Load prediction method of power system based on big data

Abstract

The invention provides a big data-based load prediction method for a power system, and relates to the technical field of power system dispatching control. The big data-based load prediction method for the power system comprises the following steps: data collection: collecting historical load data, weather data, economic activity data and special event data of the power system; data preprocessing: cleaning and normalizing the collected data to ensure the data quality; feature extraction: extracting key features influencing the load from the original data by using a big data analysis technology; model construction: adopting a machine learning algorithm to construct a load prediction model; and model training and verification: training the model by using historical data, and verifying the accuracy of the model through a verification method. Through multi-source data fusion, the historical load data, the weather data, the economic activity data and the special event data are integrated, and the factors influencing the power load can be captured more comprehensively, so that the prediction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system dispatching and control, and in particular to a load forecasting method for a power system based on big data. Background Art

[0002] The power system is an electric energy production and consumption system consisting of power plants, power transmission and transformation lines, power distribution stations and power consumption. Its function is to convert primary energy in nature into electric energy through power generation devices, and then supply electric energy to various users through transmission, transformation and distribution. To achieve this function, the power system also has corresponding information and control systems at various links and levels to measure, regulate, control, protect, communicate and dispatch the production process of electric energy to ensure that users obtain safe and high-quality electric energy.

[0003] With the rapid development of social economy and population growth, electricity demand continues to rise, and the stability and reliability of power systems are becoming more and more important. Accurate prediction of power load is crucial to the planning, operation and management of power systems. Traditional load forecasting methods mainly rely on historical load data and simple statistical models. These methods have limitations when processing large-scale, multi-dimensional data, and it is difficult to capture the complex factors that affect power load, such as weather changes, economic activities, special events, etc. In addition, these methods often lack sufficient flexibility and adaptability, and it is difficult to cope with the growing complexity and dynamic changes of power systems. Therefore, technicians in this field provide a load forecasting method for power systems based on big data to solve the problems raised in the above background technology. Summary of the invention

[0004] 1. Technical issues to be solved

[0005] In view of the shortcomings of the prior art, the present invention provides a load forecasting method for a power system based on big data, which solves the problems of the prior art lacking sufficient flexibility and adaptability and having limitations in processing data.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: A load forecasting method for a power system based on big data, comprising:

[0008] Data collection, collecting historical load data, weather data, economic activity data and special event data of the power system;

[0009] Data preprocessing: cleaning and normalizing the collected data to ensure data quality;

[0010] Feature extraction: using big data analysis technology to extract key features that affect load from raw data;

[0011] Model construction, using machine learning algorithms to build load forecasting models;

[0012] Model training and verification: Use historical data to train the model and verify the accuracy of the model through verification methods;

[0013] Forecast implementation, real-time data is input into the trained model to perform load forecasting, and power dispatch and resource allocation are carried out based on the forecast results.

[0014] Preferably, the data collection comprises the following steps:

[0015] S1. Determine the data sources, which usually include power load data, weather data, economic activity data and special event data. The power load data is obtained from the power company's SCADA system, EMS or historical database; the weather data is obtained from the Meteorological Bureau or online weather API, including temperature, humidity, wind speed and precipitation; the economic activity data can be obtained from government statistical departments, industry reports or online economic databases; special event data is collected through news media, social media or public event calendar channels;

[0016] S2. Data collection can be divided into automatic collection and manual collection. You can use API, crawler technology or cooperate with data providers to achieve automatic data collection. For some data that cannot be obtained automatically, you may need to collect and enter it manually;

[0017] S3. Data formatting: The collected data needs to be formatted to ensure its consistency, time synchronization: ensure that the timestamps of all data are consistent to facilitate subsequent analysis, and data type conversion: convert the data into a format suitable for analysis;

[0018] S4. Data cleaning: For missing data, filling, deletion or interpolation methods can be used to identify and process outliers, including deletion, correction or retention of data points and marking, including the following calculation methods:

[0019] Mean: The sum of all data points divided by the number of data points,

[0020]

[0021] Standard Deviation: A measure of how far a data point deviates from the mean.

[0022]

[0023] Variance: The square of the standard deviation, which is also an indicator of the degree of dispersion of data;

[0024]

[0025] S5. Data integration, integrating data from different sources for joint analysis, merging data together based on timestamps or other key fields, and ensuring that data from different data sources are aligned in terms of events.

