Power load prediction method, system and equipment based on user classification and medium

Abstract

The invention discloses a power load prediction method, system and device based on user classification and a medium. The method comprises the following steps: acquiring historical power load time sequence data of a target area; performing user classification on the historical power load time sequence data to obtain large industrial user power load data, resident and agricultural user power load data and general industrial and commercial user power load data; inputting the resident and agricultural user power load data into a first load prediction model for prediction, inputting the general industrial and commercial user power load data into a second load prediction model for prediction, and inputting the large industrial user power load data into a third load prediction model for prediction; and carrying out superposition and summation on the load prediction results to obtain a total load prediction result of the target region. The users are finely classified, and the special prediction model is made for the power consumption behavior characteristics of the users, so that the prediction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power load forecasting, and in particular to a power load forecasting method, system, device and medium based on user classification. Background Art

[0002] With the rapid development of the power industry, power load forecasting plays a vital role in the planning, operation and dispatching of power systems. Accurate power load forecasting can not only improve the safety and economy of the power system, but also optimize the configuration of power resources. However, traditional power load forecasting methods, such as time series analysis, regression analysis and neural network methods, usually focus on the overall prediction of the power load of the entire user group, ignoring the differences between users. These traditional methods have certain limitations in user classification and cannot fully consider the power consumption behavior characteristics of different user groups, resulting in reduced prediction accuracy.

[0003] In the existing technology, most power load forecasting methods fail to effectively combine the individual differences of users, especially when facing a diverse user group, the accuracy of the forecast is often difficult to meet actual needs. In addition, with the continuous changes in power equipment and user behavior patterns, the model parameters of traditional forecasting methods are difficult to update in real time, resulting in insufficient adaptability of the model. Existing power load forecasting systems often fail to fully combine big data technology and artificial intelligence algorithms, and thus fail to fully explore the potential laws in users' historical power consumption data, resulting in large errors in the prediction of future load change trends. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the defects of the power load forecasting method in the prior art, which has poor accuracy and robustness due to the lack of pertinence to the differences of user groups and insufficient intelligence, and to provide a power load forecasting method, system, device and medium based on user classification.

[0005] The present invention solves the above technical problems through the following technical solutions:

[0006] The present invention provides a method for predicting electric load based on user classification, the method comprising the following steps:

[0007] Obtain the historical power load time series data of the target area and perform preprocessing;

[0008] Classifying the historical power load time series data by user to obtain power load data of large industrial users, power load data of residential and agricultural users, and power load data of general industrial and commercial users;

[0009] Input the residential and agricultural user power load data into the first load forecasting model for forecasting to obtain a first load forecasting result, input the general industrial and commercial user power load data into the second load forecasting model for forecasting to obtain a second load forecasting result, and input the large industrial user power load data into the third load forecasting model for forecasting to obtain a third load forecasting result;

[0010] The first load forecast result, the second load forecast result and the third load forecast result are superimposed and summed to obtain a total load forecast result for the target area.

[0011] Preferably, the step of preprocessing the historical power load time series data includes:

[0012] The historical power load time series data is normalized using the interval normalization method according to the following formula:

[0013]

[0014] Among them, P i is the original value of the load, P min is the minimum daily load, P max is the maximum daily load, and P′ is the normalized value.

[0015] Preferably, the step of classifying the historical power load time series data by users comprises:

[0016] Calculate the proportion of each user's average load to the average total load based on historical power load time series data, and classify several users with the highest proportion and cumulative load proportion exceeding a preset proportion as large industrial electricity users;

[0017] The historical power load time series data, excluding the power load data of large industrial users, is classified by a fuzzy C clustering algorithm to obtain residential and agricultural users and general industrial and commercial users.

[0018] Preferably, user clustering is performed according to the following formula:

[0019]

[0020]

[0021] Among them, the power load time series data set X = {x1,x2,···,x n} is a data set consisting of n samples. The data set X is divided into c categories, C = {c1, c2, ···, c c} is the class center of each category, u ijis the membership of sample j to category i; the objective function (2) is the product of the distance between each sample and the center of each class and the corresponding membership, m is the fuzzy weighting index, which is any real number greater than 1; n constraints (3) are used to characterize that the sum of the membership of each sample is 1.

[0022] Preferably, the step of inputting the residential and agricultural user power load data into the first load forecasting model for forecasting to obtain the first load forecasting result includes:

[0023] The residential and agricultural user power load data are input into a support vector regression (SVR) model for load forecasting, and the significant difference between the characteristic value and the actual load value is determined by variance analysis to select significant features to obtain a first load forecasting result.

