A method, apparatus, electronic device, and medium for predicting short-term power load.

Abstract

This invention provides a method, apparatus, electronic device, and medium for predicting short-term power load. After acquiring and preprocessing historical load data and influencing factors, it first uses a bidirectional LSTM model with an attention fusion mechanism for initial prediction. Then, a load pattern state classifier identifies the current stage: stable period, transition period, and abnormal period. Next, it selects an appropriate model from three independently trained error correction models: a low-correction-strength boosting tree model for the stable period, a medium-correction model supporting incremental learning for the transition period, and a high-correction-strength model supporting online learning for the abnormal period. Targeted error compensation is then applied to the initial prediction value, ultimately outputting an optimized load prediction result. This method significantly improves the accuracy and anti-interference capability of power load prediction through a phased adaptive correction mechanism, especially performing better in scenarios with sudden load pattern changes or abnormalities.

Description

Technical Field

[0001] This invention relates to the field of power load forecasting technology, and in particular to a method, apparatus, electronic device and medium for forecasting short-term power load. Background Technology

[0002] Energy consumption forecasting is a crucial component of macro-level planning for sustainable energy development in the industrial and energy sectors. Accurate electricity load forecasting helps decision-makers better schedule and plan the operation of energy supply systems. Therefore, using forecasting models to accurately predict the changing trends of nonlinear load data satisfies the power grid security considerations of steel enterprises. It allows for advance prediction of the electricity consumption required for a specific production process, helping production workshop managers to better schedule and plan the operation of the power supply system. The steel production process involves many large and complex production equipment, such as blast furnaces, converters, electric furnaces, and heating furnaces. Since improving production processes or upgrading equipment is costly and time-consuming, from the perspective of steel enterprises, accurately analyzing historical data of steel production processes and improving the accuracy of electricity load forecasting can reduce operating costs and provide a scientific reference for reducing electricity consumption and production planning in steel enterprises. For enterprises, accurate electricity load forecasting allows for reasonable management of the electricity load required for complex and highly coupled production processes. Currently, there are relatively few studies on prediction models and the prediction accuracy of prediction models is not high. In order to improve the accuracy of short-term power load prediction for steel enterprises more effectively, a method for predicting ultra-short-term power load for steel enterprises based on BiLSTM error correction with attention mechanism is proposed. This method can effectively improve the utilization rate of the power system, realize green manufacturing, reduce enterprise production costs and increase enterprise benefits.

[0003] Current methods for predicting short-term power load in steel enterprises include applied research on dynamic prediction modeling of power load in steel enterprises, discussion on power load analysis and prediction models for steel enterprises, multi-model fusion simulation research on power load prediction in steel enterprises, power load prediction methods for steel enterprises based on improved long short-term memory neural networks, and ultra-short-term power load prediction for steel enterprises based on load characteristics. However, the above-mentioned power load prediction methods have limited analysis of the factors influencing power load and do not consider the randomness, dynamism, and complexity of historical power load data, resulting in poor prediction performance. Summary of the Invention

[0004] The purpose of this invention is to provide a method, apparatus, electronic device and medium for predicting short-term power load, so as to alleviate the technical problem of poor prediction effect caused by low prediction accuracy in the prior art, thereby improving the accuracy of prediction.

[0005] In a first aspect, embodiments of the present invention provide a method for predicting short-term power load, comprising: acquiring first power load data and first power load influencing factors of an enterprise within a first preset time range; wherein the first preset time range is earlier than the current time; preprocessing the first power load data and the first power load influencing factors to obtain processed first preprocessed data; inputting the first preprocessed data into a trained prediction model to output an initial power load prediction value of the enterprise within a second preset time range; wherein the second preset time range is later than the current time; inputting the first preprocessed data and the initial power load prediction value into a trained load pattern state classifier to output a current load pattern state identifier; wherein the load pattern state identifier is used to indicate the load pattern stage to which the current time belongs, including at least a stable period, a transition period, and an abnormal period; and, based on the current load pattern state identifier, retrieving data from pre-stored multiple... Select the corresponding target error correction model from the error correction models; the above multiple error correction models are trained independently for different load pattern stages, wherein: the first error correction model corresponding to the stable period adopts model parameters with low correction strength; the second error correction model corresponding to the transition period adopts model parameters with medium correction strength and support for incremental learning; the third error correction model corresponding to the abnormal period adopts model parameters with high correction strength and support for online learning; input the above first preprocessed data into the above target error correction model, and output the prediction error correction value corresponding to the above initial power load prediction value; based on the above initial power load prediction value and the above prediction error correction value, determine the final power load prediction value of the above enterprise within the above second preset time range; wherein the above prediction model is trained based on a bidirectional long short-term memory network model that incorporates an attention mechanism; the above error correction model is trained based on a boosting tree model.

