-party electricity consumption big data parameters;
[0018] Collecting electricity market price data to obtain fourth electricity consumption big data parameters;
[0019] Collect real-time power load data, including power load data in different regions and time periods, to obtain the fifth power consumption big data parameters;
[0020] The first electricity consumption big data parameter, the second electricity consumption big data parameter, the third electricity consumption big data parameter, the fourth electricity consumption big data parameter and the fifth electricity consumption big data parameter are taken as the electricity consumption big data set.
[0021] The step 2 of preprocessing the collected electricity consumption big data set includes the following steps:
[0022] Step 2.1: Use data cleaning methods based on statistical principles to remove duplicate data and fill in missing values;
[0023] Step 2.2: Arrange the data and normalize it to make the data of different dimensions comparable;
[0024] Step 2.3: Standardize the data and transform it into data with zero mean and unit variance.
[0025] The step 3 of extracting features related to power load from the data processing set includes:
[0026] Extracting time series features from historical power load data, such as daily, monthly, and annual power consumption, to obtain first data feature parameters;
[0027] Extracting weather characteristics, such as temperature, humidity, and rainfall, from the meteorological data to obtain second data characteristic parameters;
[0028] Extract holiday features from holiday data, such as whether it is a statutory holiday or a weekend, to obtain third data feature parameters;
[0029] Extracting price characteristics, such as price fluctuation, from the electricity market price data to obtain a fourth data characteristic parameter;
[0030] Extracting load characteristics from real-time power load data, such as power load data in different regions and different time periods, to obtain fifth data characteristic parameters;
[0031] The first data feature parameter, the second data feature parameter, the third data feature parameter, the fourth data feature parameter and the fifth data feature parameter are taken as the data feature set.
[0032] In step 4, multiple prediction models are used according to the data feature set, multiple different prediction models are integrated and learned, and the prediction results of multiple models are integrated, including:
[0033] Using the data feature set, a linear regression model is built to predict future power load;
[0034] Using the data feature set, a support vector machine (SVM) model is built to predict future power load;
[0035] Using data feature sets, a neural network model is built to predict future power loads;
[0036] Use data feature sets to build time series models and predict future power loads;
[0037] The above multiple models are integrated and learned, and the prediction results of each model are weighted averaged, where the weight is determined by the performance of each model on the training set. The prediction results of multiple groups of models are fused to obtain the final test result.
[0038] The setting of evaluation conditions in step 6 includes:
[0039] Compare the prediction results with the actual load data, compare the prediction results with the actual load data, and obtain the error rate, which is used as the evaluation condition set. The error rate formula is: Where ρ is the error rate; T is the decision value; μ i is the actual load data; w i is the predicted load data; R is the conversion symbol.
[0040] The step 1 of optimizing the analysis result set based on the evaluation condition set includes:
[0041] Adjust and optimize each prediction model, add features, and adjust model parameters;
[0042] Cross-validate each prediction model, divide the data set into multiple subsets, and perform model training and prediction on each subset to evaluate the prediction performance of the model;
[0043] Optimize the preferred prediction model set, select the appropriate fusion strategy, and improve the prediction accuracy.
[0044] A power load forecasting system, the system comprising a data acquisition module, a data processing module, a feature extraction module, a model training module, a prediction output module, an evaluation module, an optimization analysis module, an optimization prediction output module and a learning module, the power load forecasting system is used to execute the above-mentioned power load forecasting method.
