A power battery disassembly path optimization method based on an autoregressive generation strategy

Abstract

The application relates to the technical field of battery disassembly and recycling, and discloses a power battery disassembly path optimization method based on a self-recurrence generation strategy. Multi-source data of a battery to be disassembled is acquired, and the multi-source data is normalized and then input into a multi-decoding layer Transformer model; the multi-decoding layer Transformer model performs dimension reduction processing on the battery data, performs splicing and fusion processing to obtain fusion features, acquires the task ID and score of the battery to be disassembled according to the fusion features, and completes the next step of disassembly of the battery according to the score and the task ID; based on the self-recurrence generation strategy, the task ID and the fusion features acquired this time are repeatedly taken as the input of the multi-decoding layer Transformer model, the score and the task ID for the next step of disassembly of the battery are acquired, and the disassembly of the battery is completed until the disassembly of the battery is completed. The application solves the problem that the existing battery disassembly path method optimization cannot effectively plan an optimal path.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of battery disassembly and recycling, and in particular to a power battery disassembly path optimization method based on an autoregressive generation strategy. BACKGROUND

[0002] With the rapid development of the electric vehicle market, the problem of retired power batteries is becoming increasingly serious. The disassembly and recycling of retired batteries are not only crucial for environmental protection, but also have important significance for recycling valuable materials, reducing resource waste and costs. Traditional battery disassembly processes usually rely on manual experience and fixed disassembly path arrangements, which lack flexibility and intelligence, resulting in low disassembly efficiency, low equipment utilization, and difficulty in meeting the optimization needs of different battery types, disassembly tasks and environmental conditions. Therefore, how to design an efficient, flexible and schedulable disassembly path optimization method has become a problem to be solved in the field of battery recycling.

[0003] Traditional battery disassembly path optimization methods mainly rely on heuristic algorithms such as genetic algorithms and particle swarm optimization algorithms. However, genetic algorithms and particle swarm optimization algorithms still have some problems, such as being prone to local optimal solutions, especially when the complexity of the disassembly task is high, which may lead to low search efficiency, unstable optimization results when processing multi-source data in the disassembly task, and the inability to consider the complex dependency relationship between tasks in all directions, ignoring how to effectively fuse and optimize static disassembly data and dynamic task scheduling data. SUMMARY

[0004] The present application provides a power battery disassembly path optimization method based on an autoregressive generation strategy to solve the problem that existing battery disassembly path optimization methods cannot effectively plan the optimal path.

[0005] To achieve the above purpose, the present application realizes the following technical solutions:

[0006] In a first aspect, the present application provides a power battery disassembly path optimization method based on an autoregressive generation strategy, comprising the following steps:

[0007] Step 1: Obtain multi-source data of the battery to be disassembled, and normalize the multi-source data to obtain battery data;

[0008] Step 2: input the battery data into a multi-decoding layer Transformer model, perform dimensionality reduction processing on the battery data through the multi-decoding layer Transformer model, obtain multiple pieces of dimensionality reduction data, and perform splicing and fusion processing on the multiple pieces of dimensionality reduction data to obtain fusion features, the multi-decoding layer Transformer model obtains the task ID and score of the battery to be disassembled according to the fusion features, and completes the next step of disassembly of the battery according to the score and task ID;

[0009] Step 3: Based on the autoregressive generation strategy, repeatedly input the task ID obtained this time and the fusion feature into the multi-decoding layer Transformer model to obtain the score and task ID for the next battery disassembly until the battery disassembly is completed.

[0010] The multi-decoding layer Transformer model combines sine and cosine functions to encode the fusion feature to obtain the position information of each data point in the multi-source data to gradually generate the task ID;

[0011] The combination of the sine and cosine functions to encode the fusion feature includes: using sine and cosine functions to generate encoding formulas, using the sine encoding formula to encode the position of each position and even index of the embedding dimension in the fusion feature, and using the cosine encoding formula to encode the position of each position and odd index of the embedding dimension;

[0012] The sine encoding formula is represented by the following formula:

[0013] ;

[0014] Wherein, represents the fusion feature encoded by the sine encoding formula, represents the position, represents the even index of the embedding dimension, represents the dimension of the input vector;

[0015] The cosine encoding formula is represented by the following formula:

[0016] ;

[0017] Wherein, represents the fusion feature encoded by the cosine encoding formula, represents the odd index of the embedding dimension.

[0018] Regarding the task ID and score output by the multi-decoding layer Transformer model, it is determined whether to perform the next step of disassembly according to the task ID based on the comparison of the score and the preset threshold. If the preset threshold is exceeded, the task ID is compared with the target scheduling data file collected additionally to determine that the task ID belongs to the battery disassembly action corresponding to the execution in the target scheduling file, and the battery disassembly is performed based on the determined battery disassembly action. If the preset threshold is not exceeded, the task ID and the score are obtained again.