[0006] In a preferred embodiment of the present invention, the aforementioned plurality of error correction models and the aforementioned load pattern state classifier are collaboratively constructed based on the following steps: acquiring historical power load data and historical power load influencing factors; preprocessing the aforementioned historical power load data and the aforementioned historical power load influencing factors to obtain processed second preprocessed data; predicting the historical power load data and historical power load influencing factors in the aforementioned second preprocessed data based on the aforementioned trained prediction model to obtain initial results; determining the prediction error sequence based on the aforementioned initial results and the actual results of the historical periods corresponding to the aforementioned historical power load data; generating a historical load pattern state identifier sequence based on the aforementioned prediction error sequence, the aforementioned historical power load influencing factors in the aforementioned second preprocessed data, and the preset load pattern stage determination rules; training a plurality of error correction models corresponding to different load pattern stages based on the aforementioned historical load pattern state identifier sequence and the aforementioned prediction error sequence, wherein incremental learning and online learning mechanisms are enabled when training the aforementioned second error correction model and the aforementioned third error correction model; training the aforementioned load pattern state classifier based on the aforementioned second preprocessed data, the aforementioned initial results, the aforementioned prediction error sequence, and the aforementioned historical load pattern state identifier sequence until a preset first training standard is reached.

[0007] In a preferred embodiment of the present invention, the prediction model is constructed based on the following steps: a preset initial bidirectional long short-term memory network model is trained based on the second preprocessed data until a preset second training standard is reached, thereby obtaining the trained prediction model.

[0008] In a preferred embodiment of the present invention, the step of preprocessing the aforementioned historical power load data and the aforementioned historical power load influencing factors to obtain processed second preprocessed data includes: performing correlation analysis on the aforementioned historical power load influencing factors, screening out strongly correlated factors that meet preset standards to obtain first sub-preprocessed data; and removing outliers from the aforementioned historical power load data to obtain second sub-preprocessed data; and determining the aforementioned first sub-preprocessed data and the aforementioned second sub-preprocessed data as the aforementioned second preprocessed data.

[0009] In a preferred embodiment of the present invention, the step of performing correlation analysis on the aforementioned historical power load influencing factors and screening strongly correlated factors that meet preset criteria to obtain the first sub-preprocessed data includes: calculating the correlation between sub-data of different time series among the aforementioned historical power load influencing factors using the Pearson correlation coefficient to obtain multiple correlation coefficients; screening the aforementioned strongly correlated factors from the aforementioned historical power load influencing factors based on the aforementioned correlation coefficients; determining the aforementioned strongly correlated factors as the aforementioned first sub-preprocessed data; and removing outliers from the aforementioned historical power load data to obtain the second sub-preprocessed data includes: removing outliers from the aforementioned historical power load data based on Mahalanobis distance to obtain the second sub-preprocessed data.

[0010] In a preferred embodiment of the present invention, the initial bidirectional long short-term memory network model includes an optimized loss function; the optimized loss function is obtained by optimizing a preset mean square error based on the information entropy formula of a convex function.

[0011] The information entropy expression for the above convex function is as follows:

[0012]

[0013] in, H Standard information entropy; This refers to the power load prediction vector in the second preprocessed data mentioned above. This refers to the actual observed power load vector corresponding to the aforementioned power load prediction vector. is the length of the window; n is the total length of the sequence of the second preprocessed data mentioned above.

[0014] The above optimization loss function is expressed as follows:

[0015]

[0016] Where loss is the loss function; Z is the preset sequence length in the second preprocessed data mentioned above.

[0017] In a preferred embodiment of the present invention, the step of determining the final power load forecast value of the enterprise within the second preset time range based on the initial power load forecast value and the forecast error correction value includes: calculating the sum of the initial power load forecast value and the forecast error correction value to obtain the final power load forecast value.