[0045] The data collection module is used to collect multi-dimensional big data parameters of electricity consumption data to obtain a large set of electricity consumption data;
[0046] The data processing module is used to pre-process the collected electricity consumption big data set to obtain a data processing set;
[0047] The feature extraction module is used to extract features related to the power load from the data processing set to obtain a data feature set;
[0048] The model training module is used to adopt multiple prediction models, integrate and learn multiple different prediction models, and fuse the prediction results of multiple models to obtain a set of optimal prediction models;
[0049] The prediction output module is used to obtain a set of prediction results;
[0050] The evaluation module is used to set evaluation conditions;
[0051] The optimization analysis module is used to output the prediction results when the prediction result set satisfies the evaluation condition set, and, if the prediction result set does not satisfy the evaluation condition set, optimize and adjust the prediction model set based on the evaluation condition set to obtain an optimization result set and an optimization analysis result set;
[0052] The optimization prediction output module is used to output the optimization result set and the optimization analysis result set;
[0053] The learning module is used to receive new data and update model parameters in real time using online learning and incremental learning methods.
[0054] Beneficial effects of the present invention: The present invention is a method and system for predicting electricity load. During use, the present invention adopts multiple prediction models. Through the method of integrated learning, the prediction results of multiple models are integrated, which can significantly improve the accuracy and stability of the prediction. The prediction results of each model are weighted averaged, wherein the weight is determined by the performance of each model on the training set. In this way, the advantages of each model can be fully utilized while avoiding their respective limitations. In addition, a cross-validation method is also used to evaluate the accuracy and stability of the prediction. By dividing the data set into multiple subsets and performing model training and prediction on each subset respectively, the performance of the model can be more accurately evaluated, providing more reliable support for practical applications. The present invention has the advantages of integrating the prediction results of multiple models, high prediction accuracy and good prediction stability. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] Figure 1 It is a flow chart of the present invention.
[0056] Figure 2 It is a system block diagram of the present invention. DETAILED DESCRIPTION
[0057] The present invention will be further described below in conjunction with the accompanying drawings.
[0058] Example 1
[0059] like Figure 1 As shown, a method for predicting power load is applied to a power load prediction system, and the method comprises the following steps:
[0060] Step S100: Collect multi-dimensional big data parameters of electricity consumption data to obtain a set of electricity consumption big data;
[0061] In this embodiment, collecting electricity consumption data is a crucial step. The historical power load data that needs to be collected usually includes daily or hourly power consumption, and other factors related to power consumption, such as weather, holidays, etc. These data can usually be obtained through the power company's historical database or a public data platform. In addition, if there are some factors that have a special impact on the power load, such as major policy changes, economic events, etc., relevant data also need to be collected. In addition to the above conventional data collection methods, more detailed and targeted data can be obtained through questionnaires, field tests, etc. For example, questionnaires can be conducted on different types of enterprises and residents in different regions to understand their power usage habits, equipment usage, etc. Through field tests, the power usage of different types of equipment under various conditions can be understood;
[0062] Step S100 includes:
[0063] S110: Collect daily, weekly, and monthly historical power load data from the power company's database to obtain a first power load big data parameter;
[0064] S120: Collect meteorological data, including temperature, humidity, and rainfall, to obtain second electricity consumption big data parameters;
[0065] S130: Collect holiday data, including statutory holidays and weekends, to obtain third electricity consumption big data parameters;
[0066] S140: Collecting electricity market price data to obtain fourth electricity consumption big data parameters;
[0067] S150: Collecting real-time power load data, including power load data in different regions and different time periods, to obtain a fifth power consumption big data parameter;
[0068] S160: taking the first power consumption big data parameter, the second power consumption big data parameter, the third power consumption big data parameter, the fourth power consumption big data parameter and the fifth power consumption big data parameter as the power consumption big data set;
[0069] In this embodiment, the following points need to be noted when collecting electricity consumption data:
[0070] a. The data needs to be representative and comprehensive to a certain extent, and be able to reflect various factors that affect power load;
[0071] b. When collecting data, it is necessary to clarify the purpose and purpose of the data to avoid collecting too much useless data;
[0072] c. The collection of data must comply with relevant laws, regulations and privacy policies to ensure the legal and compliant use of data;
[0073] Step S200: pre-processing the collected electricity consumption big data set to obtain a data processing set;
[0074] In this embodiment, the data needs to be cleaned and processed to eliminate errors and outliers, which may include processing missing values, deleting duplicate data, checking the consistency and validity of the data, etc.
