Power system load prediction method and device, computer equipment, readable storage medium and program product

Abstract

The invention relates to a power system load prediction method and device, computer equipment, a readable storage medium and a program product, and relates to the technical field of smart power grids. The method comprises the following steps: obtaining a power load time sequence diagram according to historical power load data of a power system; according to the power load time sequence diagram, respectively determining a periodic fluctuation property and a seasonal fluctuation property of the power load time sequence diagram; the periodic fluctuation property refers to existence of unfixed periodic fluctuation or absence of unfixed periodic fluctuation, and the seasonal fluctuation property refers to existence of fixed seasonal fluctuation or absence of fixed seasonal fluctuation; according to different combinations of the periodic fluctuation property and the seasonal fluctuation property, determining different target prediction models from a plurality of preset prediction models based on a time sequence; and obtaining predicted load data of the power system according to the historical power load data and the target prediction model. The method can improve the prediction accuracy.

Description

Technical Field

[0001] The present application relates to the technical field of smart grids, and in particular, to a method, device, computer device, readable storage medium, and program product for predicting the load of a power system. Background Art

[0002] Power system load (power demand, i.e., active power) prediction refers to making a prediction of the system load for a future period by fully considering the influence of factors such as historical system load, economic conditions, meteorological conditions, and social events. In related technologies, a prediction model is used to predict the load of a power system, such as using regression prediction techniques based on machine learning and deep learning, support vector products, random forests, etc. However, uncertain factors such as complex and changeable meteorological conditions and social events will all have a certain impact on the power system load. Therefore, traditional prediction methods have the problem of low accuracy. Summary of the Invention

[0003] Based on this, in view of the above technical problems, it is necessary to provide a method, device, computer device, computer-readable storage medium, and computer program product for predicting the load of a power system that can improve accuracy.

[0004] In a first aspect, the present application provides a method for predicting the load of a power system, including:

[0005] Obtaining a power load time series graph according to the historical power load data of the power system;

[0006] According to the power load time series graph, respectively determining the periodic fluctuation property and seasonal fluctuation property of the power load time series graph; the periodic fluctuation property is the presence or absence of an irregular periodic fluctuation, and the seasonal fluctuation property is the presence or absence of a fixed seasonal fluctuation;

[0007] According to different combinations of the periodic fluctuation property and the seasonal fluctuation property, determining different target prediction models from a plurality of preset time series-based prediction models;

[0008] Obtaining the predicted load data of the power system according to the historical power load data and the target prediction model.

[0009] In one of the embodiments, the step of respectively determining the seasonal fluctuation property and the periodic fluctuation property of the power load time series graph according to the power load time series graph includes:

[0010] Performing an irregular period detection on the power load time series graph to obtain the periodic fluctuation property of the power load time series graph;

[0011] Perform fixed-period detection on the power load time series graph to obtain the seasonal fluctuation property of the power load time series graph.

[0012] In one embodiment, the performing fixed-period detection on the power load time series graph to obtain the seasonal fluctuation property of the power load time series graph includes:

[0013] Divide the power load time series graph according to the preset fixed period to obtain multiple power load time series sub-graphs;

[0014] Perform correlation analysis on the multiple power load time series sub-graphs to obtain the seasonal fluctuation property of the power load time series graph.

[0015] In one embodiment, the determining different target prediction models from multiple preset time series-based prediction models according to different combinations of the periodic fluctuation property and the seasonal fluctuation property includes:

[0016] When the periodic fluctuation property is the existence of non-fixed periodic fluctuations, determine the target prediction model as the triple exponential smoothing model from multiple preset time series-based prediction models;

[0017] When the periodic fluctuation property is the non-existence of non-fixed periodic fluctuations and the seasonal fluctuation property is the existence of fixed seasonal fluctuations, determine the target prediction model as the seasonal autoregressive integrated moving average model from multiple preset time series-based prediction models;

[0018] When the periodic fluctuation property is the non-existence of non-fixed periodic fluctuations and the seasonal fluctuation property is the non-existence of fixed seasonal fluctuations, determine the target prediction model as the autoregressive integrated moving average model from multiple preset time series-based prediction models.

[0019] In one embodiment, the triple exponential smoothing model includes: the triple exponential smoothing additive model and the triple exponential smoothing multiplicative model. The determining the target prediction model as the triple exponential smoothing model from multiple preset time series-based prediction models when the periodic fluctuation property is the existence of non-fixed periodic fluctuations includes:

[0020] When the periodic fluctuation property is the existence of non-fixed periodic fluctuations and the seasonal fluctuation property is the existence of fixed seasonal fluctuations, determine the target prediction model as the triple exponential smoothing multiplicative model from multiple preset time series-based prediction models;

[0021] When the periodic fluctuation property is that there is an unfixed periodic fluctuation and the seasonal fluctuation property is that there is no fixed seasonal fluctuation, the target prediction model is determined as the triple exponential smoothing additive model from a plurality of preset prediction models based on time series.

[0022] In one embodiment, obtaining the power load time series diagram according to the historical power load data of the power system includes:

[0023] Obtain the historical power load data of the power system;

[0024] Preprocess the historical power load data and arrange it according to the date to obtain the historical power load sequence;

[0025] Obtain the power load time series diagram according to the historical power load sequence.

[0026] In a second aspect, the present application also provides a power system load prediction device, including:

[0027] A sequence diagram generation module, configured to obtain a power load time series diagram according to the historical power load data of the power system;

[0028] A fluctuation property determination module, configured to respectively determine the periodic fluctuation property and the seasonal fluctuation property of the power load time series diagram according to the power load time series diagram; the periodic fluctuation property is that there is an unfixed periodic fluctuation or there is no unfixed periodic fluctuation, and the seasonal fluctuation property is that there is a fixed seasonal fluctuation or there is no fixed seasonal fluctuation;

[0029] A prediction model determination module, configured to determine different target prediction models from a plurality of preset prediction models based on time series according to different combinations of the periodic fluctuation property and the seasonal fluctuation property;

[0030] A load data prediction module, configured to obtain the predicted load data of the power system according to the historical power load data and the target prediction model.

