Power demand forecasting method, system and medium based on intelligent data mining

Abstract

The present invention discloses a method, system and medium for predicting power demand based on intelligent data mining, including: obtaining historical data of power demand in a certain area and data of influencing factors thereof; preprocessing each item of data in the influencing factor data; performing correlation analysis on the preprocessed data; constructing corresponding first factor feature extraction sub-model, second factor feature extraction sub-model, third factor feature extraction sub-model and fourth factor feature extraction sub-model based on each influencing factor data subset in the data set after the correlation analysis and extracting corresponding key features; constructing a comprehensive influencing factor model for the power industry based on an integrated learning framework of XGBoost; inputting the extracted corresponding key features into the comprehensive influencing factor model for the power industry for training; and performing real-time power demand prediction based on the trained model. The present invention improves the accuracy of power demand prediction through comprehensive analysis of multiple factors.

Description

Technical Field

[0001] The present invention relates to the technical field of power system management and intelligent data analysis, and in particular to a method, system and medium for predicting power demand based on intelligent data mining. Background Art

[0002] With the rapid development of modern social economy, the volatility and complexity of electricity demand are increasing, and the factors affecting electricity demand are becoming increasingly diverse, including fluctuations in economic levels, changes in meteorological conditions, population growth, the impact of social activities such as holidays, and the popularization of new energy technologies. Traditional electricity demand forecasting methods are difficult to effectively handle the nonlinear relationship between multiple factors, and cannot respond to the dynamic impact of different factors on electricity demand in real time. Inaccurate electricity demand forecasting will directly affect the operating efficiency of the power system, resulting in waste of power generation resources or insufficient power supply, and even causing instability in the power grid.

[0003] Existing power demand forecasting methods are mainly divided into traditional statistical methods and machine learning-based methods. Traditional statistical methods include time series analysis, regression analysis, etc. These methods rely on the stationarity and linearity assumptions of data and are difficult to effectively capture the complex nonlinear effects of various external factors such as economy, meteorology and society. Although machine learning-based forecasting methods, such as support vector machines and decision trees, can handle nonlinear problems to a certain extent, they usually model based on a single factor or a few factors, fail to fully integrate the interactions between various factors, and the accuracy and stability of the model are still limited.

[0004] In recent years, with the development of big data and deep learning technology, some studies have tried to use deep neural networks, LSTM, etc. for power demand forecasting, which can better capture time series characteristics and nonlinear relationships. However, these methods usually lack specificity in feature selection and extraction, and it is difficult to effectively integrate the key features of multiple factors, and the interpretation of the forecast results is poor.

[0005] Therefore, the existing technology lacks a multi-factor integrated intelligent data mining method to simultaneously analyze the four major influencing factors of economy, meteorology, society and technology, and thus achieve high-precision prediction of electricity demand. Summary of the invention

[0006] The purpose of the present invention is to provide a method, system and medium for power demand forecasting based on intelligent data mining, and propose a comprehensive influencing factor model for the power industry, which integrates four major external factors: economic, meteorological, social and technological, and uses machine learning and deep learning technology to automatically extract key features from various factors and analyze their direct and indirect impact on power demand. The present invention improves the accuracy of power demand forecasting through comprehensive analysis of multiple factors.

[0007] Through the present invention, power system managers can better understand and predict the changing trend of power demand, optimize power resource scheduling, reduce power generation and supply costs, and enhance the stability and reliability of the power system. At the same time, the model can adapt to a variety of power demand forecasting scenarios, such as short-term load forecasting, long-term power demand planning, demand forecasting under extreme weather conditions, etc., providing strong technical support for the intelligent and sustainable development of the power system.

[0008] The present invention is achieved through the following technical solutions:

[0009] In a first aspect, the present invention provides a method for predicting power demand based on intelligent data mining, the method comprising:

[0010] Obtain historical data on electricity demand in a certain area and data on its influencing factors, including the first factor, the second factor, the third factor and the fourth factor;

[0011] Preprocess each item of the influencing factor data to obtain preprocessed data;

[0012] Perform correlation analysis on the preprocessed data to obtain a reduced-dimensional data set;

[0013] Based on each influencing factor data subset in the dimensionality-reduced data set, the corresponding first factor feature extraction sub-model, second factor feature extraction sub-model, third factor feature extraction sub-model and fourth factor feature extraction sub-model are constructed and the corresponding key features are extracted;

[0014] Based on the XGBoost integrated learning framework, a comprehensive influencing factor model for the power industry is constructed; the corresponding key features are extracted and input into the comprehensive influencing factor model for the power industry for training to obtain the XGBoost comprehensive influencing factor model;

[0015] Real-time power demand forecasting is performed based on the XGBoost comprehensive influencing factor model.

[0016] Furthermore, the first factor refers to economic factors, the second factor refers to meteorological factors, the third factor refers to social factors, and the fourth factor refers to technical factors.