[0075] Step S200 includes:
[0076] S210: Use data cleaning methods based on statistical principles to remove duplicate data and fill in missing values;
[0077] S220: sorting the data and normalizing the data so that data of different dimensions are comparable;
[0078] S230: Standardize the data to convert the data into data with zero mean and unit variance;
[0079] In this embodiment, the purpose of data cleaning is to remove duplicate data, erroneous data and outliers to improve the quality of data; the purpose of data standardization is to unify the data of different indicators under one standard to facilitate subsequent analysis and comparison; the purpose of removing outliers is to prevent individual data from having too great an impact on the overall analysis;
[0080] Step S300: extracting features related to power load from the data processing set to obtain a data feature set;
[0081] In this embodiment, by observing the distribution and trend of the data, the characteristics of the power load can be preliminarily understood. For example, methods such as histograms and box plots can be used to observe the distribution of the data, or time series analysis can be used to observe the trend of the power load. At this stage, it is necessary to extract features related to the power load from the data. These features may include time series data (such as daily, weekly, monthly and other periodic data), weather data (such as temperature, humidity, air pressure, etc.) and other factors that may affect the power load;
[0082] Step S300 includes:
[0083] S310: extracting time series features from historical power load data, such as daily, monthly, and annual power consumption, to obtain a first data feature parameter;
[0084] S320: extracting weather characteristics, such as temperature, humidity, and rainfall, from the meteorological data to obtain second data characteristic parameters;
[0085] S330: extracting holiday features from the holiday data, such as whether it is a statutory holiday or a weekend, and obtaining a third data feature parameter;
[0086] S340: extracting price characteristics, such as price fluctuation, from the power market price data to obtain a fourth data characteristic parameter;
[0087] S350: extracting load characteristics from the real-time power load data, such as power load data in different regions and different time periods, to obtain a fifth data characteristic parameter;
[0088] S360: taking the first data feature parameter, the second data feature parameter, the third data feature parameter, the fourth data feature parameter and the fifth data feature parameter as the data feature set;
[0089] In this embodiment, there may be many extracted features, but not all of them have a direct impact on the power load. Therefore, feature selection is required to determine which features can best reflect the changes in the power load. This usually requires the use of some statistical methods or machine learning algorithms for feature selection.
[0090] Step S400: Based on the data feature set, multiple prediction models are adopted to perform integrated learning on multiple different prediction models, and the prediction results of multiple models are integrated to obtain a preferred prediction model set;
[0091] In this embodiment, using multiple prediction models and performing ensemble learning is an effective strategy to increase the diversity and accuracy of the model. The basic idea of ensemble learning is to fuse the prediction results of multiple models to obtain a better result than a single model. The prediction result of each model is multiplied by a weight coefficient, and then the sum is obtained to obtain the final prediction result. The weight coefficient can be determined based on the accuracy of the model or other metrics;
[0092] Step S400 includes:
[0093] S410: Using the data feature set, establish a linear regression model to predict future power load;
[0094] S420: Using the data feature set, establishing a support vector machine (SVM) model to predict future power load;
[0095] S430: Using the data feature set, establishing a neural network model to predict future power load;
[0096] S440: Use the data feature set to establish a time series model to predict future power load;
[0097] S450: performing integrated learning on the above multiple models, performing weighted average on the prediction results of each model, wherein the weight is determined by the performance of each model on the training set, and fusing the prediction results of multiple groups of models to obtain the final test result;
[0098] In this embodiment, a bagging method can be used to implement ensemble learning. Bagging (bootstrap aggregation) generates multiple data sets by introducing resampling technology, and then trains a model for each data set. When predicting, each model predicts the sample, and then a weighted method is used to fuse the results.