[0018] Secondly, embodiments of the present invention also provide a short-term power load forecasting device, comprising: a data acquisition module, configured to acquire first power load data and first power load influencing factors of an enterprise within a first preset time range; the first preset time range being earlier than the current time; preprocessing the first power load data and the first power load influencing factors to obtain processed first preprocessed data; a forecasting module, configured to input the first preprocessed data into a trained forecasting model and output an initial power load forecast value of the enterprise within a second preset time range; the second preset time range being later than the current time; inputting the first preprocessed data and the initial power load forecast value into a trained load mode state classifier and outputting a current load mode state identifier; the load mode state identifier indicating the load mode stage to which the current time belongs, including at least a stable period, a transition period, and an abnormal period; and based on the current load mode state identifier, The system selects a target error correction model from a set of pre-stored error correction models. These models are trained independently for different load pattern phases: the first error correction model for the stable period uses low-correction-strength model parameters; the second error correction model for the transition period uses medium-correction-strength model parameters that support incremental learning; and the third error correction model for the abnormal period uses high-correction-strength model parameters that support online learning. The system inputs the first pre-processed data into the target error correction model and outputs a prediction error correction value corresponding to the initial power load forecast. The output module determines the final power load forecast for the enterprise within the second preset time range based on the initial power load forecast and the prediction error correction value. The prediction model is trained using a bidirectional long short-term memory network model incorporating an attention mechanism, and the error correction model is trained using a boosting tree model.

[0019] Thirdly, embodiments of the present invention provide an electronic device, which includes a processor and a memory. The memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the short-term power load prediction method.

[0020] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a short-term power load prediction program, wherein the short-term power load prediction program implements the short-term power load prediction method described above when it is run.

[0021] The embodiments of the present invention have the following beneficial technical effects:

[0022] This invention provides a method, apparatus, electronic device, and medium for predicting short-term power load, comprising: acquiring first power load data and first power load influencing factors of an enterprise within a first preset time range; the first preset time range being earlier than the current time; preprocessing the first power load data and the first power load influencing factors to obtain processed first preprocessed data; inputting the first preprocessed data into a trained prediction model to output an initial power load prediction value of the enterprise within a second preset time range; the second preset time range being later than the current time; inputting the first preprocessed data and the initial power load prediction value into a trained load pattern state classifier to output a current load pattern state identifier; the load pattern state identifier indicating the load pattern stage to which the current time belongs, including at least a stable period, a transition period, and an abnormal period; and, based on the current load pattern state identifier, predicting the short-term power load from the pre-... The method selects a target error correction model from multiple existing error correction models. These models are trained independently for different load pattern stages: the first error correction model for the stable period uses low-correction-strength model parameters; the second error correction model for the transition period uses medium-correction-strength model parameters that support incremental learning; and the third error correction model for the abnormal period uses high-correction-strength model parameters that support online learning. The first preprocessed data is input into the target error correction model, which outputs a prediction error correction value corresponding to the initial power load forecast. Based on the initial power load forecast and the prediction error correction value, the final power load forecast for the enterprise within the second preset time range is determined. The prediction model is trained using a bidirectional long short-term memory network model incorporating an attention mechanism, and the error correction model is trained using a boosting tree model. This method, through a phased dynamic error correction mechanism combined with initial prediction using attention LSTM and error compensation using boosting trees, achieves adaptive accuracy optimization of power load forecasting during stable, transition, and abnormal periods, significantly improving the accuracy and anti-interference capability of short-term forecasts. Attached Figure Description

[0023] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 A flowchart illustrating a short-term power load forecasting method provided in an embodiment of the present invention;

[0025] Figure 2A flowchart illustrating a method for constructing a prediction model according to an embodiment of the present invention;

[0026] Figure 3 A schematic diagram of the structure of a short-term power load prediction device provided in an embodiment of the present invention;

[0027] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0028] Icons: 31-Data acquisition module; 32-Prediction module; 33-Output module; 41-Processor; 42-Memory; 43-Bus; 44-Communication interface. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0030] Currently, methods for predicting short-term power load in steel enterprises include applied research on dynamic prediction modeling of power load in steel enterprises, discussion on power load analysis and prediction models for steel enterprises, multi-model fusion simulation research on power load prediction in steel enterprises, power load prediction methods for steel enterprises based on improved long short-term memory neural networks, and ultra-short-term power load prediction for steel enterprises based on load characteristics. However, the above-mentioned power load prediction methods have limited analysis of the factors influencing power load and do not consider the randomness, dynamism, and complexity of historical power load data, resulting in poor prediction performance.