[0017] Furthermore, correlation analysis is performed on the preprocessed data to obtain a reduced-dimensional data set, including:

[0018] Convert the values ​​of two different features to obtain the rankings corresponding to the two different features;

[0019] Calculate the ranking difference of each data point based on the rankings corresponding to two different features;

[0020] Based on the ranking difference, the Spearman correlation coefficient ρ was calculated;

[0021] If the absolute value of the Spearman correlation coefficient ρ exceeds the set threshold, they are considered to be highly correlated; otherwise, they are not highly correlated;

[0022] And the highly correlated features are combined into new features through dimensionality reduction methods to obtain a reduced-dimensionality data set;

[0023] Among them, the dimensionality reduction method is to synthesize the highly correlated feature sets using principal component analysis to generate principal components representing these features.

[0024] Furthermore, based on each influencing factor data subset in the dimensionality-reduced data set, the corresponding first factor feature extraction sub-model, second factor feature extraction sub-model, third factor feature extraction sub-model and fourth factor feature extraction sub-model are constructed and the corresponding key features are extracted, including:

[0025] Based on the first factor data subset in the data set after dimensionality reduction, a first factor feature extraction sub-model is constructed; the first factor feature extraction sub-model is an economic factor feature extraction model based on Lasso regression;

[0026] Based on the second factor data subset in the data set after dimensionality reduction, a second factor feature extraction sub-model is constructed; the second factor feature extraction sub-model is a meteorological factor feature extraction model based on the LSTM network;

[0027] Based on the third factor data subset in the dimensionality-reduced data set, a third factor feature extraction sub-model is constructed; the third factor feature extraction sub-model is a social factor feature extraction model based on the random forest algorithm;

[0028] Based on the fourth factor data subset in the dimensionality-reduced data set, a fourth factor feature extraction sub-model is constructed; the fourth factor feature extraction sub-model is a technical factor feature extraction model based on Granger causal analysis;

[0029] The first factor feature affecting the electricity demand is extracted through the first factor feature extraction sub-model;

[0030] The second factor characteristics that affect electricity demand are extracted through the second factor characteristic extraction sub-model;

[0031] The third factor characteristics that affect electricity demand are extracted through the third factor characteristic extraction sub-model;

[0032] The fourth factor characteristics that affect electricity demand are extracted through the fourth factor feature extraction sub-model.

[0033] Furthermore, the specific steps of feature extraction of the fourth factor feature extraction sub-model are as follows:

[0034] Based on the information criterion AIC optimization method, the lag order l is determined, that is, how many steps are delayed using the historical data of technical factors to predict the current electricity demand;

[0035] According to the determined lag order l, for each technical factor characteristic X i Construct two regression models, including an autoregressive model and an extended regression model;

[0036] Perform F test on autoregressive model and extended regression model to determine the technical factor characteristics X i Whether there is Granger causality for electricity demand y;

[0037] By analyzing each technical factor feature X i Granger causality test is performed to screen out technical features that have significant Granger causality on electricity demand y.

[0038] Furthermore, the statistic of the F test is:

[0039]

[0040] Where n is the number of samples, l is the lag order; RSS restricted is the residual sum of squares of the autoregressive model; RSS unrestricted is the residual sum of squares of the extended regression model; if the calculated F value is greater than the critical value under the significance level, the null hypothesis is rejected and it is considered that the technical factor characteristic X i There is a Granger causality relationship between the power demand y and the

[0041] Furthermore, the prediction output expression of the XGBoost comprehensive influencing factor model is:

[0042]

[0043] in, is the predicted value, representing the model's prediction of electricity demand at the current or future moment; x is an input vector composed of four types of factor features, including the real-time data of the first factor, the second factor, the third factor, and the fourth factor at the current moment; k is the index number of each decision tree; K is the total number of all decision trees in the model.

[0044] In a second aspect, the present invention further provides a power demand forecasting system based on intelligent data mining, the system comprising:

[0045] An acquisition unit, used for acquiring historical data of power demand in a certain area and data of influencing factors thereof, including a first factor, a second factor, a third factor and a fourth factor;

[0046] A preprocessing unit, used for preprocessing each item of the influencing factor data to obtain preprocessed data;

[0047] A correlation analysis unit, used to perform correlation analysis on the preprocessed data to obtain a data set after dimensionality reduction;

[0048] A sub-model construction and extraction unit, for constructing corresponding first factor feature extraction sub-model, second factor feature extraction sub-model, third factor feature extraction sub-model and fourth factor feature extraction sub-model based on each influencing factor data subset in the data set after dimensionality reduction, and extracting corresponding key features;

[0049] The comprehensive model building and training unit is used to build a comprehensive influencing factor model for the power industry based on the XGBoost integrated learning framework; and extract the corresponding key features and input them into the comprehensive influencing factor model for the power industry for training to obtain the XGBoost comprehensive influencing factor model;

[0050] The power demand forecasting unit is used to perform real-time power demand forecasting based on the XGBoost comprehensive influencing factor model.

