Comprehensive energy system multi-energy prediction method based on effective information extraction annunciator model

By adopting the method based on the effective information extraction notification model in the integrated energy system, the problem of insufficient prediction accuracy of multi-energy load sequences is solved, and high-precision and high-efficiency load prediction are achieved, which improves the reliability of prediction.

CN119026736APending Publication Date: 2024-11-26STATE GRID HEBEI ELECTRIC POWER CO LTD +2
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411064104.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-05
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The prior art has insufficient prediction accuracy of multi-energy load sequences in integrated energy systems, making it difficult to achieve efficient, accurate and reliable load prediction.

Method used

Using the method of extracting the notification model based on the effective information, the Efficient Information Extracting Informer network architecture is built by collecting and sorting the timing data of electrical, thermal and gas loads, and the neural network is trained using the Adam solution algorithm, and performance evaluation and hyperparameter debugging are carried out until The model performance meets application requirements.

Benefits of technology

It realizes high-precision and high-efficiency prediction of multi-energy load sequences in an integrated energy system, and improves the reliability of prediction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119026736A_ABST
    Figure CN119026736A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of energy, and discloses a comprehensive energy system multi-energy prediction method based on an effective information extraction annunciator model, and the method comprises the steps: 1, collecting and sorting power, heat and gas load time sequence data in a comprehensive energy system, and enabling the data to be used for subsequent model training and debugging; the method comprises the following steps of: (1) constructing an Ethernet (Ethernet) network architecture, (2) constructing an Ethernet (Ethernet) network architecture, and (3) constructing an Ethernet (Ethernet) network architecture, 3, based on the regular training data set and the constructed neural network architecture, selecting an Adam solving algorithm to train the network, and storing optimal parameters and models; and step 4, performing performance evaluation and hyper-parameter debugging of the neural network until the model performance meets the application requirements. According to the comprehensive energy system multi-energy prediction method based on the effective information extraction annunciator model, comprehensive energy multi-energy load sequence prediction is carried out based on an improved Informer neural network architecture, and the problem that the load sequence prediction precision of a previous method is insufficient is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of energy technology, and in particular to a multi-energy prediction method for an integrated energy system based on an effective information extraction notifier model. Background Art

[0002] The integrated energy system integrates various forms of energy, such as electricity, cooling and heating, and involves the conversion, distribution and coordination of energy. It is an important physical carrier for realizing the multi-energy complementary characteristics of the energy Internet. Based on the spatial scale, the integrated energy system can be divided into inter-regional, regional and user levels. Among them, the user-level IES is smaller in scale, but the load fluctuation is larger. In the complex energy coupling relationship, it is necessary to consider the coordination and complementarity of electricity, cooling and heating. Therefore, an efficient, accurate and reliable multi-energy load forecasting method is urgently needed. Summary of the invention

[0003] 1. Technical issues to be resolved

[0004] In view of the deficiencies in the prior art, the present invention provides a multi-energy prediction method for an integrated energy system based on an effective information extraction notifier model, which solves the problem of insufficient load sequence prediction accuracy of previous methods.

[0005] (II) Technical solution

[0006] To achieve the above-mentioned purpose, the present invention provides the following technical solution: a multi-energy prediction method for an integrated energy system based on an effective information extraction notifier model, comprising the following steps:

[0007] Step 1: Collect and organize the time series data of electricity, heat and gas loads in the integrated energy system for subsequent model training and debugging;

[0008] Step 2: Build the Efficient Information Extracting Informer network architecture;

[0009] Step 3: Based on the regular training data set and the constructed neural network architecture, select the Adam algorithm to train the network and save the optimal parameters and model;

[0010] Step 4: Perform performance evaluation and hyperparameter debugging of the neural network until the model performance meets application requirements.

