Load prediction method, device and energy management system

Abstract

The application discloses a load prediction method, device and energy management system, and belongs to the field of power systems. The load prediction method comprises the following steps: acquiring first historical load data corresponding to a first time length before a target acquisition time and second historical load data corresponding to a second time length before the target acquisition time; predicting first load corresponding to a third time length after the target acquisition time based on the first historical load data and a preset ultra-short-term load prediction model; predicting second load corresponding to a fourth time length after the target acquisition time based on the second historical load data and a preset short-term load prediction model; and fusing the first load and the second load to obtain third load corresponding to the fourth time length. The load prediction method can capture and apply high-frequency power consumption load changes of user power consumption behaviors, and improves the power consumption load prediction accuracy.

Description

Technical Field

[0001] This application belongs to the field of power systems, and in particular relates to a load forecasting method, apparatus and energy management system. Background Technology

[0002] Load forecasting is a crucial aspect of power grid energy management, providing key data for this purpose. With the widespread adoption of Internet of Things (IoT) technology, the number of smart devices in the power grid is rapidly increasing, generating a large amount of high-frequency, real-time load data (e.g., time intervals of seconds or milliseconds). Traditional load forecasting methods, however, are mostly based on low-frequency load fluctuations or low-frequency instantaneous load data (e.g., time intervals of hours or minutes), making it difficult to capture high-frequency load changes in user electricity consumption behavior. This results in poor real-time performance of load forecasting results, impacting their accuracy. Summary of the Invention

[0003] This application aims to address at least one of the technical problems existing in related technologies. To this end, this application proposes a load forecasting method, apparatus, and energy management system, which can capture and apply high-frequency changes in user electricity consumption behavior, thereby improving the accuracy of user electricity load forecasting.

[0004] In a first aspect, embodiments of this application provide a load forecasting method, including:

[0005] Acquire first historical load data within a first time period before the target acquisition time, and second historical load data within a second time period before the target acquisition time; the first historical load data and the second historical load data have different time scales, and the time frequency of the first historical load data is higher than the time frequency of the second historical load data;

[0006] Based on the first historical load data and the preset ultra-short-term load prediction model, the first load corresponding to the third time period after the target collection time is predicted.

[0007] Based on the second historical load data and the preset short-term load prediction model, the second load corresponding to the fourth time period after the target collection time is predicted; the fourth time period is longer than the third time period.

[0008] The first load and the second load are merged to obtain the third load corresponding to the fourth time period; wherein the third load and the second load have the same time scale.

[0009] According to the load forecasting method of this application embodiment, ultra-short-term load is predicted by acquiring high-frequency historical load data, and short-term load is predicted by acquiring low-frequency historical load data. Then, based on the ultra-short-term load and short-term load, the final short-term load forecasting result is determined. This method can capture and predict high-frequency load changes in user electricity consumption behavior, and can perform load forecasting from multiple different time intervals. By combining the load forecasting results corresponding to multiple different time intervals, the final load forecasting result is obtained, which improves the comprehensiveness and reliability of the final load forecasting result. This method is applicable to load forecasting in complex, rapidly changing electricity consumption scenarios.

[0010] According to one embodiment of this application, the step of predicting the first load within a third time period after the target data collection time based on the first historical load data and a preset ultra-short-term load prediction model includes:

[0011] Based on the first historical load data corresponding to the target acquisition time, the first load corresponding to the third time period is predicted;

[0012] Based on the first load corresponding to the third time period and the first load predicted based on the first historical load data corresponding to other collection times, update the first load corresponding to the third time period at the same time.

[0013] The other acquisition time is at least one acquisition time located before the target acquisition time.

[0014] According to one embodiment of this application, obtaining the second historical load data corresponding to the second time period before the target collection time includes:

[0015] Acquire the first historical load data corresponding to the second time period prior to the target acquisition time;

[0016] The first historical load data is extracted using a sliding window method to obtain multiple sets of the first historical load data.

[0017] The first historical load data in each group is processed to obtain the second historical load data corresponding to each group.

[0018] According to one embodiment of this application, the step of fusing the first load and the second load to obtain the third load corresponding to the fourth time period includes:

[0019] Based on the first load, a fourth load is determined within the fourth time period after the target acquisition time; the fourth load and the second load have the same time scale.

[0020] The third load is determined based on the second load and the fourth load.

[0021] According to one embodiment of this application, determining the third load based on the second load and the fourth load includes:

[0022] Based on a preset weight sequence, the second load and the fourth load are weighted and fused to obtain the third load.

[0023] According to one embodiment of this application, after weighted fusion of the second load and the fourth load to obtain the third load, the method further includes:

[0024] Based on the actual short-term load and the third load within the fourth time period after the target acquisition time, the weight sequence is optimized and updated; wherein, the actual short-term load and the third load have the same time scale, and the optimized and updated weight sequence is used for the next weighted fusion.

[0025] Secondly, embodiments of this application provide a load forecasting device, comprising:

[0026] The first processing module is used to acquire first historical load data within a first time period before the target acquisition time, and second historical load data within a second time period before the target acquisition time; the first historical load data and the second historical load data have different time scales, and the time frequency of the first historical load data is higher than the time frequency of the second historical load data.

[0027] The second processing module is used to predict the first load within a third time period after the target acquisition time based on the first historical load data and the preset ultra-short-term load prediction model.

[0028] The third processing module is used to predict the second load within a fourth time period after the target acquisition time based on the second historical load data and a preset short-term load prediction model; the fourth time period is longer than the third time period.

[0029] The fourth processing module is used to merge the first load and the second load to obtain the third load corresponding to the fourth time period; wherein the third load and the second load have the same time scale.

[0030] Thirdly, embodiments of this application provide an energy management system, including:

[0031] A local energy management device, wherein the local energy management device is equipped with the modules of the load forecasting device described in the second aspect above, so as to forecast the third load by the load forecasting device and generate energy management instructions based on the third load; the energy management instructions are used to perform energy scheduling on the execution device;

[0032] or,

[0033] It also includes cloud servers;

[0034] The local energy management device is equipped with the first processing module and the second processing module of the load forecasting device described in the second aspect above, which are used to send the second historical load data output by the first processing module and the first load output by the second processing module to the cloud server.

[0035] The cloud server is equipped with the third processing module and the fourth processing module of the load forecasting device described in the second aspect above, so as to obtain the third load predicted by the fourth processing module; the third load is used to generate the energy management instruction.

[0036] Fourthly, embodiments of this application provide a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the load forecasting method as described in the first aspect above.