[0024] Preferably, the step of inputting the power load data of general industrial and commercial users into the second load forecasting model for forecasting to obtain the second load forecasting result includes:

[0025] The power load data of the general industrial and commercial users is stabilized by difference processing, and is input into an ARIMA (AutoRegressive Integrated Moving Average) model for prediction to obtain a second load prediction result.

[0026] Preferably, the step of inputting the large industrial user power load data into the third load forecasting model for forecasting to obtain the third load forecasting result comprises:

[0027] Inputting the large industrial user power load data into the Holt-Winters model for load forecasting to obtain a third load forecasting result;

[0028] The smoothing coefficients of the Holt-Winters model are tuned by cross-validation and / or grid search to minimize the prediction error.

[0029] The present invention also provides a power load forecasting system based on user classification, the power load forecasting system comprising:

[0030] A data acquisition module is used to obtain the historical power load time series data of the target area and perform preprocessing;

[0031] A data classification module is used to classify the historical power load time series data by user, so as to obtain power load data of large industrial users, power load data of residential and agricultural users, and power load data of general industrial and commercial users;

[0032] A load forecasting module, used for inputting the power load data of the residential and agricultural users into a first load forecasting model for forecasting to obtain a first load forecasting result, inputting the power load data of the general industrial and commercial users into a second load forecasting model for forecasting to obtain a second load forecasting result, and inputting the power load data of the large industrial users into a third load forecasting model for forecasting to obtain a third load forecasting result;

[0033] The prediction result processing module is used to superimpose and sum the first load prediction result, the second load prediction result and the third load prediction result to obtain a total load prediction result of the target area.

[0034] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method for predicting electric load based on user classification as described above is implemented.

[0035] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for predicting power load based on user classification when executing the computer program.

[0036] The present invention also provides a computer-readable medium having computer instructions stored thereon, and when the computer instructions are executed by a processor, the method for predicting electric load based on user classification as described above is implemented.

[0037] The positive and progressive effects of the present invention are:

[0038] The power load forecasting method based on user classification provided by the present invention pre-processes the historical power load time series data of a target area, and adopts the fuzzy C clustering algorithm to classify users into three categories: large industrial users, residential and agricultural users, and general industrial and commercial users, and establishes load forecasting models for users of different categories. For large industrial users, the Holt-Winters seasonal method is used for modeling to capture the trend and seasonal characteristics of their load changes; for residential and agricultural users, the support vector regression (SVR) model is used to use historical data for forecasting, taking into account short-term fluctuations and long-term trends; for general industrial and commercial users, the ARIMA model is used for forecasting, which is suitable for relatively stable time series loads, and can make some non-stationary data stable through differentiation. In addition, the system ensures the accuracy of the forecast results and the robustness of the model through steps such as data pre-processing, feature selection, model training and verification tuning. The method of the present invention has strong practicality and adaptability, can be effectively applied to the power load forecasting of different types of users, and is particularly suitable for the analysis and forecasting of large-scale electricity consumption data. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of this specification, the following is a brief introduction to the drawings required for the description of the embodiments. Obviously, the drawings described below are only some examples or embodiments of this specification. For ordinary technicians in this field, this specification can also be applied to other similar scenarios based on these drawings without creative work. Unless it is obvious from the language environment or otherwise explained, the same reference numerals in the figures represent the same structure or operation.

[0040] Figure 1 This is a first flow chart of the power load forecasting method based on user classification in this embodiment 1.

[0041] Figure 2 This is a second flow chart of the power load forecasting method based on user classification in this embodiment 1.

[0042] Figure 3 Schematic diagram of the architecture of the power load forecasting method based on user classification in this embodiment 1.

[0043] Figure 4 Schematic diagram of the structure of the electronic device in the third embodiment. DETAILED DESCRIPTION

[0044] The present invention is further described below by way of examples, but the present invention is not limited to the scope of the examples.

[0045] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various places in the text does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0046] It should be understood that the "system", "system", "unit" and / or "module" used herein are a method for distinguishing different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, the words can be replaced by other expressions.

[0047] As shown in this document, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural, unless the context clearly indicates an exception. Generally speaking, the terms "include" and "comprise" only indicate that the steps and elements that have been clearly identified are included, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0048] The definitions of inclusion in this document, such as the terms "having", "may have", "include" or "may include" used herein, indicate the existence of the corresponding functions, operations, elements, etc. herein, and do not limit the existence of one or more other functions, operations, elements, etc. In addition, it should be understood that the terms "including" or "having" used herein indicate the existence of the features, numbers, steps, operations, elements, components, or a combination thereof described in the specification, without excluding the existence or addition of one or more other features, numbers, steps, operations, elements, components, or a combination thereof.