[0011] Preferably, the implementation steps of step 1 are as follows:

[0012] Step 1.1: Assume that the collected multi-energy data contains N time points in total, then the historical data can be expressed as:

[0013]

[0014] In the formula, x 1,t , t∈[1,N] represents the electric load data, x 2,t , t∈[1,N] represents the heat load data, x 3,t , t∈[1,N] represents the natural gas load data, T t , t∈[1,N] represents time information, usually in the format of "year month time", and performs outlier and missing value detection on the load data in the data set X. For outliers, firstly, outliers are detected based on statistical methods, and then, for outliers and missing values, the average value of the four time steps before and after is used to replace them. That is, assuming that x at time t t is an outlier or missing value, then:

[0015]

[0016] Step 1.2: Normalize the electricity, heat and gas data separately:

[0017]

[0018] Among them, m and s are x t , the mean and variance of t∈[1,N];

[0019] Step 1.3: T t ,t∈[1,N] is encoded as a timestamp. Taking the time information resolution as hours as an example, each dimension of the converted 4-dimensional variable represents [year, month, date, week, hour], and the preprocessed data set is constructed:

[0020]

[0021] Step 1.4: Set the multi-energy task of the integrated energy system to use the load data of the first 72 hours to predict the load data of the next 24 hours, and organize the data according to this standard to form a sample, that is: [X in ,X out ], where X in ∈R 4×72 , X out ∈R 4×24 , divide the above-organized samples into training set and test set at a ratio of 7:3 for subsequent model training and verification debugging.

[0022] Preferably, the implementation steps of step 2 are as follows:

[0023] Step 2.1: Temporally embed the input data. The embedding of temporal data is divided into two parts.

[0024] First, encode the location of the data. The formula is as follows:

[0025]

[0026] Where p is the position of x in the time series, i is the i-th dimension in the embedding vector, and d_model is the dimension of the embedding vector, usually 256 or 512;

[0027] Secondly, the timestamp is encoded and mapped into a 256-dimensional vector S using a fully connected neural network. e (x), then the data input to the subsequent neural network is:

[0028] X embedd =x+P e (x)+S e (x) (6)

[0029] Where x is the load time series data;

[0030] Step 2.2: Build the encoder network, build the attention layer, and receive X obtained in step 1 embedd As input, multiply by a weight matrix to get Q∈R m×d , K∈R n×d , V∈R n×d , the attention layer first samples K to get K_sample, and for each q i ∈Q finds the value of M with respect to K_sample:

[0031]

[0032] Among them, L pred For the required load forecast sequence length, find the u q with the largest M value i , facing Top-u q i Regarding K to find the score value:

[0033]

[0034] in, It is Top-u q i The matrix composed of the input data X embedd Add it to A(Q,K,V) and normalize it to get variable X', then input it into the designed high-dimensional fully connected layer to get variable X", add X' and X" and normalize it to get X"' as the output of the encoder;

[0035] Step 2.3: Construct a decoder network. During the training process, the decoder input is real historical data. During the deployment process, its input data is the sequence data predicted by the decoder in the previous step, which is the same as the encoder. It is temporally embedded to obtain X e ' mbedd, and obtain Q∈R by multiplying it with the weight matrix m×d , K∈R n×d , V∈R n×d , input into the attention network with the same structure as the encoder, and obtain the variable Then X e ' mbedd and Add and normalize to get As the V value, it is input to the next layer of attention network. This layer of attention network receives the output of the encoder and forms Q and K values ​​after being processed by the weight matrix. And X d 2 and Add and normalize to get Then input it into the designed high-dimensional fully connected layer to obtain the variable and will and Add and normalize as the output of the decoder

[0036] Step 2.4: Build an output module to receive the output of the decoder Construct linear layer and softmax layer to get the predicted value

[0037] Preferably, the implementation steps of step 3 are as follows:

[0038] Based on the constructed training data set and the encoder, decoder and output module, select the Adam solution algorithm, set the network hyperparameters, train the network, and save the optimal weight parameters and model. The hyperparameters include but are not limited to train_epochs, batch_size and learning_rate.