[0037] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the load forecasting method as described in the first aspect above. Attached Figure Description

[0038] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0039] Figure 1 This is a schematic flowchart of the load forecasting method provided in the embodiments of this application;

[0040] Figure 2 This is a schematic diagram of the load forecasting system provided in the embodiments of this application;

[0041] Figure 3 This is a schematic diagram of the load forecasting device provided in the embodiments of this application;

[0042] Figure 4 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0043] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0044] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0045] The load forecasting method, load forecasting device, energy management system, and readable storage medium provided in this application will be described in detail below with reference to the accompanying drawings and through specific embodiments and application scenarios.

[0046] The load forecasting method can be applied to the terminal, and can be executed by the hardware or software in the terminal.

[0047] The terminal includes, but is not limited to, an energy management system, or a portable communication device such as a mobile phone or tablet. It should also be understood that, in some embodiments, the terminal may not be a portable communication device, but rather a desktop computer.

[0048] The load forecasting method provided in this application can be executed by a cloud server; or by an edge computing device or a local energy management device; or by a combination of a cloud server and an edge computing device.

[0049] In actual implementation, the choice can be made flexibly based on the hardware performance of the local energy management device or the cloud server, and this application embodiment does not limit it.

[0050] The energy management system provided in this application embodiment can be used for load forecasting and power dispatching in distributed power systems or centralized power systems. These power systems employ one or more forms of power supply, including energy storage devices, photovoltaic power generation devices, tidal power generation devices, and wind power generation devices.

[0051] like Figure 1 As shown, the load forecasting method includes steps 110, 120, 130 and 140.

[0052] Step 110: Obtain the first historical load data within the first time period before the target collection time, and the second historical load data within the second time period before the target collection time;

[0053] In this step, the load data is used to represent the power consumption data in the power system.

[0054] In actual implementation, load data can be obtained based on electrical data such as current, voltage, and power; among which, electrical data such as current, voltage, and power can be collected based on smart meters or sensors.

[0055] The first historical load data is high-frequency historical load data at or below the second level.

[0056] The second historical load data consists of low-frequency historical load data at the minute level or longer.

[0057] Among them, "below the second level" can include the second level, millisecond level, or microsecond level, etc.; "above the minute level" can include the minute level or hour level, etc.; microsecond level, millisecond level, second level, minute level, and hour level are all time scales (i.e. time intervals).

[0058] It is understandable that the first and second historical load data have different time scales.

[0059] For example, the first historical load data can be high-frequency historical load data on time scales such as every 1 second, every 3 seconds, every 5 milliseconds, or every 7 microseconds; the second historical load data can be low-frequency historical load data on time scales such as every 1 minute, every 5 minutes, every 1 hour, or every 2 hours.

[0060] Historical load data corresponding to different time scales can be represented by different time frequencies; the smaller the time scale, the higher the corresponding time frequency.

[0061] It is understandable that the sampling frequency of the first historical load data is higher than that of the second historical load data. Therefore, within the same time period, the number of first historical load data is greater than the number of second historical load data.

[0062] The target acquisition time can be any time when the load data is acquired, such as the current acquisition time.

[0063] The first duration (i.e., the length of the first time period) is shorter than the second duration (i.e., the length of the second time period). The specific durations of the first and second durations can be customized based on user needs; for example, the first duration can be set to 3 days, 5 days, or one week; and the second duration can be set to 14 days, half a month, or one month, etc.

[0064] In some embodiments, the first duration may also be equal to the second duration.

[0065] The first historical load data corresponding to the first time period can be high-frequency historical load data within the first time period before the target acquisition time; or it can be high-frequency historical load data within the first time period before the target acquisition time, which is adjacent to or close to the target acquisition time and located before the target acquisition time.

[0066] The second historical load data corresponding to the second time period can be low-frequency historical load data within the second time period before the target acquisition time; or it can be low-frequency historical load data within the second time period before the target acquisition time, which is adjacent to or close to the target acquisition time and located before the target acquisition time.

[0067] For example, if the target collection time is 3:20 PM on the 20th of a certain month in a certain year, then the first historical load data corresponding to the first time period can be the high-frequency historical load data from 3:20 PM on the 13th to 3:20 PM on the 20th of that month, or it can also be the high-frequency historical load data from 3:15 PM on the 13th to 3:15 PM on the 20th of that month; the second time period can be the low-frequency historical load data from 3:18 PM on the 6th to 3:18 PM on the 20th of that month, or it can also be the low-frequency historical load data from 3:16 PM on the 6th to 3:16 PM on the 20th of that month.

[0068] In some embodiments, the specific duration of the first duration or the second duration can be determined based on the density of the first historical load data or the second historical load data, as well as the actual situation such as the requirements for subsequent prediction accuracy; wherein, the density of the first historical load data or the second historical load data is related to their respective time frequencies. It can be understood that within the same time range, load data with higher time frequencies are more dense.

[0069] For example, when the load data is relatively dense, i.e., the time frequency is high, and the requirement for subsequent prediction accuracy is low, the specific duration of the first or second time period can be shorter; conversely, when the load data is relatively dense, i.e., the time frequency is low, and the requirement for subsequent prediction accuracy is high, the specific duration of the first or second time period can be longer to increase the amount of first or second historical load data within the first or second time period, thereby improving the accuracy of subsequent prediction.

[0070] In some embodiments, after acquiring the first historical load data and the second historical load data, the data can be preprocessed, such as cleaning, denoising, and anomaly detection, to improve data quality and thereby improve the accuracy of the first and second loads predicted based on the first and second historical load data.

[0071] Step 120: Based on the first historical load data and the preset ultra-short-term load prediction model, predict the first load corresponding to the third time period after the target collection time;

[0072] In this step, the third duration is shorter than the first duration, and the specific duration of the third duration can be based on user-defined settings; for example, the third duration can be set to 2 hours or 4 hours, etc.

[0073] The first load corresponding to the third time period is the high-frequency load prediction data for the third time period after the target acquisition time, i.e., the ultra-short-term load.

[0074] It is understandable that, based on the high-frequency historical load data corresponding to a relatively long period of time before the target acquisition time, the high-frequency load prediction data for a relatively short period of time after the target acquisition time can be obtained, i.e., the first load.

[0075] For example, if the target collection time is 3 PM on the 20th of a certain month, the high-frequency load prediction data for the time range from 2:55 PM on the 13th to 2:55 PM on the 20th of that month can be used to predict the high-frequency load prediction data for the time range from 3 PM to 7 PM on the 20th of that month. In other words, based on the first historical load data of the week before the target collection time, the first load in the four hours after the target collection time can be predicted.

[0076] The preset ultra-short-term load forecasting model is a pre-set forecasting model used to predict the first load.

[0077] The input to the ultra-short-term load forecasting model can be high-frequency load data, and the output can be high-frequency load forecasting data.

