Energy storage system load prediction method and apparatus, and electronic device

Abstract

The invention discloses an energy storage system load prediction method and apparatus, and an electronic device. The method comprises the steps of constructing a multi-dimensional feature based on historical load data of an energy storage system; creating a multi-dimensional feature sequence by using a sliding window technology and taking the multi-dimensional features as a data basis; a pre-trained LSTM model is adopted to predict the multi-dimensional feature sequence, and an energy storage system load prediction result in a specific period in the future is obtained; and carrying out data smoothing processing, abnormal data detection, data correction and other preprocessing on the energy storage system load prediction result to obtain a target load prediction result. According to the energy storage system load prediction scheme provided by the invention, the load of the energy storage system in multiple time steps in the next day can be accurately and comprehensively predicted.

Description

Technical Field

[0001] The present invention relates to the field of new energy technology, and in particular to a method and device for predicting load of an energy storage system, and electronic equipment. Background Art

[0002] With the widespread application of renewable energy and the development of smart grids, energy storage systems are becoming increasingly important as a key component for balancing power supply and demand and improving grid stability and efficiency. Accurate load forecasting is of great significance for optimizing the scheduling strategy of energy storage systems and reducing operating costs.

[0003] Existing energy storage system load forecasting methods, such as statistical regression models, have limitations when dealing with nonlinear relationships. They often only consider historical load data in a single dimension and fail to fully utilize the various factors that affect load changes, resulting in limited accuracy of energy storage system load forecasting results. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide a method and device for predicting load of an energy storage system, and an electronic device, which can solve the problem of low accuracy of load prediction results of an energy storage system existing in the prior art.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] An embodiment of the present invention provides a method for predicting load of an energy storage system, wherein the method comprises:

[0007] Constructing multi-dimensional features based on the historical load data of the energy storage system, wherein the multi-dimensional features include at least two of the following: time and date features, seasonal features, time step features, periodic features, rolling window statistical features, fixed window statistical features, and special date features;

[0008] Using sliding window technology to create a multidimensional feature sequence based on the multidimensional features;

[0009] A pre-trained LSTM (Long Short-Term Memory) model is used to predict the multi-dimensional feature sequence to obtain a load forecast result of the energy storage system in a specific future period, wherein the load forecast result includes load values ​​of multiple time steps, each time step corresponding to a time period;

[0010] The energy storage system load forecast result is subjected to preset processing to obtain a target load forecast result, wherein the preset processing includes at least one of the following: data smoothing processing, abnormal data detection, and data correction processing.

[0011] Optionally, before the step of constructing multi-dimensional features based on historical load data of the energy storage system, the method further includes:

[0012] Taking the specific period as a unit, obtaining the historical load data of the energy storage system in each unit;

[0013] For the system historical load data under each unit, generating a first multi-dimensional feature sequence;

[0014] The LSTM model is trained according to the first multidimensional feature sequence to obtain a load prediction result of the energy storage system corresponding to the next unit;

[0015] Determine a prediction error according to a prediction result of the energy storage system load corresponding to the next unit and actual energy storage system load data corresponding to the next unit;

[0016] The hyperparameters in the LSTM model are tuned based on the prediction error to complete a single training of the LSTM model.

[0017] Optionally, the step of using a sliding window technique to create a multidimensional feature sequence based on the multidimensional features includes:

[0018] Using the sliding window technology, with the total time steps contained in a day as the moving step length, features are extracted day by day from the multi-dimensional features constructed from multi-day historical load data to construct a multi-dimensional feature sequence corresponding to each day.

[0019] Optionally, a day contains a total of 96 time steps, each corresponding to 15 minutes.

[0020] Optionally, the multi-dimensional features further include: sine and cosine conversion columns of months, sine and cosine conversion columns of time steps, and feature columns obtained by decomposing the seasonal characteristics.