[0031] In a third aspect, the present application also provides a computer device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0032] Obtain a power load time series diagram according to the historical power load data of the power system;

[0033] According to the power load time series diagram, respectively determine the periodic fluctuation property and the seasonal fluctuation property of the power load time series diagram; the periodic fluctuation property is the existence of an irregular periodic fluctuation or the non-existence of an irregular periodic fluctuation, and the seasonal fluctuation property is the existence of a fixed seasonal fluctuation or the non-existence of a fixed seasonal fluctuation;

[0034] According to different combinations of the periodic fluctuation property and the seasonal fluctuation property, determine different target prediction models from a plurality of preset time series-based prediction models;

[0035] According to the historical power load data and the target prediction model, obtain the predicted load data of the power system.

[0036] In a fourth aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0037] According to the historical power load data of the power system, obtain a power load time series diagram;

[0038] According to the power load time series diagram, respectively determine the periodic fluctuation property and the seasonal fluctuation property of the power load time series diagram; the periodic fluctuation property is the existence of an irregular periodic fluctuation or the non-existence of an irregular periodic fluctuation, and the seasonal fluctuation property is the existence of a fixed seasonal fluctuation or the non-existence of a fixed seasonal fluctuation;

[0039] According to different combinations of the periodic fluctuation property and the seasonal fluctuation property, determine different target prediction models from a plurality of preset time series-based prediction models;

[0040] According to the historical power load data and the target prediction model, obtain the predicted load data of the power system.

[0041] In a fifth aspect, the present application also provides a computer program product, including a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0042] According to the historical power load data of the power system, obtain a power load time series diagram;

[0043] According to the power load time series diagram, respectively determine the periodic fluctuation property and the seasonal fluctuation property of the power load time series diagram; the periodic fluctuation property is the existence of an irregular periodic fluctuation or the non-existence of an irregular periodic fluctuation, and the seasonal fluctuation property is the existence of a fixed seasonal fluctuation or the non-existence of a fixed seasonal fluctuation;

[0044] According to different combinations of the periodic fluctuation property and the seasonal fluctuation property, determine different target prediction models from a plurality of preset time series-based prediction models;

[0045] Based on the historical power load data and the target prediction model, the predicted load data of the power system is obtained.

[0046] The above power system load forecasting method, device, computer equipment, computer-readable storage medium and computer program product. The method obtains a power load time series graph based on the historical power load data of the power system, which is convenient for intuitively performing time series analysis of the historical power load data. According to the power load time series graph, the periodic fluctuation property and the seasonal fluctuation property of the power load time series graph are respectively determined. Among them, the periodic fluctuation property is the existence of non-fixed periodic fluctuations or the non-existence of non-fixed periodic fluctuations, and the seasonal fluctuation property is the existence of fixed seasonal fluctuations or the non-existence of fixed seasonal fluctuations, which serves as a preparation for subsequently selecting a prediction model adapted to the power system. Further, according to different combinations of the periodic fluctuation property and the seasonal fluctuation property, different target prediction models are determined from a plurality of preset time series-based prediction models. Based on the historical power load data and the target prediction model, the predicted load data of the power system is obtained. By determining the corresponding target prediction model according to different time characteristics and different scenarios to predict the load data of the power system, the prediction accuracy is improved. Brief Description of the Drawings

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required to be used in the description of the embodiments of the present application or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0048] Figure 1 It is a schematic flowchart of a power system load forecasting method in an embodiment;

[0049] Figure 2 It is a schematic flowchart of the target prediction model determination step in an embodiment;

[0050] Figure 3 It is a schematic block diagram of a time series-based power system load forecasting system in an embodiment;

[0051] Figure 4 It is a schematic flowchart of a time series-based power system load forecasting method in an embodiment;

[0052] Figure 5 It is a schematic flowchart of a time series-based power system load forecasting method in another embodiment;

[0053] Figure 6It is a structural block diagram of a power system load forecasting device in an embodiment;

[0054] Figure 7 It is an internal structure diagram of a computer device in an embodiment. Specific implementation manners

[0055] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0056] As described in the background art, the power system load data prediction method of the related art has the problem of low accuracy. After research by the inventor, it is found that the reason for this problem is that the prediction of the power system load (power demand, i.e., active power) refers to fully considering the influence of factors such as historical system load, economic conditions, meteorological conditions, and social events to predict the system load for a period of time in the future. Load prediction is an important part of power system planning and scheduling. Short-term (within two weeks) prediction is the basis for the start-stop, scheduling, and operation plan formulation of internal units in the power grid; medium-term (in the coming months) prediction can provide support for ensuring the electricity consumption of enterprise production and social life and reasonably arranging the operation and maintenance decisions of the power grid; long-term (in the coming years) prediction can provide reference for the formulation of plans such as power grid transformation and expansion to improve the economic and social benefits of the power system. And because a large number of new factors such as new energy and electric vehicles are integrated into the load side, and in addition, the power market reform has led to the emergence of a variety of new roles, the future load of the new power system will show more complex new characteristics. Therefore, the load prediction of the new power system will also face greater challenges and at the same time has more important research significance. In this regard, many scholars and engineers have done a lot of research on power system load prediction. Common prediction models include those based on statistical models, such as regression prediction; those based on machine learning and deep learning, such as support vector products, random forests, etc. The support vector machine method uses statistical learning theory to predict future load by establishing a linear classifier. This method requires the selection of appropriate kernel functions and parameters, and the selection of the number of parameters is difficult. The random forest method is an ensemble algorithm composed of decision trees. When the random forest deals with regression problems, it is called random forest regression. By randomly sampling samples and features, multiple uncorrelated decision trees are established, and the results of all decision trees are combined to obtain the final prediction result. The disadvantage is that when the data is unbalanced, it will lead to a decrease in the classification accuracy. The neural network method uses the self-learning ability of the neural network to establish a non-linear mapping relationship between the load and related factors through learning historical data, and then predicts the future load. This method requires a large amount of training and adjustment of the neural network, making the parameters of the neural network redundant, the convergence speed slow, and the local minimization problem cannot be ignored. Commonly used machine learning prediction methods are very sensitive to the quality and magnitude of data. If the data quality is poor or the data volume is insufficient, it may affect the prediction ability and stability of the model, and is not applicable to complex non-linear relationships. For non-linear and high-dimensional data, traditional machine learning methods such as linear regression and support vector machines may not be able to well capture complex data patterns and relationships. And deep learning prediction methods require a large amount of labeled data for training and a large amount of computing resources for training and inference. This may be a challenge for scenarios with low data acquisition and processing capabilities. Moreover, due to the different selections of prediction data, many load prediction models are dependent on data. Therefore, they generally only show relatively good prediction performance for prediction problems with specific data characteristics under specific circumstances.However, in practical applications, the artificial intelligence-based load forecasting model usually uses the method of "offline training and online application" for forecasting. However, the load pattern of the power system often exists in a dynamically changing environment, and the potential distribution of various data will also change dynamically over time, resulting in a phenomenon of low forecasting accuracy when the well-trained forecasting model is applied online. Therefore, the artificial intelligence-based load forecasting model is prone to problems such as a single applicable scenario and weak generalization ability. Uncertain factors such as complex and changeable meteorological conditions and social events will all have a certain impact on the power system load, making the application of traditional load forecasting models have certain limitations. At the same time, with the diversification of the power system load structure, the application effect of the model is also reduced. Therefore, the power system load forecasting problem urgently needs further research.