[0051] Furthermore, the execution process of the sub-model construction and extraction unit is as follows:

[0052] Based on the first factor data subset in the data set after dimensionality reduction, a first factor feature extraction sub-model is constructed; the first factor feature extraction sub-model is an economic factor feature extraction model based on Lasso regression;

[0053] Based on the second factor data subset in the data set after dimensionality reduction, a second factor feature extraction sub-model is constructed; the second factor feature extraction sub-model is a meteorological factor feature extraction model based on the LSTM network;

[0054] Based on the third factor data subset in the dimensionality-reduced data set, a third factor feature extraction sub-model is constructed; the third factor feature extraction sub-model is a social factor feature extraction model based on the random forest algorithm;

[0055] Based on the fourth factor data subset in the dimensionality-reduced data set, a fourth factor feature extraction sub-model is constructed; the fourth factor feature extraction sub-model is a technical factor feature extraction model based on Granger causal analysis;

[0056] The first factor feature affecting the electricity demand is extracted through the first factor feature extraction sub-model;

[0057] The second factor characteristics that affect electricity demand are extracted through the second factor characteristic extraction sub-model;

[0058] The third factor characteristics that affect electricity demand are extracted through the third factor characteristic extraction sub-model;

[0059] The fourth factor characteristics that affect electricity demand are extracted through the fourth factor feature extraction sub-model.

[0060] In a third aspect, the present invention further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned power demand forecasting method based on intelligent data mining is implemented.

[0061] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0062] The present invention is based on the power demand forecasting method, system and medium of intelligent data mining, and proposes a comprehensive influencing factor model for the power industry. The model integrates four major external factors: economic, meteorological, social and technological. It uses machine learning and deep learning technology to automatically extract key features from various factors and analyze their direct and indirect impact on power demand. The present invention improves the accuracy of power demand forecasting through comprehensive analysis of multiple factors, and provides data support for the planning, scheduling and resource optimization of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0064] Figure 1 This is a flow chart of the power demand forecasting method based on intelligent data mining of the present invention;

[0065] Figure 2 It is a detailed flow chart of the power demand forecasting method based on intelligent data mining of the present invention;

[0066] Figure 3 This is a structural block diagram of the power demand forecasting system based on intelligent data mining of the present invention. DETAILED DESCRIPTION

[0067] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0068] Some research attempts to use deep neural networks, LSTM, etc. for power demand forecasting can better capture time series characteristics and nonlinear relationships; however, these existing methods usually lack specificity in feature selection and extraction, and it is difficult to effectively integrate the key features of multiple factors, resulting in poor interpretability of prediction results.

[0069] Therefore, it is an urgent problem for the power industry to build an intelligent power demand forecasting model that can comprehensively consider the influence of multiple factors and accurately analyze and predict the trend of power demand changes. This can not only improve the economy and stability of the power system, but also provide strong support for the rational dispatch of new energy and the development of green power grids.

[0070] Based on the above, the present invention proposes an electricity demand forecasting and influencing factor analysis model based on intelligent data mining. This model analyzes various external factors such as economy, meteorology, society and technology, and uses machine learning and deep learning technologies to achieve accurate prediction of electricity demand, providing data support for power system planning, scheduling and resource optimization.

[0071] Example 1

[0072] like Figure 1 As shown, the power demand forecasting method based on intelligent data mining of the present invention comprises:

[0073] Step 1, obtaining historical data of power demand in a certain area and data of its influencing factors, including the first factor, the second factor, the third factor and the fourth factor;

[0074] In this embodiment, the first factor refers to economic factors, the second factor refers to meteorological factors, the third factor refers to social factors, and the fourth factor refers to technical factors. See Table 1 for details:

[0075] Table 1 Influencing factors

[0076] type Influencing factors Economic factors GDP growth rate, industrial production index, residents' income level, etc. Meteorological factors Temperature, humidity, precipitation, wind speed, sunshine duration, etc. Social factors Population growth rate, urbanization level, holiday information, traffic flow, etc. Technical factors Renewable energy generation ratio, power generation equipment efficiency, technology upgrade indicators, etc.

[0077] Step 2: preprocess each item of the influencing factor data to obtain preprocessed data; the preprocessing includes detecting outliers and filling missing values.

[0078] Step 2 specifically includes:

[0079] First, the Z-score method is used to detect outliers in the data set. The Z-score method identifies possible outliers by calculating the standardized score of each data point relative to the entire data distribution. Specifically, suppose a data item is X=

[0080] {x1,x2,...,x n}, with a mean of μ and a standard deviation of σ. First, calculate the Z-score of each data point, namely:

[0081]

[0082] Among them, z i For data point x iZ-score. i |>3, x i Determined as an outlier.

