Power load prediction method, system and dispatching method
By using singular spectral decomposition and multiple regression models, combined with users' historical electricity consumption data, the problem of insufficient accuracy in predicting power load in small areas has been solved, enabling accurate prediction of electricity consumption for individual users. This method is suitable for power load management in small areas such as industrial parks.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- MAANSHAN POWER SUPPLY COMPANY STATE GRID ANHUI ELECTRIC POWER
- Filing Date
- 2026-03-20
- Publication Date
- 2026-06-23
AI Technical Summary
Existing power load forecasting methods lack accuracy in small areas such as industrial parks and residential communities, and cannot effectively account for the differences in electricity consumption among individual users, thus limiting their applicability.
By collecting historical electricity consumption data of regions and users, and using singular spectral decomposition and multiple regression models, user cycle, trend and fluctuation models are constructed to accurately predict the electricity consumption of individual users. Combined with historical electricity consumption sequences of users, data cleaning and outlier processing are performed to establish an accurate power load forecasting method.
It enables accurate forecasting of power load in small areas, takes into account the individual differences of each user, improves the refined management capability of power consumption forecasting, and is applicable to power load forecasting in even smaller areas.
Smart Images
Figure CN122267722A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power load forecasting technology, and more specifically, to a power load forecasting method, system, and dispatching method. Background Technology
[0002] As a crucial link connecting power generation and consumption, the stability and economy of power transmission and distribution networks directly impact energy supply efficiency. Research on power load forecasting and management methods is of significant practical importance for improving the flexibility of transmission and distribution systems and reducing operating costs. Power load forecasting is the foundation of load management; its core lies in scientifically predicting future load change trends through historical data and influencing factor analysis. The core issue in power load forecasting research is how to utilize existing historical data to establish predictive models to forecast load values for future moments or time periods, thereby effectively supporting various operations such as grid dispatching, maintenance planning, stability analysis, and renewable energy consumption analysis.
[0003] Currently, power system load forecasting typically considers factors such as historical system load, economic conditions, meteorological conditions, and social events to predict system load for a future period. Common forecasting models include statistical models such as time series forecasting and regression forecasting, and machine learning and deep learning models such as support vector machines and random forests.
[0004] Current electricity load forecasting methods mostly use historical load data for a specific area as the basis for prediction. By combining this data with climate data for the predicted area, the aim is to study the regularity and growth of the predicted area and thus predict the electricity load. This forecasting method assumes that the same area has similar electricity consumption regularity and periodicity, and that the random electricity consumption behavior of individual users will cancel each other out. Therefore, this forecasting method is more suitable for predicting the electricity load of larger areas with basically consistent socio-economic structures and environmental changes, and is not suitable for forecasting the electricity load of smaller areas such as industrial parks or residential communities. Summary of the Invention
[0005] This disclosure provides a power load forecasting method, system, and dispatching method that can overcome some or all of the shortcomings of the prior art.
[0006] According to the power load forecasting method disclosed herein, it includes, Historical electricity consumption data of the region to be predicted and historical electricity consumption data of each user in the region to be predicted are collected with the same sampling period. Based on the historical electricity consumption data of users, a historical electricity consumption sequence of users is constructed. Based on singular spectral decomposition, regional periodic sequences and regional trend sequences are obtained from regional historical electricity consumption data, first user periodic sequences, first user trend sequences and user fluctuation sequences are obtained from user historical electricity consumption data, and second user periodic sequences and second user trend sequences are obtained from user historical electricity consumption sequences. A first multiple regression model is established using the regional periodic sequence as the dependent variable and the first user periodic sequence of each user as the independent variable; a second multiple regression model is established using the regional trend sequence as the dependent variable and the first user trend sequence of each user as the independent variable. Based on curve fitting of each second user's cycle sequence, a user cycle model corresponding to each user is constructed; based on linear fitting of each second user's trend sequence, a user trend model corresponding to each user is constructed; based on processing of user fluctuation sequences, a user fluctuation model is obtained. The cycle prediction is obtained based on the user cycle model and the first multiple regression model; the trend prediction is obtained based on the user trend model and the second multiple regression model; and the fluctuation prediction is obtained based on the user fluctuation model. Based on periodic forecasts, trend forecasts, and fluctuation forecasts, the power load forecast for the area to be predicted is completed.