[0031] Based on this, embodiments of the present invention provide a method, apparatus, electronic device, and medium for predicting short-term power load. This method, through a phased dynamic error correction mechanism combined with initial prediction using attention LSTM and boosting tree error compensation, achieves adaptive accuracy optimization of power load prediction during stable, transitional, and abnormal periods, significantly improving the accuracy and anti-interference capability of short-term prediction. For ease of understanding, a method for predicting short-term power load is first introduced.

[0032] Example 1

[0033] In this embodiment, Figure 1 This is a flowchart illustrating a short-term power load prediction method provided in an embodiment of the present invention.

[0034] Depend on Figure 1 As seen, the method includes:

[0035] Step S101: Obtain the first power load data and the first power load influencing factors of the enterprise within a first preset time range; the first preset time range is earlier than the current time.

[0036] Step S102: Preprocess the above-mentioned first power load data and the above-mentioned first power load influencing factors to obtain the processed first preprocessed data.

[0037] Step S103: Input the first preprocessed data into the trained prediction model and output the initial power load prediction value of the enterprise in the second preset time range; the second preset time range is later than the current time.

[0038] Here, the first preset time range is generally the first power load data and the first power load influencing factors of the previous month at the current moment; the second preset time range is the first power load data and the first power load influencing factors of the next month.

[0039] Step S104: Input the first preprocessed data and the initial power load forecast value into the trained load mode state classifier and output the current load mode state identifier; the load mode state identifier is used to indicate the load mode stage at the current time, including at least the stable period, the transition period and the abnormal period.

[0040] Step S105: Based on the current load mode status identifier, select the corresponding target error correction model from the pre-stored multiple error correction models; the multiple error correction models are trained independently for different load mode stages, wherein: the first error correction model corresponding to the stable period uses model parameters with low correction strength; the second error correction model corresponding to the transition period uses model parameters with medium correction strength and support for incremental learning; and the third error correction model corresponding to the abnormal period uses model parameters with high correction strength and support for online learning.

[0041] Step S106: Input the first preprocessed data into the target error correction model and output the prediction error correction value corresponding to the initial power load prediction value.

[0042] Step S107: Based on the initial power load forecast value and the forecast error correction value, determine the final power load forecast value of the enterprise within the second preset time range.

[0043] The prediction model was trained based on a bidirectional long short-term memory network model that incorporates an attention mechanism; the error correction model was trained based on a boosting tree model.

[0044] Furthermore, the step of determining the final power load forecast value of the enterprise within the second preset time range based on the initial power load forecast value and the forecast error correction value includes: calculating the sum of the initial power load forecast value and the forecast error correction value to obtain the final power load forecast value.

[0045] Furthermore, the aforementioned multiple error correction models and the aforementioned load pattern state classifier are collaboratively constructed based on the following steps: acquiring historical power load data and historical power load influencing factors; preprocessing the aforementioned historical power load data and the aforementioned historical power load influencing factors to obtain processed second preprocessed data; predicting the historical power load data and historical power load influencing factors in the aforementioned second preprocessed data based on the aforementioned trained prediction model to obtain initial results; determining the prediction error sequence based on the aforementioned initial results and the actual results of the historical periods corresponding to the aforementioned historical power load data; generating a historical load pattern state identifier sequence based on the aforementioned prediction error sequence, the aforementioned historical power load influencing factors in the aforementioned second preprocessed data, and the preset load pattern stage determination rules; training multiple error correction models corresponding to different load pattern stages based on the aforementioned historical load pattern state identifier sequence and the aforementioned prediction error sequence, wherein incremental learning and online learning mechanisms are enabled when training the aforementioned second error correction model and the aforementioned third error correction model; training the aforementioned load pattern state classifier based on the aforementioned second preprocessed data, the aforementioned initial results, the aforementioned prediction error sequence, and the aforementioned historical load pattern state identifier sequence until a preset first training standard is reached.

