A method and system for short-term user load prediction based on high-frequency data

Abstract

The present invention belongs to the technical field of power systems, and specifically relates to a method and system for short-term user load prediction based on high-frequency data. The method includes collecting user load data to obtain the user's load-time curve; obtaining the maximum and average values of the user's load-time curve, and dividing the users into energy-saving users and non-energy-saving users based on the maximum and average values of the user's load-time curve; dividing the users into stable users and random users based on whether the load-time curve of the non-energy-saving user has flexible loads; performing load predictions on energy-saving users, random users, and random users respectively, summing the short-term predicted load-time curves of all users in the distribution area, and obtaining the short-term predicted load-time curve of the distribution area. The present invention divides the load into steady-state load and flexible load, and predicts the two separately according to their respective characteristics, making the load prediction more accurate.

Description

Technical Field

[0001] The present invention belongs to the technical field of power systems, and in particular relates to a method and system for short-term prediction of user load based on high-frequency data. Background Art

[0002] User load forecasting is a crucial tool for power demand-side management and a prerequisite for power system production and dispatch planning. Existing short-term load forecasting primarily utilizes meteorological factors and holiday types to construct a mapping relationship with the load curve, using information such as weather forecasts and holidays to predict future load curves. However, existing load forecasting techniques fail to consider the impact of flexible loads, which have a significant instantaneous impact on grid load and fluctuate randomly. Furthermore, these techniques utilize data from upper-level sources such as feeders and substations as samples, failing to consider the impact of individual user load fluctuations on the forecast results. Furthermore, due to the high volume of user data, even large load fluctuations for a few users can be difficult to detect or may be treated as line losses. Therefore, existing forecasting methods ignore user-level characteristics and the impact of flexible loads, resulting in inaccurate load forecasts. Summary of the Invention

[0003] To address the deficiencies of the prior art, the present invention provides a method and system for short-term user load prediction based on high-frequency data, which divides the load into steady-state load and flexible load, and predicts the two separately according to their respective characteristics, making the load prediction more accurate.

[0004] In order to solve the deficiencies of the prior art, the present invention provides the following technical solutions:

[0005] A short-term user load forecasting method based on high-frequency data includes:

[0006] Collect user load data to obtain the user's load-time curve;

[0007] Calculate the maximum and average load values of users based on their load-time curves, and classify users into energy-saving users and non-energy-saving users based on their maximum and average load values.

[0008] According to whether there is flexible load in the load-time curve of non-energy-saving users, users are divided into stable users and random users;

[0009] The short-term predicted load-time curve of the energy-saving user is obtained based on the load-time curve of the energy-saving user; the short-term predicted load-time curve of the stable user is obtained based on the load-time curve of the stable user; the load-time curve of the random user is decomposed to obtain a stable load-time curve and a flexible load-time curve, a short-term predicted stable load-time curve is obtained based on the stable load-time curve, a short-term predicted flexible load-time curve is obtained based on the flexible load-time curve, and the short-term predicted stable load-time curve and the short-term predicted flexible load-time curve are summed to obtain the short-term predicted load-time curve of the random user;

[0010] The short-term forecast load-time curves of all users in the distribution area are summed to obtain the short-term forecast load-time curve of the distribution area.

[0011] Preferably, the user's load data is collected once every minute.

[0012] Preferably, the step of obtaining the maximum load value and the average load value of the user according to the user's load-time curve, and classifying the user into energy-saving users and non-energy-saving users according to the maximum load value and the average load value of the user, includes:

[0013] The maximum load value per minute and the average load value per minute of the user within a fixed period are calculated based on the user's load-time curve; users whose maximum load value per minute within a fixed period is less than a first judgment threshold and whose average load value per minute within a fixed period is less than a second judgment threshold are judged as energy-saving users, and the remaining users are non-energy-saving users.

