Multi-dimensional electric quantity prediction method, system and device and readable storage medium

Abstract

According to the multi-dimensional electric quantity prediction method, system and device and the readable storage medium provided by the invention, the quality of internal and external data is comprehensively collected and optimized, key features are accurately screened by applying principal component analysis, and effective dimension reduction of the data is realized. A deep learning time sequence analysis model constructed in combination with a long short-term memory network and an attention mechanism can accurately capture complex dynamic and key information in a time sequence. And through detailed classification training and scientific model evaluation, the accuracy and reliability of a prediction result are ensured. The method not only flexibly adapts to different user groups and regional requirements, but also has strong expandability, and provides strong technical support for intelligent management and decision-making of the power industry.

Description

Technical Field

[0001] The present invention relates to the field of electric power technology, and more specifically to a multi-dimensional electricity prediction method, system, device and readable storage medium. Background Art

[0002] As the core basis for multiple core business links such as power dispatch, electricity price strategy planning and energy trading, the accuracy of power market electricity forecast is directly related to the stable operation of the power system and the efficiency of resource allocation. In recent years, with the rapid development of big data and artificial intelligence technology, machine learning algorithms have gradually become a research hotspot in the field of electricity forecasting with their powerful data processing and pattern recognition capabilities. Advanced models such as support vector machines, random forests and neural networks have effectively improved the foresight and accuracy of electricity forecasting by deeply learning the potential laws in historical electricity data.

[0003] However, although machine learning technology has shown great potential in the field of electricity forecasting, it still faces many challenges. The primary problem is data quality and accuracy, which is the basis of the performance of the forecasting model. Errors and omissions in historical data are like stumbling blocks on the road to forecasting, which directly weaken the training effect and forecasting accuracy of the model. In addition, feature engineering, as a key bridge connecting data and models, its complexity cannot be ignored. In the feature selection process, the introduction of too many irrelevant or redundant features can easily lead to model overfitting, while insufficient features may miss key information, both of which will damage the accuracy of the forecast.

[0004] Traditional power forecasting methods, such as statistical models or simple time series analysis, can meet basic needs to a certain extent, but their limitations are becoming increasingly prominent. They often lack an in-depth understanding of the multi-level characteristics of data and find it difficult to fully capture the complex dynamics of power changes. In addition, the limitations of data quality and feature extraction make the forecast results inadequate in the face of modern power markets with frequent power fluctuations and numerous influencing factors. Summary of the invention

[0005] In view of the above problems, the purpose of the present invention is to provide a multi-dimensional electricity forecasting method, system, device and readable storage medium, which performs multi-dimensional electricity forecasting based on the LSTM-Attention network, solves the problems of insufficient understanding of the multi-level nature of data, data quality and accuracy, insufficient feature extraction, etc. in traditional electricity forecasting methods, and improves the accuracy and fine-grained analysis capability of electricity forecasting.

[0006] In order to achieve the above object, the present invention is implemented through the following technical solutions: In a first aspect, the present invention discloses a multi-dimensional electricity prediction method, comprising: Collect internal and external data of the power system as sample data and evaluate the quality of the sample data. The internal data includes power data, business expansion registration data, time-of-use electricity price information data, and electric vehicle charging pile data; Combined with the quality evaluation results, the sample data is corrected to eliminate noise, dimensional differences and abnormal data; The principal component analysis method is used to extract features from sample data and screen out feature samples that have an impact on power forecasting to reduce redundancy and improve forecasting efficiency. Fusion feature samples as input data for model training, and classification of the input data for model training according to different user groups or different regions; A time series analysis model based on deep learning is constructed using a long short-term memory network model. The input data of the model training is used as input, and the target power is used as output. The attention mechanism technology is introduced to train the time series analysis model based on deep learning. The average relative percentage error and root mean square error are used as evaluation criteria, and the optimal parameters of the time series analysis model based on deep learning are determined according to the evaluation criteria; The internal and external data of the power system are obtained according to time, user group and region, and the prediction results are output using a time series analysis model based on deep learning.

[0007] Furthermore, the external data includes: meteorological data, economic information, policy data, emergency data, and holiday schedule data; the meteorological data includes temperature, humidity, rainfall, wind speed, wind direction, and air pressure; the economic information includes regional GDP, added value of the primary, secondary, and tertiary industries, total population, and per capita consumer expenditure of residents; the policy data includes policy documents, announcements, and notices; the emergency data includes information on natural disasters, accidents, disasters, and public health events.