[0083] Once an outlier is detected, it is repaired using Newton interpolation. Newton interpolation estimates the reasonable value of the outlier location through adjacent data points to ensure data continuity. Suppose there is an outlier x k In order to fill this value, we select several adjacent data points (x1,y1),(x2,y2),...,(x n ,y n ) as the base point for interpolation. The polynomial form of Newton interpolation is:

[0084]

[0085] Among them, Δ j y1 is the difference quotient, calculated as:

[0086]

[0087] P(x) is calculated by formula (3) k ), replace the original abnormal data point x with this value k This process ensures that abnormal data does not negatively impact model training while preserving the smoothness of the data series.

[0088] Step 3: perform correlation analysis on the preprocessed data to identify and eliminate redundant relationships between features and obtain a reduced-dimensional data set;

[0089] Step 3 specifically includes:

[0090] Convert the values ​​of two different features to obtain the rankings corresponding to the two different features;

[0091] Calculate the ranking difference of each data point based on the rankings corresponding to two different features;

[0092] Based on the ranking difference, the Spearman correlation coefficient ρ was calculated;

[0093] If the absolute value of the Spearman correlation coefficient ρ exceeds the set threshold, they are considered to be highly correlated; otherwise, they are not highly correlated;

[0094] And the highly correlated features are combined into new features through dimensionality reduction methods to obtain a reduced-dimensionality data set;

[0095] Among them, the dimensionality reduction method is to synthesize the highly correlated feature sets using principal component analysis to generate principal components representing these features.

[0096] In specific implementation, the detailed process of step 3 is as follows:

[0097] First, the correlation between the features is calculated. For the power industry influencing factor analysis model, the present invention uses the Spearman correlation coefficient to measure the monotonic correlation between different features.

[0098] Assume there are two features X and Y, each of which contains n data points. First, convert the values ​​of X and Y into rankings to obtain R(X) and R(Y). Next, calculate the ranking difference d of each data point i :

[0099] d i =R(x i )-R(y i ) (4)

[0100] Next, based on the ranking difference d i , calculate the Spearman correlation coefficient ρ:

[0101]

[0102] in It represents the square of the ranking difference of the i-th sample, and n is the total number of samples. The value range of the Spearman correlation coefficient is [-1,1]. When ρ approaches 1, it means that the two features are highly positively correlated; when ρ approaches -1, it means that the two features are highly negatively correlated; when ρ approaches 0, it means that the two features are not correlated.

[0103] Next, we set the correlation threshold |ρ|>0.7 to determine which features have a strong correlation. If the absolute value of the Spearman correlation coefficient of two features exceeds the threshold, they are considered to be highly correlated. In this case, the highly correlated features are combined into new features through dimensionality reduction methods.

[0104] In the process of dimensionality reduction, principal component analysis is used to synthesize the highly correlated feature sets to generate principal components representing these features. Specifically, the relevant feature data is first standardized to make its mean 0 and standard deviation 1 to eliminate the dimensional differences between different features. For the standardized data matrix X, the covariance matrix ∑ is calculated:

[0105]

[0106] Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors. Select the first few principal components based on the cumulative variance contribution rate to retain the main information of the data.

[0107] Through the above steps, after the correlation analysis is completed, a dimensionality-reduced data set is generated.

[0108] Step 4, based on each influencing factor data subset in the data set after dimensionality reduction, construct the corresponding first factor feature extraction sub-model, second factor feature extraction sub-model, third factor feature extraction sub-model and fourth factor feature extraction sub-model and extract the corresponding key features;

[0109] Step 4 specifically includes:

[0110] Step 41, based on the first factor data subset in the data set after dimensionality reduction, construct a first factor feature extraction sub-model; the first factor feature extraction sub-model is an economic factor feature extraction model based on Lasso regression; and through the first factor feature extraction sub-model, extract the first factor feature affecting the power demand;

[0111] Step 42, based on the second factor data subset in the data set after dimensionality reduction, construct a second factor feature extraction sub-model; the second factor feature extraction sub-model is a meteorological factor feature extraction model based on the LSTM network; and extract the second factor feature affecting the power demand through the second factor feature extraction sub-model;

[0112] Step 43, based on the third factor data subset in the data set after dimensionality reduction, construct a third factor feature extraction sub-model; the third factor feature extraction sub-model is a social factor feature extraction model based on the random forest algorithm; and through the third factor feature extraction sub-model, extract the third factor features affecting power demand;

[0113] Step 44, based on the fourth factor data subset in the data set after dimensionality reduction, construct a fourth factor feature extraction sub-model; the fourth factor feature extraction sub-model is a technical factor feature extraction model based on Granger causality analysis; and through the fourth factor feature extraction sub-model, extract the fourth factor characteristics that affect electricity demand.