[0014] Preferably, the method of dividing users into stable users and random users based on whether there is a flexible load in the load-time curve of the non-energy-saving user includes calculating the autoregressive coefficient of the load-time curve of the non-energy-saving user within a fixed period and performing a unit root check on the load-time curve. If the autoregressive coefficient of the load-time curve does not have periodicity and / or the load-time curve has a unit root, then there is a flexible load and the user is divided into a random user; otherwise, the user is divided into a stable user.

[0015] Preferably, obtaining the short-term predicted load-time curve of the energy-saving user based on the load-time curve of the energy-saving user includes removing noise from the load-time curve of the energy-saving user by a moving smoothing filter algorithm to obtain the short-term predicted load-time curve of the energy-saving user on the prediction day;

[0016] The method of obtaining a short-term predicted load-time curve of a stable user based on the load-time curve of the stable user includes removing noise from the load-time curve of the random user through a moving smoothing filter algorithm to obtain a short-term predicted load-time curve of the random user on the prediction day.

[0017] Preferably, the decomposing the load-time curve of the random user to obtain the stable load-time curve and the flexible load-time curve includes decomposing the load-time curve of the random user by a moving smoothing filter algorithm to obtain the stable load-time curve that satisfies the periodicity, and decomposing the flexible load-time curve that does not satisfy the periodicity by obtaining the Nth-order difference of the load-time curve of the random user;

[0018] The obtaining of the short-term predicted stable load-time curve according to the stable load-time curve includes removing noise from the stable load-time curve by a moving smoothing filter algorithm to obtain the short-term predicted stable load-time curve for the prediction day;

[0019] The short-term predicted flexible load-time curve is obtained based on the flexible load-time curve, including taking historical meteorological data and calendar data as input data and historical flexible load as output data to train a neural network model; after the neural network model training is completed, meteorological forecast data and calendar data of the forecast day are input, and the short-term predicted flexible load-time curve of the forecast day is output through the neural network model.

[0020] A user load short-term prediction system based on high-frequency data includes a user load data receiving unit, a data storage unit, a load data processing unit and a prediction unit;

[0021] The user load data receiving unit is used to collect user load data and obtain a user load-time curve;

[0022] The data storage unit is used to store the user's load-time curve, historical meteorological data, meteorological forecast data, and calendar data;

[0023] The load data processing unit is configured to obtain a maximum load and an average load of a user based on the user's load-time curve, and to classify the user into an energy-saving user and a non-energy-saving user based on the maximum load and the average load; to classify the user into a stable user and a random user based on whether the load-time curve of the non-energy-saving user has a flexible load; and, for a random user, to decompose the random user's load-time curve to obtain a stable load-time curve and a flexible load-time curve.

[0024] The prediction unit is used to obtain a short-term predicted load-time curve for energy-saving users based on their load-time curves; obtain a short-term predicted load-time curve for stable users based on their load-time curves; obtain a short-term predicted stable load-time curve based on the stable load-time curve, obtain a short-term predicted flexible load-time curve based on the flexible load-time curve, sum the short-term predicted stable load-time curve and the short-term predicted flexible load-time curve to obtain a short-term predicted load-time curve for random users; and sum the short-term predicted load-time curves for all users in the distribution area to obtain a short-term predicted load-time curve for the distribution area.

[0025] Preferably, the user load data receiving unit collects user load data once every minute.

[0026] Preferably, the load data processing unit is used to:

[0027] Calculate the user's maximum load per minute and average load per minute within a fixed period based on the user's load-time curve; determine the user whose maximum load per minute within the fixed period is less than a first judgment threshold and whose average load per minute within the fixed period is less than a second judgment threshold as an energy-saving user, and the remaining users as non-energy-saving users;

[0028] Calculate the autoregressive coefficient of the load-time curve of non-energy-saving users within a fixed period and perform a unit root check on the load-time curve. If the autoregressive coefficient of the load-time curve does not have periodicity and / or the load-time curve has a unit root, there is a flexible load and the user is classified as a random user. Otherwise, the user is classified as a stable user.