[0021] Optionally, the step of performing preset processing on the energy storage system load forecast result to obtain a target load forecast result includes:

[0022] Determine the date type corresponding to the date to be predicted;

[0023] Adopting the benchmark load determination rule matching the date type, and determining the benchmark load value according to the energy storage system load forecast result;

[0024] The energy storage system load forecast result is corrected based on the reference load value, and the corrected energy storage system load forecast result is subjected to data smoothing to obtain a target load forecast result.

[0025] Optionally, the step of using the benchmark load determination rule for date type matching to determine the benchmark load value according to the energy storage system load forecast result includes:

[0026] When the date type is the first type, selecting the data of the day before the date to be predicted as the benchmark load value;

[0027] When the date type is the second type, the average of data at one-week intervals and two-week intervals is selected as the reference load value.

[0028] Optionally, the step of using a pre-trained long short-term memory network LSTM model to predict the multi-dimensional feature sequence to obtain a load prediction result of the energy storage system in a future specific period includes:

[0029] Using a pre-trained LSTM model, the multi-dimensional feature sequence is predicted in batches to obtain the load value of each time step corresponding to multiple days;

[0030] From the load values ​​of each time step corresponding to the multiple days, the load value of each time step of the last day is extracted as the energy storage system prediction result.

[0031] An embodiment of the present invention further provides a device for predicting load of an energy storage system, wherein the device comprises:

[0032] A first construction module is used to construct multi-dimensional features based on the historical load data of the energy storage system, wherein the multi-dimensional features include at least two of the following: time and date features, seasonal features, time step features, periodic features, rolling window statistical features, fixed window statistical features, and special date features;

[0033] The second building module is used to create a multi-dimensional feature sequence using the multi-dimensional features as a data basis by using a sliding window technology;

[0034] A prediction module, used to use a pre-trained long short-term memory network LSTM model to predict the multi-dimensional feature sequence to obtain a load prediction result of the energy storage system in a specific future period, wherein the load prediction result includes load values ​​of multiple time steps, each time step corresponding to a time period;

[0035] The correction module is used to perform preset processing on the load forecast result of the energy storage system to obtain a target load forecast result, wherein the preset processing includes at least one of the following: data smoothing processing, abnormal data detection and data correction processing.

[0036] An embodiment of the present invention also provides an electronic device, wherein the electronic device includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; the memory is used to store computer programs; the processor is used to implement the energy storage system load prediction method as described in any of the above embodiments when executing the program stored in the memory.

[0037] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the energy storage system load prediction method as described in any of the above embodiments is implemented.

[0038] The present invention discloses a load forecasting scheme for an energy storage system, which constructs multi-dimensional features based on the historical load data of the energy storage system; uses sliding window technology to create a multi-dimensional feature sequence based on the multi-dimensional features as the data basis; uses a pre-trained long short-term memory network LSTM model to predict the multi-dimensional feature sequence to obtain the load forecasting result of the energy storage system in a specific future period; and performs preset processing on the prediction result of the energy storage system to obtain the target load forecasting result. On the one hand, based on the multi-dimensional features as the data basis, a multi-dimensional feature sequence is created to predict the load of the energy storage system, and the obtained prediction result is more comprehensive and accurate because it comprehensively considers the historical load data characteristics of multiple dimensions. On the other hand, based on the LSTM model, the load of the energy storage system in each specific period can be flexibly predicted, which is more flexible and efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a flowchart showing a method for predicting load of an energy storage system according to an embodiment of the present application;

[0040] Figure 2 is a flowchart showing a method for predicting load of an energy storage system according to an embodiment of the present application;

[0041] Figure 3 is a structural block diagram of a load prediction device for an energy storage system according to an embodiment of the present application;

[0042] Figure 4 It is a structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0043] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0044] Existing energy storage system load forecasting methods often only consider historical load data of a single dimension, and fail to fully utilize the various factors that affect load changes, resulting in limited prediction accuracy. The present invention improves the accuracy of prediction by introducing multi-dimensional features extracted from historical load data. The present invention can predict the load value of the energy storage system for 96 time steps in the next day, which is not only crucial for the real-time scheduling of the energy storage system, but also extremely critical for optimizing energy management and improving economic benefits. Existing forecasting methods often perform poorly when dealing with such short-term forecasting tasks because they lack a full understanding of the laws of load changes in the short term, such as predicting the load value of 96 time steps (one time step every 15 minutes) in the next day at one time.