[0114] In this embodiment, step 41 is to construct an economic factor feature extraction model based on Lasso regression based on the economic influencing factor data set (i.e., the first factor data subset) output in step 3 to identify and screen out economic features that have a significant impact on electricity demand.

[0115] The core of Lasso regression lies in its L1 regularization term, which controls the strength of feature selection through the parameter α. A larger α value will cause more feature coefficients to be compressed to 0, while a smaller α value will allow more features to enter the model. In order to determine the best α, a cross-validation method is used. The feature matrix X of economic factors and the target variable y are extracted from the data set. Assume that there is a series of candidate parameters α1, α2, ..., α k , evaluate the performance of each α on the training set through k-fold cross validation, and select the α with the smallest mean square error (MSE) as the optimal parameter:

[0116]

[0117] The optimal regularization parameter α is determined * After that, the parameter is used to perform Lasso regression training on the feature matrix X and the target variable y. The goal of Lasso regression is to minimize the following loss function:

[0118]

[0119] Among them, β0 is the intercept, β=(β1,β2,...,β p ) is the regression coefficient vector, n is the number of samples, and p is the number of features. Regularization term The introduction of makes the unimportant feature coefficients tend to zero, thus achieving feature selection.

[0120] After the model training is completed, the regression coefficient β is analyzed. j =0, it means that the feature has little impact on electricity demand and can be discarded. Finally, all features with non-zero coefficients are retained to form a simplified feature set of economic factors, which are economic factors that have a significant impact on electricity demand.

[0121] In this embodiment, step 42 is to build a meteorological factor feature extraction model based on an LSTM network based on the meteorological influencing factor data set (i.e., the second factor data subset) output in step 3, so as to extract the dynamic impact characteristics of meteorological factors on electricity demand.

[0122] First, the meteorological factor data is set as the feature matrix X, and the power demand data is set as the target variable y. The meteorological data contains multidimensional time series features such as daily average temperature, precipitation, humidity, and wind speed. In order to make full use of the time series characteristics of meteorological data, the meteorological factors are sorted in chronological order to form sequence data input into the LSTM network.

[0123] Next, design the LSTM network structure. In this model, each LSTM unit contains an input gate, a forget gate, and an output gate, which respectively control the inflow, retention, and output of information, so that important information can be effectively extracted from long time series. Assume that the input sequence length is T, the feature dimension is d, and the input shape of the LSTM layer is (T, d).

[0124] For each time t in the sequence, input the current state x t and the hidden state h at the previous time step t-1 and cell state c t-1 The status of each door is updated by the following formula:

[0125] f t =σ(Wf ·[h t-1 ,x t ]+b f ) (9)

[0126] i t =σ(W i ·[h t-1 ,x t ]+b i ) (10)

[0127]

[0128] o t =σ(W o ·[h t-1 ,x t ]+b o ) (13)

[0129] h t =o t ·tan h(c t ) (14)

[0130] Among them, f t 、i t , o t Represent the activation values ​​of the forget gate, input gate, and output gate, respectively, c t is the cell state, h t is the hidden state, W and b are the weight matrix and bias term. Through the combination of these gates, LSTM can remember or forget the feature information of a specific time step and effectively extract the key information of the time series.

[0131] The LSTM network training uses the back propagation algorithm to adjust the network parameters by optimizing the mean square error. After the training is completed, the LSTM network outputs time series features, which contain the forecast information of meteorological data on future electricity demand.

[0132] In this embodiment, step 43 is to construct a social factor feature extraction model based on the random forest algorithm based on the social influence factor data set (i.e., the third factor data subset) output in step 3. Random forest can effectively process nonlinear relationships and high-dimensional data, and is particularly suitable for identifying key features in social factors that have a significant impact on electricity demand.

[0133] Assume that the random forest is a model composed of M decision trees. Given the social factor dataset as the feature matrix X and the electricity demand data as the target variable y. In the regression task, the prediction value of the random forest model is:

[0134]

[0135] Among them, fm (X) represents the prediction of the mth decision tree for the input feature X. For power demand prediction, the output of the random forest model is the predicted value of power demand.

[0136] The trained random forest model can automatically calculate the importance of each feature. Feature importance measures the contribution of each feature to the target variable in the random forest. Social factor features are sorted according to the feature importance score, and features with importance scores higher than the set threshold are selected as key features. Redundant or irrelevant features in social factors are removed, and features with important value for power demand forecasting are retained.

[0137] In this embodiment, step 44 is to construct a technical factor feature extraction model based on Granger causality analysis based on the technical influencing factor data set (ie, the fourth factor data subset) in the data set outputted in step 3.