[0078] In actual implementation, the ultra-short-term load prediction model can be trained based on any implementable neural network model.

[0079] For example, CNN models, RNN models, etc., can be used to train ultra-short-term load prediction models.

[0080] In some embodiments, the preset ultra-short-term load prediction model can be trained using sample first load data and sample first load corresponding to the sample first load data as sample labels.

[0081] Among them, the first load of the sample is the first historical load data within the third time period after the first load data of the sample.

[0082] The first load data of the sample can be high-frequency historical load data within the seventh time period prior to the current collection time.

[0083] In actual implementation, high-frequency historical load data within the seventh hour before the target collection time can be collected in advance.

[0084] The seventh duration is longer than the first duration, and the specific duration of the seventh duration can be set by the user.

[0085] In actual implementation, the high-frequency historical load data within the seventh time period can be divided into multiple groups of high-frequency sample training data; among them, the high-frequency sample training data consists of the first load data as samples and the first load data as sample labels.

[0086] The time scale of the first load data in the sample is the same as that of the first historical load data.

[0087] The first sample load data can be the high-frequency historical load data within the first time period before a certain collection time, within the time range formed by the start and end points of the seventh time period; correspondingly, the first sample load is the high-frequency historical load data within the third time period after that certain collection time.

[0088] In actual implementation, a sliding window of [1 * the time value corresponding to the first duration] can be set, with a step size set. Starting from the beginning of the seventh duration and ending at the end of the seventh duration, multiple sample first load data are extracted from the high-frequency historical load data corresponding to the time range of the seventh duration based on this sliding window. Similarly, a sliding window of [1 * the time value corresponding to the third duration] can be set, with the same step size set. Starting from the end of the first duration after the beginning of the seventh duration and ending at the end of the seventh duration, multiple sample first loads are extracted from the high-frequency historical load data corresponding to the time range of the seventh duration based on this sliding window. These sample first load data are then matched one-to-one to form multiple sets of high-frequency sample training data to train the ultra-short-term load prediction model.

[0089] The step size of the sliding window can be customized. The specific setting can be determined based on the actual sample size requirements or model accuracy requirements, etc., and this application does not limit it here.

[0090] Understandably, for example, when there is a large demand for actual sample size and a high requirement for model accuracy, the step size of the sliding window can be set to 1, thereby obtaining more training samples and their corresponding labels, thus improving model accuracy.

[0091] In some embodiments, the ultra-short-term load forecasting model can be trained based on the Transformer model.

[0092] In this embodiment, during actual execution, a lightweight design can be implemented based on the standard Transformer model to reduce the number of encoder and decoder layers while maintaining the core functionality of the model. The ultra-short-term load prediction model can then be trained based on the lightweight Transformer model, which facilitates the subsequent adaptation and optimization of the ultra-short-term load prediction model based on the deployed equipment.

[0093] According to the load forecasting method provided in the embodiments of this application, the first load is forecasted based on the ultra-short-term load forecasting model, which can improve the forecasting accuracy and efficiency of the first load. In subsequent use, the forecasting accuracy of the first load can be further improved by inputting new samples and sample labels into the ultra-short-term load forecasting model for training or by using the model's adaptive optimization and update capabilities.

[0094] In practical applications, the first historical load data can be input into the ultra-short-term load forecasting model to obtain the first load output by the ultra-short-term load forecasting model.

[0095] In some embodiments, the predicted first load can be directly used for energy dispatch in ultra-short time periods to achieve emergency dispatch for ultra-short-term emergencies.

[0096] In some embodiments, the first load corresponding to the third time period after the target acquisition time can be predicted directly based on the first historical load data corresponding to the first time period before the target acquisition time, and the first load can be used for subsequent processing.

[0097] In other embodiments, the first load corresponding to the third time period after the target acquisition time can be updated and optimized based on the first load corresponding to the third time period after the target acquisition time and the first historical load data corresponding to other acquisition times before the target acquisition time, so as to improve the accuracy of the first load and then use it for subsequent processing. The specific implementation method will be described in the embodiments below, and will not be elaborated here.

[0098] Step 130: Based on the second historical load data and the preset short-term load prediction model, predict the second load corresponding to the fourth time period after the target collection time;

[0099] In this step, the fourth duration is longer than the third duration but shorter than the second duration. The specific duration of the fourth duration can be customized by the user; for example, the fourth duration can be set to 24 hours, 36 hours, or 48 hours, etc.

[0100] The second load corresponding to the fourth time period is the low-frequency load prediction data within the fourth time period after the target acquisition time is taken as the starting point.

[0101] Understandably, based on the second historical load data corresponding to a relatively long time range before the target acquisition time, the low-frequency load prediction data for a relatively short time range after the target acquisition time can be obtained, i.e., the second load.

[0102] For example, if the target collection time is 5 pm on the 20th of a certain month, the low-frequency load prediction data for the time range from 4:55 pm on the 6th of that month to 4:55 pm on the 20th of that month can be used to predict the low-frequency load prediction data for the time range from 5 pm on the 20th of that month to 5 pm on the 21st of that month. That is, based on the first historical load data corresponding to the two weeks before the target collection time, the second load for the day after the target collection time can be predicted.

[0103] The preset short-term load forecasting model is a pre-set forecasting model used to predict the second load.

[0104] In some embodiments, a short-term load prediction model can be trained in advance using sample second load data, meteorological data, and time information as samples, and sample second load as sample labels. In practical applications, the second historical load data, meteorological data, and time information corresponding to the second time period before the target collection time can be input into the short-term load prediction model to obtain the second load corresponding to the fourth time period after the target collection time.

[0105] In this embodiment, low-frequency historical load data within a six-hour period prior to the target acquisition time can be collected in advance.

[0106] The sixth duration is longer than the second duration, and the specific duration of the sixth duration can be set by the user.

[0107] In actual implementation, the low-frequency historical load data within the sixth time period can be divided into multiple groups of low-frequency sample training data; among them, the sample training data consists of the sample second load data as samples and the sample second load as sample labels.

[0108] The time scale of the second load data in the sample is the same as that of the second historical load data.

[0109] The second sample load data can be low-frequency historical load data within the second time period before a certain acquisition time, within the time range formed by the start and end points of the sixth time period; correspondingly, the second sample load is low-frequency historical load data within the fourth time period after that certain acquisition time.

