A time series prediction method, device and medium for energy field

Abstract

The present application relates to the field of energy forecasting technology, and to a time series forecasting method, device, and medium for the energy field. The method comprises: collecting power generation time series data and trend images, converting the trend images into time series data; constructing and training a Lasso regression model, saving the difference time series data between the time series and the prediction results of the Lasso regression model; preprocessing the time series data and the difference time series data; inputting the preprocessed data into a large time series model for model training; and using the trained large time series model to forecast power generation time series. The present invention makes a preliminary forecast through a regression model, and uses a time series model to perform difference correction on the forecast results, which effectively improves the accuracy of the forecast results and the interpretability of the model.

Description

Technical Field

[0001] The present application relates to the technical field of energy forecasting, and to a time series forecasting method, device and medium for use in the energy field. Background Art

[0002] At present, with the popularization of renewable energy, especially the widespread application of solar power generation systems, accurate prediction of electricity consumption and solar power generation has become crucial for power grid management and scheduling. Traditional forecasting methods, such as ARIMA based on statistical models, can handle time series data to a certain extent, but their forecasting effects are often unsatisfactory when faced with nonlinear, high-dimensional data and complex spatiotemporal dependencies. These methods are difficult to capture complex patterns and dynamic changes in the data, resulting in large errors in the forecast results.

[0003] On the other hand, deep learning methods, such as LSTM and other recurrent neural networks, although they have powerful nonlinear modeling capabilities, face problems such as poor generalization and long training time in practical applications. Whenever encountering new data sets or changes in data patterns, deep learning models usually need to be retrained, which not only consumes a lot of time and computing resources, but also limits its application in real-time prediction scenarios.

[0004] In addition, traditional forecasting methods often only accept time series data as input, and data presented in the form of trend graphs or line graphs need to be converted manually, which undoubtedly increases the complexity and workload of data processing. At the same time, many data in the energy field contain not only time series information, but also covariate information that is closely related to the forecast results (such as light, temperature, pressure, etc.). How to effectively use this information to improve the accuracy of the forecast is also a problem that traditional methods are difficult to solve.

[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present application, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0006] In order to provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. The summary is not an extensive review, nor is it intended to identify key / critical elements or delineate the scope of protection of these embodiments, but rather serves as a prelude to the detailed description that follows.

[0007] The embodiments of the present disclosure provide a time series prediction method, device, and medium for the energy field, which increase the accuracy of the prediction and the interpretability of the model.

[0008] In some embodiments, the method comprises:

[0009] Collect power generation time series data and trend images, and convert trend images into time series data;

[0010] Construct and train the Lasso regression model, and save the difference time series data between the time series and the prediction results of the Lasso regression model;

[0011] Preprocess time series data and difference time series data, including data cleaning and standardization;

[0012] The preprocessed data is input into the time series big model for model training, and the time series big model is an Encoder-Decoder architecture;

[0013] The trained time series model is used to predict the power generation time series, including time series without covariates and time series with covariates.

[0014] Preferably, the time series prediction process is as follows:

[0015] For time series without covariates, input the historical time series to the time series forecasting model and output the predicted time series;

[0016] For time series data with covariates, input the historical time series and the corresponding covariates into the Lasso regression model for training to obtain the prediction results for historical data and the prediction results for future data; subtract the historical time series from the prediction results of the Lasso regression model for historical data, and send the difference to the time series prediction model to obtain the predicted difference; add the prediction results of the Lasso regression model for future data and the predicted difference to output the final prediction result.

[0017] Preferably, the specific method of converting the trend image into time series data is as follows:

[0018] The trend graph is grayscaled and binarized to generate a binary image; the Hough transform of OpenCV is used to detect straight lines in the binary image and identify the x-axis and y-axis; OCR is used to identify the scales and marks of the coordinate axes in the binary image; Cancy edge detection and flux analysis are used to track the lines of the line graph; the intersection points of the lines corresponding to the x-axis and y-axis directions are identified to calculate the coordinate value of each point; the extracted image is converted into a time series.

[0019] Preferably, the time series large model encoder part includes a multi-layer Transformer encoder block, each block including a multi-head attention mechanism and a feed-forward neural network;

[0020] The decoder part consists of multiple layers of Transformer decoder blocks, each of which includes multi-head attention, contextual attention, and a feed-forward neural network.

[0021] Preferably, the encoder further includes a position encoding module, which implements position encoding through a sine function, and the formula is as follows:

[0022] ,

[0023] in, represents the embedding dimension, Represents a dimension index.