[0057] For the above reasons, the present application provides a power system load forecasting method, aiming to improve the accuracy of power system load forecasting.

[0058] In one embodiment, as Figure 1 shown, a power system load forecasting method is provided. In this embodiment, an example is given where the method is applied to a server system. It can be understood that the method can also be applied to a server or a terminal, and can also be applied to a system including a terminal, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0059] Step S102, obtain a power load time series graph according to the historical power load data of the power system.

[0060] Among them, the historical power load data can be the time series data of the electricity demand of each node or area of the power system within a historical time period, which reflects the power consumption at different time points and is usually recorded in units of hours, days, months or years.

[0061] Among them, the power load time series graph can be a graph drawn by power load data within a period of time (usually with the power demand as the vertical axis and time as the horizontal axis), which intuitively shows the trend and pattern of the power load changing with time.

[0062] Optionally, the system records the power load data of the power system within a historical time period as the historical power load data, marks the historical power load data with the corresponding time, and processes the historical power load data using a drawing software to obtain a power load time series graph.

[0063] Step S104, respectively determine the periodic fluctuation property and seasonal fluctuation property of the power load time series graph according to the power load time series graph.

[0064] Among them, the nature of periodic fluctuations is the existence or non-existence of irregular periodic fluctuations, and the nature of seasonal fluctuations is the existence or non-existence of fixed seasonal fluctuations. Among them, the periodic pattern may not be a fixed period, and the period lengths are not exactly the same, but will show similar fluctuations. Seasonality is a special case of periodicity, with a fixed period length and usually related to the calendar.

[0065] Optionally, the system analyzes the change patterns of historical power load data in different seasons or periods based on the power load time series graph, and respectively determines the nature of periodic fluctuations and the nature of seasonal fluctuations of the power load time series graph as a reference basis for subsequent selection of the prediction model.

[0066] Step S106, according to different combinations of the nature of periodic fluctuations and the nature of seasonal fluctuations, determine different target prediction models from a plurality of preset time series-based prediction models.

[0067] Among them, the time series-based prediction model can be a statistical model or a machine learning model that uses historical time series data to predict future values. By analyzing the patterns (such as trends, seasonality, periodicity) of data points changing over time in the time series, potential relationships are identified and used to predict future data points.

[0068] Optionally, the system determines a target prediction model adapted to the combination of the nature of periodic fluctuations and the nature of seasonal fluctuations of the power load time series graph from a plurality of preset time series-based prediction models according to different combinations of the nature of periodic fluctuations and the nature of seasonal fluctuations existing in the power load time series graph.

[0069] Step S108, obtain the predicted load data of the power system according to the historical power load data and the target prediction model.

[0070] Among them, the predicted load data can be the predicted value of the power load data at the current moment or in the current period.

[0071] Optionally, the system inputs the historical power load data into the target prediction model to obtain the predicted load data of the power system output by the target prediction model.

[0072] In the above power system load forecasting method, based on the historical power load data of the power system, a power load time series graph is obtained, which facilitates the intuitive time series analysis of the historical power load data. According to the power load time series graph, the periodic fluctuation property and the seasonal fluctuation property of the power load time series graph are respectively determined. Among them, the periodic fluctuation property is the existence or non-existence of an irregular period fluctuation, and the seasonal fluctuation property is the existence or non-existence of a fixed seasonal fluctuation, laying a foundation for subsequently selecting a forecasting model adapted to the power system. Further, according to different combinations of the periodic fluctuation property and the seasonal fluctuation property, different target forecasting models are determined from a plurality of preset time series-based forecasting models. According to the historical power load data and the target forecasting model, the predicted load data of the power system is obtained. By determining the corresponding target forecasting model according to different time characteristics and different scenarios to predict the load data of the power system, the forecasting accuracy is improved.

[0073] In an exemplary embodiment, step S104 of respectively determining the seasonal fluctuation property and the periodic fluctuation property of the power load time series graph according to the power load time series graph includes:

[0074] Perform an irregular period detection on the power load time series graph to obtain the periodic fluctuation property of the power load time series graph; perform a fixed period detection on the power load time series graph to obtain the seasonal fluctuation property of the power load time series graph.