[0138] Assume that the technical factor dataset is the feature matrix X and the power demand data is the target variable y. The technical factor dataset contains multiple features, and the time series of each feature is represented by X1, X2, ..., X p , where p is the number of characteristics of the technical factor.

[0139] When performing Granger causality analysis, we first need to select an appropriate lag order l, that is, we need to determine how many steps to delay the historical data of the technical factor to predict the current electricity demand. The selection of the lag order can be optimized by the information criterion AIC. According to the AIC values ​​of different lag orders, the lag order with the smallest AIC value is selected as the optimal value. Once the lag order l is determined, we can i Two regression models are constructed. The first is an autoregressive model that uses only the historical data of the electricity demand time series y to predict the current electricity demand value. This autoregressive model is:

[0140]

[0141] Among them, α0 is a constant term, α j is the autoregressive coefficient, ∈ t is the error term.

[0142] The second model is an extended regression model, which adds technical factor feature X to the autoregressive model. i The historical data of the model is expanded to:

[0143]

[0144] Among them, β0 is a constant term, β j is the autoregressive coefficient, γ jis the regression coefficient of the technical factor characteristics, v t is the error term.

[0145] In order to determine whether there is a significant causal relationship, the F test is performed on the two models (autoregression model and extended regression model). The residual sum of squares RSS of the autoregression model is calculated restricted and the residual sum of squares RSS of the extended regression model unrestricted , the F-test statistic is:

[0146]

[0147] Where n is the number of samples and l is the lag order. If the calculated F value is greater than the critical value at the significance level, the null hypothesis is rejected and the feature X is considered i There is a Granger causality relationship between the power demand y and the

[0148] By analyzing each technical factor feature X i By conducting Granger causality test, we screen out technical features that have significant Granger causality on electricity demand y.

[0149] Step 5: Based on the XGBoost integrated learning framework, a comprehensive influencing factor model for the power industry is constructed; and the corresponding key features are extracted and input into the comprehensive influencing factor model for the power industry for training to obtain the XGBoost comprehensive influencing factor model;

[0150] Based on the XGBoost integrated learning framework, a comprehensive influencing factor model for the power industry is constructed and trained to achieve multi-factor comprehensive forecasting. This model takes key features extracted from four categories of factors: economic, meteorological, social, and technical as input, and learns the complex nonlinear relationship between each feature and power demand through the XGBoost model, which is ultimately used for real-time forecasting of power demand.

[0151] The objective function of XGBoost includes the loss function and the regularization term:

[0152]

[0153] in, is the loss function, which is used to measure the prediction value of the model With the actual value y i The regularization term Ω(f k ) is used to control the complexity of the model and prevent overfitting, usually expressed as:

[0154]

[0155] Where T is the number of leaf nodes in the decision tree, ω jis the weight of the jth leaf node, γ and λ are regularization parameters used to balance the complexity and prediction accuracy of the model.

[0156] Assume x i is a feature in the input vector, and the feature importance score I(x i ) can be expressed as:

[0157]

[0158] Among them, g t (x i ) represents the feature x i The importance contribution in the t-th tree, and the final feature importance score is the cumulative result of all trees.

[0159] The XGBoost comprehensive influencing factor model can predict electricity demand based on real-time data of economic, meteorological, social and technical factors. The predicted output value is It is based on the combined effect of four types of factor feature inputs. The model output is in the following form:

[0160]

[0161] in, is the predicted value, representing the model's prediction of electricity demand at the current or future moment; x is an input vector composed of four types of factor features, including the real-time data of the first factor, the second factor, the third factor, and the fourth factor at the current moment; k is the index number of each decision tree; K is the total number of all decision trees in the model.

[0162] Step 6: Based on the XGBoost comprehensive influencing factor model, real-time power demand forecasting is performed.

[0163] The XGBoost comprehensive influencing factor model trained by the present invention through the above step 5 can be used for real-time power demand prediction.

[0164] The model can receive real-time updated data of four factors and generate instant power demand forecast results. This result can support the dispatch decision, demand response and resource allocation of the power industry, helping power companies to more effectively manage the balance between supply and demand, reduce operating costs and improve resource utilization efficiency.

[0165] Example 2

[0166] like Figure 3 As shown, the difference between this embodiment and embodiment 1 is that this embodiment provides a power demand forecasting system based on intelligent data mining, which corresponds to the power demand forecasting method based on intelligent data mining in embodiment 1 in a one-to-one manner; the system includes:

[0167] An acquisition unit, used for acquiring historical data of power demand in a certain area and data of influencing factors thereof, including a first factor, a second factor, a third factor and a fourth factor;

[0168] A preprocessing unit, used for preprocessing each item of the influencing factor data to obtain preprocessed data;

[0169] A correlation analysis unit, used to perform correlation analysis on the preprocessed data to obtain a data set after dimensionality reduction;

[0170] A sub-model construction and extraction unit, for constructing corresponding first factor feature extraction sub-model, second factor feature extraction sub-model, third factor feature extraction sub-model and fourth factor feature extraction sub-model based on each influencing factor data subset in the data set after dimensionality reduction, and extracting corresponding key features;

[0171] The comprehensive model building and training unit is used to build a comprehensive influencing factor model for the power industry based on the XGBoost integrated learning framework; and extract the corresponding key features and input them into the comprehensive influencing factor model for the power industry for training to obtain the XGBoost comprehensive influencing factor model;

[0172] The power demand forecasting unit is used to perform real-time power demand forecasting based on the XGBoost comprehensive influencing factor model.

