Load prediction method and system of energy storage system, and electronic equipment
By acquiring and encoding the timestamps and coding periods of the load data, and combining the prediction model to perform load prediction, the problem of single-step prediction distortion in the prior art is solved, and the accuracy of load prediction is improved.
Patent Information
- Application Number
- CN202411021302.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-29
- Publication Date
- 2024-11-22
AI Technical Summary
In the prior art, loads are only predicted in a single step, resulting in distortion of the prediction results and it is difficult to effectively deal with the complex relationship between multiple load factors.
By obtaining the timestamp of the load data to be processed, determining the encoding cycle, encoding the load data to be processed, obtaining the feature data set, and inputting it into the pre-constructed prediction model to obtain the load prediction results after the current moment, considering the time periodicity With load periodicity, long-span sequence-to-sequence prediction is performed.
The accuracy of load prediction is improved, and the periodic characteristics of the energy storage system load can be further considered compared with single-step prediction, and long-span load prediction can be carried out.
Smart Images

Figure CN119009954A_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the technical field of load forecasting, and specifically relates to a load forecasting method, system and electronic equipment for an energy storage system. Background Art
[0002] When forecasting loads with multiple load factors, the complex relationships among the loads make it difficult to learn the data relationships. As a result, when forecasting data, most people only perform single-step forecasts, which leads to distorted forecast results. Summary of the invention
[0003] Purpose of the invention: This application develops a load prediction method, system and electronic equipment for an energy storage system, aiming to solve the technical problem in the prior art that only a single-step prediction of the load is performed, resulting in distorted prediction results.
[0004] Technical solution: In a first aspect, an embodiment of the present application provides a load forecasting method for an energy storage system, comprising:
[0005] Obtaining load data to be processed, wherein the load data to be processed is load data within a preset period before the current moment;
[0006] Determining a timestamp of the load data to be processed;
[0007] Encoding the load data to be processed based on the timestamp and the encoding period to obtain a feature data set;
[0008] The characteristic data set is input into a pre-built prediction model to obtain load prediction results after the current moment.
[0009] In some embodiments, encoding the load data to be processed is performed based on the timestamp and the encoding period.
[0010] The characterization formula for obtaining the feature data set includes:
[0011]
[0012]
[0013] Wherein, T is the encoding period; timestamp_s is the timestamp; is the periodic sorting value of the timestamp relative to the encoding period; 2Π is the radian of a full circle; T_sin is the characteristic data of the sine mapping of the periodic sorting value in the unit circle; T_cos is the characteristic data of the cosine mapping of the periodic sorting value in the unit circle.
[0014] In some embodiments, the encoding period is the number of seconds corresponding to an hour, the number of seconds corresponding to a day, or the number of seconds corresponding to a week.
[0015] In some embodiments, the step of determining the timestamp of the load data to be processed includes:
[0016] Determining the date and time of the load data to be processed;
[0017] Concatenate the date and the time to obtain a first two-dimensional table, wherein the rows of the first two-dimensional table are used to represent the date, and the columns of the first two-dimensional table are used to represent the time;
[0018] The first two-dimensional table is converted into a time data set, and the time data set is converted into the timestamp.
[0019] In some embodiments, the prediction model includes:
[0020] An input sub-model, the input sub-model being configured to determine input data in the feature data set based on the first time length and integrate the input data into third data;
[0021] a data processing sub-model, wherein the data processing sub-model is configured to perform feature extraction and category prediction on the third data and output fourth data;
[0022] The output sub-model is configured to integrate the fourth data based on the second time length and output a load prediction result.
[0023] In some embodiments, the data processing sub-model includes:
[0024] A first substructure, comprising a fully connected layer, a normalization layer, an activation layer, and a maximum pooling layer connected in series, and the first substructure is configured to receive the third data and output first processed data;
[0025] A second substructure, the second substructure comprising a convolution layer, a normalization layer, an activation layer, and a maximum pooling layer connected in series, the second substructure being configured to receive the first processed data and output second processed data;
[0026] A third substructure includes a fully connected layer, and the third substructure is configured to receive the second processed data and output the fourth data.