[0075] Among them, the irregular period detection can detect the regularity and similarity of the power data in the power load time series graph with an irregular time length as the period. Among them, the fixed period detection can detect the regularity and similarity of the power data in the power load time series graph with a fixed period, such as year, quarter, month, etc. as the period.

[0076] Optionally, the system performs an irregular period detection on the power load time series graph to obtain the periodic fluctuation property of the power load time series graph, such as directly observing the fluctuation of the power load time series graph with an irregular period, or using methods such as Fourier transform and autocorrelation analysis for detection; perform a fixed period detection on the power load time series graph to obtain the seasonal fluctuation property of the power load time series graph. Specifically, by dividing the historical power load data into segments according to a fixed period (such as year, quarter, month, day) and plotting them into a graph, the change pattern of the data in different seasons or periods can be intuitively observed. If the data shows similar fluctuation patterns in different seasons or periods, and these fluctuation patterns are regular and repetitive, and the regularity and repeatability of the fluctuation patterns are strong, it indicates that the seasonal fluctuation property is the existence of a fixed seasonal fluctuation.

[0077] In this embodiment, by detecting the periodic fluctuation properties with an unfixed period and the seasonal fluctuation properties with a fixed period for the historical power load data respectively, the irregular fluctuations in the power load caused by special events (such as holidays, large-scale events, abnormal weather, etc.) can be captured to understand the influence of these unfixed periods, which can improve the accuracy of short-term and medium-term load forecasting and identify the seasonal patterns of the power load. Understanding the peak and trough periods of electricity consumption in different seasons, months, and even days of the year lays the foundation for selecting an appropriate prediction model subsequently.

[0078] In an exemplary embodiment, the steps of the above embodiment for performing a fixed-period detection on the power load time series graph to obtain the seasonal fluctuation properties of the power load time series graph include:

[0079] Dividing the power load time series graph according to a preset fixed period to obtain a plurality of power load time series sub-graphs; performing a correlation analysis on the plurality of power load time series sub-graphs to obtain the seasonal fluctuation properties of the power load time series graph.

[0080] Among them, the correlation analysis can be a statistical method used to measure the strength and direction of the relationship between two or more variables.

[0081] Optionally, the system divides the power load time series graph according to a preset fixed period (such as year, quarter, month, and day) to obtain a plurality of power load time series sub-graphs, performs a correlation analysis between the plurality of power load time series sub-graphs to obtain the seasonal fluctuation properties of the power load time series graph. Specifically, the seasonal index of the power load time series sub-graph can be calculated, and the threshold for the seasonal index to deviate from 100% is ±10%. That is, if the seasonal index of a certain quarter is greater than 110% or less than 90%, it can be considered that there is an obvious seasonal fluctuation in that quarter, and the seasonal fluctuation property is the existence of a fixed seasonal fluctuation. It can also be analyzed according to the p-value (a concept in hypothesis testing). If the p-value is much smaller than the significance level (such as 0.05), the seasonality is obvious. If the p-value is higher than the threshold, it indicates that the seasonal fluctuation is not significant.

[0082] In this embodiment, the seasonal fluctuation information of the power load time series graph can be incorporated into the load forecasting model, thereby improving the prediction accuracy, and the parameters of the model can be adjusted for different seasons, or a seasonal decomposition model can be used to predict the load to further improve the prediction accuracy.

[0083] In an exemplary embodiment, such as Figure 2As shown, step S106 determines different target prediction models from a plurality of preset time series-based prediction models according to different combinations of periodic fluctuation properties and seasonal fluctuation properties, including the following steps S202 to S206. Among them:

[0084] Step S202, when the periodic fluctuation property is the existence of an unfixed periodic fluctuation, determine the target prediction model as the triple exponential smoothing model from a plurality of preset time series-based prediction models.

[0085] Among them, the triple exponential smoothing model (Holt-Winters) is a prediction method that predicts trends and seasonal changes in future time series by weighted averaging historical data. It extends the simple exponential smoothing model by introducing additional parameters to better handle trends and seasonal components in time series. It uses three parameters: smoothing coefficients, and, which are used to smooth the level, trend, and seasonality of the data respectively. These smoothing operations enable the model to adapt to data changes and predict future trends and seasonal variations.

[0086] Optionally, when the system determines that the periodic fluctuation property is the existence of an unfixed periodic fluctuation, it determines the target prediction model as the triple exponential smoothing model from a plurality of preset time series-based prediction models to obtain power load prediction data of the power system.

[0087] Step S204, when the periodic fluctuation property is the non-existence of an unfixed periodic fluctuation and the seasonal fluctuation property is the existence of a fixed seasonal fluctuation, determine the target prediction model as the seasonal autoregressive integrated moving average model from a plurality of preset time series-based prediction models.

[0088] Among them, the seasonal autoregressive integrated moving average model (SARIMA, Seasonal AutoregressiveIntegrated Moving Average) model is a time series analysis and prediction method that adds a seasonal component to the ARIMA (AutoRegressive Integrated MovingAverage, autoregressive integrated moving average) model.

[0089] Optionally, when the system determines that the nature of the periodic fluctuation is the absence of an irregular periodic fluctuation and the nature of the seasonal fluctuation is the presence of a fixed seasonal fluctuation, it determines the target prediction model as the Seasonal AutoRegressive Integrated Moving Average (SARIMA) model from a plurality of preset time-series-based prediction models, and inputs the historical power load data into this model to obtain the power load prediction data. Specifically, the expression of the SARIMA model is as follows: , where the lowercase letters p, d, and g are the orders of the non-seasonal autoregressive, differencing, and moving average terms respectively; the uppercase letters P, D, Q, and S are the orders of the seasonal autoregressive, differencing, moving average terms, and the length of the seasonal period respectively. The expression of the SARIMA model can be divided into two parts: the non-seasonal part and the seasonal part. Among them, the expression of the non-seasonal part is:

[0090] )

[0091] where B is the Backshift Operator; B, B 2 to B P are the p-th order of the Backshift Operator; p is the order of the non-seasonal autoregression; d is the order of the non-seasonal differencing; , to and to are the coefficients of the non-seasonal autoregressive and moving average terms respectively; , is the observed value of the time series at time t; is the white noise error term.