[0008] Further, the quality evaluation of the sample data includes: Evaluate the data integrity of data samples, the rationality of daily electricity fluctuations, the rationality of the proportion of electricity in each user group, the deviation of total electricity and electricity by industry, the rationality of daily electricity and issued electricity, the rationality of high-voltage user fluctuations, and the quality of meteorological data.

[0009] Furthermore, the sample data is corrected in combination with the quality evaluation result to eliminate noise, dimensional differences and abnormal data, including: For meteorological data, observational data are used for data correction, and interpolation methods are used to supplement missing values; The MinMaxScaler method is used to normalize the sample data and map the data to the range of [0,1]. The normalization is performed using the following formula:

[0010] in, and Respectively represent the sample data before and after normalization, and Respectively represent the maximum and minimum values ​​of a sequence before normalization.

[0011] Furthermore, the principal component analysis method is used to extract features from the sample data and screen out feature samples that have an impact on power prediction, including: Standardize the sample data to generate a data matrix so that the mean of each variable is 0 and the variance is 1; Compute the covariance matrix of the data matrix; Compute the eigenvalues ​​and eigenvectors of the covariance matrix; Select the first k principal components according to the size of the eigenvalue; The data of k principal components are used as feature samples that have an impact on electricity prediction.

[0012] Furthermore, the fused feature samples are used as input data for model training, and the input data for model training is classified according to different user groups or different regions, including: Based on feature samples, feature samples of multi-level influencing factors of the same user group or the same region are merged to construct input data for model training; Classify the input data for model training according to different user groups or different regions.

[0013] Furthermore, the long short-term memory network model is used to construct a time series analysis model based on deep learning, including: Construct an LSTM layer. There are multiple LSTM units in the LSTM layer. Each LSTM unit contains a forget gate, an input gate, and an output gate. The mathematical expression of the LSTM unit algorithm is:

[0014]

[0015]

[0016]

[0017]

[0018]

[0019] in, For the Gate of Oblivion; is the input gate; is the output gate, , They are the state and output of the current LSTM unit respectively; is the input node; is the sigmod function; is the input vector of the current input network; , , , is the weight; , , , is bias; Construct the attention layer and first calculate the attention weight of each hidden layer state , and then multiply the different outputs by their weights and sum them up to get the final output of the attention layer. The specific formula is:

[0020]

[0021]

[0022] in, is the output of the attention layer; for The attention weight of the moment; To calculate the weight; For LSTM Get the hidden layer state value at all times; is bias; is the total period; Build an LSTM-Attention network model and design a two-layer LSTM network layer with 256 and 32 neuron data respectively; add an attention mechanism layer to increase the weight of the preset part in the prediction output results of the LSTM-Attention network model; introduce a dropout layer to prevent the model from overfitting, and output a fixed-format vector through a fully connected layer; Set the loss function and optimizer for the LSTM-Attention network model, use the fit method to train the model, and set the batch size, number of training rounds, and validation data to generate a time series analysis model based on deep learning. Furthermore, the mean relative percentage error and RMS error The calculation method is as follows:

[0023]

[0024] in, is the actual value, is the predicted value, is the sample size.

[0025] In a second aspect, the present invention further discloses a multi-dimensional electricity forecasting system, comprising: The data collection module is used to collect the internal and external data of the power system as sample data and conduct quality evaluation on the sample data. The internal data includes power data, business expansion registration data, time-of-use electricity price information data, and electric vehicle charging pile data. The data correction module is used to perform data correction processing on the sample data in combination with the quality evaluation results to eliminate noise, dimensional differences and abnormal data; The feature extraction module is used to extract features from sample data using the principal component analysis method, and screen out feature samples that have an impact on power forecasting, so as to reduce redundancy and improve forecasting efficiency; The feature fusion and classification module is used to fuse feature samples as input data for model training and classify the input data for model training according to different user groups or different regions; The model building and training module is used to build a time series analysis model based on deep learning using a long short-term memory network model. The input data of the model training is used as input and the target power is used as output. The attention mechanism technology is introduced to train the time series analysis model based on deep learning. The model evaluation and optimization module is used to use the average relative percentage error and the root mean square error as evaluation criteria, and determine the optimal parameters of the time series analysis model based on deep learning according to the evaluation criteria; The prediction module is used to obtain internal and external data of the power system according to time, user group, and region, and output prediction results using a time series analysis model based on deep learning.