[0046] This invention provides a method for predicting short-term power load, comprising: acquiring first power load data and first power load influencing factors of an enterprise within a first preset time range; the first preset time range being earlier than the current time; preprocessing the first power load data and the first power load influencing factors to obtain processed first preprocessed data; inputting the first preprocessed data into a trained prediction model to output an initial power load prediction value of the enterprise within a second preset time range; the second preset time range being later than the current time; inputting the first preprocessed data and the initial power load prediction value into a trained load pattern state classifier to output a current load pattern state identifier; the load pattern state identifier indicating the load pattern stage to which the current time belongs, including at least a stable period, a transition period, and an abnormal period; and, based on the current load pattern state identifier, selecting from multiple pre-stored error... The corrected model selects the corresponding target error correction model; the multiple error correction models are trained independently for different load pattern stages, wherein: the first error correction model corresponding to the stable period uses model parameters with low correction strength; the second error correction model corresponding to the transition period uses model parameters with medium correction strength and support for incremental learning; and the third error correction model corresponding to the abnormal period uses model parameters with high correction strength and support for online learning. The first preprocessed data is input into the target error correction model, and the output is the prediction error correction value corresponding to the initial power load prediction value. Based on the initial power load prediction value and the prediction error correction value, the final power load prediction value of the enterprise within the second preset time range is determined. The prediction model is trained based on a bidirectional long short-term memory network model incorporating an attention mechanism; the error correction model is trained based on a boosting tree model. This method, through a phased dynamic error correction mechanism combined with the initial prediction of attention LSTM and the error compensation of boosting tree, achieves adaptive accuracy optimization of power load prediction during the stable, transition, and abnormal periods, significantly improving the accuracy and anti-interference capability of short-term prediction.

[0047] Example 2

[0048] Based on the above embodiments, Figure 2 This is a flowchart illustrating a method for constructing a prediction model according to an embodiment of the present invention.

[0049] Depend on Figure 2 As seen, the method for constructing a prediction model includes the following steps:

[0050] Step S201: Obtain historical power load data and factors affecting historical power load.

[0051] Step S202: Preprocess the above-mentioned historical power load data and the above-mentioned historical power load influencing factors to obtain the second preprocessed data.

[0052] Step S203: Based on the second preprocessed data mentioned above, train the preset initial bidirectional long short-term memory network model until the preset second training standard is reached, and obtain the trained prediction model mentioned above.

[0053] Here, the initial bidirectional long short-term memory network model is an excellent variant of RNN, compensating for the vanishing and exploding gradients of RNNs and overcoming the lack of long-term memory capacity by replacing the hidden layers with LSTM model cell units. LSTM neural networks have three gates: input gate, output gate, and forget gate.

[0054] Furthermore, the step of preprocessing the aforementioned historical power load data and the aforementioned historical power load influencing factors to obtain the processed second preprocessed data includes: performing correlation analysis on the aforementioned historical power load influencing factors, screening out strongly correlated factors that meet preset standards to obtain the first sub-preprocessed data; and removing outliers from the aforementioned historical power load data to obtain the second sub-preprocessed data; and determining the aforementioned first sub-preprocessed data and the aforementioned second sub-preprocessed data as the aforementioned second preprocessed data.

[0055] Furthermore, the steps of performing correlation analysis on the aforementioned historical power load influencing factors and screening for strongly correlated factors that meet preset criteria to obtain the first sub-preprocessed data include: calculating the correlation between the sub-data of different time series among the aforementioned historical power load influencing factors using the Pearson correlation coefficient to obtain multiple correlation coefficients; screening for strongly correlated factors from the aforementioned historical power load influencing factors based on the aforementioned correlation coefficients; determining the aforementioned strongly correlated factors as the aforementioned first sub-preprocessed data; and removing outliers from the aforementioned historical power load data to obtain the second sub-preprocessed data include: removing outliers from the aforementioned historical power load data based on Mahalanobis distance to obtain the second sub-preprocessed data.

[0056] The initial bidirectional long short-term memory network model includes an optimized loss function; the optimized loss function is obtained by optimizing the preset mean square error based on the information entropy formula of a convex function.

[0057] The information entropy expression for the above convex function is as follows:

[0058]

[0059] in, H Standard information entropy; This refers to the power load prediction vector in the second preprocessed data mentioned above. This refers to the actual observed power load vector corresponding to the aforementioned power load prediction vector. is the length of the window; n is the total length of the sequence of the second preprocessed data mentioned above.

[0060] The above optimization loss function is expressed as follows:

[0061]

[0062] Where loss is the loss function; Z is the preset sequence length in the second preprocessed data mentioned above.