[0092] The expression of the seasonal part is:

[0093] )

[0094] where B is the Backshift Operator; B ds , B qs are the ds-th and qs-th orders of the Backshift Operator B respectively; , to and to are the coefficients of the non-seasonal autoregressive and moving average terms respectively; , is the observed value of the time series at time t; s is the length of the seasonal period; Q is the order of the seasonal moving average term; D is the order of the seasonal differencing; is the white noise error term.

[0095] Step S206, in the case where the periodic fluctuation property is that there is no unfixed periodic fluctuation and the seasonal fluctuation property is that there is no fixed seasonal fluctuation, determine the target prediction model as the autoregressive integrated moving average (ARIMA) model from a plurality of preset time series-based prediction models.

[0096] Among them, the autoregressive integrated moving average (ARIMA) model is a commonly used time series analysis and prediction method. It combines the characteristics of the autoregressive (AR) model, differencing, and the moving average (MA) model to capture the autocorrelation, trend, and seasonality of the data and predict future trends and changes.

[0097] Optionally, when the system determines that the periodic fluctuation property is that there is no unfixed periodic fluctuation and the seasonal fluctuation property is that there is no fixed seasonal fluctuation, it determines the target prediction model as the autoregressive integrated moving average (ARIMA) model from a plurality of preset time series-based prediction models, and inputs the historical power load data into this model to obtain power load prediction data. Specifically, the expression corresponding to the autoregressive integrated moving average (ARIMA) model is:

[0098]

[0099] Among them, B is the backward lag operator; B, B 2 to B p is the p-th order of the backward lag operator; is the time series of t; is the white noise error term; , to and , to are the coefficients of the non-seasonal autoregressive and moving average terms respectively; is 's d-th difference; p is the order of the autoregressive term; q is the order of the moving average term; d is the number of differencing times.

[0100] It should be noted that the establishment of the autoregressive integrated moving average (ARIMA) model generally includes three main steps: determining the model order, estimating the model parameters, and model diagnosis. First, determine the model order. The ARIMA model consists of three parameters, p, d, and g, representing the autoregressive order (AR), the differencing order (I), and the moving average order (MA), respectively. By observing the autocorrelation function (ACF) and partial autocorrelation function (PACF) of the time series data, the model order can be preliminarily determined. The second step is to estimate the model parameters. Using methods such as maximum likelihood estimation, the parameters of AR, I, and MA are estimated based on historical data. These parameters reflect the degree of autocorrelation, trend, and seasonal influence of the data. The last step is model diagnosis, including: statistical significance test: checking the statistical significance of the model parameters, that is, whether the t-value or z-value of the parameters is significantly different from zero; goodness-of-fit test: evaluating the goodness-of-fit of the model through some statistical indicators. The smaller these indicators are, the better the model fit is usually; residual sum of squares (RSS): is also an important indicator to evaluate the model fitting effect. The smaller the RSS, the smaller the fitting error of the model. By testing the residual sequence, the fitting effect of the model and whether the residual sequence conforms to the model assumptions are judged, including: white noise: the residuals should exhibit white noise, that is, the elements in the residual sequence are not correlated with each other. This can be verified by the Q test (Ljung-Box test); normality: the residual distribution should be close to a normal distribution. This can be evaluated by methods such as plotting the residual histogram and performing a normality test (such as the Shapiro-Wilk test); no autocorrelation: the autocorrelation function (ACF) and partial autocorrelation function (PACF) of the residuals should have no obvious correlation. This can be achieved by observing the ACF and PACF plots.

[0101] In this embodiment, when there are irregular periodic fluctuations, the triple exponential smoothing model can better adapt to this non-regular change, capture the short-term trends and level changes, and thus improve the prediction accuracy. When there are no irregular periodic fluctuations but there are fixed seasonal fluctuations, the SARIMA model can make full use of the seasonal information in the historical data, eliminate the seasonal influence through seasonal differencing, and combine the modeling capabilities of autoregression and moving average to more accurately predict the future load. When there are neither irregular periodic fluctuations nor fixed seasonal fluctuations, the ARIMA model can eliminate the non-stationarity of the time series through differencing and then use the modeling capabilities of autoregression and moving average to predict the future load. By selecting the most suitable prediction model according to different fluctuation properties, the information in the historical data can be maximally utilized, the prediction error can be reduced, and the prediction accuracy can be improved.

[0102] In an exemplary embodiment, the triple exponential smoothing model includes: a triple exponential smoothing additive model and a triple exponential smoothing multiplicative model. Step S202, in the case where the periodic fluctuation property is an unfixed periodic fluctuation, determines the target prediction model as the triple exponential smoothing model from a plurality of preset time series-based prediction models, including:

[0103] In the case where the periodic fluctuation property is an unfixed periodic fluctuation and the seasonal fluctuation property is a fixed seasonal fluctuation, determines the target prediction model as the triple exponential smoothing multiplicative model from a plurality of preset time series-based prediction models; in the case where the periodic fluctuation property is an unfixed periodic fluctuation and the seasonal fluctuation property is no fixed seasonal fluctuation, determines the target prediction model as the triple exponential smoothing additive model from a plurality of preset time series-based prediction models.

[0104] Among them, the triple exponential smoothing multiplicative model can be a triple exponential smoothing model based on multiplication, which assumes that the trend and seasonal components of the time series are multiplied together, that is, the magnitude of the seasonal fluctuation is proportional to the level of the time series. Among them, the triple exponential smoothing multiplicative model can be a triple exponential smoothing model based on addition, which assumes that the trend and seasonal components of the time series are superimposed together, that is, the magnitude of the seasonal fluctuation is independent of the level of the time series.