[0045] The present invention proposes a daily multi-time-step energy storage system load forecasting method based on multi-dimensional features. The method can utilize multi-dimensional features extracted from historical load data (such as date and time related features, periodic features, seasonal features, rolling window statistical features, fixed window statistical features, special date features, etc.), create a multi-dimensional feature sequence through sliding window technology, and combine it with LSTM model for training and forecasting, so as to realize the load forecasting of the energy storage system in a specific future period (such as 96 time steps in the future day) to improve the accuracy of the forecast.

[0046] The energy storage system load forecasting solution provided in the embodiment of the present application is described in detail below through specific embodiments and their application scenarios in conjunction with the accompanying drawings.

[0047] As attached Figure 1 As shown, the energy storage system load prediction method of the embodiment of the present application includes the following steps:

[0048] Step 101: Construct multi-dimensional features based on historical load data of the energy storage system.

[0049] The historical load data may be the energy storage system load data of the previous day, or the energy storage system historical load data of the previous 10 days, previous 15 days or even longer. In the embodiment of the present application, there is no specific limitation on the time length corresponding to the historical data.

[0050] The multidimensional features include at least two of the following: time and date features, seasonal features, time step features, periodic features, rolling window statistical features, fixed window statistical features, and special date features. In an optional embodiment, the multidimensional features may also include: sine and cosine conversion columns of months, sine and cosine conversion columns of time steps, and feature columns obtained by decomposing seasonal characteristics (such as trend, seasonality, and residual columns obtained by seasonal decomposition).

[0051] The above multi-dimensional features can be mainly divided into: basic time features, periodic features, historical load features, rolling and fixed window statistical features, and special date features. Basic time features: extract date and time information from historical load data, including year, month, day, day of the week, and hour, and add sine and cosine conversion columns of month and week to capture seasonal changes. Periodic features: In view of the periodic characteristics of the energy storage system load, sine and cosine conversion columns of 96 time steps (one time step every 15 minutes) are introduced, as well as trend, seasonality and residual columns obtained by seasonal decomposition. Historical load features: include load data of the previous moment and the previous hour to capture short-term change trends of the load. Rolling and fixed window statistical features: by setting rolling and fixed windows of different lengths, calculate the average, maximum, minimum and other statistical quantities of the load to reflect the fluctuation characteristics of the load. Special date features: mark special dates such as weekends and holidays, and consider their potential impact on the load.

[0052] The energy storage system load prediction method provided in the embodiment of the present application is applied to electronic devices with computing and analysis capabilities to predict the energy storage system load within a specific future period. The existing energy storage system prediction method fails to fully utilize the various factors that affect load changes, resulting in limited prediction accuracy. The present invention uses multi-dimensional features extracted from historical load data to perform energy storage system load prediction, which can improve the accuracy of the prediction results.

[0053] Step 102: Use sliding window technology to create a multi-dimensional feature sequence based on multi-dimensional features.

[0054] In a feasible implementation, a sliding window technique can be used, with the total time steps contained in a day as the moving step length, to extract features day by day from the multi-dimensional features constructed from multi-day historical load data, and construct a multi-dimensional feature sequence corresponding to each day.

[0055] For example, the total time steps in a day are 96 time steps, and each time step corresponds to 15 minutes. Then, the sliding window technology can be used to construct a multidimensional feature sequence based on the time step features corresponding to the previous day or multiple days, sliding once every 96 time steps. Assuming that the time step features corresponding to N days are used as the basis, the basic data is N×96 time step features. The target sequence corresponding to the energy storage system load forecast result expected to be predicted in a specific future period is the 96 time step load values ​​of the next day.