[0110] In actual execution, a sliding window of [1 * the time value corresponding to the second duration] can be set, with a step size set. Starting from the beginning of the sixth duration and ending at the end of the sixth duration, multiple sample second load data can be extracted from the low-frequency historical load data within the time range corresponding to the sixth duration based on this sliding window. Similarly, a sliding window of [1 * the time value corresponding to the fourth duration] can be set, with the same step size set. Starting from the end of the second duration after the beginning of the sixth duration and ending at the end of the sixth duration, multiple sample second loads can be extracted from the low-frequency historical load data within the time range corresponding to the sixth duration. These sample second load data can then be matched one-to-one to form multiple sets of low-frequency sample training data.

[0111] The step size of the sliding window can be customized. The specific setting can be determined based on the actual sample size requirements or model accuracy requirements, etc., and this application does not limit it here.

[0112] Meteorological data can include temperature, humidity, and wind speed, among other things.

[0113] Time information can include weekdays or holidays, such as Monday, Tuesday, weekend, Labor Day, or Mid-Autumn Festival.

[0114] In actual implementation, the meteorological data is the meteorological data corresponding to the second load data of the sample, that is, the meteorological data and the second load data of the sample have the same time scale; the time information is the time information corresponding to the second load data of the sample, which can be understood as the time information corresponding to the second load data of the sample on the same day is consistent.

[0115] For example, if the second load data of the sample is load data per minute, then the corresponding meteorological data is also meteorological data per minute; the time information of all sample second load data on Labor Day is "Labor Day".

[0116] Understandably, in actual implementation, meteorological data and time information corresponding to the sample second load data can be extracted using the same extraction method as the sample second load data, thereby forming more complete multi-set sample training data to improve the training effect of the short-term load prediction model.

[0117] In some embodiments, the short-term load forecasting model can be trained based on the Transformer model.

[0118] Transformer is a deep learning model architecture for natural language processing (NLP) and other sequence-to-sequence tasks.

[0119] In actual implementation, during the training of the short-term load forecasting model based on the Transformer model, the model parameters and structure can be adjusted according to the model validation results, including increasing the number of layers, adjusting the learning rate, and introducing regularization, in order to improve the model's prediction accuracy and generalization ability. Furthermore, the input samples can be forward-propagated through the network based on the backpropagation algorithm to calculate the network's output value. Then, the network parameters are updated based on the error to minimize the error, thereby obtaining a short-term load forecasting model with higher accuracy.

[0120] Among them, the backpropagation algorithm (BP algorithm) is a learning algorithm suitable for multi-layer neural networks and is the core technology for training deep learning models.

[0121] According to the load forecasting method provided in the embodiments of this application, by pre-training a short-term load forecasting model based on a Transformer model, using sample second load data, meteorological data, time information, etc. as samples, and sample second load as sample labels, the forecasting efficiency of short-term load can be improved. Furthermore, in subsequent use, by inputting new samples and sample labels into the model for training or by using the model's adaptive optimization and update capabilities, the forecasting accuracy of short-term load can be further improved.

[0122] Of course, in other embodiments, the short-term load forecasting model can also be trained based on other neural network models, which is not limited here.

[0123] Step 140: Combine the first load and the second load to obtain the third load corresponding to the fourth time period.

[0124] In this step, the third load is the better short-term load forecast result, where better can be understood as the energy dispatch based on the third load pair is more effective.

[0125] It is understandable that the time scale of the third load is consistent with that of the second load.

[0126] For example, if the third load can be the low-frequency load forecast result per minute within the fourth duration, then the second load can be the low-frequency load forecast result per minute within the fourth duration.

[0127] In some embodiments, fusing the first load and the second load can be achieved by replacing the low-frequency load prediction results in the second load within the same time range as the first load with the high-frequency load prediction results included in the first load, thereby obtaining the third load and improving the accuracy of the short-term load prediction results.

[0128] In some embodiments, the first load and the second load can be weighted and fused to obtain the third load, thereby improving the accuracy of short-term load forecasting results; the specific implementation will be described in the embodiments of the method of obtaining the third load below, and will not be elaborated here.

[0129] It should be noted that, in actual execution, before performing a precision comparison or weighted fusion of the first load and the second load, it is necessary to convert the first load in terms of time scale, that is, to convert the first load into the same time scale as the second load. The specific implementation method will be described in the following embodiment of the method for obtaining the third load, and will not be elaborated here.

[0130] In this application, by acquiring the first historical load data within a first time period before the current acquisition time and the second historical load data within a second time period before the current acquisition time, and based on the first historical load data, the first load within a third time period after the current acquisition time is predicted, which can realize the capture of high-frequency changes in user electricity consumption behavior and improve short-term energy dispatch capability; then, based on the second historical load data, the second load within a fourth time period after the current acquisition time is predicted, and thus, based on the first and second loads, the third load within the fourth time period is determined, which can realize the optimization processing of short-term load prediction results, thereby improving the accuracy and real-time performance of short-term load prediction results.

[0131] According to the load forecasting method provided in the embodiments of this application, ultra-short-term load is predicted by acquiring high-frequency historical load data, and short-term load is predicted by acquiring low-frequency historical load data. Then, based on the ultra-short-term load and short-term load, the final short-term load forecasting result is determined. This method can capture and predict high-frequency load changes in user electricity consumption behavior, and can perform load forecasting from multiple different time intervals. By combining the load forecasting results corresponding to multiple different time intervals, the final load forecasting result is obtained, which improves the comprehensiveness and reliability of the final load forecasting result. This method is applicable to load forecasting in complex, rapidly changing electricity consumption scenarios.

[0132] The first load update and optimization process is described below.

[0133] In some embodiments, step 120 may include:

[0134] Based on the first historical load data corresponding to the target acquisition time, the first load corresponding to the third time period is predicted;

[0135] Based on the first load corresponding to the third time period and the first load predicted based on the first historical load data corresponding to other collection times, the first load at the same time is updated to the first load corresponding to the third time period after the target collection time.

[0136] In this embodiment, the other acquisition time is at least one acquisition time located before the target acquisition time.

[0137] Among them, the target acquisition time and other acquisition times can be continuous acquisition times based on the fifth time interval, wherein the fifth time interval is less than the first time interval.

[0138] For example, the fifth duration can be 2 minutes, 5 minutes, or 10 minutes, etc.

[0139] Understandably, in actual execution, it is possible to obtain the first historical load data corresponding to the first duration before multiple consecutive collection times based on the fifth duration time interval, including the target collection time. Based on the first historical load data corresponding to each collection time, the corresponding first load is predicted. Understandably, the time ranges corresponding to each first load are different; and since the fifth duration is shorter than the first duration, the time ranges corresponding to the multiple predicted first loads have overlapping time ranges, that is, there are the same time, and there is at least one high-frequency load prediction data under each same time.

