An informer-based lithium-ion battery state of charge estimation method

Abstract

The application discloses a lithium ion battery state of charge estimation method based on an Informer, which comprises the following steps: collecting battery sample data, constructing an Informer network model, processing the battery sample data by using the Informer network model, and obtaining a state of charge estimation value of the lithium ion battery. The Informer network model used in the application reduces the point product calculation of the attention mechanism and improves the encoder-decoder architecture, thereby reducing the calculation complexity and spatial complexity, improving the model training and calculation speed, and to some extent, solving the cumulative error problem existing in the recurrent neural network, achieving higher SOC estimation accuracy, showing superiority in MAE and RMSE error indicators, and better solving the estimation difficulty under low temperature conditions and the problem of insufficient temperature expansion capability. The application is widely applied in the technical field of batteries.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of batteries, and in particular to a lithium ion battery state of charge estimation method based on an Informer. BACKGROUND

[0002] In recent years, with the development of economy, environmental pollution is becoming increasingly serious, and traditional fossil energy has been unable to meet the needs of human production and life. Therefore, many vehicles such as trains and cars have begun to use electricity as energy. Electric energy can be stored in various media, and the most widely used in daily life is the battery. The state of charge (SOC) estimation of the battery is a foundation for the application research of the battery. The deep charge and discharge of the battery in use can greatly reduce its service life, and accurate SOC estimation can avoid this situation; accurate display of the remaining power can also help the vehicle control system to calculate the driving range, and the driver can better plan the driving route. However, since the state of charge of the battery is related to many factors, it is difficult to accurately determine the state of charge of the battery by only measuring the voltage and current of the battery and the like.

[0003] The existing SOC estimation methods include traditional methods such as model-based filtering algorithms such as extended Kalman filter, unscented Kalman filter, particle filter, and deep learning methods represented by convolutional neural network (CNN), recurrent neural network (RNN), feedforward neural network (FFNN), and Transformer. The traditional model-based method often has the shortcomings of modeling difficulty, insufficient generalization ability, large error, certain limitations, etc. Although the deep learning method can solve some problems of the traditional method, it is a big challenge to improve the accuracy of the deep learning method, and the model training time of the deep learning network is relatively long, and a large amount of data set is required, which requires a certain training cost. And there is a lack of temperature expansion capability, and most of the above methods have poor SOC estimation performance at low temperature.

[0004] Explanation of terms:

[0005] State of charge: the state of charge (SOC) is a relative measure of the energy stored in the battery, defined as the amount of charge available at a specific time point from the battery cell to the total capacity.

[0006] Deep learning: derived from the research of artificial neural networks. By combining low-level features to form more abstract high-level representation attributes or features, the distributed feature representation of data is discovered.

[0007] Convolutional neural network: Convolutional neural network (CNN) is a kind of feedforward neural network containing convolution calculation and having deep structure, which is one of the representative algorithms of deep learning.

[0008] Recurrent neural network: Recurrent neural network (RNN) is a kind of recursive neural network with sequence data as input, recursion in the evolution direction of sequence and all nodes (recurrent units) connected in chain.

[0009] Long short-term memory network: Long short-term memory network (LSTM) is a kind of time recurrent neural network, which is specially designed to solve the long-term dependence problem existing in general recurrent neural network (RNN)

[0010] Kalman filtering: Kalman filtering is an algorithm for optimal estimation of system state by using linear system state equation and observing data of system input and output. Since the observation data includes the influence of noise and interference in the system, the optimal estimation can also be regarded as a filtering process.

[0011] Feedforward neural network: Feedforward neural network (FFNN) is the simplest neural network, in which each neuron is arranged in layers, and each neuron is connected only to the neurons of the previous layer. The output of the previous layer is received and output to the next layer, and there is no feedback between layers. It is one of the most widely used and fastest developing artificial neural networks. Research began in the 1960s, and theoretical research and practical application have reached a high level. SUMMARY

[0012] In view of the technical problems of the existing SOC estimation method, such as modeling difficulty, insufficient generalization ability, large error and certain limitations, the purpose of the present application is to provide a lithium ion battery state of charge estimation method based on Informer, comprising:

[0013] Collecting battery sample data; the battery sample data is used to represent the historical state of charge of the lithium ion battery;

[0014] constructing an Informer network model; the Informer network model is used to perform a probabilistic self-attention mechanism, a distillation operation and a dynamic decoding;

[0015] processing the battery sample data using the Informer network model to obtain a state of charge estimation value of a lithium ion battery.