[0007] Preferably, the historical electricity consumption sequence of a user is obtained by data cleaning of the user's historical electricity consumption data, specifically including: For any sampling point, select the first n sampling points and the last m sampling points of that sampling point to construct the decision domain; The mean of the absolute values of the differences between each sampling point and its adjacent sampling points within the decision domain is obtained and used as the decision index. Obtain the ratio of the judgment index of any sampling point to the mean of the judgment indices of the remaining sampling points within the judgment domain, and use it as the judgment factor; If the decision factor exceeds the decision threshold, then any sampling point is an outlier, and the value of that sampling point is replaced by the mean of each sampling point within the decision domain; otherwise, the original value is retained.
[0008] As preferred options, both the first and second multiple regression models employ multiple linear regression models.
[0009] Preferably, Fourier series is used to perform curve fitting on each second user periodic sequence, and linear regression is used to perform linear fitting on each second user trend sequence.
[0010] As a preferred method, a user fluctuation model is obtained based on a clustering algorithm, specifically including... Obtain the standard deviation and mean of each user's fluctuation sequence; Clustering algorithms are used to cluster the standard deviation of each user's fluctuation series; Based on the mean of the standard deviation and mean of the user fluctuation sequence corresponding to each cluster of data points, a value range that conforms to the 3σ principle is constructed. A user fluctuation model is constructed using a random function.
[0011] Preferably, when performing power load forecasting for the area to be forecasted, the following steps are included: The user cycle prediction is obtained for each user based on the user cycle model. The user cycle prediction is then substituted into the first multiple regression model to obtain the cycle prediction. The user trend prediction is obtained for each user based on the user trend model, and the user trend prediction is substituted into the second multiple regression model to obtain the trend prediction. Volatility predictions are obtained based on user volatility models; The sum of the periodic forecast, trend forecast, and fluctuation forecast is used as the power load forecast for the area to be predicted.
[0012] According to the power load forecasting system disclosed herein, it includes, At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform any of the methods described above.
[0013] According to the dispatching method disclosed herein, power load forecast data of the area to be predicted is obtained based on any of the above-mentioned methods, and dispatching instructions are issued based on power supply data.
[0014] This disclosure has the following beneficial effects: It can predict the regional power load based on the contribution of a single user to the regional power load, thus fully taking into account the individual differences of each power user; at the same time, by analyzing the power consumption patterns of a single user, it can achieve accurate prediction of the power consumption of a single user based on the assumption that the power consumption of a single user has stable continuity, periodicity and regularity on the time axis. This method can accurately predict electricity consumption by considering each user's contribution to the regional electricity load. It does not need to consider parameters such as regional economic structure or regional environmental changes. Instead, it returns to the essence that the user is the smallest unit of electricity consumption and that each user's daily electricity consumption behavior is related to their work and rest patterns. Therefore, it is better suited for predicting the electricity load of smaller areas and is conducive to improving the refined management of current electricity load prediction. Attached Figure Description
[0015] Figure 1This is a schematic diagram of a power load forecasting method disclosed herein; Figure 2 This is a schematic diagram of a power load forecasting system disclosed herein; Figure 3 A line graph showing the electricity consumption of 30 households in a certain industrial park in East China over 730 days; Figure 4 This is a line graph showing the daily total electricity consumption of a certain industrial park in East China, along with periodic, trend, and fluctuation data obtained through singular spectrum analysis. Figure 5 This is a line graph showing the daily electricity consumption of a user in a certain industrial park in East China, along with periodic, trend, and fluctuation data obtained through singular spectrum analysis. Figure 6 A fitted plot of the first multiple regression model obtained in a specific instance based on the method of this disclosure; Figure 7 This is a fitted graph of a user trend model for a specific user obtained in a specific instance using the method of this disclosure. Detailed Implementation
[0016] To further understand the content of this invention, the invention will be described in detail with reference to the embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention.