[0026] Preferably, the data preprocessing comprises the following steps:

[0027] S1. Missing value processing, deletion: delete records with missing values

[0028] Delete records = delete all rows with missing values

[0029] Fill: Fill missing values ​​with statistics or predictive models

[0030]

[0031]

[0032] S2. Data conversion, normalization, scaling the data to a standard range,

[0033]

[0034] Among them, x is the original value, x min and x max are the minimum and maximum values ​​of the feature respectively;

[0035] Standardization, converting the data into a distribution with a mean of 0 and a standard deviation of 1;

[0036]

[0037] Where μ is the mean of the feature and σ is the standard deviation;

[0038] S3. Data discretization, converting continuous variables into discrete variables. Commonly used methods include equal-width binning and equal-frequency binning. Divide the data range into several intervals, assign a value to each interval, and divide the data into several intervals according to the frequency of the data points;

[0039] S4. Feature encoding, converting categorical variables into a set of binary variables;

[0040] If x = i, then x one-hot =[0, 0, ..., 1, ..., 0]

[0041] The i-th position is 1, and the rest are 0;

[0042] Label encoding, which maps each category of a categorical variable to a unique integer;

[0043] Category A → 0, Category B → 1, ...

[0044] S5. Data dimensionality reduction, principal component analysis, converting possibly correlated variables into a set of linearly uncorrelated variables through orthogonal transformation;

[0045] PCA = Projecting data onto principal components

[0046] S6. Data enhancement, using feature crossover to combine two or more features to generate new features;

[0047] New feature = Feature A × Feature B

[0048] S7. Outlier processing: Data points with Z=±3 at Z-score identification are regarded as outliers.

[0049]

[0050] Preferably, the feature extraction includes the following calculation formula:

[0051] S1. Statistical feature extraction:

[0052] Mean, the average value of a data set, reflects the central tendency of the data;

[0053]

[0054] Variance, a measure of the dispersion of the data;

[0055]

[0056] The standard deviation, the square root of the variance, is also a measure of the dispersion of the data;

[0057]

[0058] Skewness, which describes the symmetry of the data distribution;

[0059]

[0060] Kurtosis, which describes how sharp or flat the data distribution is;

[0061]

[0062] S2. Time series feature extraction, calculating the statistical characteristics of the time series through sliding windows;

[0063] The autocorrelation coefficient measures the correlation between a time series and itself at different time lags;

[0064]

[0065] where k is the time lag;

[0066] S3. Frequency-based feature extraction, Fourier transform, convert time series data into frequency domain and extract frequency features;

[0067]

[0068] Where F(ω) is the frequency domain representation and f(t) is the time domain signal;

[0069] S4. Model-based feature extraction, principal component analysis, and dimensionality reduction techniques to extract the most important features;

[0070] PCA = Projecting data onto principal components

[0071] S5. Feature extraction based on machine learning, using random forest to select important features based on the average reduced impurity of feature splits in the tree model;

[0072] S6. Custom features, features built based on specific domain knowledge;

[0073] S7. Feature construction, feature intersection, combining two or more features to generate new features.

[0074] Preferably, the model building comprises the following steps:

[0075] S1. Select the model by selecting the random forest model, ensemble learning method, building multiple decision trees and voting, each tree is built based on a randomly selected feature subset;

[0076] S2. Train the model, use the training data set to train the model and find the optimal values ​​of the model parameters;

[0077] Loss function: measures the difference between the model's predicted value and the true value;

[0078] Mean Square Error:

[0079]

[0080] Logarithmic loss:

[0081]

[0082] S3. Model evaluation, use cross validation to evaluate the performance of the model, divide the dataset into k subsets, take turns using one of the subsets as the test set and the rest as the training set;

[0083] S4. Hyperparameter tuning, optimize model performance by adjusting the model's hyperparameters, using grid search and random search;

[0084] S5. Model testing, using an independent test set to evaluate the generalization ability of the model;

[0085] S6. Model deployment: deploy the trained model to the production environment to perform actual prediction tasks.

[0086] Preferably, the model training and verification comprises the following steps:

[0087] S1. Dataset division: divided into training set, validation set and test set. The training set is used for model learning, the validation set is used for model tuning and preventing overfitting, and the test set is used for evaluating the final performance of the model.