[0027] In some embodiments, before the step of determining the timestamp of the load data to be processed, the method further includes:
[0028] Determine sampling granularity;
[0029] The load data to be processed is sampled based on the sampling granularity to obtain a second two-dimensional table; wherein the rows of the second two-dimensional table are used to represent dates, and the columns of the second two-dimensional table are used to represent moments, and the time interval of the moments is the sampling granularity;
[0030] determining an abnormal time, wherein the abnormal time includes a time corresponding to the abnormal load value in the second two-dimensional table;
[0031] determining a first load, the first load comprising a load mean of normal load values corresponding to the abnormal moment in the second two-dimensional table;
[0032] The abnormal load value corresponding to the abnormal moment in the second two-dimensional table is updated based on the first load.
[0033] In some embodiments, before the step of determining the timestamp of the load data to be processed, the method further includes:
[0034] determining a vacant date in the second two-dimensional table;
[0035] Determine a second load, the second load comprising a load average at each sampling time in a plurality of dates before the vacant date;
[0036] The load value at each of the sampling moments in the missing date is filled in based on the second load.
[0037] In a second aspect, an embodiment of the present application further provides a load forecasting system for an energy storage system, comprising:
[0038] A first module, the first module is configured to obtain load data to be processed, the load data to be processed is load data within a preset period before the current moment;
[0039] A second module, the second module is configured to determine a timestamp of the load data to be processed;
[0040] A third module, the third module is configured to encode the load data to be processed based on the timestamp and the encoding period to obtain a feature data set;
[0041] The fourth module is configured to input the characteristic data set into a pre-built prediction model to obtain a load prediction result after the current moment.
[0042] In a third aspect, an embodiment of the present application further provides an electronic device, including:
[0043] A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the load forecasting method for the energy storage system as described in any one of the first aspects by executing the computer instructions.
[0044] Beneficial effect: Compared with the prior art, the load forecasting method for an energy storage system provided in the embodiment of the present application first obtains the timestamp of the load data to be processed before the current moment, then determines the coding period, encodes the load data to be processed based on the timestamp and coding period, obtains a feature data set related to time, and then inputs the feature data set into a pre-built prediction model to obtain the load forecasting result after the current moment. The present application encodes the load data to be processed by the timestamp and coding period, associates the time periodicity with the load periodicity, further considers the periodic characteristics of the energy storage system load compared to the single-step prediction, and can perform sequence-to-sequence prediction over a long time span, thereby improving the accuracy of load prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.
[0046] Figure 1 A flow chart of the steps of a load prediction method for an energy storage system provided in an embodiment of the present application;
[0047] Figure 2 A specific flow chart of step 100 in the load forecasting method for the energy storage system provided in an embodiment of the present application;
[0048] Figure 3 A specific flow chart of step 110 in the load forecasting method for the energy storage system provided in an embodiment of the present application;
[0049] Figure 4 A specific flow chart of step 120 in the load forecasting method for the energy storage system provided in an embodiment of the present application;
[0050] Figure 5 A specific flow chart of step 300 in the load forecasting method for the energy storage system provided in an embodiment of the present application;
[0051] Figure 6 A specific flow chart of step 400 in the load forecasting method for the energy storage system provided in an embodiment of the present application;
[0052] Figure 7A first structural diagram of a prediction model in a load prediction method for an energy storage system provided in an embodiment of the present application;
[0053] Figure 8 A second structural diagram of a prediction model in a load prediction method for an energy storage system provided in an embodiment of the present application;
[0054] Fig. 9 A module connection diagram of a load prediction system for an energy storage system provided in an embodiment of the present application;
[0055] Fig.10 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application;
[0056] Reference numerals: 10, first module; 20, second module; 30, third module; 40, fourth module; 100, memory; 200, processor. DETAILED DESCRIPTION
[0057] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.
[0058] The embodiment of the present application provides a load forecasting method for an energy storage system, which first obtains the timestamp of the load data to be processed before the current moment, then determines the coding period, encodes the load data to be processed based on the timestamp and coding period, obtains a feature data set related to time, and then inputs the feature data set into a pre-built prediction model to obtain the load forecasting result after the current moment. The present application encodes the load data to be processed by the timestamp and coding period, associates the time periodicity with the load periodicity, further considers the periodic characteristics of the energy storage system load compared to the single-step prediction, and can perform sequence-to-sequence prediction over a long time span, thereby improving the accuracy of load prediction.