[0016] Further, the battery sample data includes:

[0017] Obtain current data, voltage data and battery surface temperature data of the lithium ion battery, and the current data, the voltage data and the battery surface temperature data constitute the battery sample data;

[0018] Pretreatment of the battery sample data;

[0019] The battery sample data is processed by dimensionality.

[0020] Further, the pretreatment of the battery sample data includes:

[0021] The battery sample data is subjected to Z-Score standardization.

[0022] Further, the dimensionality processing of the battery sample data includes:

[0023] The two-dimensional battery sample data is converted into a three-dimensional format; the three-dimensional format is [time_step, input_size, batch_size]; wherein, time_step represents time step, input_size represents input feature dimension, and batch_size represents sample number.

[0024] Further, the construction of the Informer network model includes:

[0025] Building a first encoding layer, a second encoding layer, a first decoding layer, a second decoding layer and a fully connected layer;

[0026] The first encoding layer is used to perform a probabilistic self-attention mechanism and a distillation operation on the training data set input to the Informer network model in turn;

[0027] The second encoding layer is used to perform a probabilistic self-attention mechanism and a distillation operation on the training data set input to the Informer network model in turn and the output result of the first encoding layer;

[0028] The first decoding layer is used to perform a mask probabilistic self-attention mechanism and a standard self-attention mechanism on the battery sample data input to the Informer network model in turn.

[0029] The second decoding layer is configured to sequentially perform a masked probability self-attention mechanism and a standard self-attention mechanism on the output result of the second encoding layer and the output result of the first decoding layer.

[0030] The full connection layer is configured to perform full connection processing on the output result of the second decoding layer.

[0031] Further, the probability self-attention mechanism comprises:

[0032] is processed by the following formula:

[0033] Q = X enc · W Q

[0034] K = X enc · W K

[0035] V = X enc · W V

[0036]

[0037] a = score · V

[0038] wherein, X enc is the training data set, W Q , W K and W V are parameter matrices, q i is the i-th data in the training data set, p(k j | q i ) represents a deviation from a uniform distribution related to q i , d is a parameter, and a is the output result of the probability self-attention mechanism.

[0039] Further, the distillation operation comprises:

[0040] The data is sequentially processed by a convolution layer and a pooling layer for dimension reduction; the dimension reduction reduces the length of the data to half of the original length.

[0041] Further, the masked probability self-attention mechanism comprises:

[0042] is processed by the following formula:

[0043]

[0044] wherein, is the battery sample data, represents a sequence to be predicted, Output of the mask probability self-attention mechanism. token Input data representing the decoding layer Length in the time dimension, L y Data to be estimated Length in the time dimension (how many time points of battery state of charge do you want to estimate in the future), d model Data format conversion to a data format that the model can receive (not shown in the figure), and the conversion method is to perform convolution processing on the data.

[0045] Further, the standard self-attention mechanism comprises:

[0046] The output of the mask probability self-attention mechanism is processed by the probability self-attention mechanism, and the processing result is taken as Q, the output of the second encoding layer is taken as K and V, and attention scoring is performed.

[0047] Further, the lithium ion battery state of charge estimation method based on the Informer further comprises:

[0048] The back propagation and update of the Informer network model parameters are realized using the Adam optimization algorithm;

[0049] The learning rate of the Informer network model is defined;

[0050] The root mean square error (RMSE) is used as the standard for evaluating the Informer network model;

[0051] Multiple training iterations are performed; in each iteration, the root mean square error between the estimation result of the Informer network model and the true value is back propagated to adjust the parameters of the Informer network model.