[0049] Flowcharts are used herein to illustrate the operations performed by the systems of the embodiments of this invention. It should be understood that the preceding or following operations are not necessarily performed precisely in order. Instead, the various steps may be processed in reverse order or simultaneously. At the same time, other operations may also be added to these processes, or one or more operations may be removed from these processes.

[0050] Example 1

[0051] Please refer to Figure 1 , which is a first flow chart of the power load forecasting method based on user classification in this embodiment. Specifically, Figure 1 As shown, the power load forecasting method includes:

[0052] S11, obtaining the historical power load time series data of the target area and performing preprocessing;

[0053] S12, classifying the historical power load time series data by users to obtain power load data of large industrial users, power load data of residential and agricultural users, and power load data of general industrial and commercial users;

[0054] S13, inputting the power load data of residential and agricultural users into the first load forecasting model for forecasting to obtain a first load forecasting result, inputting the power load data of general industrial and commercial users into the second load forecasting model for forecasting to obtain a second load forecasting result, and inputting the power load data of large industrial users into the third load forecasting model for forecasting to obtain a third load forecasting result;

[0055] S14. Superimpose and sum the first load forecast result, the second load forecast result, and the third load forecast result to obtain a total load forecast result for the target area.

[0056] Specifically, in this embodiment, the preprocessing of the historical power load time series data can be performed by the following method:

[0057] The historical power load time series data set has a time span of 5 years and a time interval of 15 minutes. Users name it with the first letter of the region plus a number, such as the data set of Shanghai Minhang District is SHMH-001, SHMH-002, ... The unit of the data set is kilowatt, and the time is local time. The data set is read as a pandas DataFrame type, which is a matrix of XXX×YYY. The first column is time, and some data consists of load and a string of numbers separated by commas. First, delete the string of useless numbers after the load, by reading the data file into a string format, then converting it into a list, and deleting it with a comma as the delimiter, and then converting it back to the DataFrame type after completion. The time information in the first column is not needed for clustering, so it can be deleted directly, and then the remaining data is uniformly converted to float type. At this point, the load data of all users in the past five years is obtained and saved in a XXX×YYY matrix.

[0058] In addition, step S11 may further include:

[0059] The historical power load time series data is normalized using the interval normalization method according to the following formula:

[0060]

[0061] Among them, P i is the original value of the load, P min is the minimum daily load, P max is the maximum daily load, and P′ is the normalized value.

[0062] Please refer to Figure 2 , which is a second flow chart of the power load forecasting method based on user classification in this embodiment. Please refer to Figure 3 , which is a schematic diagram of the architecture of the power load prediction method based on user classification in this embodiment. Specifically, step S12 includes:

[0063] S121. Calculate the proportion of each user's average load to the average total load based on the historical power load time series data, and classify several users with the highest proportion and a cumulative load proportion exceeding a preset proportion as large industrial power users;

[0064] Specifically, the data of all XXX users in the past week are selected to draw a line graph. It can be seen from the figure that there is a certain regularity in the data. Most of the user loads are concentrated in a certain range, such as within 5000kW, but there are a small number of users with extremely large loads. The average load of the users is arranged in order from high to low, and the proportion of the average load of each user to the average total load is calculated. Based on this, the cumulative load ratio is calculated in the order from high to low, and the graph is drawn. The following results are obtained by calculation: the first few users with the highest average load are SHMH-XX1, ..., SHMH-XXM, etc., and the cumulative load ratio of these M users exceeds 45% to 90% (the value of the first-tier cities in the south is about 45% to 50%, the value of the heavy industrial cities is 80% to 90%, and the value of other cities is 60% to 70%). The first M users are classified as large industrial electricity users, and a separate prediction model is established for large industrial electricity users.

[0065] The present invention adopts the SG filtering method (Savitzky Golay Filter), performs weighted filtering on the data in the window, and obtains weighted weights by least square fitting with a high-order polynomial. The advantage of this method is that the data information can be better retained when filtering and smoothing the data, so it is suitable for smoothing the load curve. In Python, the smoothing process is realized by calling the Savitzky-Golay filter in the scipy library. When the window length is set to 5 and the k value is 1, the smoothing process not only eliminates the noise of the curve, but also retains the shape of the original load curve as much as possible.

[0066] S122. Using a fuzzy C clustering algorithm, the historical power load time series data excluding the power load data of large industrial users is classified into user categories to obtain residential and agricultural users and general industrial and commercial users.

[0067] In an optional implementation, user clustering is performed according to the following formula:

[0068]

[0069] Among them, the power load time series data set X = {x1,x2,···,x n} is a data set consisting of n samples. The data set X is divided into c categories, C = {c1, c2, ···, c c} is the class center of each category, u ij is the membership of sample j to category i; the objective function (2) is the product of the distance between each sample and the center of each class and the corresponding membership, m is the fuzzy weighting index, which can be any real number greater than 1; n constraints (3) are used to characterize that the sum of the membership of each sample is 1.