[0105] Optionally, when the system determines that the periodic fluctuation property is an unfixed periodic fluctuation and the seasonal fluctuation property is a fixed seasonal fluctuation, it determines the target prediction model as the triple exponential smoothing multiplicative model from a plurality of preset time series-based prediction models. The expression corresponding to the triple exponential smoothing multiplicative model is:

[0106]

[0107]

[0108]

[0109] Among them, 、 and , are the level, trend, and seasonality of the smoothed data respectively; , and are the estimated levels at time t for the smoothed, trend, and season respectively; is the estimated level of smoothing at time t - 1; is the estimated level of trend at time t - 1; is at time t - The estimated level of the moment season; is the time observation value at time t.

[0110] In addition, when the system has non-fixed periodic fluctuations in the periodic fluctuation property and no fixed seasonal fluctuations in the seasonal fluctuation property, the target prediction model is determined to be the triple exponential smoothing additive model from a plurality of preset time series-based prediction models. The expression corresponding to the triple exponential smoothing additive model is:

[0111]

[0112]

[0113]

[0114] Wherein, 、 and are the level, trend, and seasonality of the smoothed data respectively; , and are the estimated levels of the smoothed, trend, and season at time t respectively; is the smoothed estimated level at time t-1; is the estimated level of the trend at time t-1; is t- the estimated level of the moment season; is the time observation value at time t.

[0115] In this embodiment, the periodic fluctuation property has non-fixed periodic fluctuations, and the seasonal fluctuation property has fixed seasonal fluctuations. Selecting the triple exponential smoothing multiplicative model can better fit the power load data with both irregular fluctuations and proportional seasonal fluctuations, improving the prediction accuracy. When the periodic fluctuation property has non-fixed periodic fluctuations and the seasonal fluctuation property has no fixed seasonal fluctuations, the triple exponential smoothing additive model is selected. The additive model avoids overfitting the seasonal component and focuses more on capturing the fluctuations of non-fixed periods. This strategy of selecting a model based on the fluctuation property can improve the robustness of the model and enable it to maintain good prediction performance in different load scenarios.

[0116] In an exemplary embodiment, step S102 obtains a power load time series graph according to the historical power load data of the power system, including:

[0117] Obtain the historical power load data of the power system; preprocess the historical power load data and arrange it according to the date to obtain the historical power load sequence; obtain the power load time series graph based on the historical power load sequence.

[0118] Optionally, arrange the historical power load data into a historical power load sequence in the form of a two-dimensional data set according to the date, and preprocess the historical power load sequence. Specifically, first, use a density-based method to globally identify and correct the abnormal data in the historical power load sequence in two dimensions due to reasons such as channel errors, maximum and minimum loads, and sudden accidents. Then, on this basis, for the possible impact loads and spike loads caused by channel noise in the historical power load sequence, use a two-dimensional wavelet threshold denoising method to denoise the historical power load sequence in two dimensions to further improve the quality of the load data and obtain the preprocessed historical power load sequence. The system further uses drawing software to draw based on the preprocessed historical power load sequence to obtain the power load time series graph. For example, the preprocessed historical power load sequence is imported into Python (a computer programming language), and the pandas (a library for data processing and analysis in Python) library is the preferred tool for processing time series data. Pandas provides a DataFrame (a two-dimensional tabular data structure) data structure, which can easily import, clean, transform, and analyze time series data. Use the read_csv (a function for reading CSV files) function of pandas to import the time series data in csv format, and then use the to_datetime (a function for converting data to datetime format) function to convert the date column to the pateTimeIndex (an index format for processing time series data) format of pandas, which can facilitate time series analysis more conveniently.

[0119] In this embodiment, plotting the power load data as a time series graph can intuitively observe the overall trend, seasonal fluctuations, periodic changes, and abnormal conditions of the power load. It helps to quickly understand the basic characteristics of the data and discover potential patterns and problems.

[0120] In one embodiment, as Figure 3 shown, a power system load prediction system based on time series is provided, including a data processing module, a graph drawing module, a periodicity and seasonality judgment module, and a time series-based prediction module.

[0121] Among them, the data processing module includes a historical load processing module, the periodic and seasonal judgment module includes a periodic judgment module and a seasonal judgment module, and the time series-based prediction module includes a Holt-Winters three-parameter exponential smoothing model module, an ARIMA model module, and a SARIMA model module.

[0122] In one embodiment, as Figure 4 shown, a time series-based power system load forecasting method is provided, which is applied to Figure 3 the time series-based power system load forecasting system shown, and includes:

[0123] Step 1: Make a time series graph based on historical data to obtain historical power load data. Power load forecasting depends on a large amount of historical data and related factor analysis. The forecasting result depends to a great extent on the reliability of the collected historical data and the detailed accuracy of the related factor analysis data. Process the collected historical data. Since there are generally some abnormal data in power load data, by considering both the horizontal and vertical continuity of the load, arrange the load data into a two-dimensional data set according to the date. First, use a density-based method to globally identify and correct the abnormal data in the load data due to channel errors, maximum and minimum loads, and sudden accidents in two dimensions. Then, on this basis, for the possible impact loads and burr loads caused by channel noise in the load data, use a two-dimensional wavelet threshold denoising method to denoise the load data in two dimensions to further improve the quality of the load data. Draw a time series graph. A simple and effective way to analyze time series data is to visualize the time series data on a chart. Import the processed power consumption data into Python. The pandas library in it is the preferred tool for processing time series data. Pandas provides a DataFrame data structure that can easily import, clean, transform, and analyze time series data. Use the read_csv function of pandas to import the time series data in csv format, and then use the to_datetime function to convert the date column to the pateTimeIndex format of pandas, which can facilitate time series analysis more conveniently.

[0124] Step 2: According to the made time series graph, judge the degree of periodic fluctuation of the data. If the data shows a large degree of periodic fluctuation, use the Holt-winters model. This method takes into account the seasonal and periodic effects in the data and is suitable for data with obvious seasonality and periodicity. If the degree of periodic fluctuation of the data is small, select other models through the next judgment.