[0052] The beneficial effects of the present application are: the lithium ion battery state of charge estimation method based on the Informer in the embodiment uses the Informer network model, which reduces the computational complexity and spatial complexity by reducing the dot product calculation of the attention mechanism and improving the encoder-decoder architecture, improves the speed of model training and calculation, and to some extent solves the cumulative error problem of the recurrent neural network, can achieve higher SOC estimation accuracy, and has superior performance in MAE and RMSE error indicators, and can better solve the problems of estimation difficulty under low temperature conditions and insufficient temperature expansion capability. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1A schematic diagram of steps of an embodiment of a lithium-ion battery state-of-charge estimation method based on Informer;

[0054] Figure 2 A schematic diagram of a flow of an embodiment of a lithium-ion battery state-of-charge estimation method based on Informer;

[0055] Figure 3 and Figure 4 A schematic diagram of structure and principle of an Informer network model in an embodiment. DETAILED DESCRIPTION

[0056] In this embodiment, one of the variants based on Transformer, namely the Informer model suitable for long time series estimation, is used to enable its use for SOC estimation of lithium-ion batteries, to improve the accuracy of SOC estimation, and to improve the temperature expansion capability of the model.

[0057] With reference to Figure 1 , the lithium-ion battery state-of-charge estimation method based on Informer includes the following steps:

[0058] S1. Collecting battery sample data; the battery sample data is used to represent the historical state-of-charge of the lithium-ion battery;

[0059] S2. Constructing an Informer network model; the Informer network model is used to perform a probabilistic self-attention mechanism, a distillation operation, and a dynamic decoding;

[0060] S3. Using the Informer network model to process the battery sample data to obtain the state-of-charge estimation value of the lithium-ion battery.

[0061] In this embodiment, the flow of steps S1-S3 is as shown in Figure 2 .

[0062] When performing step S1, that is, the step of collecting battery sample data, the following steps can be performed:

[0063] S101. Obtaining current data, voltage data, and battery surface temperature data of the lithium-ion battery, to form the battery sample data with the current data, the voltage data, and the battery surface temperature data;

[0064] S102. Preprocessing the battery sample data;

[0065] S103. Dimensionally upgrading the battery sample data.

[0066] In step S102, the battery sample data P is preprocessed, and specifically, the current, voltage, and battery surface temperature in the battery sample data P are standardized by Z-Score, and the conversion function is:

[0067]

[0068] wherein μ is the mean of the sample data (current, voltage, and battery surface temperature, respectively), and σ is the standard deviation of the sample data.

[0069] In step S103, the original two-dimensional feature data format battery sample data is constructed into a three-dimensional data format for use as the input of the network. The three-dimensional data format is [time_step, input_size, batch_size], wherein time_step represents the time step, for example, using the data of the first 6 time points to predict the fourth time point, then time_step = 6. input_size represents the input feature dimension, in this embodiment, input_size = 3, representing three dimensions of current, voltage, and battery surface temperature. batch_size represents the number of samples, i.e., the number of samples put into the network.

[0070] In this embodiment, the structure and principle of the constructed Informer network model are as shown in Figure 3 and Figure 4 . Referring to Figure 3 and Figure 4 , the Informer network model includes a first encoding layer, a second encoding layer, a first decoding layer, a second decoding layer, and a fully connected layer. The first encoding layer is used to sequentially perform a probabilistic self-attention mechanism and a distillation operation on the training data set input to the Informer network model; the second encoding layer is used to sequentially perform a probabilistic self-attention mechanism and a distillation operation on the output result of the first encoding layer and the training data set input to the Informer network model; the first decoding layer is used to sequentially perform a masked probabilistic self-attention mechanism and a standard self-attention mechanism on the battery sample data input to the Informer network model; the second decoding layer is used to sequentially perform a masked probabilistic self-attention mechanism and a standard self-attention mechanism on the output result of the second encoding layer and the output result of the first decoding layer; and the fully connected layer is used to perform fully connected processing on the output result of the second decoding layer.

[0071] In this embodiment, for the probability self-attention mechanism: the attention mechanism can focus on the key content from the global information, focus on the key features, and has the characteristics of relatively few parameters, fast speed and good effect. ProbSaprse Self-Attention solves the problem of attention calculation complexity and reduces the dot product operation of attention. For the score obtained by the dot product calculation of the attention layer, it can be regarded as a probability conditional probability distribution matrix:

[0072] Q=X enc ·W Q

[0073] K=X enc ·W K

[0074] V=X enc ·W V

[0075]

[0076] a=score·V

[0077] Some of the attention weights corresponding to the p(k j |q i ) deviate from the uniform distribution. If p(k j |q i ) is close to a uniform distribution , then Self-Attention becomes the average of score, that is, redundant. Therefore, only the attention weights that play a significant role dominate the attention calculation, and as long as this part is extracted, the network operation speed can be improved without sacrificing performance.