[0026] In a third aspect, the present invention further discloses a multi-dimensional power prediction device, comprising: A memory, used for storing a multi-dimensional power prediction program; A processor is used to implement the steps of the multi-dimensional electricity prediction method as described in any one of the above items when executing the multi-dimensional electricity prediction program.

[0027] In a fourth aspect, the present invention further discloses a readable storage medium, on which a multi-dimensional electricity prediction program is stored. When the multi-dimensional electricity prediction program is executed by a processor, the steps of the multi-dimensional electricity prediction method as described in any one of the above items are implemented.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention discloses a multi-dimensional electricity forecasting method, system, device and readable storage medium, which realizes effective dimensionality reduction of data by comprehensively collecting and optimizing the quality of internal and external data and using principal component analysis to accurately screen key features. The deep learning time series analysis model constructed by combining long short-term memory network and attention mechanism can accurately capture the complex dynamics and key information in time series. The accuracy and reliability of the prediction results are ensured through meticulous classification training and scientific model evaluation. This method not only flexibly adapts to the needs of different user groups and regions, but also has strong scalability, providing strong technical support for intelligent management and decision-making in the power industry.

[0029] It can be seen that compared with the prior art, the present invention has outstanding substantive features and significant progress, and the beneficial effects of its implementation are also obvious. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0031] Figure 1 It is a method flow chart of a specific implementation mode of the present invention.

[0032] Figure 2 It is a LSTM network structure diagram of a specific implementation mode of the present invention.

[0033] Figure 3 This is a network structure diagram of the attention mechanism of a specific implementation method of the present invention.

[0034] Figure 4 It is a system structure diagram of a specific implementation mode of the present invention. DETAILED DESCRIPTION

[0035] In order to enable those skilled in the art to better understand the scheme of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0036] See also Figure 1 As shown, this embodiment provides a multi-dimensional power prediction method, including the following steps: S1: Collect internal and external data of the power system as sample data and evaluate the quality of the sample data. The internal data includes electricity data, business expansion registration data, time-of-use electricity price information data, and electric vehicle charging pile data.

[0037] In a specific implementation method, first, internal data is collected, including power data, business expansion registration data, time-of-use electricity price information data, electric vehicle charging pile data, etc.

[0038] At the same time, external data is collected, including meteorological data, economic information, policy data, emergencies, holiday arrangements, etc. Meteorological data consists of temperature, humidity, rainfall, wind speed, wind direction, air pressure, etc., and is usually provided by the meteorological department; economic information includes regional GDP, added value of the primary, secondary and tertiary industries, total population, per capita consumer expenditure of residents, etc., provided by relevant statistical agencies; policy data usually includes various policy documents, announcements, notices, etc., such as energy policies and electricity price policies. Emergency data usually includes relevant information on emergencies such as natural disasters, accidents and disasters, public health events, etc., which can be obtained through emergency management departments, news media and other channels; holiday arrangement data are usually released by relevant agencies, including statutory holidays, adjustment arrangements, etc.

[0039] In this step, data quality evaluation includes evaluating the data integrity of data samples, the rationality of daily electricity fluctuations, the rationality of the electricity proportion of each user group, the deviation of total electricity and industry electricity, the rationality of daily electricity and issued electricity, the rationality of high-voltage user fluctuations, and the quality of meteorological data.

[0040] S2: Perform data correction on the sample data in combination with the quality evaluation results to eliminate noise, dimensional differences and abnormal data.

[0041] In a specific implementation, data correction includes preprocessing operations such as cleaning and normalization of the data to eliminate noise and dimensional differences. For missing and abnormal data, the mean of the time period data is generally used to replace it.

[0042] For meteorological data, secondary corrections of various meteorological elements (irradiance, wind speed, temperature, etc.) are carried out by combining observation data from micro observation stations and centralized stations. According to numerical weather forecasting research, meteorological elements within a range of 10km*10km are basically the same. According to the location of micro observation stations and centralized stations, the area is divided into several grid groups (grid size is 10km*10km), and the grid groups are divided as units. The historical meteorological data of centralized stations in the grid groups are used to correct the meteorology to obtain more accurate meteorological forecasts. For grid points with missing meteorological data, appropriate interpolation methods are selected, such as linear interpolation, bilinear interpolation, bicubic interpolation, etc., and the data of neighboring grid points are used to estimate the missing data.