[0056] Step 103: Use the pre-trained LSTM model to predict the multi-dimensional feature sequence to obtain the energy storage system load prediction result within a specific future period.

[0057] The load forecast result includes load values ​​of multiple time steps, each time step corresponding to a time period. Preferably, each time step corresponds to 15 minutes, and accordingly, one day includes 96 time steps. The future specific period can be set to 1 day, 2 days or multiple days, etc.

[0058] In the embodiment of the present application, the LSTM model is used to process the long-term dependencies in the multidimensional feature sequence generated by the time series data, that is, the historical load data, so as to improve the prediction performance. In the actual implementation process, the sliding window technology is used to create a multidimensional feature sequence, and combined with the LSTM model, the load value of 96 time steps in the next day can be predicted. It is also possible to dynamically obtain the real data of the previous 10 days or even 15 days, batch predict the load value of 96 time steps of each day, and cut the load value of the last 96 time steps from it as the load prediction value of 96 time steps in the next day.

[0059] Step 104: Perform preset processing on the energy storage system load forecast result to obtain a target load forecast result.

[0060] In order to improve the accuracy and reliability of the prediction results, in the embodiment of the present application, after the energy storage system load prediction result is obtained by the LSTM model prediction, it can also be preset. The preset processing includes at least one of the following: data smoothing processing, abnormal data detection, and data correction processing. It should be noted that step 104 is an optional step and may not be performed in the actual implementation process.

[0061] The energy storage system load forecasting method provided in the embodiment of the present application constructs multi-dimensional features based on the historical load data of the energy storage system; uses sliding window technology to create a multi-dimensional feature sequence based on the multi-dimensional features as the data basis; uses a pre-trained long short-term memory network LSTM model to predict the multi-dimensional feature sequence to obtain the energy storage system load forecast result in a specific future period; and performs preset processing on the energy storage system forecast result to obtain the target load forecast result. On the one hand, based on the multi-dimensional features as the data basis, a multi-dimensional feature sequence is created to forecast the energy storage system load, and the obtained forecast result is more comprehensive and accurate because it comprehensively considers the historical load data characteristics of multiple dimensions; on the second hand, based on the LSTM model, the energy storage system load of each specific period can be flexibly predicted, which is more flexible and efficient; on the third hand, the trend of energy storage system load changes can be more accurately analyzed, thereby providing more effective decision support for the scheduling and management of the energy storage system.

[0062] As attached Figure 2 As shown, a method for predicting energy storage system load in another embodiment of the present application includes the following steps:

[0063] Step 201: Train the LSTM model based on the historical load data of the energy storage system.

[0064] An optional method for training the LSTM model based on historical load data of the energy storage system may include the following sub-steps:

[0065] Sub-step 1: Taking a specific period as a unit, obtain the historical load data of the energy storage system under each unit.

[0066] In the embodiment of the present application, the LSTM model needs to be trained based on the historical load data of the energy storage system. In the actual implementation process, the energy storage system load data of the past year can be collected and sorted, including the load records every 15 minutes, and the sorted historical load data can be pre-processed by data cleaning, missing value filling, standardization and other pre-processing operations to ensure the quality of historical load data.

[0067] Sub-step 2: Generate a first multi-dimensional feature sequence for the system historical load data under each unit.

[0068] Construct multi-dimensional features, including but not limited to: date and time related columns (such as month, day of the week, hour, etc.); seasonal features (such as season columns); sine and cosine transformation columns of the month to capture seasonal patterns; time step features (a total of 96 time steps in a day); sine and cosine transformation columns of the time step; trend, seasonality and residual columns obtained by seasonal decomposition; load data columns at the previous moment; load data for the previous hour; rolling window statistical features; fixed window statistical features; special date features (such as weekends, holidays, etc.).

[0069] The sliding window technique is used to create a sequence in the form of a multidimensional feature sequence. The sequence slides once every 96 time steps to obtain multiple first multidimensional feature sequences.