[0099] Step S500: using the preferred prediction model set, inputting prediction parameters, and obtaining a prediction result set;
[0100] In this embodiment, the constructed optimal prediction model set is used to predict the future power load, for example, predict the power consumption in the next day or week, and the prediction results are output in the form of charts or reports to facilitate decision-makers to make decisions;
[0101] Step S600: Evaluate the prediction results, set evaluation conditions, and determine whether the prediction structure set meets the evaluation condition set;
[0102] In this embodiment, the prediction result and the actual load data are compared to obtain an error rate, and the error rate is used as an evaluation condition set. The error rate formula is: Where ρ is the error rate; T is the decision value (for example, if the power load is too high, the power will be cut off); μ i is the actual load data; w i is the predicted load data; R is the conversion symbol; the smaller the error rate, the closer the predicted result is to the actual load data;
[0103] Step S700: if the prediction result set satisfies the evaluation condition set, the prediction result is output; if the prediction result set does not satisfy the evaluation condition set, the prediction model set is optimized and adjusted based on the evaluation condition set to obtain an optimization result set and an optimization analysis result set;
[0104] In this embodiment, step S700 includes:
[0105] S710: Adjust and optimize each prediction model, add features, and adjust model parameters;
[0106] S720: Perform cross-validation on each prediction model, divide the data set into multiple subsets, and perform model training and prediction on each subset to evaluate the prediction performance of the model;
[0107] S730: Optimize the preferred prediction model set, select a suitable fusion strategy, and improve prediction accuracy;
[0108] Cross-validation can help evaluate the generalization ability of the model and prevent overfitting. In addition, regularization techniques, feature selection and other methods can be used to optimize the model to reduce the complexity of the model and improve the robustness of the model. During the optimization process, care should be taken not to over-optimize. Over-optimization will cause the model to be too complex, resulting in overfitting and affecting the generalization ability of the model. Do not ignore errors. During the optimization process, factors such as prediction error and computational complexity need to be considered at the same time to find a balance. The actual situation should be considered. The goal of optimization is to improve prediction accuracy, but the actual situation also needs to be considered, such as computing resources, time cost and other factors.
[0109] Step S800: outputting the optimization result set and the optimization analysis result set;
[0110] Step S900: Using online learning and incremental learning methods, receiving new data in real time and updating model parameters.
[0111] In this embodiment, the online learning and incremental learning methods allow the model parameters to be updated in real time when new data is received, rather than the need to retrain the entire model. This method is particularly effective when processing large-scale data sets because it can save computing resources and time; the model parameters need to be initialized and then new data begins to be received. When new data is received, the new data will be used to update the parameters of the preferred prediction model set. This is usually achieved through some optimization algorithms, such as the gradient descent method. After each update of the model parameters, the performance of the model needs to be evaluated. This process will continue in a loop, that is, when we have new data, the above steps will be repeated.
[0112] The present invention is a method and system for predicting electricity load. In use, the present invention adopts multiple prediction models. Through the method of integrated learning, the prediction results of multiple models are integrated, which can significantly improve the accuracy and stability of the prediction. The prediction results of each model are weighted averaged, wherein the weight is determined by the performance of each model on the training set. In this way, the advantages of each model can be fully utilized while avoiding their respective limitations. In addition, a cross-validation method is also used to evaluate the accuracy and stability of the prediction. By dividing the data set into multiple subsets and performing model training and prediction on each subset respectively, the performance of the model can be more accurately evaluated, providing more reliable support for practical applications. The present invention has the advantages of integrating the prediction results of multiple models, high prediction accuracy and good prediction stability.
[0113] Example 2
[0114] like Figure 2As shown, a power load forecasting system includes a data acquisition module, a data processing module, a feature extraction module, a model training module, a prediction output module, an evaluation module, an optimization analysis module, an optimization prediction output module and a learning module. The power load forecasting system is used to execute the above-mentioned power load forecasting method.