[0043] The MinMaxScaler method is used to normalize the sample data and map the data to the range of [0,1]. The normalization is performed using the following formula:

[0044] in, and Respectively represent the sample data before and after normalization, and Respectively represent the maximum and minimum values ​​of a sequence before normalization.

[0045] S3: Use the principal component analysis method to extract features from sample data and screen out feature samples that have an impact on power forecasting to reduce redundancy and improve forecasting efficiency.

[0046] In a specific implementation manner, this step specifically includes: S301: Standardize the sample data so that the mean of each variable is 0 and the variance is 1.

[0047] S302: Calculate the covariance matrix of the standardized data matrix.

[0048] S303: Calculate the eigenvalues ​​and eigenvectors of the covariance matrix.

[0049] S304: Select the first k principal components according to the size of the eigenvalue (or the corresponding variance contribution rate). These principal components can retain the information of the original data to the greatest extent.

[0050] S305: Select the data of the principal component as the input feature data set of the LSTM model.

[0051] S4: Fusion feature samples as input data for model training, and classification of the input data for model training according to different user groups or different regions.

[0052] In a specific implementation, first, the characteristic data of multi-level influencing factors of the same user group or the same region, such as weather, region, time dimension, etc., are integrated to build a multi-factor electricity forecasting model. Then, the multi-level factor data of different user groups or different regions are divided as input data for training the forecasting model.

[0053] S5: A long short-term memory network model is used to build a timing analysis model based on deep learning. The input data of the model training is used as input and the target power is used as output. The attention mechanism technology is introduced to train the timing analysis model based on deep learning.

[0054] In a specific implementation, a time series analysis model based on deep learning is constructed, and a long short-term memory network (LSTM) model is adopted. The extracted features are used as input and the target power is used as output to train the model. The attention mechanism (Attention) technology is introduced to enhance the model's attention to important features and improve the prediction accuracy.

[0055] As an example, this step specifically includes the following sub-steps: S501: Divide feature data sets of different user groups or different regions into training sets and test sets respectively; S502: Build an LSTM model, introduce an attention mechanism, and use an attention layer (such as an Attention layer) to give greater weights to important features in the LSTM output results to improve prediction accuracy. The specific construction process of the neural network model is as follows: 1. Construct LSTM layer: Reference Figure 2 Each LSTM unit contains a forget gate, an input gate, and an output gate, and has three input data: the output of the previous LSTM unit at the previous moment , the state of the current LSTM unit at the previous moment and the current time input The forget gate is used to calculate the information to be forgotten, the input gate is used to calculate the information saved in the state unit, and the output gate calculates the information to be output through the sigmoid function, and then multiplies it by the value of the current unit state through the tanh function to get the output. The mathematical expression of the LSTM unit algorithm is:

[0056]

[0057]

[0058]

[0059]

[0060]

[0061] in, For the Gate of Oblivion; is the input gate; is the output gate, , They are the state and output of the current LSTM unit respectively; is the input node; is the sigmod function; is the input vector of the current input network; , , , is the weight; , , , For bias.

[0062] 2. Construct the attention layer: Reference Figure 3 , first calculate the attention weight of each hidden layer state , and then multiply the different outputs by their weights and sum them up to get the final output of the attention layer. The specific formula is:

[0063]

[0064]

[0065] in, is the output of the attention layer; for The attention weight of the moment; To calculate the weight; For LSTM Get the hidden layer state value at all times; is bias; For the total period.

[0066] 3. Build LSTM-Attention network model: Design a 2-layer LSTM network layer with 256 and 32 neurons respectively to improve the accuracy of the prediction model; add an attention mechanism layer to give greater weight to the important parts of the LSTM prediction output results; introduce a dropout layer to prevent the model from overfitting, and randomly select neurons not to participate in training. Finally, output a vector in a fixed format through the fully connected layer. Set a suitable loss function (such as mean square error MSE) and optimizer (such as Adam optimizer), use the fit method to train the model, and set appropriate parameters such as batch size, number of training rounds, and verification data.

[0067] S503: Use the training sets of different user groups or different regions to train separately to obtain multiple models, use the test set data to test, evaluate the prediction results of the LSTM-Attention network model, and generate a time series analysis model based on deep learning.