[0173] As a further implementation, the execution process of the sub-model construction and extraction unit is:

[0174] Based on the first factor data subset in the data set after dimensionality reduction, a first factor feature extraction sub-model is constructed; the first factor feature extraction sub-model is an economic factor feature extraction model based on Lasso regression;

[0175] Based on the second factor data subset in the data set after dimensionality reduction, a second factor feature extraction sub-model is constructed; the second factor feature extraction sub-model is a meteorological factor feature extraction model based on the LSTM network;

[0176] Based on the third factor data subset in the dimensionality-reduced data set, a third factor feature extraction sub-model is constructed; the third factor feature extraction sub-model is a social factor feature extraction model based on the random forest algorithm;

[0177] Based on the fourth factor data subset in the dimensionality-reduced data set, a fourth factor feature extraction sub-model is constructed; the fourth factor feature extraction sub-model is a technical factor feature extraction model based on Granger causal analysis;

[0178] The first factor feature affecting the electricity demand is extracted through the first factor feature extraction sub-model;

[0179] The second factor characteristics that affect electricity demand are extracted through the second factor characteristic extraction sub-model;

[0180] The third factor characteristics that affect electricity demand are extracted through the third factor characteristic extraction sub-model;

[0181] The fourth factor characteristics that affect electricity demand are extracted through the fourth factor feature extraction sub-model.

[0182] Among them, the execution process of each unit can be executed according to the process steps of the power demand prediction method based on intelligent data mining in Example 1, and will not be repeated one by one in this embodiment.

[0183] At the same time, the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned power demand forecasting method based on intelligent data mining is implemented.

[0184] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0185] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0186] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0187] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0188] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting power demand based on intelligent data mining, characterized in that: The method includes: Obtain historical data on electricity demand in a certain area and data on its influencing factors, including the first factor, the second factor, the third factor and the fourth factor; Preprocessing each item of the influencing factor data to obtain preprocessed data; Performing correlation analysis on the preprocessed data to obtain a data set after dimensionality reduction; Based on each influencing factor data subset in the dimensionality-reduced data set, construct a corresponding first factor feature extraction sub-model, a second factor feature extraction sub-model, a third factor feature extraction sub-model and a fourth factor feature extraction sub-model and extract corresponding key features; Based on the XGBoost integrated learning framework, a comprehensive influencing factor model for the power industry is constructed; and the corresponding key features are extracted and input into the comprehensive influencing factor model for the power industry for training to obtain the XGBoost comprehensive influencing factor model; Based on the XGBoost comprehensive influencing factor model, real-time power demand forecasting is performed.

2. The method for predicting power demand based on intelligent data mining according to claim 1, characterized in that: The first factor refers to economic factors, the second factor refers to meteorological factors, the third factor refers to social factors, and the fourth factor refers to technical factors.

3. The method for predicting power demand based on intelligent data mining according to claim 1, characterized in that: Performing correlation analysis on the preprocessed data to obtain a reduced-dimensional data set includes: Convert the values ​​of two different features to obtain the rankings corresponding to the two different features; Calculate the ranking difference of each data point based on the rankings corresponding to two different features; Based on the ranking difference, the Spearman correlation coefficient ρ is calculated; If the absolute value of the Spearman correlation coefficient ρ exceeds the set threshold, they are considered to be highly correlated; otherwise, they are not highly correlated; And the highly correlated features are combined into new features through dimensionality reduction methods to obtain a reduced-dimensionality data set; The dimension reduction method is to synthesize a set of highly correlated features using principal component analysis to generate principal components representing these features.