[0059] In some embodiments, see Figure 1 The load prediction method of the energy storage system provided in the embodiment of the present application is specifically implemented through steps 100 to 400:
[0060] Step 100: Obtain the load data to be processed, where the load data to be processed is the load data within a preset period before the current moment.
[0061] In some embodiments, the embodiment of the present application collects load data collected by a certain energy storage system in the past year as load data to be processed, please refer to Figure 2For the load data to be processed, it is also necessary to fill in the abnormal data, which is specifically achieved through steps 110 to 120:
[0062] Step 110: pre-process the load data to be processed to obtain a second two-dimensional table.
[0063] In some embodiments, see Figure 3 , step 110 is specifically implemented through steps 111 to 112:
[0064] Step 111: Determine the sampling granularity.
[0065] In some embodiments, the sampling granularity of the determined data is 15 minutes, ie, there are a total of 96 load data points per day.
[0066] Step 112: Sample the load data to be processed based on the sampling granularity to obtain a second two-dimensional table.
[0067] In some embodiments, the rows of the second two-dimensional table are used to represent dates, and the columns of the second two-dimensional table are used to represent moments, and the time intervals between the moments are sampling granularity.
[0068] It can be understood that, for example, the load data in the above-mentioned year includes data for 330 days. After sampling in step 100, a second two-dimensional table of 330*97 is generated.
[0069] Step 120: Fill in the load data to be processed based on the second two-dimensional table.
[0070] In some embodiments, see Figure 4 , step 120 is specifically implemented through steps 121 to 126:
[0071] Step 121: Determine the abnormal time, where the abnormal time includes the time corresponding to the abnormal load value in the second two-dimensional table.
[0072] In some embodiments, the abnormal load value may include abnormal values such as null values and zero values. It can be understood that in the second two-dimensional table, the null values and zero values are found, and the corresponding moments are determined to be abnormal moments.
[0073] Step 122: Determine a first load, where the first load includes a load mean value of a normal load value corresponding to the abnormal moment in the second two-dimensional table.
[0074] It can be understood that, for example, at time a, there are b zero values and null values and c normal values remaining, then the average value of these c normal values is obtained as the load average value at time a.
[0075] Step 123: Update the abnormal load value corresponding to the abnormal time in the second two-dimensional table based on the first load.
[0076] It can be understood that, taking the load average value at time a in step 120 as an example, the load average value at time a is assigned to b zero value and null value positions at time a, thereby completing the correction of the abnormal load at the abnormal time.
[0077] Step 124: Determine the vacant dates in the second two-dimensional table.
[0078] It can be understood that the load data in the above-mentioned one year includes 330 days of data, but there are missing dates therein.
[0079] Step 125: determining a second load, the second load comprising load averages at various times on a plurality of dates before the vacant date;
[0080] In some embodiments, for a vacant date, the data of three days before the vacant date is used to obtain the load average at each time in the three days before the vacant date.
[0081] Step 126: Fill in the load values at each time in the missing date based on the second load.
[0082] It can be understood that the load mean values at various moments obtained in step 160 are assigned to the corresponding moments in the missing dates respectively, so as to complete the filling of the missing dates.
[0083] In some embodiments, due to the periodicity factor in the load data, the present application performs cyclic encoding on the time in the load data to be processed and maps the time dimension to the range [-1, 1]. Figure 1 , specifically implemented by step 200 and step 300:
[0084] Step 200: Determine the timestamp of the load data to be processed.
[0085] In some embodiments, the timestamp is a UNIX timestamp, which is the number of seconds that have passed since January 1, 1970 (midnight UTC / GMT). Step 200 is specifically implemented through steps 210 to 230:
[0086] Step 210: Determine the date and time of the load data to be processed.
[0087] Step 220: concatenate the date and time to obtain a first two-dimensional table.
[0088] In some embodiments, the rows of the first two-dimensional table are used to represent dates, and the columns of the first two-dimensional table are used to represent time.
[0089] Step 230: Convert the first two-dimensional table into a time data set, and convert the time data set into a timestamp.
[0090] It can be understood that the date of each day and the time of each moment in the load data to be processed are concatenated and used as a new two-dimensional table, namely the first two-dimensional table. The first two-dimensional table is then converted to obtain a time data set of (365*96, 1), and then the time in the data set is converted into a UNIX timestamp.