[0063] This invention provides a method for constructing a prediction model, comprising: acquiring historical power load data and historical power load influencing factors; preprocessing the historical power load data and historical power load influencing factors to obtain second preprocessed data; training an initial long short-term memory network using an attention mechanism based on the second preprocessed data until a preset first training standard is reached to obtain the trained prediction model. This method, by introducing an attention mechanism to optimize the long short-term memory network (LSTM), effectively improves the sensitivity of power load prediction to the temporal characteristics of historical data and multi-dimensional influencing factors, significantly enhancing the prediction accuracy and stability of the model under complex operating conditions.

[0064] Example 3

[0065] Based on the above embodiments, Figure 3 This is a schematic diagram of a short-term power load prediction device provided in an embodiment of the present invention.

[0066] Depend on Figure 3 As seen, the device includes:

[0067] The data acquisition module 31 is used to acquire the first power load data and the first power load influencing factors of the enterprise within a first preset time range; the first preset time range is earlier than the current time; and to preprocess the first power load data and the first power load influencing factors to obtain the processed first preprocessed data.

[0068] The prediction module 32 is used to input the first preprocessed data into a trained prediction model and output the initial power load prediction value of the enterprise within a second preset time range; the second preset time range is later than the current time; input the first preprocessed data and the initial power load prediction value into a trained load mode state classifier and output the current load mode state identifier; the load mode state identifier is used to indicate the load mode stage to which the current time belongs, including at least a stable period, a transition period, and an abnormal period; select the corresponding target error correction model from a plurality of pre-stored error correction models according to the current load mode state identifier; the plurality of error correction models are trained independently for different load mode stages, wherein: the first error correction model corresponding to the stable period uses model parameters with low correction intensity; the second error correction model corresponding to the transition period uses model parameters with medium correction intensity and support for incremental learning; the third error correction model corresponding to the abnormal period uses model parameters with high correction intensity and support for online learning; input the first preprocessed data into the target error correction model and output the prediction error correction value corresponding to the initial power load prediction value.

[0069] The output module 33 is used to determine the final power load forecast value of the enterprise within the second preset time range based on the initial power load forecast value and the forecast error correction value; wherein the forecast model is trained based on a bidirectional long short-term memory network model that incorporates an attention mechanism; and the error correction model is trained based on a boosting tree model.

[0070] The data acquisition module 31, the prediction module 32, and the output module 33 are connected in sequence.

[0071] In one embodiment, the prediction module 32 is further configured to acquire historical power load data and historical power load influencing factors; preprocess the historical power load data and historical power load influencing factors to obtain processed second preprocessed data; predict the historical power load data and historical power load influencing factors in the second preprocessed data based on the trained prediction model to obtain initial results; determine the prediction error sequence based on the initial results and the actual results of the historical periods corresponding to the historical power load data; generate a historical load mode state identifier sequence based on the prediction error sequence, the historical power load influencing factors in the second preprocessed data, and the preset load mode stage determination rules; train multiple error correction models corresponding to different load mode stages based on the historical load mode state identifier sequence and the prediction error sequence, wherein incremental learning and online learning mechanisms are enabled when training the second error correction model and the third error correction model; and train the load mode state classifier based on the second preprocessed data, the initial results, the prediction error sequence, and the historical load mode state identifier sequence until a preset first training standard is reached.

[0072] In one embodiment, the prediction module 32 is further used to train a preset initial bidirectional long short-term memory network model based on the second preprocessed data until a preset second training standard is reached, thereby obtaining the trained prediction model.

[0073] In one embodiment, the prediction module 32 is further configured to perform correlation analysis on the aforementioned historical power load influencing factors, screen out strongly correlated factors that meet preset standards, and obtain first sub-preprocessed data; and remove outliers in the aforementioned historical power load data to obtain second sub-preprocessed data; and determine the aforementioned first sub-preprocessed data and the aforementioned second sub-preprocessed data as the aforementioned second preprocessed data.

[0074] In one embodiment, the prediction module 32 is further configured to perform correlation analysis on the aforementioned historical power load influencing factors, and to screen strongly correlated factors that meet preset criteria to obtain the first sub-preprocessed data. The steps include: calculating the correlation between the sub-data of different time series among the aforementioned historical power load influencing factors using the Pearson correlation coefficient to obtain multiple correlation coefficients; screening the aforementioned strongly correlated factors from the aforementioned historical power load influencing factors based on the aforementioned correlation coefficients; determining the aforementioned strongly correlated factors as the aforementioned first sub-preprocessed data; and removing outliers from the aforementioned historical power load data to obtain the second sub-preprocessed data. The steps include: removing outliers from the aforementioned historical power load data based on Mahalanobis distance to obtain the second sub-preprocessed data.