[0017] Seen in Figure 1 This disclosure proposes a method for predicting electricity load, which includes, Historical electricity consumption data of the region to be predicted and historical electricity consumption data of each user in the region to be predicted are collected with the same sampling period. Based on the historical electricity consumption data of users, a historical electricity consumption sequence of users is constructed. Based on singular spectral decomposition, regional periodic sequences and regional trend sequences are obtained from regional historical electricity consumption data, first user periodic sequences, first user trend sequences and user fluctuation sequences are obtained from user historical electricity consumption data, and second user periodic sequences and second user trend sequences are obtained from user historical electricity consumption sequences. A first multiple regression model is established using the regional periodic sequence as the dependent variable and the first user periodic sequence of each user as the independent variable; a second multiple regression model is established using the regional trend sequence as the dependent variable and the first user trend sequence of each user as the independent variable. Based on curve fitting of each second user's cycle sequence, a user cycle model corresponding to each user is constructed; based on linear fitting of each second user's trend sequence, a user trend model corresponding to each user is constructed; based on processing of user fluctuation sequences, a user fluctuation model is obtained. The cycle prediction is obtained based on the user cycle model and the first multiple regression model; the trend prediction is obtained based on the user trend model and the second multiple regression model; and the fluctuation prediction is obtained based on the user fluctuation model. Based on periodic forecasts, trend forecasts, and fluctuation forecasts, the power load forecast for the area to be predicted is completed.
[0018] The method disclosed herein can predict the regional power load based on the contribution of a single user to the regional power load, thereby fully taking into account the individual differences of each power user. At the same time, by analyzing the power consumption patterns of a single user, it can achieve accurate prediction of the power consumption of a single user based on the assumption that the power consumption of a single user has stable continuity, periodicity, and regularity on the time axis.
[0019] In this disclosure, regional historical electricity consumption data and user historical electricity consumption data can be collected with a sampling step size such as a daily cycle. That is, regional historical electricity consumption data, user historical electricity consumption sequences, and user historical electricity consumption data are all time series.
[0020] Since the method disclosed herein can analyze the regularity of regional electricity consumption and the regularity of user electricity consumption based on regional historical electricity consumption data, user historical electricity consumption data, and user historical electricity consumption sequences, the impact of random events of individual users on the overall regional electricity consumption can be ignored when the number of users is large enough. However, random electricity consumption events of individual users can have a significant impact on the regularity of user historical electricity consumption sequences. Therefore, in this disclosure, outlier processing is performed on user historical electricity consumption data to obtain user historical electricity consumption sequences, and based on the singular spectrum decomposition of user historical electricity consumption sequences, the obtained second user periodic sequences and second user trend sequences can better analyze the regularity of user electricity consumption.
[0021] Among them, the regional historical electricity consumption data P, the user's historical electricity consumption data U, and the user's historical electricity consumption sequence V can be expressed as follows: ; ; ; in, This represents the regional power consumption at the i-th sampling point. This represents the electricity consumption of the j-th user at the i-th sampling point. This represents the electricity consumption sequence value of the j-th user at the i-th sampling point. It is understood that when the sampling period is a daily cycle, in order to express the annual cycle variation pattern, the number of sampling points should have at least one complete natural year of data. In fact, in this disclosure, data for three complete natural years can be collected, which can better express the electricity consumption trend and electricity consumption cycle.
[0022] In other words, in this disclosure, the acquisition of users' historical electricity consumption sequences based on data cleaning of users' historical electricity consumption data specifically includes, For any sampling point, select the first n sampling points and the last m sampling points of that sampling point to construct the decision domain; The mean of the absolute values of the differences between each sampling point and its adjacent sampling points within the decision domain is obtained and used as the decision index. Obtain the ratio of the judgment index of any sampling point to the mean of the judgment indices of the remaining sampling points within the judgment domain, and use it as the judgment factor; If the decision factor exceeds the decision threshold, then any sampling point is an outlier, and the value of that sampling point is replaced by the mean of each sampling point within the decision domain; otherwise, the original value is retained.
[0023] That is, ; in, ; When the sampling period is a daily period, both m and n can be 7. It can be set to 3.