[0091] Step 300: Encode the load data to be processed based on the timestamp and the encoding period to obtain a feature data set.
[0092] In some embodiments, see Figure 5 , step 300 is specifically implemented through steps 310 to 330:
[0093] Step 310: Determine first data, where the first data includes a ratio of a timestamp to a coding period.
[0094] It can be understood that the timestamp is the UNIX timestamp corresponding to the current moment, and the encoding period is the number of seconds of the timestamp contained in an hour, a day, or a week. Comparing the timestamp with the encoding period indicates which period the current timestamp belongs to, the hour, day, or week. The time period is cyclical, and an hour today is similar to an hour tomorrow, and a day this week is similar to a day next week.
[0095] In some embodiments, the ratio of the timestamp to the coding period is the period ranking value of the timestamp relative to the coding period, and its characterization formula includes: Among them, T is the encoding period; timestamp_s is the timestamp.
[0096] In some embodiments, multiple coding cycles can be set. For example, the number of seconds per hour, day, and week can correspond to multiple coding cycles. Specifically, one hour is 3600 seconds, one day is 86400 seconds, and one week is 604800 seconds.
[0097] Step 320: Determine second data, where the second data includes a product value of the first data and 2π.
[0098] It can be understood that 2Π is the radian of a whole circle, and multiplying the timestamp by 2Π converts the timestamp into radians to represent its position in the cycle. Its characterization formula includes:
[0099] Step 330: Perform sine calculation and cosine calculation on the second data to obtain a feature data set.
[0100] It can be understood that using sine and cosine to map time to the unit circle and assigning corresponding sine and cosine values, which are always between [-1, 1], can well reflect the characteristics of periodic time of hours, days, and weeks. Moreover, according to the analysis of daily and weekly load data, the load value changes periodically. Therefore, the time period encoding is combined with the load data to better utilize the periodicity of the load data. The characterization formula includes Specifically, sine and cosine calculations are performed on these three values respectively, so six time characteristic values are obtained at each moment to constitute a characteristic data set. These six time characteristic values are all in the range of [-1, 1], and a time data set of (365*96, 6) is obtained. Then, the load data for the whole year is dimensionally converted and added to the data set to obtain a load data set with a dimension of (365*96, 7).
[0101] It can be understood that based on steps 310 to 330, encoding is performed on the load data to be processed based on the timestamp and the encoding period, and the characterization formula for obtaining the characteristic data set includes:
[0102]
[0103] Wherein, T is the encoding period; timestamp_s is the timestamp; is the periodic sorting value of the timestamp relative to the encoding period; 2∏ is the radian of a full circle; T_sin is the characteristic data of the sine mapping of the periodic sorting value in the unit circle, and T_cos is the characteristic data of the cosine mapping of the periodic sorting value in the unit circle.
[0104] Specifically, when the coding period is in hours, the characterization formula includes:
[0105]
[0106]
[0107] Among them, Hour is the coding period corresponding to the number of seconds in an hour;
[0108] When one day is used as the coding cycle, the characterization formula includes:
[0109]
[0110] Among them, Day is the coding period corresponding to the number of seconds in a day;
[0111] When one week is used as the coding cycle, the characterization formula includes:
[0112]
[0113] Wherein, Week is a coding period corresponding to the number of seconds in a week.
[0114] Step 400: Input the characteristic data set into the pre-built prediction model to obtain the load prediction results after the current moment.
[0115] In some embodiments, see Figure 7 and Figure 8 , the prediction model includes input sub-model, data processing sub-model and output sub-model. The following is a detailed introduction to the three sub-models:
[0116] In some embodiments, the input sub-model is used to determine the input data in the feature data set based on the first time length and integrate it into the third data. Specifically, the input sub-model provides a fixed sliding window, and the length of the first time is adjusted by adjusting the size of the sliding window to determine the number of days used for load forecasting. For example, the size of the sliding window is IW, and the data dimension of each input is his_win*96*7, where his_win is the number of days included in the input data. In this embodiment, his_win is set to 3, and 96*7 is the data feature dimension of each day, that is, there are 96 moments every day, and each moment has 7 feature points.