4. The method for predicting power demand based on intelligent data mining according to claim 1, characterized in that: Based on each influencing factor data subset in the dimensionality-reduced data set, a corresponding first factor feature extraction sub-model, a second factor feature extraction sub-model, a third factor feature extraction sub-model and a fourth factor feature extraction sub-model are constructed and corresponding key features are extracted, including: Based on the first factor data subset in the dimension-reduced data set, constructing a first factor feature extraction sub-model; the first factor feature extraction sub-model is an economic factor feature extraction model based on Lasso regression; Based on the second factor data subset in the dimensionally reduced data set, construct a second factor feature extraction sub-model; the second factor feature extraction sub-model is a meteorological factor feature extraction model based on an LSTM network; Based on the third factor data subset in the dimension-reduced data set, a third factor feature extraction sub-model is constructed; the third factor feature extraction sub-model is a social factor feature extraction model based on a random forest algorithm; Based on the fourth factor data subset in the dimension-reduced data set, constructing a fourth factor feature extraction sub-model; the fourth factor feature extraction sub-model is a technical factor feature extraction model based on Granger causality analysis; The first factor feature affecting the electricity demand is extracted through the first factor feature extraction sub-model; The second factor characteristics that affect electricity demand are extracted through the second factor characteristic extraction sub-model; The third factor characteristics that affect electricity demand are extracted through the third factor characteristic extraction sub-model; The fourth factor characteristics that affect electricity demand are extracted through the fourth factor feature extraction sub-model.

5. The method for predicting power demand based on intelligent data mining according to claim 4 is characterized in that: The specific steps of feature extraction by the fourth factor feature extraction sub-model are as follows: Based on the information criterion AIC optimization method, the lag order l is determined, that is, how many steps are delayed using the historical data of technical factors to predict the current electricity demand; According to the determined lag order l, for each technical factor characteristic X i Construct two regression models, including an autoregressive model and an extended regression model; The F test is performed on the autoregressive model and the extended regression model to determine the technical factor characteristics X i Whether there is Granger causality for electricity demand y; By analyzing each technical factor feature X i Granger causality test is performed to screen out technical features that have significant Granger causality on electricity demand y.

6. The method for predicting power demand based on intelligent data mining according to claim 5, characterized in that: The statistic of the F test is: Where n is the number of samples, l is the lag order; RSS restricted is the residual sum of squares of the autoregressive model; RSS unrestricted is the residual sum of squares of the extended regression model; if the calculated F value is greater than the critical value under the significance level, the null hypothesis is rejected and it is considered that the technical factor characteristic X i There is Granger causality on the electricity demand y.

7. The method for predicting power demand based on intelligent data mining according to claim 1, characterized in that: The prediction output expression of the XGBoost comprehensive influencing factor model is: in, is the predicted value, representing the model's prediction of electricity demand at the current or future moment; x is an input vector composed of four types of factor features, including the real-time data of the first factor, the second factor, the third factor, and the fourth factor at the current moment; k is the index number of each decision tree; K is the total number of all decision trees in the model.

8. The power demand forecasting system based on intelligent data mining is characterized by: The system includes: An acquisition unit, used for acquiring historical data of power demand in a certain area and data of influencing factors thereof, including a first factor, a second factor, a third factor and a fourth factor; A preprocessing unit, used for preprocessing each item of the influencing factor data to obtain preprocessed data; A correlation analysis unit, used to perform correlation analysis on the preprocessed data to obtain a data set after dimension reduction; A sub-model construction and extraction unit, used to construct a corresponding first factor feature extraction sub-model, a second factor feature extraction sub-model, a third factor feature extraction sub-model and a fourth factor feature extraction sub-model based on each influencing factor data subset in the reduced dimension data set and extract corresponding key features; A comprehensive model building and training unit is used to build a comprehensive influencing factor model for the power industry based on the XGBoost integrated learning framework; and extract the corresponding key features and input them into the comprehensive influencing factor model for the power industry for training to obtain the XGBoost comprehensive influencing factor model; The power demand forecasting unit is used to perform real-time power demand forecasting based on the XGBoost comprehensive influencing factor model.

9. The power demand forecasting system based on intelligent data mining according to claim 8, characterized in that: The execution process of the sub-model construction and extraction unit is as follows: Based on the first factor data subset in the dimension-reduced data set, construct a first factor feature extraction sub-model; The first factor feature extraction sub-model is an economic factor feature extraction model based on Lasso regression; Based on the second factor data subset in the dimension-reduced data set, construct a second factor feature extraction sub-model; The second factor feature extraction sub-model is a meteorological factor feature extraction model based on LSTM network; Based on the third factor data subset in the dimension-reduced data set, construct a third factor feature extraction sub-model; The third factor feature extraction sub-model is a social factor feature extraction model based on a random forest algorithm; Based on the fourth factor data subset in the dimension-reduced data set, constructing a fourth factor feature extraction sub-model; The fourth factor feature extraction sub-model is a technical factor feature extraction model based on Granger causality analysis; The first factor feature affecting the electricity demand is extracted through the first factor feature extraction sub-model; The second factor characteristics that affect electricity demand are extracted through the second factor characteristic extraction sub-model; The third factor characteristics that affect electricity demand are extracted through the third factor characteristic extraction sub-model; The fourth factor characteristics that affect electricity demand are extracted through the fourth factor feature extraction sub-model.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for predicting power demand based on intelligent data mining as described in any one of claims 1 to 7 is implemented.

Images