[0024] Based on the above, abnormal electricity consumption in users' historical electricity consumption data can be smoothed, so that the obtained user historical electricity consumption sequence can better reflect periodicity and trend.
[0025] In this disclosure, by using the singular spectrum decomposition of regional historical electricity consumption data, user historical electricity consumption data, and user historical electricity consumption sequences, it is possible to decompose and reconstruct the time series of electricity consumption into trend components, periodic components, and / or fluctuation components without relying on prior assumptions.
[0026] In this disclosure, when the sampling period is daily, a trajectory matrix can be constructed with a window length of 90 (90 days covers a season, which can better present the seasonal regularity of electricity consumption). Then, by performing singular value decomposition on the trajectory matrix, the relevant elementary matrices can be obtained after obtaining the singular values. Then, by grouping and fusing the elementary matrices representing periodicity, trend, and volatility, and performing diagonal averaging, the corresponding periodic sequence, trend sequence, and volatility sequence can be reconstructed.
[0027] To further illustrate this, this disclosure briefly introduces the singular spectrum decomposition of regional historical electricity consumption data. The regional historical electricity consumption data can be 2 years in length, that is, it has 365*2=730 sampling points. When the window width is 90, the constructed trajectory matrix is 90×641 in size. That is, for regional historical electricity consumption data The trajectory matrix X it constructs is, ; Then, the trajectory matrix X is decomposed, that is, ; Where U is a 90×90 orthogonal matrix. It is a diagonal matrix of 90×641. Given a 641×641 orthogonal matrix, it can be understood that the column vectors of the orthogonal matrix U are... orthogonal matrix The column vector is Then the corresponding singular value is k. Elementary matrices for, ; That is, through singular value decomposition, the trajectory matrix X can be decomposed into at most 90 elementary matrices. ,Right now, ; Where r is the rank of the trajectory matrix X, that is, the number of non-zero singular values, which is at most 90.
[0028] Next, for elementary matrices By grouping and merging the data, a fusion matrix representing the period can be obtained. A fusion matrix representing trends and the fusion matrix representing fluctuations ,Right now ; It is understandable that elementary matrices are all 90×641 matrices. Grouping and merging refers to performing addition operations on elementary matrices representing cycles, trends, or noise. After obtaining the fusion matrix representing the period... A fusion matrix representing trends And the fusion matrix representing fluctuations Then, by performing diagonal averaging, we can obtain the corresponding reconstructed sequences representing the period, the trend, and the fluctuation.
[0029] It is understood that, in the method disclosed herein, the singular spectral decomposition of regional historical electricity consumption data only requires taking the reconstructed sequence representing the period and the reconstructed sequence representing the trend, i.e., the regional periodic sequence. and regional trend sequence Singular spectral decomposition of historical electricity consumption data requires extracting reconstructed sequences representing the period, the trend, and the fluctuations, i.e., the first user period sequence. First User Trend Sequence and user fluctuation sequence For the singular spectrum decomposition of a user's historical electricity consumption sequence, only the reconstructed sequence representing the period and the reconstructed sequence representing the trend are needed, i.e., the second user period sequence. Second User Trend Sequence .
[0030] In this disclosure, both the first and second multiple regression models employ multiple linear regression models; that is... The first multiple regression model is, ; The second multiple regression model is, ; Here, A and B are the corresponding estimated parameter vectors, and a and b are the corresponding error terms. Using methods such as least squares and gradient descent, the estimated parameter vectors and error terms can be solved, thus obtaining the first multiple regression model. Second Multiple Regression Model .
[0031] In this disclosure, Fourier series is used to perform curve fitting on each second user periodic sequence, and linear regression is used to perform linear fitting on each second user trend sequence.
[0032] The user lifecycle model can be expressed as: User trend models can be expressed as It is understandable that electricity load has a strong periodicity that follows the cycle. Therefore, when the sampling period is the daily cycle, Fourier fitting can be performed with 1 / 365 as the fundamental frequency, thus obtaining a better user cycle model. The trend part obtained by singular spectrum analysis has strong linearity, so the user trend model can be obtained better through linear regression.