[0117] In some embodiments, the output submodel is used to integrate the fourth data based on the second time length and output the load forecast result. Specifically, the output submodel can customize the time step of the forecast data as needed to adjust the second time length and determine the number of days to be forecasted. For example, the time step of the forecast data is OW, and the output data dimension is fut_win*96*1, where fut_win is the output forecast load days. In this embodiment, fut_win is set to 1, and 96 means that the load value of 96 moments is forecasted every day.
[0118] In some embodiments, the data processing submodel is used to perform feature extraction and category prediction on the third data, and output the fourth data. Specifically, the present application transforms the convolutional neural network to construct a data processing submodel. The data processing submodel specifically includes a first substructure, a second substructure, and a third substructure, wherein the first substructure includes two fully connected layers stacked in sequence, a normalization layer, an activation layer, and a maximum pooling layer, and the first substructure is used to receive the third data and output the first processed data; the second substructure includes a convolutional layer, a normalization layer, an activation layer, and a maximum pooling layer stacked in sequence, and the second substructure is used to receive the first processed data and output the second processed data; the third substructure includes a fully connected layer, and the third substructure is used to receive the second processed data and output the fourth data.
[0119] In some embodiments, a plurality of second substructures may be provided. In the present application, two second substructures are provided, and the two second substructures are connected in series in sequence.
[0120] In some embodiments, the representation formula of the convolutional layer includes:
[0121]
[0122] Where Y is the output of the convolution window; n is the dimension of the convolution kernel; i is the horizontal subscript of the convolution kernel window; j is the vertical subscript of the convolution kernel window; W i,j is the parameter matrix corresponding to the convolution kernel; X i,j is the data in the convolution kernel window; b is the bias;
[0123] In some embodiments, the characterization formula of the maximum pooling layer includes:
[0124] P=Max(X i,j )(i,j∈(1,n));
[0125] Where P is the output of the pooling window; i is the horizontal subscript of the pooling window; j is the vertical subscript of the pooling window; X i,j is the data in the pooling window; n is the size of the pooling window.
[0126] In some embodiments, after building the model architecture of the data processing sub-model, the model is trained. Figure 6 , the model training is specifically implemented through steps 410 to 430:
[0127] Step 410: Obtain a training data set and perform normalization processing on the training data set.
[0128] In some embodiments, the training data set is processed through steps 100 to 400 to obtain a data set that meets the input requirements of the input submodule, and then normalized according to the maximum-minimum normalization formula, and the normalization in the formula is saved as a pkl file, and then the denormalization processing of the model output is performed, wherein the characterization formula of the maximum-minimum normalization is as follows:
[0129]
[0130] Among them, x out is the normalized feature sequence; X in is the feature sequence before normalization; is the average value of the feature sequence in the training data set; X std is the standard deviation of the feature sequence in the training data set.
[0131] Step 420: Set training hyperparameters.
[0132] In some embodiments, the present application sets epochs=1000, where epoch is a complete traversal of the training data set once; batch_size=64, where batch_size is the number of training samples in a single pass; lr=0.005, where lr is the initial learning rate; the number of output channels of the convolutional layer is 16; the model optimizer is SGD; the loss function is MSE, and the characterization formula of the loss function includes:
[0133]
[0134] Among them, MSE is the loss function; n is the number of training samples; i is the subscript of the training sample, indicating the i-th sample; y is the true value; is the predicted value.
[0135] Step 430: Use the training data set to train the data processing sub-model, and perform model verification after the training is completed.
[0136] In some embodiments, the training data set processed in step 410 is used to train the data processing sub-model. After the training is completed, the accuracy of the model is verified by testing the validation set. Specifically, the evaluation criteria of this application adopt the mean absolute error (MAE), root mean square error (RMSE), mean relative error (MAPE) and goodness of fit (R 2 ) are compared with the four indicators, and their respective characterization formulas are as follows:
[0137]
[0138] Among them, y i is the actual value at time i; is the predicted value at time i; is the mean of the true value sequence; m is the length of the predicted sequence.
[0139] It can be understood that the embodiment of the present application provides a load forecasting method for an energy storage system, which first obtains the timestamp of the load data to be processed before the current moment, then determines the coding period, encodes the load data to be processed based on the timestamp and coding period, obtains a time-related feature data set, and then inputs the feature data set into a pre-built prediction model to obtain the load forecast result after the current moment. The present application encodes the load data to be processed by the timestamp and coding period, associates the time periodicity with the load periodicity, and further considers the periodic characteristics of the energy storage system load compared to the single-step prediction, and can perform sequence-to-sequence predictions with a long time span, thereby improving the accuracy of load predictions.