[0068] S6: The average relative percentage error and root mean square error are used as evaluation criteria, and the optimal parameters of the deep learning-based time series analysis model are determined according to the evaluation criteria.

[0069] In a specific embodiment, the average relative percentage error , Root Mean Square Error The two indicators are used as evaluation criteria to determine the optimal parameters of the deep learning-based time series analysis model.

[0070] The average relative percentage error , Root Mean Square Error The calculation method is as follows:

[0071]

[0072] in, is the actual value, is the predicted value, is the sample size.

[0073] In addition, methods such as Grid Search or Random Search are used to train the model under different parameter combinations to determine the optimal parameters of the deep learning-based time series analysis model.

[0074] S7: Obtain internal and external data of the power system according to time, user group, and region, and use the time series analysis model based on deep learning to output prediction results.

[0075] In the specific implementation method, in order to adapt to different application scenarios, multi-dimensional fine-grained analysis and prediction are carried out. In terms of time, the time series analysis model based on deep learning is used to carry out power forecasting according to multiple time dimensions such as day, week, month, season, and year to meet the needs of spot market transactions; in terms of user groups, the time series analysis model based on deep learning is used to carry out power forecasting according to key user groups; in terms of region, the time series analysis model based on deep learning is used to achieve regional forecasting. Finally, the power forecast results of each user group and region can be combined to generate a total power forecast result.

[0076] The present invention provides a multi-dimensional electricity forecasting method, which performs multi-dimensional electricity forecasting based on the LSTM-Attention network, solves the problems of insufficient understanding of the multi-level nature of data, data quality and accuracy, insufficient feature extraction, etc. in traditional electricity forecasting methods, and improves the accuracy of electricity forecasting and fine-grained analysis capabilities.

[0077] See also Figure 4 As shown, the present invention also discloses a multi-dimensional electricity prediction system, including: a data collection module 1, a data correction module 2, a feature extraction module 3, a feature fusion and classification module 4, a model construction and training module 5, a model evaluation and optimization module 6 and a prediction module 7. The data collection module 1 is used to collect the internal and external data of the power system as sample data and to evaluate the quality of the sample data. The internal data includes power data, business expansion registration data, time-of-use electricity price information data, and electric vehicle charging pile data.

[0078] The data correction module 2 is used to perform data correction processing on the sample data in combination with the quality evaluation results to eliminate noise, dimensional differences and abnormal data.

[0079] The feature extraction module 3 is used to extract features from the sample data using the principal component analysis method, and screen out feature samples that have an impact on the power prediction, so as to reduce redundancy and improve prediction efficiency.

[0080] The feature fusion and classification module 4 is used to fuse feature samples as input data for model training, and classify the input data for model training according to different user groups or different regions.

[0081] The model building training module 5 is used to build a timing analysis model based on deep learning by adopting a long short-term memory network model. The input data of the model training is used as input and the target power is used as output. The attention mechanism technology is introduced to train the timing analysis model based on deep learning.

[0082] The model evaluation and optimization module 6 is used to use the average relative percentage error and the root mean square error as evaluation criteria, and determine the optimal parameters of the deep learning-based time series analysis model according to the evaluation criteria.

[0083] The prediction module 7 is used to obtain internal and external data of the power system according to time, user group, and region, and output prediction results using a time series analysis model based on deep learning.

[0084] The specific implementation of the multi-dimensional electricity prediction system of this embodiment is basically the same as the specific implementation of the multi-dimensional electricity prediction method described above, and will not be repeated here.

[0085] The present invention also discloses a multi-dimensional electricity prediction device, comprising a processor and a memory; wherein the processor implements the steps of the multi-dimensional electricity prediction method as described in any one of the above items when executing the multi-dimensional electricity prediction program stored in the memory.

[0086] Furthermore, the multi-dimensional power prediction device in this embodiment may also include: The input interface is used to obtain the multi-dimensional power prediction program imported from the outside, and save the obtained multi-dimensional power prediction program to the memory, and can also be used to obtain various instructions and parameters transmitted by the external terminal device, and transmit them to the processor, so that the processor can use the above various instructions and parameters to carry out corresponding processing. In this embodiment, the input interface can specifically include but is not limited to a USB interface, a serial interface, a voice input interface, a fingerprint input interface, a hard disk reading interface, etc.