[0075] In one embodiment, the output module 33 is further configured to calculate the sum of the initial power load forecast value and the forecast error correction value to obtain the final power load forecast value.

[0076] The short-term power load prediction device provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned short-term power load prediction method embodiment. For the sake of brevity, any parts not mentioned in the embodiments of the aforementioned short-term power load prediction device can be referred to the corresponding content in the aforementioned method embodiment.

[0077] This invention also provides an electronic device, such as... Figure 4 The diagram shows the structure of the electronic device, which includes a processor 41 and a memory 42. The memory 42 stores machine-executable instructions that can be executed by the processor 41. The processor 41 executes the machine-executable instructions to implement the aforementioned short-term power load prediction method.

[0078] exist Figure 4 In the illustrated embodiment, the electronic device further includes a bus 43 and a communication interface 44, wherein the processor 41, the communication interface 44, and the memory 42 are connected via the bus.

[0079] The memory 42 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 44 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc. The bus can be an ISA bus, PCI bus, or EISA bus, etc. The aforementioned bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0080] Processor 41 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 41 or by software instructions. Processor 41 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, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in the memory. The processor 41 reads the information in the memory 42 and, in conjunction with its hardware, completes the steps of the short-term power load prediction method of the aforementioned embodiment.

[0081] This invention also provides a machine-readable storage medium storing machine-executable instructions. When these machine-executable instructions are invoked and executed by a processor, they cause the processor to implement the aforementioned short-term power load prediction method. For specific implementation details, please refer to the foregoing method embodiments, which will not be repeated here.

[0082] The computer program products of the short-term power load prediction method, short-term power load prediction device and electronic device provided in the embodiments of the present invention include a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the short-term power load prediction method described in the preceding method embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0083] Unless otherwise specifically stated, the relative steps, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the invention.

[0084] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0085] Finally, it should be noted that the above embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the scope of the technology disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention.

Claims

1. A method for predicting short-term power load, characterized in that, include: Obtain the first power load data of the enterprise within a first preset time range and the factors affecting the first power load; The first preset time range is earlier than the current time. The first power load data and the first power load influencing factors are preprocessed to obtain the first preprocessed data. The first preprocessed data is input into the trained prediction model, and the initial power load prediction value of the enterprise is output within a second preset time range. The second preset time range is later than the current time. The first preprocessed data and the initial power load prediction value are input into the trained load pattern state classifier, and the current load pattern state identifier is output. The load mode status identifier is used to indicate the load mode phase at the current moment, including at least the stable period, the transition period, and the abnormal period; Based on the current load mode status identifier, a corresponding target error correction model is selected from multiple pre-stored error correction models. The multiple error correction models are trained independently for different load mode stages, wherein: the first error correction model corresponding to the stable period uses model parameters with low correction intensity; the second error correction model corresponding to the transition period uses model parameters with medium correction intensity and support for incremental learning; and the third error correction model corresponding to the abnormal period uses model parameters with high correction intensity and support for online learning. The first preprocessed data is input into the target error correction model, and the prediction error correction value corresponding to the initial power load prediction value is output. Based on the initial power load forecast value and the forecast error correction value, determine the final power load forecast value of the enterprise within the second preset time range; The prediction model is trained based on a bidirectional long short-term memory network model that incorporates an attention mechanism; the error correction model is trained based on a boosting tree model.

2. The method for predicting short-term power load according to claim 1, characterized in that, The multiple error correction models and the load pattern state classifier are collaboratively constructed based on the following steps: Obtain historical power load data and factors influencing historical power load; The historical power load data and the factors influencing the historical power load are preprocessed to obtain the second preprocessed data. Based on the trained prediction model, the historical power load data and historical power load influencing factors in the second preprocessed data are predicted to obtain initial results; Based on the initial results and the actual results of the corresponding historical periods of the historical power load data, a prediction error sequence is determined; Based on the prediction error sequence, the historical power load influencing factors in the second preprocessed data, and the preset load mode stage determination rules, a historical load mode status identifier sequence is generated. Based on the historical load pattern state identifier sequence and the prediction error sequence, multiple error correction models corresponding to different load pattern stages are trained respectively, wherein incremental learning and online learning mechanisms are enabled when training the second error correction model and the third error correction model. Based on the second preprocessed data, the initial results, the prediction error sequence, and the historical load pattern state identifier sequence, the load pattern state classifier is trained until a preset first training criterion is reached.