[0070] Sub-step 3: Train the LSTM model based on the first multi-dimensional feature sequence to obtain the energy storage system load prediction result corresponding to the next unit.

[0071] When building the LSTM model, set appropriate hyperparameters for the model, such as the number of hidden layers, the number of neurons, etc. Use the training set data to train the model, and use the validation set data to tune the model.

[0072] The data set is divided into a training set, a validation set, and a test set, and the ratio of the three can be 70%:15%:15%. In the actual implementation process, the model can be trained based on the training set, and the accuracy of the prediction results of the trained model can be verified based on the validation set and the test set.

[0073] Sub-step 4: Determine the prediction error based on the energy storage system load prediction result corresponding to the next unit and the energy storage system load data actually corresponding to the next unit.

[0074] In the actual implementation process, the LSTM model trained in the previous time can be used to predict the load of the next unit, such as 96 time steps in the next day. The prediction results are compared with the actual load data in the test set to calculate the prediction error. Specifically, the mean square error (MSE) and mean absolute error (MAE) can be used to evaluate the model performance, and the generalization ability of the model can be tested through the cross-validation method.

[0075] Sub-step 5: Tune the hyperparameters in the LSTM model based on the prediction error to complete a single training of the LSTM model.

[0076] The above process can be repeated multiple times to train the LSTM model multiple times. By adjusting the LSTM model hyperparameters to optimize the prediction accuracy, the LSTM model training can be determined to be complete when the prediction error of the LSTM model meets the conditions. The LSTM model can then be used to predict the load of the energy storage system in a specific time period in the future.

[0077] It should be noted that the optimization algorithms used in the LSTM model training process may include but are not limited to: gradient descent method, Adam optimizer, etc. The goal of LSTM model training is to minimize the prediction error.

[0078] Step 202: Construct multi-dimensional features based on historical load data of the energy storage system.

[0079] In the embodiment of the present application, not only is it necessary to implement it based on the historical load data of the energy storage system when training the LSTM model, but when subsequently predicting the load forecast results of the energy storage system in a specific future period based on the LSTM model, it is also necessary to construct multi-dimensional features based on the recent historical load data of the energy storage system to achieve accurate prediction.

[0080] In practical applications, the real energy storage system load data of the previous N days (N≥1) can be dynamically obtained to construct the corresponding multi-dimensional feature sequence. That is, the historical load data can be the energy storage system load data of the previous day, or the energy storage system historical load data of the previous 10 days, the previous 15 days or even longer.

[0081] Step 203: Use sliding window technology to create a multi-dimensional feature sequence based on multi-dimensional features.

[0082] For example, a sliding window technique sequence may be used, where the sequence is in the form of a multi-dimensional feature sequence that slides every 96 time steps.

[0083] Step 204: Use the pre-trained LSTM model to predict the multi-dimensional feature sequence to obtain the energy storage system load prediction result within a specific future period.

[0084] The load forecast result includes load values ​​of multiple time steps, and each time step corresponds to a time period.

[0085] When making a prediction, the trained LSTM model can be called separately, and only the features for predicting the 96 time steps of the next day can be input to make a single prediction. That is, the LSTM model predicts the load value of the 96 time steps of the next day based on the features of the 96 time steps of the previous day, and the corresponding historical load data collected in step 202 is the load data corresponding to the 96 time steps of the previous day.

[0086] Of course, it is not limited to this. A dynamic data fusion strategy can also be used to take the energy storage system load data of the previous N days as input in chronological order, perform continuous batch prediction, and extract a specific part of the final prediction result as output. Optionally, a pre-trained long short-term memory network LSTM model is used to predict the multi-dimensional feature sequence, and the method for obtaining the energy storage system load prediction result in a specific future period can be:

[0087] Using a pre-trained LSTM model, batch predictions are made on multi-dimensional feature sequences to obtain the load values ​​of each time step corresponding to multiple days; from the load values ​​of each time step corresponding to multiple days, the load values ​​of each time step of the last day are extracted as the prediction results of the energy storage system. Among them, one day can correspond to 96 time steps, and each time step corresponds to 15 minutes; of course, it is not limited to this, one time step can also correspond to 30 minutes, then one day corresponds to 48 time steps; one time step can also correspond to 10 minutes, then one day corresponds to 144 time steps, and the duration corresponding to a time step is not specifically limited in the embodiments of this application.