[0039] Preferably, the implementation steps of step 4 are as follows:

[0040] Step 4.1: Comprehensively evaluate the model performance based on the mean square error (MSE) and mean absolute percentage error (MAPE) of the model on the training set and test set. The calculation formulas for MSE and MAE are as follows:

[0041]

[0042] Step 4.2: Based on the model evaluation results, analyze whether the current model is underfitting or overfitting, adjust the model hyperparameters, return to step 3, and perform network training again until the model performance meets the application requirements.

[0043] (III) Beneficial effects

[0044] Compared with the prior art, the present invention provides a multi-energy prediction method for an integrated energy system based on an effective information extraction notifier model, which has the following beneficial effects:

[0045] This multi-energy prediction method for integrated energy systems based on the Effective Information Extracting Informer model collects time series data of electricity, heat, and gas loads, and then builds an Efficient Information Extracting Informer (Efficient Information Extracting Informer) network architecture to perform high-precision and high-efficiency predictions of integrated energy multi-energy load sequences. At the same time, it performs performance evaluation and hyperparameter debugging of the neural network until the model performance meets application requirements and has high reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A schematic diagram of a multi-energy prediction method for an integrated energy system based on an effective information extraction notifier model proposed by the present invention;

[0047] Figure 2 This is the network architecture diagram of the Efficient Information Extracting Informer proposed in the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] like Figure 1 As shown, a multi-energy prediction method for an integrated energy system based on an effective information extraction notifier model comprises the following steps:

[0050] Step 1: Collect and organize the time series data of electricity, heat and gas loads in the integrated energy system for subsequent model training and debugging. The implementation steps are as follows:

[0051] Step 1.1: Assume that the collected multi-energy data contains N time points in total, then the historical data can be expressed as:

[0052]

[0053] In the formula, x 1,t , t∈[1,N] represents the electric load data, x 2,t , t∈[1,N] represents the heat load data, x 3,t , t∈[1,N] represents the natural gas load data, T t, t∈[1,N] represents time information, usually in the format of "year month time", and performs outlier and missing value detection on the load data in the data set X. For outliers, firstly, outliers are detected based on statistical methods, and then, for outliers and missing values, the average value of the four time steps before and after is used to replace them. That is, assuming that x at time t t is an outlier or missing value, then:

[0054]

[0055] Step 1.2: Normalize the electricity, heat and gas data separately:

[0056]

[0057] Among them, m and s are x t , the mean and variance of t∈[1,N];

[0058] Step 1.3: T t ,t∈[1,N] is encoded as a timestamp, taking the time information resolution as an example, as shown in the following table:

[0059]

[0060]

[0061] Each dimension of the converted 4-dimensional variable represents [year, month, date, week, hour], and constructs the preprocessed data set:

[0062]

[0063] Step 1.4: Set the multi-energy task of the integrated energy system to use the load data of the first 72 hours to predict the load data of the next 24 hours, and organize the data according to this standard to form a sample, that is: [X in ,X out ], where X in ∈R 4×72 , X out ∈R 4×24 , divide the above-organized samples into training set and test set at a ratio of 7:3 for subsequent model training and verification debugging.

[0064] like Figure 1 As shown, step 2: build the Efficient Information Extracting Informer (Effective Information Extraction Informer model) network architecture, the implementation steps are as follows:

[0065] Step 2.1: Temporally embed the input data. The embedding of temporal data is divided into two parts.

[0066] First, encode the location of the data. The formula is as follows:

[0067]

[0068] Where p is the position of x in the time series, i is the i-th dimension in the embedding vector, and d_model is the dimension of the embedding vector, usually 256 or 512;

[0069] Secondly, the timestamp is encoded and mapped into a 256-dimensional vector S using a fully connected neural network. e (x), then the data input to the subsequent neural network is:

[0070] X embedd =x+P e (x)+S e (x) (6)

[0071] Where x is the load time series data;

[0072] Step 2.2: Build the encoder network, build the attention layer, and receive X obtained in step 1 embedd As input, multiply by a weight matrix to get Q∈R m×d , K∈R n×d , V∈R n×d , the attention layer first samples K to get K_sample, and for each q i ∈Q finds the value of M with respect to K_sample:

[0073]

[0074] Among them, L pred For the required load forecast sequence length, find the u q with the largest M value i , facing Top-u q i Regarding K to find the score value:

[0075]

[0076] in, It is Top-u q i The matrix composed of.