[0024] Preferably, the feedforward neural network uses two linear transformations and a ReLU activation function.

[0025] Preferably, in the training process of the time series large model, the teacher forcing method is used for training, and the mean square error MAE is used as the loss function. At the same time, reinforcement learning is introduced, and the model parameters are updated through the policy gradient method. The formula is as follows:

[0026] ,

[0027] in, represents the gradient at the next moment, represents the learning rate, represents the policy function, Indicates action, represents the reward function.

[0028] Preferably, the covariates include light, temperature and pressure. For time series data containing covariates, a dislocated correspondence method is adopted to input historical covariates as covariates for future predictions.

[0029] In some embodiments, the device includes: a processor and a memory storing program instructions, and the processor is configured to execute the time series prediction method for the energy field when running the program instructions.

[0030] In some embodiments, the medium stores program instructions, and when the program instructions are run, they execute the time series prediction method for the energy field.

[0031] The time series prediction method, device, and medium for the energy field provided by the embodiments of the present disclosure can achieve the following technical effects:

[0032] This method can process time series data and trend images simultaneously, and automatically convert trend images into time series data through image processing and OCR technology, which significantly improves the efficiency and automation of data processing.

[0033] By building a large time series model and combining it with a large amount of energy-related data for training, the model can discover the underlying patterns in the data, thereby improving the accuracy of predictions. In particular, this method shows better prediction performance for nonlinear, high-dimensional data and complex spatiotemporal dependencies.

[0034] After training, the large time series model of this method can be applied to multiple time series prediction tasks at one time without the need to retrain the model for each new task, which greatly improves the generalization ability and application scope of the model.

[0035] This method innovatively combines traditional regression methods (such as LASSO regression) with time series forecasting. It makes preliminary forecasts through the regression model and uses the time series model to make difference corrections to the forecast results, effectively improving the accuracy of the forecast results and the interpretability of the model.

[0036] For time series data containing covariates, this method uses past covariates for oblique prediction, avoiding the problem of unknown future covariates and improving the effective application of covariates.

[0037] The above general description and the following description are exemplary and explanatory only and are not intended to limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] One or more embodiments are exemplarily described by corresponding drawings, which do not limit the embodiments. Elements with the same reference numerals in the drawings are shown as similar elements, and the drawings do not constitute a scale limitation, and wherein:

[0039] Figure 1 is a schematic diagram of the method flow chart;

[0040] Figure 2 It is a schematic diagram of the method of converting trend pictures into time series data;

[0041] Figure 3 is a schematic diagram of a prediction process provided by an embodiment of the present disclosure;

[0042] Figure 4 It is a schematic diagram of a device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0043] In order to be able to understand the features and technical contents of the embodiments of the present disclosure in more detail, the implementation of the embodiments of the present disclosure is described in detail below in conjunction with the accompanying drawings. The attached drawings are for reference only and are not used to limit the embodiments of the present disclosure. In the following technical description, for the convenience of explanation, a full understanding of the disclosed embodiments is provided through multiple details. However, one or more embodiments can still be implemented without these details. In other cases, to simplify the drawings, well-known structures and devices can be simplified for display.

[0044] The terms "first", "second", etc. in the specification and claims of the embodiments of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchanged where appropriate, so as to describe the embodiments of the embodiments of the present disclosure described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions.

[0045] Unless otherwise stated, the term "plurality" means two or more.

[0046] Example 1

[0047] Combination Figure 1 , a time series forecasting method for energy field, comprising:

[0048] S1: Collect power generation time series data and trend images, and convert trend images into time series data.

[0049] S2: Construct and train the Lasso regression model, and save the difference time series data between the time series and the prediction results of the Lasso regression model.

[0050] S3: Preprocess the time series data and difference time series data, including data cleaning and standardization.

[0051] S4: Input the preprocessed data into the time series large model for model training, where the time series large model is an Encoder-Decoder architecture.

[0052] S5: Use the trained time series model to predict the power generation time series, including time series without covariates and time series with covariates.

[0053] As a refinement of the above embodiment, Figure 2 The specific method of converting the trend image into time series data in step S1 is as follows:

[0054] The trend graph is grayscaled and binarized to generate a binary image; the Hough transform of OpenCV is used to detect straight lines in the binary image and identify the x-axis and y-axis; OCR is used to identify the scales and marks of the coordinate axes in the binary image; Cancy edge detection and flux analysis are used to track the lines of the line graph; the intersection points of the lines corresponding to the x-axis and y-axis directions are identified to calculate the coordinate value of each point; the extracted image is converted into a time series.