[0088] This optional method of predicting the load of the energy storage system can more accurately capture the trend of load changes, thereby providing more effective decision support for the scheduling and management of the energy storage system.

[0089] The energy storage system load forecast result is subjected to preset processing to obtain a target load forecast result, wherein the preset processing includes at least one of the following: data smoothing processing, abnormal data detection, and data correction processing. An optional method of performing preset processing on the energy storage system load forecast result is shown in steps 205 to 207.

[0090] Step 205: Determine the date type corresponding to the date to be predicted.

[0091] The date types include: the first type and the second type. The type of the date is determined by the day of the week on which the date falls. If the date falls on Monday, Friday, Saturday, or Sunday, it is determined to be the second type. If the date falls on Tuesday, Wednesday, or Thursday, it is determined to be the first type.

[0092] It should be noted that the above is only an exemplary description of the first type and the second type of determination rules, which are not limited to this in the actual implementation process and can be flexibly set by those skilled in the art according to actual needs.

[0093] Step 206: adopt the benchmark load determination rule of date type matching and determine the benchmark load value according to the energy storage system load forecast result.

[0094] Specifically, when the date type is the first type, the data of the day before the predicted date is selected as the benchmark load value; when the date type is the second type, the average of the data at one week interval and two weeks interval is selected as the benchmark load value.

[0095] Step 207: Correct the energy storage system load forecast result based on the reference load value, and perform data smoothing on the corrected energy storage system load forecast result to obtain a target load forecast result.

[0096] Correcting the load forecast results of the energy storage system based on the benchmark load value can ensure that the load forecast results are within a reasonable range. Further data smoothing of the load forecast results can reduce data noise and ensure the reliability of the target load forecast results.

[0097] The energy storage system load forecasting method provided in the embodiment of the present application, on the one hand, comprehensively considers various influencing factors of the energy storage system load, thereby improving the accuracy of the forecast results output by the model; on the second hand, by dynamically acquiring multi-day data and performing batch forecasting, it fully utilizes the time continuity and correlation of historical information, further improving the accuracy and reliability of the forecast results; on the third hand, the forecast period and input characteristics can be adjusted according to actual needs to adapt to the load forecasting needs in different scenarios.

[0098] Figure 3 A structural block diagram of a load prediction device for an energy storage system to implement an embodiment of the present application.

[0099] The energy storage system load prediction device provided in the embodiment of the present application includes the following functional modules:

[0100] A first construction module 301 is used to construct multi-dimensional features based on the historical load data of the energy storage system, wherein the multi-dimensional features include at least two of the following: time and date features, seasonal features, time step features, periodic features, rolling window statistical features, fixed window statistical features, and special date features;

[0101] The second construction module 302 is used to create a multi-dimensional feature sequence using the multi-dimensional features as data basis by using a sliding window technology;

[0102] The prediction module 303 is used to use a pre-trained long short-term memory network LSTM model to predict the multi-dimensional feature sequence to obtain a load prediction result of the energy storage system in a specific future period, wherein the load prediction result includes load values ​​of multiple time steps, each time step corresponding to a time period;

[0103] The correction module 304 is used to perform preset processing on the energy storage system load forecast result to obtain a target load forecast result, wherein the preset processing includes at least one of the following: data smoothing processing, abnormal data detection and data correction processing.