[0077] like Figure 2 As shown, the input data X embedd Add it to A(Q,K,V) and normalize it to get variable X', then input it into the designed high-dimensional fully connected layer to get variable X", add X' and X" and normalize it to get X"' as the output of the encoder;

[0078] The above network architecture is stacked H times, depending on the task requirements.

[0079] like Figure 2 As shown in step 2.3: Construct a decoder network. During the training process, the decoder input is real historical data. During the deployment and application process, its input data is the sequence data predicted by the decoder in the previous step, which is the same as the encoder. It is temporally embedded to obtain X e ' mbedd , and obtain Q∈R by multiplying it with the weight matrix m×d , K∈R n×d , V∈R n×d , input into the attention network with the same structure as the encoder, and obtain the variable Then X e ' mbedd and Add and normalize to get As the V value, it is input to the next layer of attention network. This layer of attention network receives the output of the encoder and forms Q and K values ​​after being processed by the weight matrix. and will and Add and normalize to get Then input it into the designed high-dimensional fully connected layer to obtain the variable and will and Add and normalize as the output of the decoder

[0080] The above network architecture is stacked H times, depending on the task requirements.

[0081] like Figure 2 As shown, step 2.4: construct an output module to receive the output of the decoder Construct linear layer and softmax layer to get the predicted value

[0082] like Figure 1 As shown, step 3: Based on the regular training data set and the constructed neural network architecture, select the Adam solution algorithm to train the network and save the optimal parameters and model. The implementation steps are as follows:

[0083] Based on the constructed training data set and the encoder, decoder and output module, select the Adam algorithm, set the network hyperparameters, train the network, and save the optimal weight parameters and model. The hyperparameters include but are not limited to train_epochs, batch_size and learning_rate.

[0084] like Figure 1As shown, step 4: perform performance evaluation and hyperparameter debugging of the neural network until the model performance meets the application requirements. The implementation steps are as follows:

[0085] Step 4.1: Comprehensively evaluate the model performance based on the mean square error (MSE) and mean absolute percentage error (MAPE) of the model on the training set and test set. The calculation formulas for MSE and MAE are as follows:

[0086]

[0087] Step 4.2: Based on the model evaluation results, analyze whether the current model is underfitting or overfitting, adjust the model hyperparameters, return to step 3, and perform network training again until the model performance meets the application requirements.

[0088] It should be noted that the term "comprises" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, article, or device that includes a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such process, method, article, or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article, or device that includes the element.

[0089] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-energy prediction method for an integrated energy system based on an effective information extraction informant model, characterized in that: The following steps are involved: Step 1: Collect and organize the time series data of electricity, heat and gas loads in the integrated energy system for subsequent model training and debugging; Step 2: Build the Efficient Information Extracting Informer network architecture; Step 3: Based on the regular training data set and the constructed neural network architecture, select the Adam algorithm to train the network and save the optimal parameters and model; Step 4: Perform performance evaluation and hyperparameter debugging of the neural network until the model performance meets application requirements.