[0140] In actual implementation, the high-frequency load prediction data at the same time can be updated based on the high-frequency load prediction data at the same time in the first load corresponding to other acquisition times, within the same time range as the first load corresponding to the target acquisition time, thereby realizing the updating and optimization of the first load corresponding to the target acquisition time.

[0141] The updating and optimization method can be as follows: based on at least one high-frequency load prediction data from the target acquisition time and other acquisition times at the same time, perform filtering or fusion processing to obtain the optimal high-frequency load prediction data at the same time, and replace the original high-frequency load prediction data at the same time in the first load corresponding to the target acquisition time with the optimal high-frequency load prediction data at the same time, thereby obtaining the updated first load corresponding to the target acquisition time; wherein, the filtering processing can include random filtering, taking the mode or taking the median, etc.; the fusion processing can include calculating the average or weighted fusion, etc.

[0142] In some embodiments, the second load corresponding to the fourth time period after the target acquisition time can be updated based on the second load corresponding to the fourth time period and the second load predicted based on the second historical load data corresponding to other acquisition times, and the second load at the same time.

[0143] The specific implementation method can be similar to the first load update and optimization process mentioned above, but the target acquisition time and other acquisition times in the second load update and optimization process are continuous acquisition times based on the eighth time interval; where the eighth time interval is less than the second time interval and greater than the fifth time interval.

[0144] For example, the eighth duration can be 1 hour or 4 hours, etc.

[0145] According to the load prediction method provided in the embodiments of this application, the first load at the same time is updated and optimized by using the first load predicted from the first historical load data corresponding to other collection times. This can improve the accuracy of the first load at the target collection time and improve the accuracy of the third load subsequently obtained based on the first and second loads. In this way, the reliability and stability of the subsequent energy management system's energy dispatch based on the first or third load can be improved.

[0146] The process of obtaining the second historical load data is explained below.

[0147] In some embodiments, acquiring second historical load data corresponding to a second time period prior to the target acquisition time may include:

[0148] Acquire the first historical load data corresponding to the second time period before the target acquisition time;

[0149] The first historical load data is extracted using the sliding window method to obtain multiple sets of first historical load data;

[0150] The first historical load data within each group is processed to obtain the second historical load data for each group.

[0151] In this embodiment, it can be understood that the second historical load data is obtained by transforming the first historical load data on a time scale.

[0152] The transformation of the time scale is based on the sliding window extraction method.

[0153] In actual implementation, for example, when the time scale of the second historical load data is 1 minute and the time scale of the first historical load data is 1 second, a 1*60 sliding window can be set with a step size of 60. The first historical load data within the window can be extracted to obtain multiple sets of first historical load data. The first historical load data within each set can be fused to obtain the second historical load data corresponding to each window position, that is, the second historical load data corresponding to each set. Thus, the first historical load data can be converted into second historical load data with a minute-level average.

[0154] The fusion processing methods may include weighted summation or averaging.

[0155] It is understandable that the second historical load data is low-frequency load data derived from the high-frequency characteristics of the first historical load data. Compared with traditional low-frequency load data, the low-frequency load data derived from the high-frequency characteristics can cover more and more continuous load change information, thus being more representative and real-time.

[0156] Of course, in actual implementation, the size and step size of the sliding window can be defined based on the difference in time scale between the first historical load data and the second historical load data, and this application does not impose any restrictions on this.

[0157] According to the load forecasting method provided in the embodiments of this application, the first historical load data is extracted by a sliding window to obtain multiple sets of first historical load data. The first historical load data in each set is processed to obtain the corresponding second historical load data. This can make the second historical load data cover more load change information, so as to make the second historical load data more representative and real-time, thereby improving the accuracy of the second load predicted based on the second historical load data.

[0158] The method for obtaining the third load is explained below.

[0159] In some embodiments, step 140 may include:

[0160] Based on the first load, determine the fourth load corresponding to the fourth time period after the target acquisition time;

[0161] The third load is determined based on the second and fourth loads.

[0162] In this embodiment, the fourth load is a short-term load prediction result obtained by time-scale conversion of multiple consecutive first loads after the target acquisition time.

[0163] The number of first loads to be converted can be determined based on the difference between the third and fourth durations.

[0164] For example, if the second load is the load forecast result per minute over 24 hours and the first load is the load forecast result per second over 4 hours, the number of first loads to be converted is 6, that is, 6 consecutive first loads are converted on a time scale.

[0165] The specific conversion method can be similar to the above-mentioned process of extracting the second historical load data based on the first historical load data, and will not be elaborated here.

[0166] It is understandable that the time scale of the fourth load is consistent with that of the second load.

[0167] For example, if the second load is the load forecast result per minute within the fourth time period, the fourth load is also the load forecast result per minute within the fourth time period.

[0168] The method for obtaining the fourth load is explained below.

[0169] In some embodiments, determining the fourth load within a fourth time period after the target acquisition time, based on the first load, may include:

[0170] Update the target acquisition time to the end time of the third time interval after the target acquisition time;

[0171] Based on the first load, determine the high-frequency historical load data corresponding to the first time period before the updated target acquisition time;

[0172] Based on high-frequency historical load data, the first load corresponding to the third time period after the updated target acquisition time is predicted.

[0173] A fourth load is determined based on the first load corresponding to multiple target acquisition times; wherein the duration covered by the first load corresponding to multiple target acquisition times should not be less than the duration covered by the second load.

[0174] The following example uses the target data collection time as 0:00 on November 7th, the first duration as 24 hours, the third duration as 4 hours, and the fourth duration, i.e., the duration covered by the second load, as an example to illustrate this embodiment.

[0175] First, based on the high-frequency historical load data from 0:00 to 24:00 on November 6th, the ultra-short-term load from 0:00 to 4:00 on November 7th, i.e., the first load, can be predicted. Then, the target collection time is updated to 4:00 on November 7th. Next, based on the high-frequency historical load data from 4:00 on November 6th to 4:00 on November 7th, the ultra-short-term load from 4:00 to 8:00 on November 7th can be predicted. Since 0:00 to 4:00 on November 7th is the time range after the target collection time, the ultra-short-term load from 4:00 to 8:00 on November 7th is predicted. The prediction samples involved include the ultra-short-term load from 0:00 to 4:00 on November 7th and the high-frequency historical load data from 4:00 to 24:00 on November 6th.

[0176] Following this pattern, the target collection time is updated to 8:00 AM on November 7th. Based on the high-frequency historical load data from 8:00 AM to 11:59 PM on November 6th and the ultra-short-term load from 12:00 AM to 8:00 AM on November 7th, the ultra-short-term load from 8:00 AM to 12:00 PM on November 8th is predicted. This process is repeated, continuously updating the target collection time and performing subsequent predictions, until the ultra-short-term loads corresponding to multiple target collection times, i.e., the duration covered by multiple first loads, are not less than 14 hours. This yields the fourth load before the time scale transformation.