[0088] S2. Model training, which is divided into initializing parameters, forward propagation, calculating losses, and back propagation;

[0089] Initialize parameters to set initial values ​​for the model’s weights and bias-related parameters;

[0090] Forward propagation, input data is predicted by the model, for neural networks:

[0091]

[0092] where z (l) is the weighted input of layer l, a (l) is the output after activation, g is the activation function;

[0093] Calculate the loss, using the loss function to calculate the difference between the predicted value and the true value;

[0094] Mean Square Error:

[0095]

[0096] Logarithmic loss:

[0097]

[0098] Back propagation: Update model parameters according to the gradient of the loss function;

[0099] Parameter update, Where θ is the model parameter and α is the learning rate;

[0100] S3. Model validation, using cross validation, divide the data set into k subsets, take turns using one of the subsets as the validation set and the rest as the training set;

[0101] S4. Hyperparameter optimization, randomly selecting values ​​in the specified hyperparameter space;

[0102] S5. Performance evaluation,accuracy: the number of correctly predicted samples divided by the total number of samples;

[0103]

[0104] Precision: The number of samples correctly predicted as positive divided by the number of all samples predicted as positive;

[0105]

[0106] Recall: The number of samples correctly predicted as positive divided by the number of samples actually positive;

[0107]

[0108] F1 score: the harmonic mean of precision and recall;

[0109]

[0110] S6. Model selection, select the best model based on the performance indicators on the validation set;

[0111] S7. Test set evaluation,The final performance evaluation of the selected models is performed using an independent test set.

[0112] Preferably, the prediction implementation comprises the following steps:

[0113] S1. Data preparation: Data preprocessing, ensuring that the data input to the model has the same format and range as the training data, including normalization and encoding; feature extraction, extracting the same features from the real-time data as when training the model;

[0114] S2. Model loading: loading the trained model and loading the previously trained and saved model into memory;

[0115] S3. Prediction, forward propagation, input the preprocessed data into the model for forward propagation to obtain the prediction result;

[0116] S4. Result interpretation, denormalization. If the data is normalized during preprocessing, the prediction results need to be denormalized to the scale of the original data. The denormalization formula is:

[0117] x original =x scaled ×(max-min)+min

[0118] Among them, x scaled is the normalized value, max and min are the maximum and minimum values ​​of the original data;

[0119] S5. Result analysis, performance evaluation, use mean square error to evaluate the accuracy of the prediction results:

[0120]

[0121] where y i is the true value, is the predicted value, n is the number of samples;

[0122] S6. Decision support, using the forecast results for practical decision support, such as power dispatch, resource allocation and demand response;

[0123] S7. Model updating: Regularly retrain the model based on new data to maintain the accuracy and relevance of the model.

[0124] (III) Beneficial effects

[0125] The present invention provides a load forecasting method for a power system based on big data. It has the following beneficial effects:

[0126] 1. In the present invention, by integrating historical load data, weather data, economic activity data and special event data through multi-source data fusion, the factors affecting power load can be captured more comprehensively, thereby improving the accuracy of prediction.

[0127] 2. In the present invention, by adopting the methods of normalization, standardization and outlier processing, it is ensured that the model training is carried out on high-quality data, thereby improving the generalization ability of the model. At the same time, it can process real-time data streams, realize real-time updating of load forecasts, and provide support for the dynamic scheduling of power systems.

[0128] 3. In the present invention, the reason for the model prediction results can be explained through the feature importance analysis method, which helps the power system operator to understand the model decision, and can automatically extract relevant features according to the characteristics of the data to adapt to load changes in different time periods and under different conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0129] Figure 1 It is a schematic diagram of the overall system flow of the present invention;

[0130] Figure 2 It is a schematic diagram of the overall system flow of data collection in the present invention;

[0131] Figure 3 It is a schematic diagram of the system flow of feature preprocessing in the present invention;

[0132] Figure 4 It is a schematic diagram of the system flow of feature preprocessing in the present invention;

[0133] Figure 5 A schematic diagram of the system flow of the model constructed in the present invention;

[0134] Figure 6 It is a schematic diagram of the overall process of model training and verification in the present invention;

[0135] Figure 7It is a schematic diagram of the overall process of predicting implementation in the present invention. DETAILED DESCRIPTION

[0136] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0137] Embodiment 1:

[0138] like Figure 1-7 As shown, an embodiment of the present invention provides a load forecasting method for a power system based on big data, including:

[0139] Data collection, collecting historical load data, weather data, economic activity data and special event data of the power system;

[0140] Data preprocessing: cleaning and normalizing the collected data to ensure data quality;

[0141] Feature extraction: using big data analysis technology to extract key features that affect load from raw data;

[0142] Model construction, using machine learning algorithms to build load forecasting models;

[0143] Model training and verification: Use historical data to train the model and verify the accuracy of the model through verification methods;

[0144] Forecast implementation, real-time data is input into the trained model to perform load forecasting, and power dispatch and resource allocation are carried out based on the forecast results.

[0145] Data collection includes the following steps:

[0146] S1. Determine the data sources, which usually include power load data, weather data, economic activity data and special event data. The power load data is obtained from the power company's SCADA system, EMS or historical database; the weather data is obtained from the Meteorological Bureau or online weather API, including temperature, humidity, wind speed and precipitation; the economic activity data can be obtained from government statistical departments, industry reports or online economic databases; special event data is collected through news media, social media or public event calendar channels;

[0147] S2. Data collection can be divided into automatic collection and manual collection. You can use API, crawler technology or cooperate with data providers to achieve automatic data collection. For some data that cannot be obtained automatically, you may need to collect and enter it manually;

[0148] S3. Data formatting: The collected data needs to be formatted to ensure its consistency, time synchronization: ensure that the timestamps of all data are consistent to facilitate subsequent analysis, and data type conversion: convert the data into a format suitable for analysis;

[0149] S4. Data cleaning: For missing data, filling, deletion or interpolation methods can be used to identify and process outliers, including deletion, correction or retention of data points and marking, including the following calculation methods:

[0150] Mean: The sum of all data points divided by the number of data points,

[0151]

[0152] Standard Deviation: A measure of how far a data point deviates from the mean.

[0153]

[0154] Variance: The square of the standard deviation, which is also an indicator of the degree of dispersion of data;

[0155]

[0156] S5. Data integration, integrating data from different sources for joint analysis, merging data together based on timestamps or other key fields, and ensuring that data from different data sources are aligned in terms of events.

[0157] Data preprocessing includes the following steps:

[0158] S1. Missing value processing, deletion: delete records with missing values

[0159] Delete records = delete all rows with missing values

[0160] Fill: Fill missing values ​​with statistics or predictive models

[0161]

[0162]

[0163] S2. Data conversion, normalization, scaling the data to a standard range,

[0164]

[0165] Among them, x is the original value, x min and x max are the minimum and maximum values ​​of the feature respectively;

[0166] Standardization, converting the data into a distribution with a mean of 0 and a standard deviation of 1;

[0167]

[0168] Where μ is the mean of the feature and σ is the standard deviation;

[0169] S3. Data discretization, converting continuous variables into discrete variables. Commonly used methods include equal-width binning and equal-frequency binning. Divide the data range into several intervals, assign a value to each interval, and divide the data into several intervals according to the frequency of the data points;

[0170] S4. Feature encoding, converting categorical variables into a set of binary variables;

[0171] If x = i, then x one-hot =[0, 0, ..., 1, ..., 0]

[0172] The i-th position is 1, and the rest are 0;

[0173] Label encoding, which maps each category of a categorical variable to a unique integer;

[0174] Category A → 0, Category B → 1, ...

[0175] S5. Data dimensionality reduction, principal component analysis, converting possibly correlated variables into a set of linearly uncorrelated variables through orthogonal transformation;

[0176] PCA = Projecting data onto principal components

[0177] S6. Data enhancement, using feature crossover to combine two or more features to generate new features;

[0178] New feature = Feature A × Feature B

[0179] S7. Outlier processing: Data points with Z=±3 at Z-score identification are regarded as outliers.