[0029] The load-time curve of random users is decomposed into a stable load-time curve that meets the periodicity through the moving smoothing filter algorithm, and the flexible load-time curve that does not meet the periodicity is decomposed by taking the N-order difference of the load-time curve of random users.

[0030] Preferably, the prediction unit is used to:

[0031] The load-time curve of energy-saving users is subjected to moving smoothing filter algorithm to remove noise and obtain the short-term forecast load-time curve of energy-saving users on the forecast day.

[0032] The load-time curve of random users is subjected to moving smoothing filter algorithm to remove noise and obtain the short-term forecast load-time curve of random users on the forecast day.

[0033] The stable load-time curve is subjected to moving smoothing filter algorithm to remove noise and obtain the short-term predicted stable load-time curve of the forecast day;

[0034] The historical meteorological data and calendar data are used as input data, and the historical flexible load is used as output data to train the neural network model. After the neural network model training is completed, the meteorological forecast data and calendar data of the forecast day are input, and the short-term forecast flexible load-time curve of the forecast day is output through the neural network model.

[0035] Beneficial effects of the present invention:

[0036] The present invention comprehensively considers the determinism and randomness of each user's load curve, as well as the contribution coefficient to the total load, and divides the load into steady-state load and flexible load, and makes predictions according to user type and load type respectively, making load prediction more accurate.

[0037] The present invention collects minute-level user load data, thereby increasing the number of samples and the accuracy, and achieving higher prediction accuracy.

[0038] The present invention also analyzes the actual electricity consumption of residents, can discover abnormal load conditions in advance, and provide a reference for anti-electricity theft, peak-valley scheduling and other services. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A flow chart of a method for short-term load forecasting based on high-frequency data provided by the present invention;

[0040] Figure 2 It is a schematic diagram of the load-time curve of energy-saving users;

[0041] Figure 3 Schematic diagram of the load-time curve for stable users;

[0042] Figure 4 Schematic diagram of the load-time curve for random users;

[0043] Figure 5 Schematic diagram of the smart IoT system architecture. DETAILED DESCRIPTION

[0044] The present invention will be further described below in conjunction with the embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0045] The embodiment of the present invention provides a method for short-term load prediction of users based on high-frequency data. Figure 1 , including the following steps:

[0046] 1: Collect the user's load data to obtain the user's load-time curve. Preferably, the user's load data is collected once every minute.

[0047] Existing load forecasting primarily focuses on trend prediction at 96 daily load sampling points within a region (city, line). The traditional 15-minute sampling frequency is unable to detect short-term user load variations, which often include a variety of high-volume and short-duration flexible loads. Therefore, existing technologies cannot accurately predict flexible loads.

[0048] At the same time, due to the coarse granularity of meteorological characteristics (currently historical meteorological detection can reach the 5-minute level, and meteorological forecasts can reach the hourly level), the prediction method is more of a trend prediction of the overall load stability, which itself balances the differences between individual users, resulting in a visible but uncontrollable load. Therefore, existing load forecasts have the problems of low accuracy and too few samples (15 minutes). The present invention collects minute-level user load data, which can better combine historical meteorological samples, achieve an increase in sample quantity and accuracy, and achieve higher prediction accuracy.

[0049] 2. Based on the user's load-time curve, calculate the user's maximum, minimum, and average minute-by-minute load values within a fixed period. These values are then recorded in the user's profile. The fixed period is typically one month, meaning the user's maximum, minimum, and average minute-by-minute load values within the past month must be recorded. These maximum, minimum, and average minute-by-minute load values must be continuously updated; to reduce computational complexity, they can be updated weekly.

[0050] Users whose maximum load per minute is less than the first judgment threshold and whose average load per minute is less than the second judgment threshold are judged as energy-saving users, and the remaining users are non-energy-saving users. That is, the classification results of energy-saving users and non-energy-saving users are also updated synchronously once a week. The present invention uses the load per minute as an example to distinguish between energy-saving users and non-energy-saving users. Of course, the maximum load and minimum load values within other fixed time periods can also be selected for judgment, such as judging based on the load amount per hour. Directly using the load per minute for judgment can reduce the amount of calculation. The first judgment threshold and the second judgment threshold can be set as needed. The first judgment threshold can be set to 2000W and the second judgment threshold can be set to 200W.