[0115] The data collection module is used to collect multi-dimensional big data parameters of electricity consumption data to obtain a large set of electricity consumption data;
[0116] The data processing module is used to pre-process the collected electricity consumption big data set to obtain a data processing set;
[0117] The feature extraction module is used to extract features related to the power load from the data processing set to obtain a data feature set;
[0118] The model training module is used to adopt multiple prediction models, integrate and learn multiple different prediction models, and fuse the prediction results of multiple models to obtain a set of optimal prediction models;
[0119] The prediction output module is used to obtain a set of prediction results;
[0120] The evaluation module is used to set evaluation conditions;
[0121] The optimization analysis module is used to output the prediction results when the prediction result set satisfies the evaluation condition set, and, if the prediction result set does not satisfy the evaluation condition set, optimize and adjust the prediction model set based on the evaluation condition set to obtain an optimization result set and an optimization analysis result set;
[0122] The optimization prediction output module is used to output the optimization result set and the optimization analysis result set;
[0123] The learning module is used to receive new data and update model parameters in real time using online learning and incremental learning methods.
[0124] The present invention is a method and system for predicting electricity load. In use, the present invention adopts multiple prediction models. Through the method of integrated learning, the prediction results of multiple models are integrated, which can significantly improve the accuracy and stability of the prediction. The prediction results of each model are weighted averaged, wherein the weight is determined by the performance of each model on the training set. In this way, the advantages of each model can be fully utilized while avoiding their respective limitations. In addition, a cross-validation method is also used to evaluate the accuracy and stability of the prediction. By dividing the data set into multiple subsets and performing model training and prediction on each subset respectively, the performance of the model can be more accurately evaluated, providing more reliable support for practical applications. The present invention has the advantages of integrating the prediction results of multiple models, high prediction accuracy and good prediction stability.
Claims
1. A method for predicting power load, characterized in that: The method comprises the following steps: Step 1: Collect multi-dimensional big data parameters of electricity consumption data to obtain a big data set of electricity consumption; Step 2: Preprocess the collected electricity consumption big data set to obtain a data processing set; Step 3: Extracting features related to power load from the data processing set to obtain a data feature set; Step 4: Based on the data feature set, multiple prediction models are used to integrate and learn multiple different prediction models, and the prediction results of multiple models are integrated to obtain the optimal prediction model set; Step 5: Use the optimal prediction model set, input prediction parameters, and obtain a prediction result set; Step 6: Evaluate the prediction results, set evaluation conditions, and determine whether the prediction structure set meets the evaluation condition set; Step 7: If the prediction result set satisfies the evaluation condition set, the prediction result is output; if the prediction result set does not satisfy the evaluation condition set, the prediction model set is optimized and adjusted based on the evaluation condition set to obtain an optimization result set and an optimization analysis result set; Step 8: Output the optimization result set and the optimization analysis result set; Step 9: Use online learning and incremental learning methods to receive new data and update model parameters in real time.
2. A method for predicting power load according to claim 1, characterized in that: The multi-dimensional big data parameters of the electricity consumption data in step 1 include: Collect daily, weekly and monthly historical electricity load data through the power company's database to obtain the first electricity big data parameter; Collect meteorological data, including temperature, humidity, and rainfall, to obtain the second electricity consumption big data parameters; Collect holiday data, including statutory holidays and weekends, to obtain third-party electricity consumption big data parameters; Collecting electricity market price data to obtain fourth electricity consumption big data parameters; Collect real-time power load data, including power load data in different regions and time periods, to obtain the fifth power consumption big data parameters; The first electricity consumption big data parameter, the second electricity consumption big data parameter, the third electricity consumption big data parameter, the fourth electricity consumption big data parameter and the fifth electricity consumption big data parameter are taken as the electricity consumption big data set.