[0070] The objective function (2) contains two variables, membership u ij and class center c i To obtain the extreme value of the objective function, we can use the two variables u ij , c i Derived respectively. Prior to this, the constraints are processed using the Lagrange multiplier method, and the original objective function is expanded into the following formula (4).

[0071]

[0072] The above formula is for u ij The derivation is,

[0073]

[0074] Let the derivative be equal to 0 and solve for u ij , and use constraint (3) to eliminate λ j , we can get:

[0075]

[0076] Similarly, let J in the objective function (3) be i Taking the derivative, we get the following result:

[0077]

[0078]

[0079] It can be seen that u needs to be given in advance when starting the calculation ij or c i One of the initial values ​​is set, and after repeated iterative calculations, when u ij and c i It gradually tends to be stable. When the objective function J no longer changes, the clustering algorithm converges.

[0080] The characteristics of the electricity load of residents and agricultural users are: gradually decreasing to the lowest point from 0 to 5 o'clock, and gradually increasing from 5 to 10 o'clock, and then slowly decreasing to about 16 o'clock after maintaining this load for a period of time, and then continuing to climb to reach the peak load of the day at about 20 o'clock, and then rapidly decreasing. The characteristics of the electricity load of general industry and commerce are: starting to rise rapidly from about 5 o'clock, reaching a higher point at about 11 o'clock, then slightly decreasing from 11 o'clock to 15 o'clock, and then gradually climbing again, reaching the highest point at about 19 o'clock, and then the load rapidly decreasing, and the load curve is close to the double peak load curve. By finely classifying users, taking into full account the electricity consumption habits and demand changes of different user groups, and adopting targeted prediction models, the prediction error can be reduced and the prediction accuracy can be improved.

[0081] Therefore, in another optional implementation, step S13 includes:

[0082] S131. Input the residential and agricultural user power load data into the support vector regression (SVR) model for load forecasting, and determine whether there is a significant difference between the characteristic value and the actual load value through variance analysis to select significant features to obtain the first load forecasting result.

[0083] Specifically, an SVR model is established for the load data of residents and agricultural users, and the input feature x of the model is first determined. For time series data such as load curves, since there are no other features such as weather and festivals, only historical load data can be used as the input of the model. The following requirements are imposed on the input historical load data features: In order to make the model as accurate as possible, enough historical data should be input; at the same time, it is also necessary to avoid the feature dimension being too high to make the model too complicated, and to ensure that the running time of the algorithm is within an acceptable range; considering that the data cannot be obtained in time during the actual prediction process, the present invention adopts the same standard, and considers that there is a three-day lag in the acquisition of data during the prediction, that is, it is necessary to use the data three days ago for prediction. When selecting features, analysis of variance (ANOVA) is used to determine whether there is a significant difference between the feature value and the true value of the load. Analysis of variance is also called F test, which is used to find the linear relationship between two sets of data. Its null hypothesis is H0: there is no significant linear relationship between the data. When the P value is less than 0.05, the null hypothesis is rejected, and it can be considered that there is a significant linear relationship between the feature and the load value. When the P value is greater than 0.05, it is considered that there is no significant linear relationship between the feature and the load value, and the feature should be deleted. By calculating the F test number and the corresponding P value, it is found that when any number of historical data is taken as features, more than 90% of the feature P values ​​are less than 0.05. Therefore, it is difficult to use correlation filtering to reduce the number of input features. Considering the complexity of the model and the number of features required for prediction, as well as the actual need for prediction three days in advance, it is found after testing that when the historical load data of about 300 to 400 time points for 4 days is input, the model effect and algorithm running time are better. Thus, the form of the input features of the model and the corresponding load values ​​is obtained.

[0084] The advantages of using the SVR model for residential and agricultural electricity are:

[0085] (1) Strong robustness: Residential and agricultural electricity consumption data may have certain volatility and outliers, such as sudden weather changes, holiday effects, or policy adjustments. SVR can ignore a small number of abnormal data points through relaxation variables and penalty coefficients, maintain the robustness of the model, and reduce the negative impact on the prediction results.

[0086] (2) Good generalization ability: Electricity consumption prediction requires the model to perform well on future unseen data. SVR can prevent overfitting by controlling model complexity and balancing training errors, thereby improving the prediction ability of new data.

[0087] (3) Capturing long-term and short-term trends: SVR can capture both long-term trends (such as seasonal changes) and short-term fluctuations (such as the impact of temporary weather changes on electricity consumption), which is very important for accurately predicting residential and agricultural electricity demand.