[0051] Attachment Figure 2 For energy-saving users, the load-time curve is as follows: a household refrigerator typically has a cooling power of 200W, and a household air conditioner has a heating power of 2kW. If both the maximum and average values of a user's load do not reach the corresponding judgment threshold, it means that the user does not have a large load and their electricity consumption has little impact on the load of the entire grid. Forecasting these energy-saving users separately can greatly reduce subsequent computing overhead.

[0052] In order to eliminate the adverse effects caused by large numerical differences, the user's load-time curve is normalized by the maximum load value per minute and the minimum load value per minute to obtain a normalized load-time curve.

[0053] 3: Based on whether the normalized load-time curve of non-energy-saving users has flexible load, users are divided into stable users and random users.

[0054] Calculate the autoregressive coefficient of the normalized load-time curve within a fixed period and perform a unit root check on the normalized load-time curve. If the autoregressive coefficient of the normalized load-time curve is not periodic and / or the normalized load-time curve exhibits a unit root or a random trend, then a flexible load exists and the user is classified as a random user. Otherwise, the user is classified as a stable user. The fixed period is typically one month. To mitigate seasonal factors, the classification of stable and random users can be updated weekly.

[0055] 4: Forecast user loads separately according to user types:

[0056] 1) For energy-saving and stable users, electricity load changes exhibit significant periodicity. The normalized load-time curve for the user over a recent period (e.g., one month) is filtered using a moving smoothing filter algorithm to remove noise. This is then denormalized based on the maximum and minimum minute-by-minute load values to produce the short-term predicted load-time curve for the energy-saving user and the short-term predicted load-time curve for the stable user. Alternatively, the normalized load-time curve for another recent period can be used for prediction.

[0057] 2) For random users, the user's load has a lot of random changes, so it is necessary to decompose it, determine the steady-state and random characteristics, and analyze and predict them separately. The normalized load-time curve of the random user in the most recent period (such as one month) is decomposed into a stable load-time curve that meets the periodicity through the moving smoothing filter algorithm. The flexible load-time curve that does not meet the periodicity is decomposed by taking the Nth-order difference of the random user's load-time curve. N is usually 1. The stable load-time curve is decomposed into a short-term predicted stable load-time curve for the forecast day by using the moving smoothing filter algorithm to remove noise. The maximum load value and the minimum load value per minute are denormalized. To obtain a short-term forecast flexible load-time curve based on the flexible load-time curve, a neural network model is trained using historical meteorological and calendar data from a recent period (e.g., one month) as input and historical flexible loads as output. Once the neural network model is trained, the meteorological and calendar data for the forecast day are input. The neural network model then outputs a normalized short-term forecast flexible load-time curve for the forecast day. This normalized short-term forecast flexible load-time curve is then denormalized based on the user's maximum and minimum minute load values to obtain the short-term forecast flexible load-time curve. The short-term forecast stable load-time curve and the short-term forecast flexible load-time curve are summed to obtain the short-term forecast load-time curve for the random user's forecast day. Typically, to minimize seasonal influences on the neural network model, the model is updated once a month.

[0058] Specifically, the meteorological data includes the maximum temperature, minimum temperature, average temperature, precipitation and humidity of each day; the meteorological forecast data for the forecast day includes the predicted maximum temperature, predicted minimum temperature, predicted average temperature, predicted precipitation and predicted humidity for the forecast day; the calendar data includes the day types of the day before the forecast day (T-1 day), the forecast day (T day) and the day after the forecast day (T+1 day); the day types include weekdays and holidays.