[0087] The output interface is used to output various data generated by the processor to the terminal device connected thereto, so that other terminal devices connected to the output interface can obtain various data generated by the processor. In this embodiment, the output interface may specifically include but is not limited to a USB interface, a serial interface, etc.

[0088] The communication unit is used to establish a remote communication connection between the multi-dimensional power prediction device and the external server so that the multi-dimensional power prediction device can mount the image file to the external server. In this embodiment, the communication unit may specifically include but is not limited to a remote communication unit based on wireless communication technology or wired communication technology.

[0089] The keyboard is used to obtain various parameter data or instructions input by the user by tapping the keycaps in real time.

[0090] The display is used to display the relevant information of the multi-dimensional power forecasting process in real time.

[0091] The mouse can be used to assist users in inputting data and simplify user operations.

[0092] The present invention also discloses a readable storage medium, wherein the readable storage medium includes a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable hard disk, a CD-ROM, or any other form of storage medium known in the technical field. A multi-dimensional power prediction program is stored in the readable storage medium, and when the multi-dimensional power prediction program is executed by a processor, the steps of the multi-dimensional power prediction method described in any one of the above items are implemented.

[0093] In summary, the present invention achieves high-precision prediction of electricity volume by comprehensively considering multiple factors and data sources, adopting advanced data processing technology and deep learning models, and providing strong support for the optimized operation and energy management of the power system.

[0094] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. As for the method disclosed in the embodiment, since it corresponds to the system disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0095] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0096] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of systems or units, which can be electrical, mechanical or other forms.

[0097] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0098] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, or each module may exist physically separately, or two or more modules may be integrated into one unit.

[0099] Similarly, each processing unit in each embodiment of the present invention may be integrated into one functional module, or each processing unit may exist physically, or two or more processing units may be integrated into one functional module.

[0100] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0101] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0102] The above is a detailed introduction to the multi-dimensional electricity prediction method, system, device and readable storage medium provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, several improvements and modifications can be made to the present invention, and these improvements and modifications also fall within the scope of protection of the present invention.

Claims

1. A multi-dimensional electricity prediction method, characterized in that: include: Collect internal and external data of the power system as sample data and evaluate the quality of the sample data. The internal data includes power data, business expansion registration data, time-of-use electricity price information data, and electric vehicle charging pile data; Combined with the quality evaluation results, the sample data is corrected to eliminate noise, dimensional differences and abnormal data; The principal component analysis method is used to extract features from sample data and screen out feature samples that have an impact on power forecasting to reduce redundancy and improve forecasting efficiency. Fusion feature samples as input data for model training, and classification of the input data for model training according to different user groups or different regions; A time series analysis model based on deep learning is constructed using a long short-term memory network model. The input data of the model training is used as input, and the target power is used as output. The attention mechanism technology is introduced to train the time series analysis model based on deep learning. The average relative percentage error and root mean square error are used as evaluation criteria, and the optimal parameters of the time series analysis model based on deep learning are determined according to the evaluation criteria; The internal and external data of the power system are obtained according to time, user group and region, and the prediction results are output using a time series analysis model based on deep learning.

2. The multi-dimensional electricity prediction method according to claim 1, characterized in that: The external data includes: Meteorological data, economic information, policy data, emergency data, holiday schedule data; The meteorological data include temperature, humidity, rainfall, wind speed, wind direction, and air pressure; The economic information includes regional GDP, added value of the primary, secondary and tertiary industries, total population and per capita consumer expenditure of residents; The policy data includes policy documents, announcements, and notices; The emergency data includes natural disasters, accidents, disasters, and public health event information; The quality evaluation of sample data includes: Evaluate the data integrity of data samples, the rationality of daily electricity fluctuations, the rationality of the proportion of electricity in each user group, the deviation of total electricity and electricity by industry, the rationality of daily electricity and issued electricity, the rationality of high-voltage user fluctuations, and the quality of meteorological data.

3. The multi-dimensional electricity prediction method according to claim 2, characterized in that: The sample data is corrected in combination with the quality evaluation results to eliminate noise, dimensional differences and abnormal data, including: For meteorological data, observational data are used for data correction, and interpolation methods are used to supplement missing values; The MinMaxScaler method is used to normalize the sample data and map the data to the range of [0,1]. The normalization is performed using the following formula: in, and Respectively represent the sample data before and after normalization, and Respectively represent the maximum and minimum values ​​of a sequence before normalization.