[0104] Optionally, the device further comprises:

[0105] An acquisition module, used for acquiring the historical load data of the energy storage system in each unit in the specific period before the first construction module constructs the multi-dimensional features based on the historical load data of the energy storage system;

[0106] A sequence generation module, used for generating a first multi-dimensional feature sequence for the system historical load data under each unit;

[0107] A training module, used for training the LSTM model according to the first multi-dimensional feature sequence to obtain a load prediction result of the energy storage system corresponding to the next unit;

[0108] An error determination module, used to determine a prediction error according to a load prediction result of the energy storage system corresponding to the next unit and actual load data of the energy storage system corresponding to the next unit;

[0109] A tuning module is used to tune the hyperparameters in the LSTM model based on the prediction error to complete a single training of the LSTM model.

[0110] Optionally, the second construction module is specifically used to: use the sliding window technology, take the total time steps contained in a day as the moving step length, extract features day by day from the multi-dimensional features constructed from multi-day historical load data, and construct a multi-dimensional feature sequence corresponding to each day.

[0111] Optionally, a day contains a total of 96 time steps, each corresponding to 15 minutes.

[0112] Optionally, the multi-dimensional features further include: sine and cosine conversion columns of months, sine and cosine conversion columns of time steps, and feature columns obtained by decomposing the seasonal characteristics.

[0113] The correction module comprises:

[0114] The judgment submodule is used to judge the date type corresponding to the date to be predicted;

[0115] A first extraction submodule, configured to determine a reference load value according to the energy storage system load forecast result by adopting the reference load determination rule matching the date type;

[0116] The smoothing processing submodule is used to correct the energy storage system load forecast result based on the reference load value, and perform data smoothing processing on the corrected energy storage system load forecast result to obtain a target load forecast result.

[0117] Optionally, the extraction submodule is specifically used for:

[0118] When the date type is the first type, selecting the data of the day before the date to be predicted as the benchmark load value;

[0119] When the date type is the second type, the average of data at one-week intervals and two-week intervals is selected as the reference load value.

[0120] Optionally, the prediction module includes:

[0121] A batch processing submodule is used to use a pre-trained LSTM model to batch predict the multi-dimensional feature sequence to obtain the load value of each time step corresponding to multiple days;

[0122] The second extraction submodule is used to extract the load value of each time step of the last day from the load values ​​of each time step corresponding to the multiple days as the energy storage system prediction result.

[0123] The embodiments of the present application provide Figure 3 The energy storage system load forecasting device shown can achieve Figure 1 To avoid repetition, the various processes implemented by the method embodiment are not described here.

[0124] The embodiment of the present invention further provides an electronic device, such as Figure 4 As shown, it includes a processor 401 , a communication interface 402 , a memory 403 and a communication bus 404 , wherein the processor 401 , the communication interface 402 , and the memory 403 communicate with each other via the communication bus 404 .

[0125] Memory 403, used for storing computer programs;

[0126] The processor 401 is used to implement each method step in the above method embodiment when executing the program stored in the memory 403.

[0127] The communication bus mentioned in the above terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0128] The communication interface is used for communication between the above terminal and other devices.

[0129] The memory may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0130] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0131] In another embodiment provided by the present invention, a computer-readable storage medium is provided, in which instructions are stored. When the computer-readable storage medium is executed on a computer, the computer implements the energy storage system load prediction method described in any of the above embodiments.

[0132] In another embodiment provided by the present invention, a computer program product including instructions is also provided, which, when executed on a computer, enables the computer to implement the energy storage system load forecasting method described in any one of the above embodiments.

[0133] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Solid State Disk (SSD)), etc.

[0134] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0135] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for predicting energy storage system load, characterized in that: The method comprises: Constructing multi-dimensional features based on the historical load data of the energy storage system, wherein the multi-dimensional features include at least two of the following: time and date features, seasonal features, time step features, periodic features, rolling window statistical features, fixed window statistical features, and special date features; Using sliding window technology to create a multidimensional feature sequence based on the multidimensional features; A pre-trained long short-term memory network (LSTM) model is used to predict the multi-dimensional feature sequence to obtain a load prediction result of the energy storage system in a specific future period, wherein the load prediction result includes load values ​​of multiple time steps, and each time step corresponds to a time period; The energy storage system load forecast result is subjected to preset processing to obtain a target load forecast result, wherein the preset processing includes at least one of the following: data smoothing processing, abnormal data detection, and data correction processing.