[0177] In actual implementation, the fourth load obtained in the above process before time scale conversion is performed to obtain a fourth load with the same time scale as the second load.

[0178] In some embodiments, during practical application, the accuracy of the fourth load and the second load can be compared, and the better one can be determined as the third load; or, it can also be based on user-defined settings, for example, the user can determine the one with the best performance as the third load based on actual usage.

[0179] According to the load prediction method provided in the embodiments of this application, based on the first load, the fourth load corresponding to the fourth time period after the target acquisition time is determined, and then based on the second load and the fourth load, the third load is determined. This can achieve the uniformity of the time scale between the first load and the second load, thereby facilitating the comparison of the accuracy of the first load and the second load, and facilitating the optimization of the second load by the first load, thereby improving the accuracy of the determined third load.

[0180] In some embodiments, determining the third load based on the second load and the fourth load may include:

[0181] Based on a preset weight sequence, the second and fourth loads are weighted and fused to obtain the third load.

[0182] In this embodiment, the weight sequence is a sequence composed of the weights of the second load and the fourth load.

[0183] In actual execution, the second and fourth loads can be weighted separately to form a weight sequence. Based on the weight sequence, the second and fourth loads can be weighted and summed to obtain the third load.

[0184] The weights of the second and fourth loads can be customized by the user.

[0185] In actual implementation, users can customize the weights of the second and fourth loads according to the actual application of the third load, thereby providing the third load as desired by the user.

[0186] According to the load forecasting method provided in the embodiments of this application, the second load and the fourth load are weighted and fused based on a weight sequence to obtain the third load. This can achieve feature fusion of the second load and the fourth load, so that the third load covers more load change features, making the third load more representative and more accurate, thereby improving the accuracy of energy dispatching based on the third load.

[0187] In some embodiments, after weighted fusion of the second load and the fourth load to obtain the third load, the method may further include:

[0188] Based on the actual short-term load and the third load corresponding to the fourth time interval after the target acquisition time, the weight sequence is optimized and updated; the actual short-term load and the third load have the same time scale; the optimized and updated weight sequence is used for the next weighted fusion.

[0189] In this embodiment, it can be understood that in practical applications, the weight sequence can be continuously optimized based on the actual short-term load and the third load, thereby improving the accuracy and reliability of the weight sequence.

[0190] In actual implementation, initial weights can be set for the second and fourth loads to form an initial weight sequence. The two initial weights can be equally weighted. Then, the initial weight sequence can be optimized based on the actual application effect.

[0191] The optimization process can be as follows: after obtaining the third load corresponding to the fourth time period after the target acquisition time, as time goes by, the actual short-term load corresponding to the fourth time period after the target acquisition time is obtained. Then, based on the actual short-term load and the previously obtained corresponding third load, the weight sequence can be adjusted, and the adjusted weight sequence can be used in the next weighted fusion of the second and fourth loads to adapt to the real-time changes of the actual load and obtain a more representative and accurate third load.

[0192] Among them, the weighting is adjusted based on the actual short-term load and the third load, or the weighting can be adjusted based on the error between the third load and the actual short-term load; where the error can be mean square error (MSE), root mean square error (RMSE), etc.

[0193] In actual implementation, the strategy and weight sequence can be adjusted based on gradient descent.

[0194] Among them, the gradient descent method adjustment strategy can be a weight adjustment strategy based on the error convergence function algorithm, which aims to minimize the error between the third load and the actual short-term load.

[0195] In some embodiments, before adjusting the weights based on the actual short-term load and the third load, the actual short-term load and the third load may be preprocessed, including cleaning, denoising, outlier handling, and standardization, to improve data quality and thus improve the weight adjustment effect.

[0196] According to the load forecasting method provided in the embodiments of this application, the weight sequence is optimized based on the actual short-term load and the third load corresponding to the fourth time period after the target collection time. This can realize real-time online optimization of the weight sequence, and the optimized weight sequence is used for the next weighted fusion, which can improve the accuracy and reliability of the subsequent weight sequence. Thus, the optimized weight sequence can adapt to the real-time changes of the actual load, thereby obtaining a more representative and accurate third load.

[0197] In practice, cloud servers possess abundant computing resources and powerful big data analytics, enabling the training and optimization of complex algorithms. However, due to the latency in data transmission and processing inherent in traditional cloud server hardware, they struggle to meet real-time requirements and may be unable to respond promptly to high-frequency load changes, resulting in slow system response and impacting energy management effectiveness, stability, and reliability. In contrast, edge computing devices are electrically connected to the energy management system and located close to the data source, reducing data transmission and latency. Furthermore, edge computing devices possess sufficient memory, processing power, and computing resources, enabling them to collect and process high-frequency load data in near real-time with the assistance of other network components.

[0198] The load forecasting method of this application will be described below using cloud servers and edge computing devices as the main execution entities.

[0199] like Figure 2 As shown, in some embodiments, the edge computing device may include multiple functional modules, including: a high-frequency data acquisition module, a data processing module, an ultra-short-term load forecasting module, and a data transmission module.

[0200] In this embodiment, the edge computing device is the physical hardware that performs data collection, processing, and execution at the network edge.

[0201] In actual implementation, the edge computing device is electrically connected to the energy management system, which can reduce data transmission and latency, and provide faster response speed, thereby enabling the collection of first historical load data. Moreover, the edge computing device has a certain amount of memory, processing power and computing resources, and can collect and process the first historical load data in near real-time with the help of other network components, thereby enabling the determination of second historical load data based on the first historical load data. It can also support the deployment of ultra-short-term load forecasting models, so that the first load can be predicted based on the first historical load data in the edge computing device.

[0202] The high-frequency data acquisition module is used to collect the first historical load data.

[0203] The data processing module can be used to preprocess the first historical load data, convert the first historical data into the second historical load data, and convert the ultra-short-term load forecast results into the same time scale as the short-term load forecast results, that is, to convert the first load into the fourth load.

[0204] The ultra-short-term load forecasting module is used to predict ultra-short-term load forecasting results based on the first historical load data, i.e., the first load.

[0205] The data transmission module is used to send the second historical load data and the ultra-short-term load forecast results after the time scale is converted, i.e., the first load, to the cloud server.

[0206] In actual implementation, edge computing devices can also directly send the first load to the energy management system to assist the energy management system in energy scheduling in a short period of time, which is convenient for emergency scheduling.