[0088] (4) Support for multi-variable input: Residential and agricultural electricity consumption forecasts usually involve multiple input variables, such as temperature, rainfall, sunshine duration, etc. SVR can effectively integrate these multi-variable information and improve the accuracy of the forecast.

[0089] Further, step S13 also includes:

[0090] S132. Stabilize the power load data of general industrial and commercial users by difference, and input it into the ARIMA model for prediction to obtain a second load prediction result.

[0091] Specifically, the ARIMA model is a classic method widely used in time series analysis. It can effectively process non-stationary time series data by combining autoregression (AR), difference (I) and moving average (MA). In power load forecasting, the ARIMA model is often used for short-term forecasting of general industrial and commercial power loads because it is simple, flexible and has strong explanatory power. The detailed forecasting steps using the ARIMA model are as follows:

[0092] (1) Data collection and preprocessing

[0093] First, collect historical electricity load data for general industrial and commercial users. These data usually include electricity consumption at each time interval (such as hour, day, week, etc.) and should cover a long enough time range to capture possible seasonal and trend changes.

[0094] Before modeling, data needs to be preprocessed, including but not limited to the following aspects:

[0095] Missing value processing: Check whether there are missing values ​​in the data, and use reasonable methods (such as linear interpolation, mean filling, etc.) to fill them.

[0096] Outlier handling: Identify and handle outliers to prevent them from biasing the model.

[0097] Stationarity test and processing: By drawing a time series diagram and calculating the autocorrelation function (ACF) and partial autocorrelation function (PACF), the stationarity of the data can be preliminarily judged. If the data is non-stationary, the difference method is usually used to convert it into a stationary series.

[0098] (2) Model identification and parameter estimation

[0099] The ARIMA model is determined by three parameters, p, d, and q, where p is the order of the autoregressive term, d is the number of differences, and q is the order of the moving average term. When determining these parameters, the following steps are usually followed:

[0100] Determine the number of differences d: Use the stationarity test to determine how many times the data needs to be differentiated to become a stationary series. Generally speaking, the number of differences will not exceed 2.

[0101] Determine the autoregressive order p: Use the Partial Autocorrelation Function (PACF) plot to determine the order of the autoregressive term. The order corresponding to the first significant non-zero lag in the PACF plot is usually p.

[0102] Determine the moving average order q: Use the autocorrelation function (ACF) plot to determine the order of the moving average term. The order corresponding to the first significant non-zero lag in the ACF plot is usually q.

[0103] By observing the ACF and PACF graphs, the parameters p and q of the ARIMA model can be preliminarily estimated. In order to further optimize the parameter combination, the Akaike Information Criterion (AIC) or Bayesian Information Criterion (BIC) can be used for comparison to select the parameter combination that minimizes the AIC or BIC value.

[0104] (3) Model fitting

[0105] Once the parameters of the ARIMA model are determined, the model can be fitted using historical data. In Python, the ARIMA class in the statsmodels library is usually used to implement this process. By calling the fit function, the parameter estimates and fitting results of the model are obtained.

[0106] (4) Model diagnosis

[0107] After the model is fitted, the model must be diagnosed to ensure its rationality. Model diagnosis mainly includes the following aspects:

[0108] Residual analysis: Check whether the residuals of the model are white noise sequences, that is, whether the autocorrelation of the residuals is significant. You can check whether the residuals are random by drawing the ACF graph of the residuals. If the residuals show randomness, it means that the model fits well; otherwise, you need to readjust the model parameters or consider other models.

[0109] Normality test: Check whether the distribution of residuals conforms to normal distribution. This can be evaluated by drawing a histogram of residuals or performing a Shapiro-Wilk normality test. If the residuals do not conform to normal distribution, it may be necessary to transform the original data (such as logarithmic transformation) and then re-model.

[0110] (5) Prediction and evaluation

[0111] By diagnosing a qualified ARIMA model, future electricity load can be predicted. The ARIMA model can not only make single-step predictions (such as predicting the electricity consumption at the next point in time), but also make multi-step predictions (such as predicting the electricity consumption in the next week).

[0112] After completing the prediction, you need to compare the prediction results with the actual values ​​to evaluate the prediction accuracy of the model. Common evaluation indicators include:

[0113] Mean Squared Error (MSE): Measures the average squared error between the predicted values ​​and the actual values.

[0114] Mean absolute percentage error (MAPE): It represents the percentage of the predicted error to the actual value and has intuitive interpretation.

[0115] Mean Absolute Error (MAE): Measures the average absolute difference between the predicted values ​​and the actual values.