[0059] The promotion and access of a large number of flexible loads such as electric vehicles has led to an increase in the demand side to control the stability of the power grid. Forecasting demand for loads, regulating flexible loads, and simulating load operating conditions can better promote the smooth operation of the power grid. The analysis of flexible loads is based on the different characteristics of users, which requires load forecasting at the substation and meter levels. The present invention analyzes user types and decomposes the loads of random users with flexible loads into stable loads and flexible loads. Combining the characteristics of users' electricity consumption cycles and seasonal meteorological working days, a neural network model is established, which can effectively improve the accuracy of users' short-term forecasts.

[0060] 5. Sum the short-term forecast load-time curves for all users in the distribution area to obtain the short-term forecast load-time curve for the distribution area. Typically, the distribution area is defined as the distribution substation. Accumulating the short-term forecast load-time curves for the distribution substations yields the short-term forecast load-time curve for the feeder.

[0061] The embodiment of the present invention further provides a user load short-term prediction system based on high-frequency data, comprising a user load data receiving unit, a data storage unit, a load data processing unit and a prediction unit;

[0062] The user load data receiving unit is used to collect the user's load data and obtain the user's load-time curve;

[0063] The data storage unit is used to store the user's load-time curve, historical meteorological data, meteorological forecast data, and calendar data;

[0064] The load data processing unit is configured to obtain a maximum load and an average load of a user based on the user's load-time curve, and to classify the user into an energy-saving user and a non-energy-saving user based on the maximum load and the average load; to classify the user into a stable user and a random user based on whether the load-time curve of the non-energy-saving user has a flexible load; and, for a random user, to decompose the random user's load-time curve to obtain a stable load-time curve and a flexible load-time curve.

[0065] The prediction unit is used to obtain a short-term predicted load-time curve of an energy-saving user based on the load-time curve of the energy-saving user; obtain a short-term predicted load-time curve of a stable user based on the load-time curve of the stable user; obtain a short-term predicted stable load-time curve based on the stable load-time curve, obtain a short-term predicted flexible load-time curve based on the flexible load-time curve, sum the short-term predicted stable load-time curve and the short-term predicted flexible load-time curve to obtain a short-term predicted load-time curve of a random user; sum the short-term predicted load-time curves of all users in the distribution area to obtain a short-term predicted load-time curve of the distribution area.

[0066] The user load data receiving unit collects user load data once every minute.

[0067] The load data processing unit is used to:

[0068] The maximum load value and average load value per minute of the user within a fixed period are calculated based on the user's load-time curve; users whose maximum load value per minute within a fixed period is less than the first judgment threshold and whose average load value per minute within a fixed period is less than the second judgment threshold are judged as energy-saving users, and the rest of the users are non-energy-saving users.

[0069] Calculate the autoregressive coefficient of the load-time curve of non-energy-saving users within a fixed period and perform a unit root check on the load-time curve. If the autoregressive coefficient of the load-time curve does not have periodicity or the load-time curve has a unit root, there is a flexible load and the user is classified as a random user. Otherwise, the user is classified as a stable user.

[0070] The load-time curve of random users is decomposed into a stable load-time curve that meets the periodicity through the moving smoothing filter algorithm, and the flexible load-time curve that does not meet the periodicity is decomposed by taking the N-order difference of the load-time curve of random users.

[0071] The prediction unit is used to

[0072] The load-time curve of energy-saving users is subjected to moving smoothing filter algorithm to remove noise and obtain the short-term forecast load-time curve of energy-saving users on the forecast day.

[0073] The load-time curve of random users is subjected to moving smoothing filter algorithm to remove noise and obtain the short-term forecast load-time curve of random users on the forecast day.

[0074] The stable load-time curve is subjected to moving smoothing filter algorithm to remove noise and obtain the short-term predicted stable load-time curve of the forecast day;

[0075] The historical meteorological data and calendar data are used as input data, and the historical flexible load is used as output data to train the neural network model. After the neural network model training is completed, the meteorological forecast data and calendar data of the forecast day are input, and the short-term forecast flexible load-time curve of the forecast day is output through the neural network model.