[0140] Accordingly, see Fig. 9 , an embodiment of the present application further provides a load forecasting system for an energy storage system, comprising:
[0141] The first module 10 is configured to obtain load data to be processed, where the load data to be processed is load data within a preset period before the current moment;
[0142] A second module 20, the second module 20 is configured to determine a timestamp of the load data to be processed;
[0143] The third module 30 is configured to encode the load data to be processed based on the timestamp and the encoding period to obtain a feature data set;
[0144] The fourth module 40 is configured to input the characteristic data set into a pre-built prediction model to obtain the load prediction result after the current moment.
[0145] In some embodiments, the second module 20 is specifically used for:
[0146] Determine the date and time of the load data to be processed;
[0147] Concatenate the date and time to obtain a first two-dimensional table, wherein the rows of the first two-dimensional table are used to represent the date, and the columns of the first two-dimensional table are used to represent the time;
[0148] Convert the first two-dimensional table to a time dataset, and convert the time dataset to timestamps.
[0149] In some embodiments, the first module 10 is specifically used for:
[0150] Determine sampling granularity;
[0151] Sampling the load data to be processed based on the sampling granularity to obtain a second two-dimensional table; wherein the rows of the second two-dimensional table are used to represent dates, and the columns of the second two-dimensional table are used to represent moments, and the time interval of the moments is the sampling granularity;
[0152] Determine the abnormal time, the abnormal time including the time corresponding to the abnormal load value in the second two-dimensional table;
[0153] Determine a first load, the first load comprising a load mean of normal load values corresponding to the abnormal moment in the second two-dimensional table;
[0154] The abnormal load value corresponding to the abnormal moment in the second two-dimensional table is updated based on the first load.
[0155] In some embodiments, the first module 10 is specifically used for:
[0156] Determine the vacant dates in the second two-dimensional table;
[0157] Determine a second load, the second load comprising load averages at each sampling time in a plurality of dates before the vacant date;
[0158] The load values at each sampling time in the missing date are filled based on the second load.
[0159] Accordingly, see Fig.10 , an embodiment of the present application further provides an electronic device, including:
[0160] The memory 100 and the processor 200 are connected to each other in communication, the memory 100 stores computer instructions, and the processor 200 executes the load forecasting method of the energy storage system provided in the embodiment of the present application by executing the computer instructions. Specifically, the memory 100 and the processor 200 are connected via a communication bus.
[0161] In some embodiments, the processor 200 may be a central processing unit (CPU). The processor 200 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.
[0162] In some embodiments, the memory 100 is a non-transitory computer-readable storage medium that can be used to store non-transitory software programs, non-transitory computer executable programs and modules, such as program instructions / modules corresponding to the methods in the method embodiments of the present application. The processor 200 executes various functional applications and data processing of the processor 200 by running the non-transitory software programs, instructions and modules stored in the memory 100, that is, implementing the methods in the above method embodiments.
[0163] In some embodiments, the memory 100 may include a program storage area and a data storage area, wherein the program storage area may store an application required for operating the device or at least one function; the data storage area may store data created by the processor 200, etc. In addition, the memory 100 may include a high-speed random access memory 100, and may also include a non-volatile memory 100, such as at least one disk memory 100, a flash memory device, or other non-volatile solid-state memory 100. In some embodiments, the memory 100 may optionally include a memory 100 remotely arranged relative to the processor 200, and these remote memories 100 may be connected to the processor 200 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0164] In some embodiments, one or more modules are stored in the memory 100 and when executed by the processor 200, the methods in the above method embodiments are performed.
[0165] In some embodiments, the specific details of the above-mentioned electronic device can be understood by referring to the corresponding related descriptions and effects in the above-mentioned method embodiments, and will not be repeated here.
[0166] The present application has introduced in detail a load prediction method, system and electronic device for an energy storage system provided by an embodiment of the present application. Specific examples are used herein to illustrate the principle and implementation method of the present application. The description of the above embodiments is only used to help understand the method and core idea of the present application. At the same time, for those skilled in the art, according to the idea of the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.