[0116] If the prediction error is large, you can try to adjust the model parameters or introduce exogenous variables (such as weather, holidays, etc.) to improve the model accuracy.

[0117] (6) Model update and maintenance

[0118] The prediction effect of the ARIMA model will gradually decrease over time, so the model needs to be updated regularly. Usually, the model can be retrained based on the latest load data to adapt to new changes in the data. In addition, a rolling forecast method can be used to continuously update the model to correct future power load forecasts in real time.

[0119] In addition, step S13 further includes:

[0120] S133. Inputting the power load data of large industrial users into the Holt-Winters model for load forecasting to obtain a third load forecasting result; the smoothing coefficient of the Holt-Winters model is optimized through cross-validation and / or grid search to minimize the prediction error.

[0121] Specifically, the Holt-Winters seasonal method is a classic method for time series forecasting, which is particularly suitable for time series data with obvious seasonality and trends. The following are the detailed steps for modeling and forecasting the electricity load of large industrial users using the Holt-Winters seasonal method:

[0122] Step 1: Data preparation

[0123] (1) Collect data:

[0124] Obtain historical electricity load data for large industrial users. The data should include a time range long enough to cover multiple seasonal cycles (at least two years of data is usually recommended). The time interval of the data should be consistent (such as hours, days, weeks, etc.).

[0125] (2) Data preprocessing:

[0126] Check the data for missing values ​​and outliers and perform appropriate processing (such as interpolation, filling or removing outliers).

[0127] If the data is significantly non-stationary, it can be made stationary through differencing or other methods.

[0128] Step 2: Model selection

[0129] Depending on the seasonality type of the data, choose the appropriate Holt-Winters model:

[0130] Additive model: suitable for time series with relatively stable seasonal fluctuations.

[0131] Multiplicative model: Applicable to time series whose seasonal fluctuations vary over time.

[0132] Step 3: Initialize model parameters

[0133] The main parameters of the Holt-Winters model include:

[0134] Level value (L): represents the average level of the time series.

[0135] Trend (T): Indicates the trend of the time series.

[0136] Seasonality (S): represents the seasonal fluctuation of the time series.

[0137] Smoothing coefficients (α, β, γ): used to update the level, trend, and seasonal components.

[0138] Initialization methods usually include:

[0139] Level and trend: Initialized by the average of the first few periods of data.

[0140] Seasonality: Initial seasonality can be determined by comparing the values ​​of the first few cycles to the average level.

[0141] Step 4: Model training

[0142] Update the model parameters recursively:

[0143] Horizontal update: L_t = α*(Y_t / S_{tp})+(1-α)*(L_{t-1}+T-{t-1})

[0144] Trend update: T_t = β*(L_t-L{t-1})+(1-β)*T{t-1}

[0145] Seasonal update: S_t = y*(Y_t / L_t)+(1-γ)*S-{tp}

[0146] Among them, Y_t is the actual electricity consumption at time t, and p is the length of the seasonal cycle.

[0147] Step 5: Model verification and tuning

[0148] Use part of the data for model training and keep part of the data for model validation.

[0149] The smoothing coefficients α, β, and γ are optimized by minimizing the prediction error (such as mean square error MSE, mean absolute error MAE, etc.).

[0150] Parameter tuning can be performed via cross-validation or grid search.

[0151] Step 6: Prediction

[0152] Use the trained Holt-Winters model to predict future electricity load. Generate forecasts based on the selected period length and model type.

[0153] The forecast results can be used to formulate power demand plans and load management strategies for industrial users.

[0154] Step 7: Evaluate and adjust

[0155] Compare the predicted values ​​with the actual values ​​to evaluate the performance of the model.

[0156] Based on the evaluation results, further adjust the model parameters or methods (such as choosing a different seasonal model or changing the smoothing coefficient).

[0157] Advantages of Holt-Winters seasonal method for large industrial users' electricity load forecasting:

[0158] (1) Dealing with seasonal changes:

[0159] The Holt-Winters method can capture both the trend and seasonal components in the time series, and is very suitable for dealing with fluctuations in the electricity load of large industrial users caused by seasonal factors (such as temperature, production cycle, etc.).

[0160] (2) Easy to understand and implement:

[0161] Compared with some complex machine learning models, the Holt-Winters method has an intuitive principle, fewer parameters, is easy to implement and explain, and is very friendly to industrial applications.

[0162] (3) Strong adaptability:

[0163] The Holt-Winters method can adapt to different degrees of trend and seasonal changes by adjusting the smoothing coefficient. It is highly flexible and can be applied to different types of large industrial users.