[0076] The user short-term forecast load-time curve system based on high-frequency data provided by the present invention can be implemented based on the smart IoT system architecture. The system architecture is as follows: Figure 5 As shown in the figure, the smart IoT architecture includes the application layer, platform layer, edge layer, and terminal layer.

[0077] The edge physical unit can serve as a user load data receiving unit to receive and process user meter data, realize high-frequency collection of user loads, and report the user's load-time curve to the IoT management platform.

[0078] The IoT management platform aggregates the load-time curves of edge agent users and forwards them to the application system.

[0079] The application system includes the above-mentioned data storage unit, load data processing unit and prediction unit, and is used for the above-mentioned storage, user analysis and load prediction functions.

[0080] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0081] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0082] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0083] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0084] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which are all protected by the present invention.

[0085] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A short-term user load prediction method based on high-frequency data, characterized in that: include Collect user load data to obtain the user's load-time curve; Calculate the maximum and average load values of users based on their load-time curves, and classify users into energy-saving users and non-energy-saving users based on their maximum and average load values. According to whether there is flexible load in the load-time curve of non-energy-saving users, users are divided into stable users and random users; The short-term predicted load-time curve of the energy-saving user is obtained based on the load-time curve of the energy-saving user; the short-term predicted load-time curve of the stable user is obtained based on the load-time curve of the stable user; the load-time curve of the random user is decomposed to obtain a stable load-time curve and a flexible load-time curve, a short-term predicted stable load-time curve is obtained based on the stable load-time curve, a short-term predicted flexible load-time curve is obtained based on the flexible load-time curve, and the short-term predicted stable load-time curve and the short-term predicted flexible load-time curve are summed to obtain the short-term predicted load-time curve of the random user; The short-term forecast load-time curves of all users in the distribution area are summed to obtain the short-term forecast load-time curve of the distribution area; The method of classifying the users into stable users and random users based on whether the load-time curve of the non-energy-saving users has a flexible load includes calculating the autoregressive coefficient of the load-time curve of the non-energy-saving users within a fixed period and performing a unit root check on the load-time curve. If the autoregressive coefficient of the load-time curve does not have periodicity and / or the load-time curve has a unit root, a flexible load exists and the user is classified as a random user. Otherwise, the user is classified as a stable user. The method of obtaining a short-term predicted load-time curve of the energy-saving user based on the load-time curve of the energy-saving user includes removing noise from the load-time curve of the energy-saving user by a moving smoothing filter algorithm to obtain a short-term predicted load-time curve of the energy-saving user on the prediction day; The method of obtaining a short-term predicted load-time curve of a stable user based on the load-time curve of the stable user includes removing noise from the load-time curve of the random user through a moving smoothing filter algorithm to obtain a short-term predicted load-time curve of the random user on the prediction day.

2. The method for short-term user load prediction based on high-frequency data according to claim 1, characterized in that: User load data is collected once every minute.

3. The method for short-term user load prediction based on high-frequency data according to claim 2, characterized in that: The method of obtaining the maximum load value and the average load value of the user according to the load-time curve of the user, and dividing the user into energy-saving users and non-energy-saving users according to the maximum load value and the average load value of the user, includes: The maximum load value per minute and the average load value per minute of the user within a fixed period are calculated based on the user's load-time curve; users whose maximum load value per minute within a fixed period is less than a first judgment threshold and whose average load value per minute within a fixed period is less than a second judgment threshold are judged as energy-saving users, and the remaining users are non-energy-saving users.