2. The method according to claim 1, characterized in that Before the step of constructing multi-dimensional features based on historical load data of the energy storage system, the method further includes: Taking the specific period as a unit, obtaining the historical load data of the energy storage system in each unit; For the system historical load data under each unit, generating a first multi-dimensional feature sequence; The LSTM model is trained according to the first multi-dimensional feature sequence to obtain a load prediction result of the energy storage system corresponding to the next unit; Determine a prediction error according to a prediction result of the energy storage system load corresponding to the next unit and actual energy storage system load data corresponding to the next unit; The hyperparameters in the LSTM model are tuned based on the prediction error to complete a single training of the LSTM model.

3. The method according to claim 1, characterized in that The step of using the sliding window technology to create a multidimensional feature sequence based on the multidimensional features includes: Using the sliding window technology, with the total time steps contained in a day as the moving step length, features are extracted day by day from the multi-dimensional features constructed from multi-day historical load data to construct a multi-dimensional feature sequence corresponding to each day.

4. The method according to claim 3, characterized in that A day contains a total of 96 time steps, each corresponding to 15 minutes.

5. The method according to claim 1, characterized in that: The multi-dimensional features also include: sine and cosine conversion columns of months, sine and cosine conversion columns of time steps, and feature columns obtained by decomposing the seasonal characteristics.

6. The method according to claim 1, characterized in that The step of performing preset processing on the energy storage system load forecast result to obtain a target load forecast result comprises: Determine the date type corresponding to the date to be predicted; Adopting the benchmark load determination rule matching the date type, and determining the benchmark load value according to the energy storage system load forecast result; The energy storage system load forecast result is corrected based on the reference load value, and the corrected energy storage system load forecast result is subjected to data smoothing to obtain a target load forecast result.

7. The method according to claim 6, characterized in that The step of adopting the benchmark load determination rule matching the date type and determining the benchmark load value according to the energy storage system load forecast result comprises: When the date type is the first type, selecting the data of the day before the date to be predicted as the benchmark load value; When the date type is the second type, the average of data at one-week intervals and two-week intervals is selected as the reference load value.

8. The method according to claim 1, characterized in that: The step of using a pre-trained long short-term memory network LSTM model to predict the multi-dimensional feature sequence to obtain a load prediction result of the energy storage system in a future specific period includes: Using a pre-trained LSTM model, the multi-dimensional feature sequence is predicted in batches to obtain the load value of each time step corresponding to multiple days; From the load values ​​of each time step corresponding to the multiple days, the load value of each time step of the last day is extracted as the energy storage system prediction result.

9. A load prediction device for an energy storage system, characterized in that: The device comprises: A first construction module is used to construct multi-dimensional features based on the historical load data of the energy storage system, wherein the multi-dimensional features include at least two of the following: time and date features, seasonal features, time step features, periodic features, rolling window statistical features, fixed window statistical features, and special date features; The second building module is used to create a multi-dimensional feature sequence using the multi-dimensional features as a data basis by using a sliding window technology; A prediction module, used to use a pre-trained long short-term memory network LSTM model to predict the multi-dimensional feature sequence to obtain a load prediction result of the energy storage system in a specific future period, wherein the load prediction result includes load values ​​of multiple time steps, each time step corresponding to a time period; The correction module is used to perform preset processing on the prediction result of the energy storage system to obtain a target load prediction result, wherein the preset processing includes at least one of the following: data smoothing processing, abnormal data detection and data correction processing.

10. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor is used to implement the energy storage system load prediction method as described in any one of claims 1-8 when executing the program stored in the memory.

Images