[0164] (4) Real-time update:

[0165] This method can be updated in real time when new data arrives, and is suitable for use as a rolling forecast, that is, continuously revising future electricity load forecasts to cope with real-time changes in electricity demand.

[0166] (5) Reduce the risk of overfitting:

[0167] The Holt-Winters method reduces the sensitivity to noise in historical data through smoothing techniques, reduces the risk of model overfitting, and makes the prediction more stable and reliable.

[0168] The power load forecasting method based on user classification in this embodiment significantly improves the accuracy, robustness and efficiency of power load forecasting, and has the following significant effects:

[0169] 1. Improve prediction accuracy: By finely classifying users and developing special prediction models based on the electricity consumption behavior characteristics of different user groups, the present invention effectively reduces the prediction error caused by user differences. Especially when facing complex and changeable user electricity demand, the present invention can significantly improve the accuracy of prediction, thereby providing more reliable data support for the optimization management of the power system.

[0170] 2. Enhance model adaptability and robustness: This invention uses big data technology and artificial intelligence algorithms to dynamically optimize and update the prediction model so that it can adapt to changes in user behavior patterns in a timely manner. This feature enables the prediction model to maintain a high degree of accuracy in long-term operation, and can effectively respond to sudden changes in electricity demand, thereby improving the operational stability of the power system.

[0171] 3. Improve data processing efficiency: This invention introduces an efficient data processing framework that can quickly process historical power consumption data and real-time data of large-scale user groups. This capability not only improves the response speed of prediction, but also provides a guarantee for the operation of the power system in a large-scale data environment.

[0172] 4. Optimize power resource allocation: Through more accurate load forecasting, the present invention can help the power system better allocate resources, reduce power generation and supply costs, and improve the economy and safety of the power system.

[0173] In general, the present invention significantly improves the overall performance of power load forecasting, provides key technical support for the development of smart grids, and promotes the further development of the power industry towards intelligence and refinement.

[0174] Example 2

[0175] This embodiment provides a power load forecasting system based on user classification, and the power load forecasting system includes:

[0176] A data acquisition module is used to obtain the historical power load time series data of the target area and perform preprocessing;

[0177] A data classification module is used to classify the historical power load time series data by user, so as to obtain power load data of large industrial users, power load data of residential and agricultural users, and power load data of general industrial and commercial users;

[0178] A load forecasting module, used for inputting the power load data of the residential and agricultural users into a first load forecasting model for forecasting to obtain a first load forecasting result, inputting the power load data of the general industrial and commercial users into a second load forecasting model for forecasting to obtain a second load forecasting result, and inputting the power load data of the large industrial users into a third load forecasting model for forecasting to obtain a third load forecasting result;

[0179] The prediction result processing module is used to superimpose and sum the first load prediction result, the second load prediction result and the third load prediction result to obtain a total load prediction result of the target area.

[0180] As for the system embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The system embodiment described above is only schematic, in which the modules described as separate components may or may not be physically separated, and the modules displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. Ordinary technicians in this field can understand and implement it without paying creative work.

[0181] Example 3

[0182] Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the method of Embodiment 1 is implemented when the processor executes the program. Figure 4 The electronic device 30 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0183] like Figure 4 As shown, the electronic device 30 may be in the form of a general-purpose computing device, for example, it may be a server device. The components of the electronic device 30 may include, but are not limited to: at least one processor 31, at least one memory 32, and a bus 33 connecting different system components (including the memory 32 and the processor 31).

[0184] The bus 33 includes a data bus, an address bus, and a control bus.

[0185] The memory 32 may include a volatile memory, such as a random access memory (RAM) 321 and / or a cache memory 322 , and may further include a read-only memory (ROM) 323 .

[0186] The memory 32 may also include a program / utility 325 having a set (at least one) of program modules 324, such program modules 324 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0187] The processor 31 executes various functional applications and data processing, such as the method of embodiment 1 of the present invention, by running the computer program stored in the memory 32 .

[0188] The electronic device 30 may also communicate with one or more external devices 34 (e.g., keyboards, pointing devices, etc.). Such communication may be performed via an input / output (I / O) interface 35. Furthermore, the model-generated device 30 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 36. As shown, the network adapter 36 communicates with other modules of the model-generated device 30 via a bus 33. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the model-generated device 30, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (RAID) systems, tape drives, and data backup storage systems.

[0189] It should be noted that although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided into multiple units / modules to be embodied.

[0190] Example 4

[0191] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method of Embodiment 1 is implemented.

[0192] The readable storage medium may include but is not limited to: a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical storage device, a magnetic storage device or any suitable combination of the above.

[0193] In a possible implementation manner, the present invention may also be implemented in the form of a program product, which includes a program code. When the program product is run on a terminal device, the program code is used to enable the terminal device to execute the method for implementing Embodiment 1.