4. The multi-dimensional electricity prediction method according to claim 3, characterized in that: The principal component analysis method is used to extract features from sample data and screen out feature samples that have an impact on power prediction, including: Standardize the sample data to generate a data matrix so that the mean of each variable is 0 and the variance is 1; Compute the covariance matrix of the data matrix; Compute the eigenvalues ​​and eigenvectors of the covariance matrix; Select the first k principal components according to the size of the eigenvalue; The data of k principal components are used as feature samples that have an impact on electricity prediction.

5. The multi-dimensional electricity prediction method according to claim 4, characterized in that: The fused feature samples are used as input data for model training, and the input data for model training is classified according to different user groups or different regions, including: Based on feature samples, feature samples of multi-level influencing factors of the same user group or the same region are merged to construct input data for model training; Classify the input data for model training according to different user groups or different regions.

6. The multi-dimensional electricity prediction method according to claim 5, characterized in that: The long short-term memory network model is used to construct a time series analysis model based on deep learning, including: Construct an LSTM layer. There are multiple LSTM units in the LSTM layer. Each LSTM unit contains a forget gate, an input gate, and an output gate. The mathematical expression of the LSTM unit algorithm is: in, For the Gate of Oblivion; is the input gate; is the output gate, , They are the state and output of the current LSTM unit respectively; is the input node; is the sigmod function; is the input vector of the current input network; , , , is the weight; , , , is bias; Construct the attention layer and first calculate the attention weight of each hidden layer state , and then multiply the different outputs by their weights and sum them up to get the final output of the attention layer. The specific formula is: in, is the output of the attention layer; for The attention weight of the moment; To calculate the weight; For LSTM Get the hidden layer state value at all times; is bias; is the total period; Build an LSTM-Attention network model and design a two-layer LSTM network layer with 256 and 32 neuron data respectively; add an attention mechanism layer to increase the weight of the preset part in the prediction output results of the LSTM-Attention network model; introduce a dropout layer to prevent the model from overfitting, and output a fixed-format vector through a fully connected layer; Set the loss function and optimizer for the LSTM-Attention network model, use the fit method to train the model, and set the batch size, number of training rounds, and validation data to generate a time series analysis model based on deep learning.

7. The multi-dimensional electricity prediction method according to claim 6, characterized in that: The mean relative percentage error and RMS error The calculation method is as follows: in, is the actual value, is the predicted value, is the sample size.

8. A multi-dimensional electricity forecasting system, characterized in that: The system adopts the multi-dimensional power prediction method according to any one of claims 1 to 7; The system comprises: The data collection module is used to collect the internal and external data of the power system as sample data and conduct quality evaluation on the sample data. The internal data includes power data, business expansion registration data, time-of-use electricity price information data, and electric vehicle charging pile data. The data correction module is used to perform data correction processing on the sample data in combination with the quality evaluation results to eliminate noise, dimensional differences and abnormal data; The feature extraction module is used to extract features from sample data using the principal component analysis method, and screen out feature samples that have an impact on power forecasting, so as to reduce redundancy and improve forecasting efficiency; The feature fusion and classification module is used to fuse feature samples as input data for model training and classify the input data for model training according to different user groups or different regions; The model building and training module is used to build a time series analysis model based on deep learning using a long short-term memory network model. The input data of the model training is used as input and the target power is used as output. The attention mechanism technology is introduced to train the time series analysis model based on deep learning. The model evaluation and optimization module is used to use the average relative percentage error and the root mean square error as evaluation criteria, and determine the optimal parameters of the time series analysis model based on deep learning according to the evaluation criteria; The prediction module is used to obtain internal and external data of the power system according to time, user group, and region, and output prediction results using a time series analysis model based on deep learning.

9. A multi-dimensional electricity prediction device, characterized in that: include: A memory, used for storing a multi-dimensional power prediction program; A processor, configured to implement the steps of the multi-dimensional electricity prediction method according to any one of claims 1 to 7 when executing the multi-dimensional electricity prediction program.

10. A readable storage medium, characterized in that: The readable storage medium stores a multi-dimensional electricity prediction program, and when the multi-dimensional electricity prediction program is executed by the processor, the steps of the multi-dimensional electricity prediction method according to any one of claims 1 to 7 are implemented.

Images