[0033] In this disclosure, a user fluctuation model is obtained based on a clustering algorithm, specifically including: Obtain the fluctuation sequence for each user Standard deviation and mean ; Clustering algorithms are used to analyze the fluctuation sequences of each user. Standard deviation Perform clustering; Based on the user fluctuation sequence corresponding to each cluster of data points Standard deviation and mean The mean, constructing a system that conforms to The range of values for the principle; A user fluctuation model is constructed using a random function.
[0034] That is, user fluctuation model It can be expressed as, ; in, Represents the fluctuation sequences of all users in cluster c. mean The mean, Represents the fluctuation sequences of all users in cluster c. Standard deviation The mean, where C is the total number of clusters. This represents the total number of data points in each cluster. This indicates that for the d-th data point in the c-th cluster, the interval... Random values are selected.
[0035] Based on the above, users with similar volatility characteristics can be clustered. By constructing a random function, the user volatility model can make users with the same volatility characteristics take probabilistic values when performing volatility prediction. This method satisfies both the randomness of volatility characteristics and the probability constraints of statistics.
[0036] Clustering algorithms can be implemented using algorithms such as k-means.
[0037] This disclosure includes the following steps when performing power load forecasting for the area to be forecasted: The user cycle prediction is obtained for each user based on the user cycle model. The user cycle prediction is then substituted into the first multiple regression model to obtain the cycle prediction. The user trend prediction is obtained for each user based on the user trend model, and the user trend prediction is substituted into the second multiple regression model to obtain the trend prediction. Volatility predictions are obtained based on user volatility models; The sum of the periodic forecast, trend forecast, and fluctuation forecast is used as the power load forecast for the area to be predicted.
[0038] Based on the method disclosed herein, the method can accurately predict electricity consumption by considering the contribution of each user to the regional electricity load. This method does not need to consider parameters such as regional economic structure or regional environmental changes. Instead, it returns to the essence that the user is the smallest electricity-consuming unit and that each user's daily electricity consumption behavior is related to their work and rest patterns. Therefore, it is better suited for predicting the electricity load of smaller areas and is conducive to improving the refined management of current electricity load prediction.
[0039] Seen in Figures 3-7 In a specific example, the electricity consumption of 30 users in a certain industrial park in East China over 730 days was analyzed. Figure 3 The graph showing the electricity consumption of these 30 households is displayed. Figure 4 The chart displays a line graph of the park's daily total electricity consumption, as well as periodic, trend, and fluctuation data obtained through singular spectrum analysis. Figure 5 The chart displays a line graph of a user's daily electricity consumption, along with periodic, trend, and fluctuation data obtained through singular spectrum analysis. Figure 6 The fitted plot of the first multiple regression model is shown.
[0040] Specifically, by performing curve fitting on the second user period sequence of a user's historical electricity consumption sequence, the corresponding user period model can be obtained. ; After linearly fitting the second user trend sequence of a certain user's historical electricity consumption sequence, the corresponding user trend model can be obtained as follows: .
[0041] in, Figure 7 The image shows a fitted plot of the user trend model for a particular user.
[0042] Furthermore, another object of this disclosure is to provide an electricity load forecasting system, which includes, At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the methods described above in this disclosure.
[0043] Specifically seen in Figure 2The power load forecasting system disclosed herein includes a data acquisition module, a data management module, and a data output module. The data acquisition module is used to collect historical electricity consumption data of the region and users. The data management module includes a data cleaning unit, a singular spectrum analysis unit, a fitting unit, a random number generation unit, and a storage unit. The data cleaning unit is used to generate historical electricity consumption sequences of users based on their historical electricity consumption data. The singular spectrum analysis unit is used to perform singular spectrum analysis on the historical electricity consumption data of the region, the historical electricity consumption data of users, and the historical electricity consumption sequences of users to output relevant periodic data, trend data, and fluctuation data. The fitting unit is used to perform fitting operations on the relevant data to construct relevant data models. The random number generation unit is used to process the fluctuation data to construct relevant fluctuation models. The storage unit is used to perform data storage and other steps. The data output module is used to perform the power load forecasting steps and output the power load forecasting results.
[0044] Furthermore, the purpose of this disclosure is to provide a power dispatching method, which obtains power load forecast data of the area to be predicted based on the above method, and issues dispatching instructions based on power supply data.