[0194] The program code for executing the present invention may be written in any combination of one or more programming languages, and may be executed entirely on a user device, partially on a user device, as an independent software package, partially on a user device and partially on a remote device, or entirely on a remote device.

[0195] Although the specific embodiments of the present invention are described above, it should be understood by those skilled in the art that this is only for illustration and the protection scope of the present invention is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, but these changes and modifications all fall within the protection scope of the present invention.

Claims

1. A method for predicting power load based on user classification, characterized in that: The power load forecasting method comprises the following steps: Obtain the historical power load time series data of the target area and perform preprocessing; Classifying the historical power load time series data by user to obtain power load data of large industrial users, power load data of residential and agricultural users, and power load data of general industrial and commercial users; Input the residential and agricultural user power load data into the first load forecasting model for forecasting to obtain a first load forecasting result, input the general industrial and commercial user power load data into the second load forecasting model for forecasting to obtain a second load forecasting result, and input the large industrial user power load data into the third load forecasting model for forecasting to obtain a third load forecasting result; The first load forecast result, the second load forecast result and the third load forecast result are superimposed and summed to obtain a total load forecast result for the target area.

2. The power load forecasting method according to claim 1, characterized in that: The steps for preprocessing the historical power load time series data include: The historical power load time series data is normalized using the interval normalization method according to the following formula: Among them, P i is the original value of the load, P min is the minimum daily load, P max is the maximum daily load, and P′ is the normalized value.

3. The power load forecasting method according to claim 1, characterized in that: The step of classifying the historical power load time series data by users comprises: Calculate the proportion of each user's average load to the average total load based on historical power load time series data, and classify several users with the highest proportion and cumulative load proportion exceeding a preset proportion as large industrial electricity users; The historical power load time series data, excluding the power load data of large industrial users, is classified by a fuzzy C clustering algorithm to obtain residential and agricultural users and general industrial and commercial users.

4. The power load forecasting method according to claim 3, characterized in that: Users are clustered according to the following formula: Among them, the power load time series data set X = {x1,x2,···,x n } is a data set consisting of n samples. The data set X is divided into c categories, C = {c1, c2, ···, c c } is the class center of each category, u ij is the membership of sample j to category i; the objective function (2) is the product of the distance between each sample and the center of each class and the corresponding membership, m is the fuzzy weighting index, which is any real number greater than 1; n constraints (3) are used to characterize that the sum of the membership of each sample is 1.

5. The power load forecasting method according to claim 1, characterized in that: The step of inputting the residential and agricultural user power load data into a first load forecasting model for forecasting to obtain a first load forecasting result comprises: The residential and agricultural user power load data are input into a support vector regression (SVR) model for load forecasting, and variance analysis is used to determine whether there is a significant difference between the characteristic value and the actual load value to select significant features to obtain a first load forecasting result.

6. The power load forecasting method according to claim 1, characterized in that: The step of inputting the power load data of general industrial and commercial users into the second load forecasting model for forecasting to obtain the second load forecasting result comprises: The power load data of the general industrial and commercial users is stabilized by difference processing and input into the ARIMA model for prediction to obtain a second load prediction result.

7. The power load forecasting method according to claim 1, characterized in that: The step of inputting the large industrial user power load data into the third load forecasting model for forecasting to obtain the third load forecasting result comprises: Inputting the large industrial user power load data into the Holt-Winters model for load forecasting to obtain a third load forecasting result; The smoothing coefficients of the Holt-Winters model are tuned by cross-validation and / or grid search to minimize the prediction error.

8. A power load forecasting system based on user classification, characterized in that: The power load forecasting system comprises: A data acquisition module is used to obtain the historical power load time series data of the target area and perform preprocessing; A data classification module is used to classify the historical power load time series data by user, so as to obtain power load data of large industrial users, power load data of residential and agricultural users, and power load data of general industrial and commercial users; A load forecasting module, used for inputting the power load data of the residential and agricultural users into a first load forecasting model for forecasting to obtain a first load forecasting result, inputting the power load data of the general industrial and commercial users into a second load forecasting model for forecasting to obtain a second load forecasting result, and inputting the power load data of the large industrial users into a third load forecasting model for forecasting to obtain a third load forecasting result; The prediction result processing module is used to superimpose and sum the first load prediction result, the second load prediction result and the third load prediction result to obtain a total load prediction result of the target area.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for predicting power load based on user classification as described in any one of claims 1 to 7 is implemented.

10. A computer readable medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by a processor, the method for predicting electric load based on user classification as claimed in any one of claims 1 to 7 is implemented.

Images