[0125] Optionally, the Holt-Winters model is used for load forecasting. The Holt-Winters three-parameter exponential smoothing model makes predictions by considering the trend, seasonality, and stationary components of the data. It uses three parameters: smoothing coefficients α, β, and γ, which are used to smooth the level, trend, and seasonality of the data respectively. These smoothing operations enable the model to adapt to changes in the data and predict future trends and seasonal variations. If the time series changes approximately equally over seasons and the amount of change is roughly the same each year, the Holt-Winters additive model is adopted; if the series fluctuates greatly over seasons, the Holt-Winters multiplicative model is used.

[0126] Step 3: By splitting the time series data into segments at fixed intervals (such as years, quarters, months, days) and plotting them, the change patterns of the data in different seasons or cycles can be visually observed. If the data shows similar fluctuation patterns in different seasons or cycles, and these fluctuation patterns are regular and repetitive, and if the regularity and repeatability of the fluctuation patterns are strong, it indicates strong seasonal fluctuations. The specific quantification criteria are as follows:

[0127] Assume that for quarterly data, the threshold for the seasonal index to deviate from 100% is ±10%. That is, if the seasonal index of a certain quarter is greater than 110% or less than 90%, it can be considered that there are obvious seasonal fluctuations in that quarter. It is also possible to analyze based on the p-value. If the p-value is much smaller than the significance level (such as 0.05), the seasonality is obvious; if the p-value is large, it indicates that the seasonal fluctuations are not significant. At the same time, the autocorrelation function (ACF) and partial autocorrelation function (PACF) plots can be combined for analysis. If the ACF plot shows significant correlations at seasonal lags (such as lags 4, 8, 12 for quarterly data), and the PACF plot decays rapidly to insignificance at these lags, there is strong seasonality. If the seasonality is not obvious, the ARIMA model is used. If there is strong seasonality, the Sarima model is used. Compared with the ARIMA model, the Sarima model performs more significantly in fitting and prediction and can better adapt to the seasonal fluctuations of the data.

[0128] Optionally, in the absence of seasonal fluctuations, the ARIMA model is selected for prediction, and in the presence of seasonal fluctuations, the SARIMA model is selected for prediction.

[0129] It should be noted that the judgment logic of this embodiment can also be carried out according to Figure 5Perform power load data prediction according to the process. According to the time series graph, first judge the seasonal fluctuation, and then judge according to the periodic fluctuation. In the case of seasonal fluctuation and periodic fluctuation, select the Holt-Winters multiplicative model; in the case of seasonal fluctuation and no periodic fluctuation, select the SARIMA model; in the case of no seasonal fluctuation and periodic fluctuation, select the Holt-Winters additive model; in the case of no seasonal fluctuation and no periodic fluctuation, select the ARIMA model.

[0130] In this embodiment, the periodic and seasonal changes of the load can be considered simultaneously, and the data can be analyzed periodically and seasonally, so as to select an appropriate time series analysis method to predict future data. This time series analysis method is a prediction method based on historical data. By analyzing and modeling the load data in the past period of time, the load situation in the future period of time can be predicted. The core idea is that the future load situation is related to the past load situation to a certain extent, and the prediction is made by analyzing this correlation. Compared with machine learning prediction methods and deep learning prediction methods, the advantage of this method is that it requires less data, can reduce the inaccuracy of prediction results caused by low data quality. And it does not require a large amount of computing resources for training and inference, and can also make accurate predictions in scenarios with low data acquisition and processing capabilities.

[0131] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise clearly stated in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least a part of other steps or steps or stages in other steps.

[0132] Based on the same inventive concept, the embodiments of the present application also provide a power system load prediction device for implementing the power system load prediction method described above. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the power system load prediction device provided below can refer to the limitations on the power system load prediction method in the above text, and will not be repeated here.

[0133] In an exemplary embodiment, as Figure 6As shown, a power system load forecasting device 600 is provided, including: a sequence diagram generation module 602, a fluctuation property determination module 604, a prediction model determination module 606, and a load data prediction module 608, where:

[0134] The sequence diagram generation module 602 is configured to obtain a power load time series diagram based on the historical power load data of the power system.

[0135] The fluctuation property determination module 604 is configured to respectively determine the periodic fluctuation property and the seasonal fluctuation property of the power load time series diagram according to the power load time series diagram; the periodic fluctuation property is the presence or absence of an unfixed periodic fluctuation, and the seasonal fluctuation property is the presence or absence of a fixed seasonal fluctuation.

[0136] The prediction model determination module 606 is configured to determine different target prediction models from a plurality of preset time series-based prediction models according to different combinations of the periodic fluctuation property and the seasonal fluctuation property.

[0137] The load data prediction module is configured to obtain the predicted load data of the power system according to the historical power load data and the target prediction model.

[0138] Further, in one embodiment, the fluctuation property determination module 604 is further configured to perform an unfixed period detection on the power load time series diagram to obtain the periodic fluctuation property of the power load time series diagram; perform a fixed period detection on the power load time series diagram to obtain the seasonal fluctuation property of the power load time series diagram.

[0139] Further, in one embodiment, the fluctuation property determination module 604 is further configured to divide the power load time series diagram according to a preset fixed period to obtain a plurality of power load time series sub-diagrams; perform a correlation analysis on the plurality of power load time series sub-diagrams to obtain the seasonal fluctuation property of the power load time series diagram.

[0140] Further, in one embodiment, the prediction model determination module 606 is further configured to, when the periodic fluctuation property is the presence of an unfixed periodic fluctuation, determine the target prediction model as a triple exponential smoothing model from a plurality of preset time series-based prediction models; when the periodic fluctuation property is the absence of an unfixed periodic fluctuation and the seasonal fluctuation property is the presence of a fixed seasonal fluctuation, determine the target prediction model as a seasonal autoregressive integrated moving average model from a plurality of preset time series-based prediction models; when the periodic fluctuation property is the absence of an unfixed periodic fluctuation and the seasonal fluctuation property is the absence of a fixed seasonal fluctuation, determine the target prediction model as an autoregressive integrated moving average model from a plurality of preset time series-based prediction models.