[0055] As a refinement of the above embodiment, in step S2, the model not only learns the time series, but also needs to learn the prediction of the difference with the linear regression. Therefore, it is necessary to construct a Lasso regression model, and train each time series data set, and then subtract the actual value from the result predicted by the training model to generate difference data, which is also time series data, and the difference data is saved as part of the training data.

[0056] As a refinement of the above embodiment, the time series data preprocessing in step S3 includes:

[0057] First, clean the data. Due to inaccurate recognition or problems with the original data, there may be null values ​​or abnormal values ​​in the data. Delete the null values ​​and abnormal data.

[0058] Then perform standardization to reduce the impact of noise; the standardization formula is as follows:

[0059]

[0060] in, is the standardized data, is the mean, is the variance, express The original value of the moment.

[0061] As a refinement of the above embodiment, the structure of the time series model in step S4 is an Encoder-Decoder architecture, in which the encoder is responsible for capturing information in the input features, and the decoder is responsible for generating the prediction sequence. The Transformer architecture is used as the basis, and component position encoding and cyclic residual connections specific to time series prediction are added.

[0062] The encoder part of the time series large model includes multi-layer Transformer encoder blocks, each of which contains a multi-head attention mechanism and a feed-forward neural network.

[0063] The encoder also includes a position encoding module, which implements position encoding through a sine function , for time series data, position encoding is necessary because the original input does not have clear position information. The formula is as follows:

[0064] ,

[0065] in, represents the embedding dimension, Represents a dimension index.

[0066] The multi-head attention mechanism is as follows:

[0067] Multi-head attention allows the model to focus on different parts of the input sequence from different position perspectives. , define the query ,key Sum for:

[0068]

[0069] Attention score Calculated as:

[0070]

[0071] in, Represents the input vector of the attention mechanism Represents the Query matrix Represents the Key matrix represents the Value matrix, Represents the Query after matrix transformation Represents the transpose of the Key after matrix transformation, is the dimension of the key vector, It is the attention distribution obtained by the softmax function.

[0072] The feedforward neural network uses two linear transformations and a ReLU activation function, and the formula is as follows:

[0073]

[0074] in, represents the feedforward neural network function, Represents the input information of the network, represents the weight parameter of linear layer 1, represents the bias parameter of linear layer 1, Represents the weight parameters of linear layer 2 represents the bias parameters of linear layer 2.

[0075] The decoder part includes multiple layers of Transformer decoder blocks, each of which includes multi-head attention, contextual attention, and a feed-forward neural network. In the first layer of the decoder, masked multi-head attention is used to ensure that the model can only see past sequence values. The contextual attention layer uses the output of the encoder as keys and values ​​to focus on the information of the entire input sequence. A recurrent residual connection is introduced in the decoder, that is, after each decoder layer, the output of the current moment is combined with the output of the previous moment. This helps to maintain the consistency of temporal information and helps alleviate the gradient vanishing problem.

[0076] Optionally, in the training process of the time series large model, the teacher forcing method is used for training, that is, the real sequence value is used as the input of the next time step during the training process. The mean square error MAE is used as the loss function, and reinforcement learning is introduced at the same time. The model parameters are updated through the policy gradient method. The formula is as follows:

[0077] ,

[0078] in, represents the gradient at the next moment, represents the learning rate, represents the policy function, Indicates action, represents the reward function.

[0079] As a refinement of the above embodiment, the covariates include light, temperature and pressure. For time series data containing covariates, a staggered correspondence method is adopted to input historical covariates as covariates for future predictions.

[0080] As a refinement of the above embodiment, Figure 3 The time series prediction process in step S5 is specifically as follows:

[0081] For time series without covariates, input the historical time series to the time series forecasting model and output the predicted time series;

[0082] For time series data with covariates, input the historical time series and the corresponding covariates into the Lasso regression model for training to obtain the prediction results for historical data and the prediction results for future data; subtract the historical time series from the prediction results of the Lasso regression model for historical data, and send the difference to the time series prediction model to obtain the predicted difference; add the prediction results of the Lasso regression model for future data and the predicted difference to output the final prediction result.