3. The method for predicting short-term power load according to claim 2, characterized in that, The prediction model is constructed based on the following steps: The preset initial bidirectional long short-term memory network model is trained based on the second preprocessed data until the preset second training standard is reached, and the trained prediction model is obtained.

4. The method for predicting short-term power load according to claim 3, characterized in that, The step of preprocessing the historical power load data and the factors influencing the historical power load to obtain the processed second preprocessed data includes: Correlation analysis is performed on the historical power load influencing factors to screen out strongly correlated factors that meet preset criteria, thus obtaining the first sub-preprocessed data; and outliers in the historical power load data are removed to obtain the second sub-preprocessed data. The first sub-preprocessed data and the second sub-preprocessed data are determined as the second preprocessed data.

5. The method for predicting short-term power load according to claim 4, characterized in that, The steps for performing correlation analysis on the historical power load influencing factors, screening for strongly correlated factors that meet preset criteria, and obtaining the first sub-preprocessed data include: The correlation between different time-series sub-data of the historical power load influencing factors is calculated using the Pearson correlation coefficient to obtain multiple correlation coefficients; based on the correlation coefficients, the strongly correlated factors are selected from the historical power load influencing factors; the strongly correlated factors are determined as the first sub-preprocessed data. The step of removing outliers from the historical power load data to obtain the second sub-preprocessed data includes: Based on Mahalanobis distance, outliers in the historical power load data are removed to obtain the second sub-preprocessed data.

6. The method for predicting short-term power load according to claim 5, characterized in that, The initial bidirectional long short-term memory network model includes an optimized loss function; the optimized loss function is obtained by optimizing a preset mean square error based on the information entropy formula of a convex function. The entropy expression of the convex function is as follows: in, H Standard information entropy; This refers to the power load prediction vector in the second preprocessed data; The actual observed power load vector corresponding to the predicted power load vector; is the length of the window; n is the total length of the second preprocessed data sequence. The optimization loss function is expressed as follows: Where loss is the loss function; Z is the preset sequence length in the second preprocessed data.

7. The method for predicting short-term power load according to claim 6, characterized in that, The step of determining the final power load forecast value for the enterprise within the second preset time range based on the initial power load forecast value and the forecast error correction value includes: The final power load forecast value is obtained by summing the initial power load forecast value and the forecast error correction value.

8. A short-term power load forecasting device, characterized in that, include: The data acquisition module is used to acquire the first power load data and the influencing factors of the first power load within a first preset time range for the enterprise; The first preset time range is earlier than the current time; the first power load data and the first power load influencing factors are preprocessed to obtain the first preprocessed data; The prediction module is used to input the first preprocessed data into the trained prediction model and output the initial power load prediction value of the enterprise within a second preset time range. The second preset time range is later than the current time. The first preprocessed data and the initial power load prediction value are input into the trained load pattern state classifier, and the current load pattern state identifier is output. The load mode status identifier is used to indicate the load mode stage at the current moment, including at least a stable period, a transition period, and an abnormal period. Based on the current load mode status identifier, a corresponding target error correction model is selected from a plurality of pre-stored error correction models. The plurality of error correction models are trained independently for different load mode stages, wherein: the first error correction model corresponding to the stable period uses model parameters with low correction strength; the second error correction model corresponding to the transition period uses model parameters with medium correction strength and support for incremental learning; and the third error correction model corresponding to the abnormal period uses model parameters with high correction strength and support for online learning. The first preprocessed data is input into the target error correction model, and the output is a prediction error correction value corresponding to the initial power load prediction value. The output module is used to determine the final power load forecast value of the enterprise within the second preset time range based on the initial power load forecast value and the forecast error correction value; wherein the forecast model is trained based on a bidirectional long short-term memory network model that incorporates an attention mechanism; and the error correction model is trained based on a boosting tree model.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory, the memory storing computer-executable instructions that can be executed by the processor, the processor executing the computer-executable instructions to implement the short-term power load prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a short-term power load forecasting program, which, when executed, implements the short-term power load forecasting method as described in any one of claims 1 to 7.

Images