[0141] Further, in one embodiment, the prediction model determination module 606 is further configured to, when the periodic fluctuation property is of an unfixed periodic fluctuation and the seasonal fluctuation property is of a fixed seasonal fluctuation, determine, from a plurality of preset time series-based prediction models, that the target prediction model is a triple exponential smoothing multiplicative model; and when the periodic fluctuation property is of an unfixed periodic fluctuation and the seasonal fluctuation property is of no fixed seasonal fluctuation, determine, from a plurality of preset time series-based prediction models, that the target prediction model is a triple exponential smoothing additive model.

[0142] Further, in one embodiment, the sequence diagram generation module 602 is further configured to obtain historical power load data of the power system; preprocess the historical power load data and arrange it according to dates to obtain a historical power load sequence; and obtain a power load time series diagram based on the historical power load sequence.

[0143] Each module in the above power system load prediction device 600 can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in the processor of the computer device in hardware form or be independent of it, or can be stored in the memory of the computer device in software form, so as to facilitate the processor to call and execute the operations corresponding to the above respective modules.

[0144] In an exemplary embodiment, a computer device is provided. The computer device can be a server, and its internal structure diagram can be as Figure 7 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as historical power load data and power load time series diagrams. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. The computer program, when executed by the processor, implements a power system load prediction method.

[0145] Those skilled in the art can understand, Figure 7The structure shown is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have a different component layout.

[0146] In one embodiment, a computer device is further provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0147] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0148] In one embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0149] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, artificial intelligence (AI) processors, etc., without limitation.

[0150] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in the present application.

[0151] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A method for predicting the load of a power system, characterized in that, The method includes: Obtaining a power load time series graph based on historical power load data of a power system; Respectively determining the periodic fluctuation property and seasonal fluctuation property of the power load time series graph according to the power load time series graph; the periodic fluctuation property is the existence or non-existence of an irregular periodic fluctuation, and the seasonal fluctuation property is the existence or non-existence of a fixed seasonal fluctuation; Determining different target prediction models from a plurality of preset time series-based prediction models according to different combinations of the periodic fluctuation property and the seasonal fluctuation property; Obtaining predicted load data of the power system according to the historical power load data and the target prediction model.

2. The method according to claim 1, wherein The respectively determining the seasonal fluctuation property and periodic fluctuation property of the power load time series graph according to the power load time series graph includes: Performing an irregular period detection on the power load time series graph to obtain the periodic fluctuation property of the power load time series graph; Performing a fixed period detection on the power load time series graph to obtain the seasonal fluctuation property of the power load time series graph.

3. The method according to claim 2, wherein The performing a fixed period detection on the power load time series graph to obtain the seasonal fluctuation property of the power load time series graph includes: Dividing the power load time series graph according to the preset fixed period to obtain a plurality of power load time series sub-graphs; Performing a correlation analysis on the plurality of power load time series sub-graphs to obtain the seasonal fluctuation property of the power load time series graph.

4. The method according to claim 1, characterized in that The determining different target prediction models from a plurality of preset time series-based prediction models according to different combinations of the periodic fluctuation property and the seasonal fluctuation property includes: When the periodic fluctuation property is the existence of an irregular periodic fluctuation, determining the target prediction model as a triple exponential smoothing model from a plurality of preset time series-based prediction models; When the periodic fluctuation property is the non-existence of an irregular periodic fluctuation and the seasonal fluctuation property is the existence of a fixed seasonal fluctuation, determining the target prediction model as a seasonal autoregressive integrated moving average model from a plurality of preset time series-based prediction models; When the periodic fluctuation property is the non-existence of an irregular periodic fluctuation and the seasonal fluctuation property is the non-existence of a fixed seasonal fluctuation, determining the target prediction model as an autoregressive integrated moving average model from a plurality of preset time series-based prediction models.

5. The method according to claim 4, wherein The triple exponential smoothing model includes: a triple exponential smoothing additive model and a triple exponential smoothing multiplicative model. The determining the target prediction model as a triple exponential smoothing model from a plurality of preset time series-based prediction models when the periodic fluctuation property is the existence of an irregular periodic fluctuation includes: When the periodic fluctuation property is the existence of an irregular periodic fluctuation and the seasonal fluctuation property is the existence of a fixed seasonal fluctuation, determining the target prediction model as a triple exponential smoothing multiplicative model from a plurality of preset time series-based prediction models; In the case where the periodic fluctuation property is the existence of non-fixed periodic fluctuations and the seasonal fluctuation property is the non-existence of fixed seasonal fluctuations, the target prediction model is determined to be the triple exponential smoothing additive model from a plurality of preset prediction models based on time series.

6. The method according to claim 1, wherein The obtaining of the power load time series diagram according to the historical power load data of the power system includes: Obtaining the historical power load data of the power system; Preprocessing the historical power load data and arranging it according to dates to obtain a historical power load sequence; Obtaining the power load time series diagram according to the historical power load sequence.

7. A power system load forecasting device, characterized in that, The device includes: A sequence diagram generation module for obtaining a power load time series diagram according to the historical power load data of the power system; A fluctuation property determination module for respectively determining the periodic fluctuation property and the seasonal fluctuation property of the power load time series diagram according to the power load time series diagram; the periodic fluctuation property is the existence of non-fixed periodic fluctuations or the non-existence of non-fixed periodic fluctuations, and the seasonal fluctuation property is the existence of fixed seasonal fluctuations or the non-existence of fixed seasonal fluctuations; A prediction model determination module for determining different target prediction models from a plurality of preset prediction models based on time series according to different combinations of the periodic fluctuation property and the seasonal fluctuation property; A load data prediction module for obtaining the predicted load data of the power system according to the historical power load data and the target prediction model.

8. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images