[0019] Further, in step 1, the multi-source data of the battery to be disassembled includes: obtaining the multi-source data of the battery to be disassembled from a specified folder;

[0020] The multi-source data is composed of data contents in a specified folder, the data contents including fromto data in a fromto folder and machining data in a machining folder, the fromto data including positions of a start module and a termination module, and the machining data including a module ID, a device ID, a disassembly time, and a disposal cost.

[0021] Further, the normalization processing includes alignment and MinMax normalization processing, the alignment being alignment processing of the fromto data and the machining data according to the number of data rows.

[0022] The MinMax normalization processing is represented by the following formula:

[0023] ;

[0024] wherein, is normalized multi-source data, is any input variable in the multi-source data to be subjected to MinMax normalization processing, is a minimum value in column data where the input variable is located, is a maximum value in column data where the input variable is located.

[0025] Through strict alignment and normalization processing, different scales of data can be uniformly converted to the same order of magnitude, avoiding the problem of gradient mismatch in model training. This process not only improves the training efficiency of the model, but also enhances the stability of the final path optimization result.

[0026] Further, in step 2, the dimension reduction processing includes Emberdding processing for the fromto data and full connection linear mapping for the machining data.

[0027] The dimension reduction processing of the fromto data includes: mapping the fromto data to a continuous vector space to obtain an embedding vector, and mapping the embedding vector to a feature space with half the dimension of the model hidden vector through a full connection layer.

[0028] The dimension reduction processing of the machining data includes: linearly mapping the machining data through a full connection layer.

[0029] Further, the splicing and fusion processing of the multi-piece dimension reduction data to obtain fusion features includes: splicing the fromto data and the machining data after the dimension reduction processing in the dimension feature, and further mapping through a full connection layer to obtain fusion features.

[0030] Through the feature processing strategy of separate mapping and subsequent fusion, the application greatly improves the information fusion effect in the feature expression of disassembly path and operation data. The combination of multi-modal data enables the system to comprehensively understand various complex factors in the disassembly process, thereby improving the accuracy and efficiency of disassembly path optimization.

[0031] Further, the multi-decoding layer Transformer model comprises a Transformer model composed of a plurality of decoding layers;

[0032] The decoding layer comprises, in sequence, a self-attention module, a cross-attention module and a feedforward neural network, and a Dropout operation is added after each sub-module of the self-attention module, the cross-attention module and the feedforward neural network.

[0033] Through the above operation, dynamic and static input information from different data sources are effectively processed, and this structure enhances the expression ability of the model, especially in complex disassembly paths and device scheduling, the complex relationship between tasks can be captured.

[0034] Further, the multi-decoding layer Transformer model constructs a loss function based on a cross-entropy loss, a uniqueness penalty and a ranking consistency penalty;

[0035] The cross-entropy loss comprises a gap between the task ID output by the multi-decoding layer Transformer model and a preset real target sequence;

[0036] The uniqueness penalty comprises a penalty on repeated task IDs in the generated sequence;

[0037] The ranking consistency penalty comprises a penalty on task IDs with inconsistent scheduling sequences.

[0038] Further, the autoregressive generation strategy comprises an initial input setting, a step-by-step generation mechanism, a repeated task shielding and a predicted update generation sequence.

[0039] The initial input setting comprises setting a dynamic input for the multi-decoding layer Transformer model, and setting the initial input of the dynamic input as a zero vector;

[0040] The step-by-step generation mechanism comprises taking the task ID output by the multi-decoding layer Transformer model at the current step as input;

[0041] The repeated task shielding comprises adjusting the corresponding value of the logits vector output by the multi-decoding layer Transformer model at the current step in the corresponding generated task ID set to negative infinity for shielding processing;

[0042] The predicted update generation sequence comprises: selecting the task ID corresponding to the highest probability value logits vector in the logits vector after the repeated task shielding processing as the output of the current step.

[0043] Wherein, logits vector is the original output of the last linear layer of the multi-decoding layer Transformer model, which is a real number vector without normalization processing, and after normalization processing, it becomes a probability distribution, and the final output task ID is obtained based on the probability distribution.

[0044] Beneficial effects:

[0045] The power battery disassembly path optimization method based on the autoregressive generation strategy provided by the application can accurately predict the order of disassembly tasks by fusing text information, machine data and location information, using a multi-decoding layer Transformer model combined with an autoregressive generation mechanism to optimize the power battery disassembly path. This accurate task order prediction effectively improves the efficiency and rationality of the disassembly process, and by generating task IDs step by step, the adaptability and reliability of the system are enhanced.

[0046] By combining static multi-source data and dynamic task IDs, the task sequence is optimized. This method not only improves the utilization efficiency of data, but also dynamically adapts to changes in disassembly tasks, enhancing the real-time performance and flexibility of the system.