2. The method for multi-energy prediction of an integrated energy system based on an effective information extraction notifier model according to claim 1 is characterized in that: The implementation steps of step 1 are as follows: Step 1.1: Assume that the collected multi-energy data contains N time points in total, then the historical data can be expressed as: In the formula, x 1,t , t∈[1,N] represents the electric load data, x 2,t , t∈[1,N] represents the heat load data, x 3,t , t∈[1,N] represents the natural gas load data, T t , t∈[1,N] represents time information, usually in the format of "year month time", and performs outlier and missing value detection on the load data in the data set X. For outliers, firstly, outliers are detected based on statistical methods, and then, for outliers and missing values, the average value of the previous and next four time steps is used to replace them. That is, assuming that x at time t t is an outlier or missing value, then: Step 1.2: Normalize the electricity, heat and gas data separately: Among them, m and s are x t , the mean and variance of t∈[1,N]; Step 1.3: T t ,t∈[1,N] is encoded as a timestamp. Taking the time information resolution as hours as an example, each dimension of the converted 4-dimensional variable represents [year, month, date, week, hour], and the preprocessed data set is constructed: Step 1.4: Set the multi-energy task of the integrated energy system to use the load data of the first 72 hours to predict the load data of the next 24 hours, and organize the data according to this standard to form a sample, that is: [X in ,X out ], where X in ∈R 4×72 , X out ∈R 4×24 , divide the above-organized samples into training set and test set at a ratio of 7:3 for subsequent model training and verification debugging.

3. The multi-energy prediction method for an integrated energy system based on an effective information extraction notifier model according to claim 1 is characterized in that: The implementation steps of step 2 are as follows: Step 2.1: Temporally embed the input data. The embedding of temporal data is divided into two parts. First, encode the location of the data. The formula is as follows: Where p is the position of x in the time series, i is the i-th dimension in the embedding vector, and d_model is the dimension of the embedding vector, usually 256 or 512; Secondly, the timestamp is encoded and mapped into a 256-dimensional vector S using a fully connected neural network. e (x), then the data input to the subsequent neural network is: X embedd =x+P e (x)+S e (x) (6) Where x is the load time series data; Step 2.2: Build the encoder network, build the attention layer, and receive X obtained in step 1 embedd As input, multiply by a weight matrix to get Q∈R m×d , K∈R n×d , V∈R n×d , the attention layer first samples K to get K_sample, and for each q i ∈Q finds the value of M with respect to K_sample: Among them, L pred For the required load forecast sequence length, find the u q with the largest M value i , facing Top-u q i Regarding K to find the score value: in, It is Top-u q i The matrix composed of the input data X embedd Add it to A(Q,K,V) and normalize it to get variable X', then input it into the designed high-dimensional fully connected layer to get variable X", add X' and X" and normalize it to get X"' as the output of the encoder; Step 2.3: Construct a decoder network. During the training process, the decoder input is real historical data. During the deployment process, its input data is the sequence data predicted by the decoder in the previous step, which is the same as the encoder. It is time-embedded to obtain X' embedd , and obtain Q∈R by multiplying it with the weight matrix m×d , K∈R n×d , V∈R n×d , input into the attention network with the same structure as the encoder, and obtain the variable Then X e ' mbedd and Add and normalize to get As the V value, it is input to the next layer of attention network. This layer of attention network receives the output of the encoder and forms Q and K values ​​after being processed by the weight matrix. and will and Add and normalize to get Then input it into the designed high-dimensional fully connected layer to obtain the variable and will and Add and normalize as the output of the decoder Step 2.4: Build an output module to receive the output of the decoder Construct linear layer and softmax layer to get the predicted value 4. The method for multi-energy prediction of an integrated energy system based on an effective information extraction notifier model according to claim 1 is characterized in that: The implementation steps of step 3 are as follows: Based on the constructed training data set and the encoder, decoder and output module, select the Adam solution algorithm, set the network hyperparameters, train the network, and save the optimal weight parameters and model. The hyperparameters include but are not limited to train_epochs, batch_size and learning_rate.

5. The method for multi-energy prediction of an integrated energy system based on an effective information extraction notifier model according to claim 1 is characterized in that: The implementation steps of step 4 are as follows: Step 4.1: Comprehensively evaluate the model performance based on the mean square error (MSE) and mean absolute percentage error (MAPE) of the model on the training set and test set. The calculation formulas for MSE and MAE are as follows: Step 4.2: Based on the model evaluation results, analyze whether the current model is underfitting or overfitting, adjust the model hyperparameters, return to step 3, and perform network training again until the model performance meets the application requirements.

Citation Information

Cited By

  • An industrial equipment health prediction method based on reinforcement density clustering and hidden Markov model

    CN122885653A