[0083] It should be noted that covariates refer to variables used to predict results. For example, light, temperature, pressure, etc. in solar power generation can all be considered covariates. The reasonable application of covariates can effectively improve the prediction effect. However, since future covariates are unknown, past covariates can be used as inputs. For example, to predict power generation in the next ten days, the data of the previous ten days can be regarded as covariates for the next ten days. This method improves the application of covariates and avoids the problem of unknown future covariates.

[0084] Example 2

[0085] The time series prediction process is described below in conjunction with a specific embodiment:

[0086] If you want to predict the solar power generation in the next 32 days, you currently have data on solar power generation, sunlight, temperature, etc. for the previous year. Select 128 time points to construct as input to the entire prediction system.

[0087] The input data is , which contains 128 data, representing the power generation data of the past 128 days.

[0088] Input data 2 is , which contains 160 data, namely the number of historical data 128 plus the number of data to be predicted 32, which represents temperature and is input as a covariate. The misaligned correspondence is adopted to avoid the problem of unobtainable future covariates.

[0089] Input data three is , which contains 160 data, i.e. the number of historical data plus the number of data to be predicted, which represents the illumination and is input as a covariate. The misaligned correspondence is adopted to avoid the problem of unobtainable future covariates.

[0090] The data is first passed through the LASSO linear regression model to train the LASSO regression model. The training time of this model is only tens of seconds. The independent variable 1 of the LASSO model training is ,represent The first 128 data of the variable 2 are ,represent The first 128 data of After training, a fitting model is generated ,Will and Input to The output obtained is , represents the power generation predicted by the regression model in the past 128 days. With the true value The difference between , then send d to the time series model, and set the number of output time points to 32. The model will output , is a sequence of 32 numbers, representing the forecast of the difference between the actual data and the regression model for the next 32 days. and Input to fit model The input is 32 data and the output is , represents the power generation in the next 32 days predicted by the regression model through covariates. Since there is a difference between the regression model and the actual value, in order to correct the deviation, the deviation predicted by the time series model is and Add them together and then output the result, is the final prediction result.

[0091] Example 3

[0092] Combination Figure 4 As shown, the embodiment of the present disclosure provides a time series prediction device 300 for the energy field, including a processor 304 and a memory 301. Optionally, the device may also include a communication interface 302 and a bus 303. The processor 304, the communication interface 302, and the memory 301 may communicate with each other through the bus 303. The communication interface 302 may be used for information transmission. The processor 304 may call the logic instructions in the memory 301 to execute the time series prediction method for the energy field of the above embodiment.

[0093] In addition, the logic instructions in the memory 301 described above can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product.

[0094] The memory 301 is a computer-readable storage medium that can be used to store software programs and computer executable programs, such as program instructions / modules corresponding to the method in the embodiment of the present disclosure. The processor 304 executes the function application and data processing by running the program instructions / modules stored in the memory 301, that is, the time series prediction method for the energy field in the above embodiment is implemented.

[0095] The memory 301 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required for at least one function; the data storage area may store data created according to the use of the terminal device, etc. In addition, the memory 301 may include a high-speed random access memory and may also include a non-volatile memory.

[0096] The embodiment of the present disclosure also provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured to execute the above-mentioned time series prediction method for the energy field.

[0097] The computer-readable storage medium mentioned above may be a transient computer-readable storage medium or a non-transitory computer-readable storage medium.

[0098] The technical solution of the embodiment of the present disclosure can be embodied in the form of a software product, which is stored in a storage medium and includes one or more instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in the embodiment of the present disclosure. The aforementioned storage medium may be a non-transient storage medium, including: a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk, and other media that can store program codes, or a transient storage medium.

[0099] The above description and the accompanying drawings fully illustrate the embodiments of the present disclosure so that those skilled in the art can practice them. Other embodiments may include structural, logical, electrical, process and other changes. The embodiments represent only possible changes. Unless explicitly required, separate components and functions are optional, and the order of operations may vary. Parts and features of some embodiments may be included in or replace parts and features of other embodiments. Moreover, the words used in this application are only used to describe the embodiments and are not used to limit the claims. As used in the description of the embodiments and the claims, unless the context clearly indicates, the singular forms of "a", "an" and "the" are intended to include plural forms as well. Similarly, the term "and / or" as used in this application refers to any and all possible combinations of one or more associated listings. In addition, when used in this application, the term "comprise" and its variants "comprises" and / or comprising refer to the presence of stated features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or groups thereof. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the presence of other identical elements in the process, method or device including the elements. In this article, each embodiment may focus on the differences from other embodiments, and the same and similar parts between the embodiments may refer to each other. For the methods, products, etc. disclosed in the embodiments, if they correspond to the method part disclosed in the embodiments, then the relevant parts can refer to the description of the method part.