[0078] The KL divergence is used to represent the degree of deviation of the weight from the uniform distribution, and the greater the KL divergence, the greater the degree of deviation of the weight from the uniform distribution, that is, the more attention is needed for this data. The calculation format of the KL divergence is as follows:

[0079]

[0080]

[0081] Where M(q i , K) is the calculated KL divergence without constant value, used to represent the sparsity measure of the i-th query, that is, the i-th sample data. The data dimension after passing through all self-attention layers does not change.

[0082] In this embodiment, for the distillation operation: in order to ensure the consistency of the sequence dimension, the sampled q i will be updated, and the unsampled q i will be replaced with a uniform distribution, so that the Attention result obtained through the Prob Sparse Self-Attention step will still produce redundancy, at which time distilling operation (Distilling) is needed to extract the main characteristic data, which can greatly shorten the time dimension of the input. The specific method of the distilling operation is to realize the data set dimension reduction through the convolution layer and then the pooling layer, and the length of the data set after distilling will become X / 2.

[0083] In this embodiment, in the first decoding layer and the second decoding layer, the Informer changes the architecture of the traditional decoding layer (Decoder), generates all prediction outputs at once, and essentially solves the dynamic decoding problem by increasing more real data to assist the Decoder decoding.

[0084] In this embodiment, the input of the first decoding layer and the second decoding layer is:

[0085]

[0086] Among them, is the real and effective data (battery sample data), represents the sequence to be estimated, and needs to be set to 0, L token represents the input data of the decoding layer in the time dimension, L y represents the data to be estimated in the time dimension (want to estimate the battery state of charge for how many time points in the future), d model represents the data format obtained by splicing the two into a data format that the model can receive (not marked in the figure), and the conversion method is to perform convolution processing on the data. The input of the Decoder first goes through the same Prob Self-Attention step as the Encoder, then takes the output as Q, the output of the Encoder as K and V, and then does an Attention scoring step, and finally splices the output into a fully connected layer, which directly outputs the estimation result.

[0087] In this embodiment, after the Informer network model is built, the Informer network model can be trained. Specifically, the back propagation and update of network parameters are realized using the Adam optimization algorithm, and the learning rate of the network is defined as 0.001. The root mean square error (RMSE) is used as the standard for model evaluation. In each iteration, the parameters of the network are adjusted according to the root mean square error between the predicted results of the network and the true values, so that after multiple training, the predicted results of the network can continuously approach the true values.

[0088] After the training of the Informer network model is completed, step S3 is performed to process the battery sample data using the Informer network model to obtain the state of charge estimation value of the lithium ion battery.

[0089] In this embodiment, the principle of the Informer-based lithium ion battery state of charge estimation method is that the Informer network model used is a variant of Transformer. The Informer network model reduces the computational complexity and spatial complexity by reducing the dot product calculation of the attention mechanism and improving the encoder-decoder architecture, improves the speed of model training and calculation, and to some extent solves the cumulative error problem of recurrent neural networks. Compared with the machine learning methods, deep learning methods, model-based filtering algorithms and other methods proposed by people before, the Informer method for lithium ion battery SOC estimation proposed by the present application can achieve higher SOC estimation accuracy and show superiority in MAE and RMSE error indicators. At the same time, compared with existing methods, the present application can better solve the estimation difficulty under low temperature conditions and the problem of insufficient temperature expansion capability.

[0090] Specifically, the structure of the Informer network model includes:

[0091] 1. The probability self-attention mechanism can focus on the key features that affect the state of charge of the battery, and compared with the ordinary attention mechanism, it can greatly reduce the operation time and improve the efficiency of the network.

[0092] 2. The distillation operation can extract the main characteristic data and eliminate the redundant data after the probability self-attention, which can greatly shorten the time dimension of the input.

[0093] 3. The dynamic decoding layer realizes the multi-time parallel estimation of the state of charge of the battery through the input assistance of the original data, and can make the prediction result continuously approach the true value and improve the prediction accuracy.