[0207] In some embodiments, when deploying an ultra-short-term load forecasting model on an edge computing device, the trained lightweight Transformer model can be converted into a format suitable for running on the edge computing device (such as TensorFlow Lite, PyTorch Mobile, etc.), and the model can be adapted and optimized according to the hardware characteristics of the edge computing device (such as CPU, memory, power consumption, etc.), thereby improving the efficiency of edge computing and enhancing the real-time performance and accuracy of subsequent ultra-short-term load forecasting models in predicting the first load.

[0208] The load forecasting method provided in the embodiments of this application can rely on the advantages of edge computing devices to achieve real-time capture of high-frequency load data, improve the real-time performance of the load forecasting system and the rapid response to sudden events, thereby improving the accuracy and reliability of the high-frequency load forecasting results based on the load forecasting system, and further improving the stability and reliability of the energy management system for energy dispatching based on the high-frequency load forecasting results.

[0209] like Figure 2 As shown, in some embodiments, the cloud server abstraction may include multiple functional modules, including: a data storage module, a data processing module, a short-term load forecasting module, and a fusion module of short-term load forecasting and ultra-short-term load forecasting.

[0210] In this embodiment, the cloud server is communicatively connected to the edge computing device; the cloud server is used to receive second historical load data and first load sent by the edge computing device, predict the second load based on the second historical load data, and determine the third load based on the first load and the second load.

[0211] Cloud servers have abundant computing resources and powerful big data analysis technology, which can train and optimize complex algorithms, thereby enabling high-precision, long-term prediction of the second load and determination of the third load with even higher accuracy.

[0212] In actual implementation, the cloud server communicates with the edge computing device, which can send the third load to the edge computing device and then forward it to the energy management system, thereby realizing long-term and high-precision energy scheduling of the energy management system.

[0213] The data storage module is used to store the second historical load data sent by the edge computing device and the ultra-short-term load prediction results after the time scale is converted, which is the fourth load.

[0214] The short-term load forecasting module is used to forecast short-term load based on the second historical load data, i.e., the second load.

[0215] The short-term load forecasting and ultra-short-term load forecasting fusion module is used to weight and fuse the second and fourth loads to obtain the third load, and to transmit the third load result to the edge computing device for forwarding to the energy management system.

[0216] The data processing module can be used for weight optimization and adjustment, as well as preprocessing of the data involved in weight adjustment.

[0217] According to the load forecasting method provided in the embodiments of this application, based on the abundant computing resources and powerful big data analysis technology of cloud servers, it can achieve high-precision, long-term forecasting of the second load, and can achieve fusion processing of the forecasting results of the first load and the second load to obtain a third load with higher accuracy. This can improve the accuracy and reliability of the low-frequency load forecasting results of the load forecasting system over a long period of time, and further improve the stability and reliability of the energy management system based on the low-frequency load forecasting results for energy dispatching.

[0218] In some embodiments, an efficient and secure cloud-edge collaborative communication protocol and interface can be established to ensure real-time data transmission and command synchronization between edge computing devices and cloud server platforms. Low-latency and high-reliability communication technologies are adopted to enable rapid interaction between edge prediction results and cloud server prediction tasks, thereby improving collaborative efficiency. The low-latency and high-reliability communication technologies may include 5G or Wi-Fi, and the communication protocols may include MQTT, CoAP, WebSocket, or HTTP / HTTPS.

[0219] The load forecasting method provided in this application can be executed by a load forecasting device. This application uses the example of a load forecasting device executing the load forecasting method to illustrate the load forecasting device provided in this application.

[0220] This application also provides a load forecasting device.

[0221] like Figure 3 As shown, the load prediction device includes: a first processing module 310, a second processing module 320, a third processing module 330 and a fourth processing module 340.

[0222] The first processing module 310 is used to acquire first historical load data within a first time period before the target acquisition time, and second historical load data within a second time period before the target acquisition time; the time scales of the first historical load data and the second historical load data are different, and the time frequency of the first historical load data is higher than the time frequency of the second historical load data.

[0223] The second processing module 320 is used to predict the first load within the third time period after the target acquisition time based on the first historical load data and the preset ultra-short-term load prediction model.

[0224] The third processing module 330 is used to predict the second load within a fourth time period after the target acquisition time based on the second historical load data and a preset short-term load prediction model; the fourth time period is longer than the third time period.

[0225] The fourth processing module 340 is used to merge the first load and the second load to obtain the third load corresponding to the fourth time period.

[0226] According to the load forecasting device provided in the embodiments of this application, ultra-short-term load is predicted by acquiring high-frequency historical load data, and short-term load is predicted by acquiring low-frequency historical load data. Then, based on the ultra-short-term load and short-term load, the final short-term load forecasting result is determined. This can realize the capture and prediction of high-frequency load changes in user electricity consumption behavior, and realize load forecasting from multiple different time intervals. By combining the load forecasting results corresponding to multiple different time intervals, the final load forecasting result is obtained, which improves the comprehensiveness and reliability of the final load forecasting result. It can be applied to load forecasting in complex, fast and changeable electricity consumption scenarios.

[0227] In some embodiments, the first processing module 310 may also be used for:

[0228] Acquire the first historical load data corresponding to the second time period before the target acquisition time;

[0229] The first historical load data is extracted using the sliding window method to obtain multiple sets of first historical load data;

[0230] The first historical load data within each group is processed to obtain the second historical load data for each group.

[0231] In some embodiments, the second processing module 320 may also be used for:

[0232] Based on the first historical load data corresponding to the target acquisition time, the first load corresponding to the third time period is predicted;

[0233] Based on the first load corresponding to the third time period and the first load predicted based on the first historical load data corresponding to other collection times, update the first load corresponding to the third time period at the same time.

[0234] Other acquisition times are at least one acquisition time that is prior to the target acquisition time.

[0235] In some embodiments, the fourth processing module 340 can also be used for:

[0236] Based on the first load, determine the fourth load corresponding to the fourth time period after the target acquisition time;

[0237] The third load is determined based on the second and fourth loads; the fourth load and the second load have the same time scale.

[0238] In some embodiments, the fourth processing module 340 can also be used for:

[0239] Based on a preset weight sequence, the second and fourth loads are weighted and fused to obtain the third load.

[0240] In some embodiments, the device may further include a fifth processing module for:

[0241] After weighted fusion of the second and fourth loads to obtain the third load, the weight sequence is optimized and updated based on the actual short-term load and the third load within the fourth time period after the target acquisition time. The actual short-term load and the third load have the same time scale, and the optimized weight sequence is used for the next weighted fusion.

[0242] The load forecasting device in this application embodiment can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit the specific operating system used.

[0243] The load forecasting device provided in this application embodiment can achieve... Figures 1 to 2 The various processes implemented in the method implementation examples will not be described again here to avoid repetition.