4. The method for short-term user load prediction based on high-frequency data according to claim 1, characterized in that: Decomposing the load-time curve of the random user to obtain a stable load-time curve and a flexible load-time curve includes decomposing the load-time curve of the random user by a moving smoothing filter algorithm to obtain a stable load-time curve that satisfies periodicity, and decomposing the flexible load-time curve that does not satisfy periodicity by taking the Nth-order difference of the load-time curve of the random user; The obtaining of the short-term predicted stable load-time curve according to the stable load-time curve includes removing noise from the stable load-time curve by a moving smoothing filter algorithm to obtain the short-term predicted stable load-time curve for the prediction day; The method of obtaining a short-term predicted flexible load-time curve based on the flexible load-time curve includes taking historical meteorological data and calendar data as input data, taking historical flexible load as output data, and training a neural network model. After the neural network model training is completed, the meteorological forecast data and calendar data of the forecast day are input, and the short-term forecast flexible load-time curve of the forecast day is output through the neural network model.

5. A system for short-term user load prediction based on high-frequency data based on the method for short-term user load prediction based on high-frequency data according to any one of claims 1 to 4, characterized in that: It includes a user load data receiving unit, a data storage unit, a load data processing unit and a prediction unit; The user load data receiving unit is used to collect user load data and obtain a user load-time curve; The data storage unit is used to store the user's load-time curve, historical meteorological data, meteorological forecast data, and calendar data; The load data processing unit is configured to obtain a maximum load and an average load of a user based on the user's load-time curve, and to classify the user into an energy-saving user and a non-energy-saving user based on the maximum load and the average load; to classify the user into a stable user and a random user based on whether the load-time curve of the non-energy-saving user has a flexible load; and, for a random user, to decompose the random user's load-time curve to obtain a stable load-time curve and a flexible load-time curve. The prediction unit is used to obtain a short-term predicted load-time curve for energy-saving users based on their load-time curves; obtain a short-term predicted load-time curve for stable users based on their load-time curves; obtain a short-term predicted stable load-time curve based on the stable load-time curve, obtain a short-term predicted flexible load-time curve based on the flexible load-time curve, sum the short-term predicted stable load-time curve and the short-term predicted flexible load-time curve to obtain a short-term predicted load-time curve for random users; and sum the short-term predicted load-time curves for all users in the distribution area to obtain a short-term predicted load-time curve for the distribution area.

6. The user load short-term forecasting system based on high-frequency data according to claim 5 is characterized in that: The user load data receiving unit collects user load data once every minute.

7. The user load short-term forecasting system based on high-frequency data according to claim 5 is characterized in that: The load data processing unit is used to: Calculate the user's maximum load per minute and average load per minute within a fixed period based on the user's load-time curve; determine the user whose maximum load per minute within the fixed period is less than a first judgment threshold and whose average load per minute within the fixed period is less than a second judgment threshold as an energy-saving user, and the remaining users as non-energy-saving users; Calculate the autoregressive coefficient of the load-time curve of non-energy-saving users within a fixed period and perform a unit root check on the load-time curve. If the autoregressive coefficient of the load-time curve does not have periodicity and / or the load-time curve has a unit root, there is a flexible load and the user is classified as a random user. Otherwise, the user is classified as a stable user. The load-time curve of random users is decomposed into a stable load-time curve that meets the periodicity through the moving smoothing filter algorithm, and the flexible load-time curve that does not meet the periodicity is decomposed by taking the N-order difference of the load-time curve of random users.

8. The user load short-term forecasting system based on high-frequency data according to claim 5 is characterized in that: The prediction unit is used to: The load-time curve of energy-saving users is subjected to moving smoothing filter algorithm to remove noise and obtain the short-term forecast load-time curve of energy-saving users on the forecast day. The load-time curve of random users is subjected to moving smoothing filter algorithm to remove noise and obtain the short-term forecast load-time curve of random users on the forecast day. The stable load-time curve is subjected to moving smoothing filter algorithm to remove noise and obtain the short-term predicted stable load-time curve of the forecast day; The neural network model is trained using historical meteorological data and calendar data as input data and historical flexible load as output data; After the neural network model training is completed, the meteorological forecast data and calendar data of the forecast day are input, and the short-term forecast flexible load-time curve of the forecast day is output through the neural network model.

Images