[0180]

[0181] Feature extraction includes the following calculation formula:

[0182] S1. Statistical feature extraction:

[0183] Mean, the average value of a data set, reflects the central tendency of the data;

[0184]

[0185] Variance, a measure of the dispersion of the data;

[0186]

[0187] The standard deviation, the square root of the variance, is also a measure of the dispersion of the data;

[0188]

[0189] Skewness, which describes the symmetry of the data distribution;

[0190]

[0191] Kurtosis, which describes how sharp or flat the data distribution is;

[0192]

[0193] S2. Time series feature extraction, calculating the statistical characteristics of the time series through sliding windows;

[0194] The autocorrelation coefficient measures the correlation between a time series and itself at different time lags;

[0195]

[0196] where k is the time lag;

[0197] S3. Frequency-based feature extraction, Fourier transform, convert time series data into frequency domain and extract frequency features;

[0198]

[0199] Where F(ω) is the frequency domain representation and f(t) is the time domain signal;

[0200] S4. Model-based feature extraction, principal component analysis, and dimensionality reduction techniques to extract the most important features;

[0201] PCA = Projecting data onto principal components

[0202] S5. Feature extraction based on machine learning, using random forest to select important features based on the average reduced impurity of feature splits in the tree model;

[0203] S6. Custom features, features built based on specific domain knowledge;

[0204] S7. Feature construction, feature intersection, combining two or more features to generate new features.

[0205] Building the model involves the following steps:

[0206] S1. Select the model by selecting the random forest model, ensemble learning method, building multiple decision trees and voting, each tree is built based on a randomly selected feature subset;

[0207] S2. Train the model, use the training data set to train the model and find the optimal values ​​of the model parameters;

[0208] Loss function: measures the difference between the model's predicted value and the true value;

[0209] Mean Square Error:

[0210]

[0211] Logarithmic loss:

[0212]

[0213] S3. Model evaluation, use cross validation to evaluate the performance of the model, divide the dataset into k subsets, take turns using one of the subsets as the test set and the rest as the training set;

[0214] S4. Hyperparameter tuning, optimize model performance by adjusting the model's hyperparameters, using grid search and random search;

[0215] S5. Model testing, using an independent test set to evaluate the generalization ability of the model;

[0216] S6. Model deployment: deploy the trained model to the production environment to perform actual prediction tasks.

[0217] Model training and validation include the following steps:

[0218] S1. Dataset division: divided into training set, validation set and test set. The training set is used for model learning, the validation set is used for model tuning and preventing overfitting, and the test set is used for evaluating the final performance of the model.

[0219] S2. Model training, which is divided into initializing parameters, forward propagation, calculating losses, and back propagation;

[0220] Initialize parameters to set initial values ​​for the model’s weights and bias-related parameters;

[0221] Forward propagation, input data is predicted by the model, for neural networks:

[0222]

[0223] where z (l) is the weighted input of layer l, a (l)is the output after activation, g is the activation function;

[0224] Calculate the loss, using the loss function to calculate the difference between the predicted value and the true value;

[0225] Mean Square Error:

[0226]

[0227] Logarithmic loss:

[0228]

[0229] Back propagation: Update model parameters according to the gradient of the loss function;

[0230] Parameter update, Where θ is the model parameter and α is the learning rate;

[0231] S3. Model validation, using cross validation, divide the data set into k subsets, take turns using one of the subsets as the validation set and the rest as the training set;

[0232] S4. Hyperparameter optimization, randomly selecting values ​​in the specified hyperparameter space;

[0233] S5. Performance evaluation,accuracy: the number of correctly predicted samples divided by the total number of samples;

[0234]

[0235] Precision: The number of samples correctly predicted as positive divided by the number of all samples predicted as positive;

[0236]

[0237] Recall: The number of samples correctly predicted as positive divided by the number of samples actually positive;

[0238]

[0239] F1 score: the harmonic mean of precision and recall;

[0240]

[0241] S6. Model selection, select the best model based on the performance indicators on the validation set;

[0242] S7. Test set evaluation,The final performance evaluation of the selected models is performed using an independent test set.

[0243] The forecast implementation consists of the following steps:

[0244] S1. Data preparation: Data preprocessing, ensuring that the data input to the model has the same format and range as the training data, including normalization and encoding; feature extraction, extracting the same features from the real-time data as when training the model;

[0245] S2. Model loading: loading the trained model and loading the previously trained and saved model into memory;

[0246] S3. Prediction, forward propagation, input the preprocessed data into the model for forward propagation to obtain the prediction result;

[0247] S4. Result interpretation, denormalization. If the data is normalized during preprocessing, the prediction results need to be denormalized to the scale of the original data. The denormalization formula is:

[0248] x original =x scaled ×(max-min)+min

[0249] Among them, x scaled is the normalized value, max and min are the maximum and minimum values ​​of the original data;

[0250] S5. Result analysis, performance evaluation, use mean square error to evaluate the accuracy of the prediction results:

[0251]

[0252] where y i is the true value, is the predicted value, n is the number of samples;

[0253] S6. Decision support, using the forecast results for practical decision support, such as power dispatch, resource allocation and demand response;

[0254] S7. Model updating: Regularly retrain the model based on new data to maintain the accuracy and relevance of the model.