[0244] This application also provides an energy management system.

[0245] In this embodiment, the energy management system includes: a local energy management device; or, it may also include a cloud server.

[0246] In the case of a local energy management device, the local energy management device is equipped with modules of the load forecasting device as described in any of the above embodiments, so that the load forecasting device can forecast a third load and generate energy management instructions based on the third load.

[0247] Among them, energy management commands are used to schedule energy for the execution equipment.

[0248] The executing equipment may include electrical equipment (equipment that consumes electrical energy, such as electric vehicle charging equipment, industrial production equipment, water electrolysis hydrogen production equipment, etc.), energy storage equipment (electrochemical energy storage equipment, supercapacitor energy storage, flywheel energy storage, compressed air energy storage, etc.), and new energy power generation equipment (photovoltaic power generation equipment, wind power generation equipment, etc.).

[0249] In cases where a cloud server is also included, the local energy management device is equipped with a first processing module 310 and a second processing module 320, as described in any of the above embodiments, for sending the second historical load data output by the first processing module 310 and the first load output by the second processing module 320 to the cloud server.

[0250] The cloud server is equipped with a third processing module 330 and a fourth processing module 340 in the load prediction device described in any of the above embodiments, so as to obtain the third load predicted by the fourth processing module 340.

[0251] The third load is used to generate energy management instructions.

[0252] In actual implementation, energy management instructions can be generated on the cloud server based on the third load and then sent to the local energy management device; alternatively, the cloud server can send the third load to the local energy management device, and the local energy management device can generate energy management instructions based on the third load.

[0253] According to the energy management system provided in the embodiments of this application, ultra-short-term load is predicted by acquiring high-frequency historical load data, and short-term load is predicted by acquiring low-frequency historical load data. Then, based on the ultra-short-term load and short-term load, the final short-term load prediction result is determined. This can realize the capture and prediction of high-frequency load changes in user electricity consumption behavior, and realize load prediction from multiple different time intervals. By combining the load prediction results corresponding to multiple different time intervals, the final load prediction result is obtained, which improves the comprehensiveness and reliability of the final load prediction result. It can be applied to load prediction in complex, fast and changeable electricity consumption scenarios.

[0254] In some embodiments, such as Figure 4 As shown, this application embodiment also provides an electronic device 400, including a processor 401, a memory 402, and a computer program stored in the memory 402 and executable on the processor 401. When the program is executed by the processor 401, it implements the various processes of the above-described load forecasting method embodiment and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0255] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0256] This application also provides a non-transitory computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the above-described load forecasting method embodiments and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0257] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0258] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described load forecasting method.

[0259] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0260] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described load prediction method embodiments and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0261] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0262] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0263] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the related technology, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0264] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

[0265] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0266] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

Claims

1. A load forecasting method, characterized in that, include: Acquire first historical load data within a first time period before the target acquisition time, and second historical load data within a second time period before the target acquisition time; the first historical load data and the second historical load data have different time scales, and the time frequency of the first historical load data is higher than the time frequency of the second historical load data; Based on the first historical load data and the preset ultra-short-term load prediction model, the first load corresponding to the third time period after the target collection time is predicted. Based on the second historical load data and the preset short-term load prediction model, the second load corresponding to the fourth time period after the target collection time is predicted; the fourth time period is longer than the third time period. The first load and the second load are merged to obtain the third load corresponding to the fourth time period; wherein the third load and the second load have the same time scale.

2. The load forecasting method according to claim 1, characterized in that, The method of predicting the first load within a third time period after the target data collection time based on the first historical load data and a preset ultra-short-term load prediction model includes: Based on the first historical load data corresponding to the target acquisition time, the first load corresponding to the third time period is predicted; Based on the first load corresponding to the third time period and the first load predicted based on the first historical load data corresponding to other collection times, update the first load corresponding to the third time period at the same time. The other acquisition time is at least one acquisition time located before the target acquisition time.

3. The load forecasting method according to claim 1 or 2, characterized in that, Acquiring the second historical load data corresponding to the second time period prior to the target acquisition time includes: Acquire the first historical load data corresponding to the second time period prior to the target acquisition time; The first historical load data is extracted using a sliding window method to obtain multiple sets of the first historical load data. The first historical load data in each group is processed to obtain the second historical load data corresponding to each group.

4. The load forecasting method according to claim 1 or 2, characterized in that, The step of fusing the first load and the second load to obtain the third load corresponding to the fourth time period includes: Based on the first load, a fourth load is determined within the fourth time period after the target acquisition time; the fourth load and the second load have the same time scale. The third load is determined based on the second load and the fourth load.

5. The load forecasting method according to claim 4, characterized in that, Determining the third load based on the second load and the fourth load includes: Based on a preset weight sequence, the second load and the fourth load are weighted and fused to obtain the third load.

6. The load forecasting method according to claim 5, characterized in that, After weighting and fusing the second load and the fourth load to obtain the third load, the method further includes: Based on the actual short-term load and the third load within the fourth time period after the target acquisition time, the weight sequence is optimized and updated; wherein, the actual short-term load and the third load have the same time scale, and the optimized and updated weight sequence is used for the next weighted fusion.

7. A load forecasting device, characterized in that, include: The first processing module is used to acquire first historical load data within a first time period before the target acquisition time, and second historical load data within a second time period before the target acquisition time; the first historical load data and the second historical load data have different time scales, and the time frequency of the first historical load data is higher than the time frequency of the second historical load data. The second processing module is used to predict the first load within a third time period after the target acquisition time based on the first historical load data and the preset ultra-short-term load prediction model. The third processing module is used to predict the second load within a fourth time period after the target acquisition time based on the second historical load data and a preset short-term load prediction model; the fourth time period is longer than the third time period. The fourth processing module is used to merge the first load and the second load to obtain the third load corresponding to the fourth time period; wherein the third load and the second load have the same time scale.

8. An energy management system, characterized in that, include: A local energy management device, wherein the local energy management device is equipped with the modules of the load forecasting device as described in claim 7, so as to forecast the third load by the load forecasting device and generate energy management instructions based on the third load; the energy management instructions are used to perform energy scheduling on the execution device; or, It also includes cloud servers; The local energy management device is equipped with the first processing module and the second processing module of the load forecasting device as described in claim 7, which are used to send the second historical load data output by the first processing module and the first load output by the second processing module to the cloud server. The cloud server is equipped with the third processing module and the fourth processing module of the load forecasting device as described in claim 7, so as to obtain the third load predicted by the fourth processing module; the third load is used to generate the energy management instruction.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the load forecasting method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the load forecasting method as described in any one of claims 1-6.

Images