[0045] It is readily understood that those skilled in the art can combine, split, or reorganize the embodiments provided in this application to obtain other embodiments, all of which do not exceed the protection scope of this application.
[0046] The present invention and its embodiments have been described above illustratively. This description is not restrictive, and the embodiments shown are only part of the embodiments of the present invention. The actual structure is not limited thereto. Therefore, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the present invention, they should all fall within the protection scope of the present invention.
Claims
1. Power load forecasting methods, which include: Historical electricity consumption data of the region to be predicted and historical electricity consumption data of each user in the region to be predicted are collected with the same sampling period. Based on the historical electricity consumption data of users, a historical electricity consumption sequence of users is constructed. Based on singular spectral decomposition, regional periodic sequences and regional trend sequences are obtained from regional historical electricity consumption data, first user periodic sequences, first user trend sequences and user fluctuation sequences are obtained from user historical electricity consumption data, and second user periodic sequences and second user trend sequences are obtained from user historical electricity consumption sequences. A first multiple regression model is established with the regional periodic sequence as the dependent variable and the first user periodic sequence of each user as the independent variable. A second multiple regression model is established using the regional trend sequence as the dependent variable and the first user trend sequence of each user as the independent variable. Based on curve fitting of each second user's cycle sequence, a user cycle model corresponding to each user is constructed; based on linear fitting of each second user's trend sequence, a user trend model corresponding to each user is constructed; based on processing of user fluctuation sequences, a user fluctuation model is obtained. The cycle prediction is obtained based on the user cycle model and the first multiple regression model; the trend prediction is obtained based on the user trend model and the second multiple regression model; and the fluctuation prediction is obtained based on the user fluctuation model. Based on periodic forecasts, trend forecasts, and fluctuation forecasts, the power load forecast for the area to be predicted is completed.
2. The power load forecasting method according to claim 1, characterized in that: Based on data cleaning of users' historical electricity consumption data, the user's historical electricity consumption sequence is obtained, specifically including: For any sampling point, select the first n sampling points and the last m sampling points of that sampling point to construct the decision domain; The mean of the absolute values of the differences between each sampling point and its adjacent sampling points within the decision domain is obtained and used as the decision index. Obtain the ratio of the judgment index of any sampling point to the mean of the judgment indices of the remaining sampling points within the judgment domain, and use it as the judgment factor; If the decision factor exceeds the decision threshold, then any sampling point is an outlier, and the value of that sampling point is replaced by the mean of each sampling point within the decision domain; otherwise, the original value is retained.
3. The power load forecasting method according to claim 1, characterized in that: Both the first and second multiple regression models use multiple linear regression models.
4. The power load forecasting method according to claim 1, characterized in that: Fourier series was used to perform curve fitting on each second user periodic sequence, and linear regression was used to perform linear fitting on each second user trend sequence.
5. The power load forecasting method according to claim 1, characterized in that: The user fluctuation model is obtained based on clustering algorithms, specifically including... Obtain the standard deviation and mean of each user's fluctuation sequence; Clustering algorithms are used to cluster the standard deviation of each user's fluctuation series; Based on the mean of the standard deviation and mean of the user fluctuation sequence corresponding to each cluster of data points, a value range that conforms to the 3σ principle is constructed. A user fluctuation model is constructed using a random function.
6. The power load forecasting method according to claim 1, characterized in that: When performing power load forecasting for the area to be forecasted, the following steps are included: The user cycle prediction is obtained for each user based on the user cycle model. The user cycle prediction is then substituted into the first multiple regression model to obtain the cycle prediction. The user trend prediction is obtained for each user based on the user trend model, and the user trend prediction is substituted into the second multiple regression model to obtain the trend prediction. Volatility predictions are obtained based on user volatility models; The sum of the periodic forecast, trend forecast, and fluctuation forecast is used as the power load forecast for the area to be predicted.
7. A power load forecasting system, comprising, At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-6.
8. A scheduling method, characterized in that: Based on the method of claims 1-6, power load forecast data for the area to be predicted is obtained, and based on the power supply data, dispatch instructions are issued.