[0094] Therefore, the Informer network model can realize accurate estimation of the state of charge of the battery at multiple future times through analysis of the current, voltage and battery surface temperature of the battery

[0095] The lithium ion battery state of charge estimation method based on Informer in the embodiment can be implemented by writing a computer program for executing the lithium ion battery state of charge estimation method based on Informer in the embodiment, writing the computer program into a storage medium or a computer device, and executing the lithium ion battery state of charge estimation method based on Informer in the embodiment when the computer program is read out and run, thereby achieving the same technical effects as the lithium ion battery state of charge estimation method based on Informer in the embodiment.

[0096] It should be noted that, unless otherwise specified, when a certain feature is said to be "fixed" or "connected" to another feature, it can be directly fixed or connected to the other feature, or indirectly fixed or connected to the other feature. In addition, the up, down, left, right and the like used in the present disclosure are only relative to the relative positional relationship of the components of the present disclosure in the drawings. The singular forms "a", "an" and "the" used in the present disclosure are also intended to include the plural forms, unless the context clearly indicates otherwise. In addition, unless otherwise defined, all technical and scientific terms used in the embodiments have the same meaning as generally understood by those skilled in the art. The terms used in the embodiments are only used to describe the specific embodiments, and are not intended to limit the present application. The term "and / or" used in the embodiments includes any combination of one or more related listed items.

[0097] It should be understood that although the terms first, second, third, etc. can be used in the present disclosure to describe various elements, these elements should not be limited to these terms. These terms are only used to distinguish one type of element from another. For example, without departing from the scope of the present disclosure, the first element can also be referred to as the second element, and similarly, the second element can also be referred to as the first element. The use of any and all examples or exemplary language (e.g., "for example", "as such", etc.) provided in the present embodiments is only intended to better illustrate the embodiments of the present application, and unless otherwise required, does not impose any limitation on the scope of the present application.

[0098] It should be appreciated that embodiments of the present application can be realized by a computer hardware, a combination of hardware and software, or by computer instructions stored on a non-transitory computer-readable storage medium. The methods can be implemented in a computer program using standard programming techniques— including non-transitory computer-readable storage medium configured with a computer program to implement the methods described herein and the accompanying drawings, wherein the storage medium so configured makes the computer operate in a specific and predefined manner. Each program can be implemented in a high level procedural or object oriented programming language to communicate with a computer system. However, the programs can be implemented in assembly or machine language, if so desired. In any case, the language can be a compiled or interpreted language. Also, the programs can be able to run on standalone systems or in conjunction with other programs or in a distributed computing environment, with portions of programs executed on multiple computers. The methods described herein can be implemented by a computer program product, which can include a computer program tangibly embodied in a non-transitory computer-readable storage medium.

[0099] Further, the operations of the processes described herein can be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. The processes described herein (or variations and / or combinations thereof) can be implemented under the control of one or more computer systems configured with executable instructions (e.g., computer programs, one or more computer programs, or one or more applications) to perform the processes described herein, and can be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) executing collectively on one or more processors, by hardware, or combinations thereof. The computer programs include a plurality of instructions executable by one or more processors.

[0100] Further, the methods can be implemented in any suitable type of computing platform operably connected to, including but not limited to, a personal computer, a mini-computer, a mainframe, a workstation, a network or distributed computing environment, a stand-alone or integrated computer platform, or in communication with a charged particle tool or other imaging device, and the like. Aspects of the present application can be implemented in machine readable code stored on a non-transitory storage medium or device, whether removable or integrated to the computing platform, such as a hard disk, an optical read and / or write storage medium, RAM, ROM, and the like, such that it is readable by a programmable computer and, when the storage medium or device is read by the computer, is used to configure and operate the computer to perform the processes described herein. Further, the machine readable code, or portions thereof, can be transmitted over a wired or wireless network. The present application described herein includes these and other different types of non-transitory computer readable storage media when such media include instructions or programs to implement the steps described above in conjunction with a microprocessor or other data processor. The present application also includes the computer itself when programmed according to the methods and techniques described herein.

[0101] A computer program can be applied to input data to perform the functions described in this embodiment, thereby transforming the input data to generate output data that is stored to non-volatile memory. The output information can also be applied to one or more output devices such as a display. In a preferred embodiment of the application, the transformed data represents a physical and tangible object, including a particular visual depiction of a physical and tangible object produced on a display.