[0255] Although 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 appended claims and their equivalents.

Claims

1. A load forecasting method for a power system based on big data, characterized by: include: Data collection, collecting historical load data, weather data, economic activity data and special event data of the power system; Data preprocessing: cleaning and normalizing the collected data to ensure data quality; Feature extraction: using big data analysis technology to extract key features that affect load from raw data; Model construction, using machine learning algorithms to build load forecasting models; Model training and verification: Use historical data to train the model and verify the accuracy of the model through verification methods; Forecast implementation, real-time data is input into the trained model to perform load forecasting, and power dispatch and resource allocation are carried out based on the forecast results.

2. The method for load forecasting of a power system based on big data according to claim 1, characterized in that: The data collection includes the following steps: S1. Determine the data sources, which usually include power load data, weather data, economic activity data and special event data. The power load data is obtained from the power company's SCADA system, EMS or historical database; the weather data is obtained from the Meteorological Bureau or online weather API, including temperature, humidity, wind speed and precipitation; the economic activity data can be obtained from government statistical departments, industry reports or online economic databases; special event data is collected through news media, social media or public event calendar channels; S2. Data collection can be divided into automatic collection and manual collection. You can use API, crawler technology or cooperate with data providers to achieve automatic data collection. For some data that cannot be obtained automatically, you may need to collect and enter it manually; S3. Data formatting: The collected data needs to be formatted to ensure its consistency, time synchronization: ensure that the timestamps of all data are consistent to facilitate subsequent analysis, and data type conversion: convert the data into a format suitable for analysis; S4. Data cleaning: For missing data, filling, deletion or interpolation methods can be used to identify and process outliers, including deletion, correction or retention of data points and marking, including the following calculation methods: Mean: The sum of all data points divided by the number of data points, Standard Deviation: A measure of how far a data point deviates from the mean. Variance: The square of the standard deviation, which is also an indicator of the degree of dispersion of data; S5. Data integration, integrating data from different sources for joint analysis, merging data together based on timestamps or other key fields, and ensuring that data from different data sources are aligned in terms of events.

3. The method for load forecasting of a power system based on big data according to claim 1, characterized in that: The data preprocessing comprises the following steps: S1. Missing value processing, deletion: delete records with missing values Delete records = delete all rows with missing values Fill: Fill missing values ​​with statistics or predictive models S2. Data conversion, normalization, scaling the data to a standard range, Among them, x is the original value, x min and x max are the minimum and maximum values ​​of the feature respectively; Standardization converts the data into a distribution with a mean of 0 and a standard deviation of 1; Where μ is the mean of the feature and σ is the standard deviation; S3. Data discretization, converting continuous variables into discrete variables. Commonly used methods include equal-width binning and equal-frequency binning. Divide the data range into several intervals, assign a value to each interval, and divide the data into several intervals according to the frequency of the data points; S4. Feature encoding, converting categorical variables into a set of binary variables; If x = i, then x one-hot =[0, 0, ..., 1, ..., 0] The i-th position is 1, and the rest are 0; Label encoding, which maps each category of a categorical variable to a unique integer; Category A → 0, Category B → 1, ... S5. Data dimensionality reduction, principal component analysis, converting possibly correlated variables into a set of linearly uncorrelated variables through orthogonal transformation; PCA = Projecting data onto principal components S6. Data enhancement, using feature crossover to combine two or more features to generate new features; New feature = Feature A × Feature B S7. Outlier processing: Data points with Z=±3 at Z-score identification are regarded as outliers.