Claims
1. A load forecasting method for an energy storage system, characterized in that: include: Obtaining load data to be processed, wherein the load data to be processed is load data within a preset period before the current moment; Determining a timestamp of the load data to be processed; Encoding the load data to be processed based on the timestamp and the encoding period to obtain a feature data set; The characteristic data set is input into a pre-built prediction model to obtain load prediction results after the current moment.
2. The load forecasting method for an energy storage system according to claim 1, characterized in that: The encoding process of the load data to be processed based on the timestamp and the encoding period to obtain the characterization formula of the characteristic data set includes: Wherein, T is the encoding period; timestamp_s is the timestamp; is the periodic sorting value of the timestamp relative to the encoding period; 2Π is the radian of a full circle; T_sin is the characteristic data of the sine mapping of the periodic sorting value in the unit circle; T_cos is the characteristic data of the cosine mapping of the periodic sorting value in the unit circle.
3. The load forecasting method for an energy storage system according to claim 1, characterized in that: The encoding period includes the number of seconds corresponding to an hour, the number of seconds corresponding to a day and / or the number of seconds corresponding to a week.
4. The load forecasting method for an energy storage system according to claim 1, characterized in that: The step of determining the timestamp of the load data to be processed comprises: Determining the date and time of the load data to be processed; Concatenate the date and the time to obtain a first two-dimensional table, wherein the rows of the first two-dimensional table are used to represent the date, and the columns of the first two-dimensional table are used to represent the time; The first two-dimensional table is converted into a time data set, and the time data set is converted into the timestamp.
5. The load forecasting method for an energy storage system according to claim 1, characterized in that: The prediction model includes: An input sub-model, the input sub-model being configured to determine input data in the feature data set based on the first time length and integrate the input data into third data; a data processing sub-model, wherein the data processing sub-model is configured to perform feature extraction and category prediction on the third data and output fourth data; The output sub-model is configured to integrate the fourth data based on the second time length and output a load prediction result.
6. The load forecasting method for an energy storage system according to claim 5, characterized in that: The data processing sub-model includes: A first substructure, comprising a fully connected layer, a normalization layer, an activation layer, and a maximum pooling layer connected in series, and the first substructure is configured to receive the third data and output first processed data; a second substructure, the second substructure comprising a convolutional layer, a normalization layer, an activation layer, and a maximum pooling layer connected in series, the second substructure being configured to receive the first processed data and output second processed data; A third substructure includes a fully connected layer, and the third substructure is configured to receive the second processed data and output the fourth data.
7. The load forecasting method for an energy storage system according to claim 1, characterized in that: Before the step of determining the timestamp of the load data to be processed, the method further includes: Determine sampling granularity; The load data to be processed is sampled based on the sampling granularity to obtain a second two-dimensional table; wherein the rows of the second two-dimensional table are used to represent dates, and the columns of the second two-dimensional table are used to represent moments, and the time interval of the moments is the sampling granularity; determining an abnormal time, wherein the abnormal time includes a time corresponding to the abnormal load value in the second two-dimensional table; determining a first load, the first load comprising a load mean of normal load values corresponding to the abnormal moment in the second two-dimensional table; The abnormal load value corresponding to the abnormal moment in the second two-dimensional table is updated based on the first load.
8. The load forecasting method for an energy storage system according to claim 7, characterized in that: Before the step of determining the timestamp of the load data to be processed, the method further includes: determining the vacant dates in the second two-dimensional table; Determine a second load, wherein the second load includes a load average at each sampling time in a plurality of dates before the vacant date; The load value at each of the sampling moments in the missing date is filled in based on the second load.
9. A load forecasting system for an energy storage system, characterized in that: include: A first module, the first module is configured to obtain load data to be processed, the load data to be processed is load data within a preset period before the current moment; A second module, the second module is configured to determine a timestamp of the load data to be processed; A third module, the third module is configured to encode the load data to be processed based on the timestamp and the encoding period to obtain a feature data set; The fourth module is configured to input the characteristic data set into a pre-built prediction model to obtain a load prediction result after the current moment.
10. An electronic device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the load forecasting method for the energy storage system according to any one of claims 1 to 8 by executing the computer instructions.