[0102] The above description is only preferred embodiments of the present application, the present application is not limited to the above-described embodiments, as long as the same means to achieve the technical effects of the present application, any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application, should be included in the scope of protection of the present application. The technical solutions and / or embodiments within the scope of protection of the present application can have various modifications and changes.

Claims

1. A method for estimating the state of charge of a lithium-ion battery based on Informer, characterized in that, The Informer-based method for estimating the state of charge of a lithium-ion battery includes: Collect battery sample data; the battery sample data is used to represent the historical state of charge of lithium-ion batteries. Construct an Informer network model; the Informer network model is used to perform probabilistic self-attention mechanism, distillation operation and dynamic decoding; The Informer network model is used to process the battery sample data to obtain the estimated state of charge of the lithium-ion battery. The construction of the Informer network model includes: Construct a first encoding layer, a second encoding layer, a first decoding layer, a second decoding layer, and a fully connected layer; The first encoding layer is used to sequentially execute a probabilistic self-attention mechanism and a distillation operation on the training dataset input to the Informer network model; The second encoding layer is used to perform probabilistic self-attention mechanism and distillation operation on the training dataset input to the Informer network model, in turn, and the output of the first encoding layer; The first decoding layer is used to sequentially execute a masked probability self-attention mechanism and a standard self-attention mechanism on the battery sample data input to the Informer network model; The second decoding layer is used to sequentially execute a mask probability self-attention mechanism and a standard self-attention mechanism on the output results of the second encoding layer and the output results of the first decoding layer; The fully connected layer is used to perform fully connected processing on the output of the second decoding layer; The mask probability self-attention mechanism includes: The following formula is used for processing: in, For the battery sample data, This represents the sequence to be predicted. This is the output of the mask probability self-attention mechanism. Indicates the input data of the decoding layer Length in the time dimension Indicates the data to be estimated Length in the time dimension This means that the data format obtained by concatenating the two data formats needs to be converted into a data format that the model can receive. The conversion method is to perform convolution processing on the data. The standard self-attention mechanism includes: The output of the mask probabilistic self-attention mechanism is processed by the probabilistic self-attention mechanism, and the processing result is used as... The output of the second coding layer is used as and Attention scores were then calculated.

2. The method for estimating the state of charge of a lithium-ion battery based on Informer as described in claim 1, characterized in that, The collected battery sample data includes: Acquire current data, voltage data, and surface temperature data of a lithium-ion battery, and use the current data, voltage data, and surface temperature data to form the battery sample data; The battery sample data is preprocessed; The battery sample data is then subjected to dimensionality upscaling.

3. The method for estimating the state of charge of a lithium-ion battery based on Informer as described in claim 2, characterized in that, The preprocessing of the battery sample data includes: The battery sample data was Z-score standardized.

4. The method for estimating the state of charge of a lithium-ion battery based on Informer as described in claim 2, characterized in that, The upsizing process of the battery sample data includes: The two-dimensional battery sample data is converted into a three-dimensional format; the three-dimensional format is [time_step, input_size, batch_size]; where time_step represents the time step, input_size represents the input feature dimension, and batch_size represents the number of samples.

5. The method for estimating the state of charge of a lithium-ion battery based on Informer as described in claim 1, characterized in that, The probabilistic self-attention mechanism includes: The following formula is used for processing: in, For the training dataset, , and For parameter matrices, For the first in the training dataset One data point, Indicates and The relevant deviation from uniform distribution, For parameters, This is the output of the probabilistic self-attention mechanism.

6. The method for estimating the state of charge of a lithium-ion battery based on Informer as described in claim 1, characterized in that, The distillation operation includes: The data is then processed by passing convolutional layers and pooling layers in sequence to reduce the length of the data to half of its original length.

7. A method for estimating the state of charge of a lithium-ion battery based on Informer according to any one of claims 1-6, characterized in that, The Informer-based lithium-ion battery state-of-charge estimation method further includes: The backpropagation and update of the Informer network model parameters are implemented using the Adam optimization algorithm. Define the learning rate of the Informer network model; The root mean square error (RMSE) was used as the evaluation criterion for the Informer network model. Multiple training iterations are performed; in each iteration, the root mean square error between the estimated result and the true value of the Informer network model is backpropagated to adjust the parameters of the Informer network model.

Images