4. The method for load forecasting of a power system based on big data according to claim 1, characterized in that: The feature extraction includes the following calculation formula: S1. Statistical feature extraction: Mean, the average value of a data set, reflects the central tendency of the data; Variance, a measure of the dispersion of the data; The standard deviation, the square root of the variance, is also a measure of the dispersion of the data; Skewness, which describes the symmetry of the data distribution; Kurtosis, which describes how sharp or flat the data distribution is; S2. Time series feature extraction, calculating the statistical characteristics of the time series through sliding windows; The autocorrelation coefficient measures the correlation between a time series and itself at different time lags; where k is the time lag; S3. Frequency-based feature extraction, Fourier transform, convert time series data into frequency domain and extract frequency features; Where F(ω) is the frequency domain representation and f(t) is the time domain signal; S4. Model-based feature extraction, principal component analysis, and dimensionality reduction techniques to extract the most important features; PCA = Projecting data onto principal components S5. Feature extraction based on machine learning, using random forest to select important features based on the average reduced impurity of feature splits in the tree model; S6. Custom features, features built based on specific domain knowledge; S7. Feature construction, feature intersection, combining two or more features to generate new features.

5. The method for load forecasting of a power system based on big data according to claim 1, characterized in that: The model building comprises the following steps: S1. Select the model by selecting the random forest model, ensemble learning method, building multiple decision trees and voting, each tree is built based on a randomly selected feature subset; S2. Train the model, use the training data set to train the model and find the optimal values ​​of the model parameters; Loss function: measures the difference between the model's predicted value and the true value; Mean Square Error: Logarithmic loss: S3. Model evaluation, use cross validation to evaluate the performance of the model, divide the dataset into k subsets, take turns using one of the subsets as the test set and the rest as the training set; S4. Hyperparameter tuning, optimize model performance by adjusting the model's hyperparameters, using grid search and random search; S5. Model testing, using an independent test set to evaluate the generalization ability of the model; S6. Model deployment: deploy the trained model to the production environment to perform actual prediction tasks.

6. The method for load forecasting of a power system based on big data according to claim 1, characterized in that: The model training and verification includes the following steps: S1. Dataset division: divided into training set, validation set and test set. The training set is used for model learning, the validation set is used for model tuning and preventing overfitting, and the test set is used for evaluating the final performance of the model. S2. Model training, which is divided into initializing parameters, forward propagation, calculating losses, and back propagation; Initialize parameters to set initial values ​​for the model’s weights and bias-related parameters; Forward propagation, input data is predicted by the model, for neural networks: a (l) =g(z (l) ) where z (l) is the weighted input of layer l, a (l) is the output after activation, g is the activation function; Calculate the loss, using the loss function to calculate the difference between the predicted value and the true value; Mean Square Error: Logarithmic loss: Back propagation: Update model parameters according to the gradient of the loss function; Parameter update, Where θ is the model parameter and α is the learning rate; S3. Model validation, using cross validation, divide the data set into k subsets, take turns using one of the subsets as the validation set and the rest as the training set; S4. Hyperparameter optimization, randomly selecting values ​​in the specified hyperparameter space; S5. Performance evaluation,accuracy: the number of correctly predicted samples divided by the total number of samples; Precision: The number of samples correctly predicted as positive divided by the number of all samples predicted as positive; Recall: The number of samples correctly predicted as positive divided by the number of samples actually positive; F1 score: the harmonic mean of precision and recall; S6. Model selection, select the best model based on the performance indicators on the validation set; S7. Test set evaluation,The final performance evaluation of the selected models is performed using an independent test set.

7. The method for load forecasting of a power system based on big data according to claim 1, characterized in that: The prediction implementation includes the following steps: S1. Data preparation: Data preprocessing, ensuring that the data input to the model has the same format and range as the training data, including normalization and encoding; feature extraction, extracting the same features from the real-time data as when training the model; S2. Model loading: loading the trained model and loading the previously trained and saved model into memory; S3. Prediction, forward propagation, input the preprocessed data into the model for forward propagation to obtain the prediction result; S4. Result interpretation, denormalization. If the data is normalized during preprocessing, the prediction results need to be denormalized to the scale of the original data. The denormalization formula is: x original =x scaled ×(max-min)+min Among them, x scaled is the normalized value, max and min are the maximum and minimum values ​​of the original data; S5. Result analysis, performance evaluation, use mean square error to evaluate the accuracy of the prediction results: where y i is the true value, is the predicted value, n is the number of samples; S6. Decision support, using the forecast results for practical decision support, such as power dispatch, resource allocation and demand response; S7. Model updating: Regularly retrain the model based on new data to maintain the accuracy and relevance of the model.

Images