3. A method for predicting power load according to claim 1, characterized in that: The step 2 of preprocessing the collected electricity consumption big data set includes the following steps: Step 2.1: Use data cleaning methods based on statistical principles to remove duplicate data and fill in missing values; Step 2.2: Arrange the data and normalize it to make the data of different dimensions comparable; Step 2.3: Standardize the data and transform it into data with zero mean and unit variance.
4. A method for predicting power load according to claim 2, characterized in that: The step 3 of extracting features related to power load from the data processing set includes: Extracting time series features from historical power load data, such as daily, monthly, and annual power consumption, to obtain first data feature parameters; Extracting weather characteristics, such as temperature, humidity, and rainfall, from the meteorological data to obtain second data characteristic parameters; Extract holiday features from holiday data, such as whether it is a statutory holiday or a weekend, to obtain third data feature parameters; Extracting price characteristics, such as price fluctuation, from the electricity market price data to obtain a fourth data characteristic parameter; Extracting load characteristics from real-time power load data, such as power load data in different regions and different time periods, to obtain fifth data characteristic parameters; The first data feature parameter, the second data feature parameter, the third data feature parameter, the fourth data feature parameter and the fifth data feature parameter are taken as the data feature set.
5. The method for predicting power load according to claim 1, characterized in that: In step 4, multiple prediction models are used according to the data feature set, multiple different prediction models are integrated and learned, and the prediction results of multiple models are integrated, including: Using the data feature set, a linear regression model is built to predict future power load; Using the data feature set, a support vector machine model is established to predict future power load; Using data feature sets, a neural network model is built to predict future power loads; Use data feature sets to build time series models and predict future power loads; The above multiple models are integrated and learned, and the prediction results of each model are weighted averaged, where the weight is determined by the performance of each model on the training set. The prediction results of multiple groups of models are fused to obtain the final test result.
6. A method for predicting power load according to claim 1, characterized in that: The setting of evaluation conditions in step 6 includes: Compare the prediction results with the actual load data, compare the prediction results with the actual load data, and obtain the error rate, which is used as the evaluation condition set. The error rate formula is: Where ρ is the error rate; T is the decision value; μ i is the actual load data; w i is the predicted load data; R is the conversion symbol.
7. A method for predicting power load according to claim 1, characterized in that: The step 1 of optimizing the analysis result set based on the evaluation condition set includes: Adjust and optimize each prediction model, add features, and adjust model parameters; Cross-validate each prediction model, divide the data set into multiple subsets, and perform model training and prediction on each subset to evaluate the prediction performance of the model; Optimize the preferred prediction model set, select the appropriate fusion strategy, and improve the prediction accuracy.
8. A power load forecasting system, characterized in that: The system includes a data acquisition module, a data processing module, a feature extraction module, a model training module, a prediction output module, an evaluation module, an optimization analysis module, an optimization prediction output module and a learning module. The power load prediction system is used to execute a power load prediction method as described in any one of claims 1-7.
9. The power load forecasting system according to claim 8, characterized in that: The data collection module is used to collect multi-dimensional big data parameters of electricity consumption data to obtain a large set of electricity consumption data; The data processing module is used to pre-process the collected electricity consumption big data set to obtain a data processing set; The feature extraction module is used to extract features related to the power load from the data processing set to obtain a data feature set; The model training module is used to adopt multiple prediction models, integrate and learn multiple different prediction models, and fuse the prediction results of multiple models to obtain a set of optimal prediction models; The prediction output module is used to obtain a set of prediction results; The evaluation module is used to set evaluation conditions; The optimization analysis module is used to output the prediction results when the prediction result set satisfies the evaluation condition set, and, if the prediction result set does not satisfy the evaluation condition set, optimize and adjust the prediction model set based on the evaluation condition set to obtain an optimization result set and an optimization analysis result set; The optimization prediction output module is used to output the optimization result set and the optimization analysis result set; The learning module is used to receive new data and update model parameters in real time using online learning and incremental learning methods.