[0100] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software may depend on the specific application and design constraints of the technical solution. The 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 embodiments of the present disclosure. The technicians may clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here.

[0101] In the embodiments disclosed herein, the disclosed methods and products (including but not limited to devices, equipment, etc.) can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units can be 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. In addition, the coupling or direct coupling or communication connection between each other shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to implement this embodiment. In addition, each functional unit in the embodiment of the present disclosure may be integrated in a processing unit, or each unit may exist physically separately, or two or more units may be integrated in one unit.

[0102] The flowchart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to the embodiment of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. In some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, which can depend on the functions involved. In the description corresponding to the flowchart and the block diagram in the accompanying drawings, the operations or steps corresponding to different boxes can also occur in a different order from the order disclosed in the description, and sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, which can depend on the functions involved. Each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented by a dedicated hardware-based system that performs the specified functions or actions, or may be implemented by a combination of dedicated hardware and computer instructions.

Claims

1. A time series prediction method for the energy field, characterized in that: The following steps are involved: Collect power generation time series data and trend images, and convert trend images into time series data; Construct and train the Lasso regression model, and save the difference time series data between the time series and the prediction results of the Lasso regression model; Preprocess time series data and difference time series data, including data cleaning and standardization; The preprocessed data is input into the time series big model for model training, and the time series big model is an Encoder-Decoder architecture; Use the trained time series model to predict the power generation time series, including time series without covariates and time series with covariates; The specific method of converting trend images into time series data is as follows: Grayscale and binarize the trend graph to generate a binary image; use OpenCV's Hough transform to detect straight lines in the binary image and identify the x-axis and y-axis; use OCR to identify the scale and marks of the coordinate axes in the binary image; track the lines of the line graph through Cancy edge detection and connectivity analysis; Identify the intersection points of the lines corresponding to the x-axis and y-axis directions, and calculate the coordinate value of each point; Convert the extracted images into time series.

2. The time series prediction method for energy field according to claim 1, characterized in that: The time series forecasting process is as follows: For time series without covariates, input the historical time series to the time series forecasting model and output the predicted time series; For time series data with covariates, input the historical time series and the corresponding covariates into the Lasso regression model for training to obtain the prediction results of the historical data and the prediction results of the future data; subtract the historical time series from the prediction results of the Lasso regression model for the historical data, and send the difference into the time series prediction model to obtain the predicted difference; Add the predicted result of the Lasso regression model for future data to the predicted difference and output the final predicted result.

3. The time series prediction method for energy field according to claim 1, characterized in that: The encoder part of the time series large model includes multi-layer Transformer encoder blocks, each of which contains a multi-head attention mechanism and a feed-forward neural network; The decoder part consists of multiple layers of Transformer decoder blocks, each of which includes multi-head attention, contextual attention, and a feed-forward neural network.

4. The time series prediction method for energy field according to claim 3, characterized in that: The encoder also includes a position encoding module, which implements position encoding through a sine function, and the formula is as follows: ， in, represents the embedding dimension, Represents a dimension index.

5. The time series prediction method for energy field according to claim 3, characterized in that: The feedforward neural network uses two linear transformations and a ReLU activation function.

6. The time series prediction method for energy field according to claim 1, characterized in that: In the training process of the time series large model, the teacher forcing method is used for training, and the mean square error MAE is used as the loss function. At the same time, reinforcement learning is introduced, and the model parameters are updated through the policy gradient method. The formula is as follows: ， in, represents the gradient at the next moment, represents the learning rate, represents the policy function, Indicates action, represents the reward function.

7. The time series prediction method for energy field according to claim 1, characterized in that: The covariates include light, temperature and pressure. For time series data containing covariates, a dislocation correspondence method is adopted to input historical covariates as covariates for future predictions.

8. A time series prediction device for the energy field, comprising a processor and a memory storing program instructions, characterized in that: The processor is configured to execute the time series prediction method for the energy field as described in any one of claims 1 to 7 when running the program instructions.

9. A storage medium storing program instructions, characterized in that: When the program instructions are executed, the time series prediction method for the energy field as described in any one of claims 1 to 7 is executed.

Images