[0047] The absolute position encoding, data preprocessing and feature fusion technology designed by the application is particularly suitable for the field of battery disassembly and recycling, and can effectively cope with complex industrial environments and production requirements. In addition, by using fixed sine and cosine position encoding, the problem of parameters falling into local optimum in training is avoided, and the stability and efficiency of the model in actual industrial application are improved. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 A flowchart of the power battery disassembly path optimization method based on the autoregressive generation strategy of the embodiment of the application. DETAILED DESCRIPTION

[0049] The technical solutions of the application will be described below in a clear and complete manner. Obviously, the described embodiments are only a part of the embodiments of the application, not all embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0050] Unless otherwise defined, technical terms and scientific terms used in the present application shall have the ordinary meanings as understood by a person with ordinary skill in the art to which the present application pertains. The terms "first", "second", and similar terms used in the present application do not denote any order, quantity, or importance, but are used to distinguish different components. Similarly, the terms "one" or "a" or similar terms do not denote a quantity of one, but rather denote the existence of at least one. The terms "connected" or "coupled" or similar terms do not limit to physical or mechanical connections or couplings, but can include electrical connections or couplings, whether direct or indirect. The terms "upper", "lower", "left", "right", and the like are used to indicate relative positions, and when the absolute positions of the described objects are changed, the relative positions are also changed accordingly.

[0051] Referring to Figure 1 The embodiment of the present application provides a power battery disassembly path optimization method based on an autoregressive generation strategy, comprising the following steps:

[0052] Step 1: Obtain multi-source data of a battery to be disassembled, and normalize the multi-source data to obtain battery data;

[0053] Obtaining the multi-source data of the battery to be disassembled comprises: obtaining the multi-source data of the battery to be disassembled from a specified folder;

[0054] The multi-source data is composed of data contents in the specified folder, and the data contents comprise fromto data in a fromto folder and machining data in a machining folder. The fromto data comprises positions of a starting module and a terminating module, and the machining data comprises a module ID, a device ID, a disassembly time, and a disposal cost.

[0055] Regarding the fromto data, each row comprises two numbers, the first number representing a starting module and the second number representing a terminating module. Regarding the machining data, each row comprises four numbers, the first number representing a module ID, the second number representing a device ID, the third number representing a disassembly time, and the fourth number representing a disposal cost.

[0056] The normalization processing comprises alignment and MinMax normalization processing. The alignment is performed on the basis of the number of data rows, and the fromto data and the machining data are aligned.

[0057] The MinMax normalization processing is represented by the following formula:

[0058] ;

[0059] wherein, is the normalized multi-source data, For any input variable to be MinMax normalized in multi-source data, is the minimum value in the column data where the input variable is located, is the maximum value in the column data where the input variable is located, such as is 1, is 5, and when the data value is 3, the normalized value is 0.5.

[0060] Step 2: input the battery data into the multi-decoding layer Transformer model, perform dimensionality reduction processing on the battery data through the multi-decoding layer Transformer model, obtain multiple pieces of dimensionality reduction data, and perform splicing and fusion processing on the multiple pieces of dimensionality reduction data to obtain fusion features, the multi-decoding layer Transformer model obtains the task ID and score of the battery to be disassembled according to the fusion features, and completes the next disassembly of the battery according to the score and the task ID;

[0061] In Figure 1 , the machine data is machining data, the from-to text is fromto data, the position is the index position of the number in the machining data and the fromto data, and the dynamic input represents the circular operation of the autoregressive generation strategy; Operation represents operation, that is, whether to perform according to the comparison between the score and the preset threshold, and the battery disassembly action determined according to the comparison between the task ID and the target scheduling data;

[0062] The dimensionality reduction processing includes Emberdding processing for the fromto data and full connection linear mapping for the machining data;

[0063] The dimensionality reduction processing of the fromto data includes: mapping the fromto data to a continuous vector space to obtain an embedding vector, and mapping the embedding vector to a feature space with half the dimension of the model hidden vector through a full connection layer. The mathematical expression of the mapping process is:

[0064] ;

[0065] Among them, is the weight matrix, is the bias vector, represents the input fromto data, represents the mapped embedding vector;

[0066] The dimensionality reduction processing of the machining data includes: linearly mapping the machining data through a full connection layer, and the mathematical expression of the linear mapping is:

[0067] ;

[0068] wherein, and are the corresponding mapping parameters, respectively, represents input machining data, represents corresponding low-dimensional data after linear mapping of the machining data.

[0069] The plurality of pieces of dimension-reduced data are spliced and fused to obtain fusion features, including splicing the fromto data and the machining data after dimension reduction in the dimension feature, and further mapping through a fully connected layer to obtain the fusion features.

[0070] In the embodiment, the input of the multi-decoding layer Transformer model also includes position data, which is the index position of the numbers in the machining data and the fromto data, which is not described here as common knowledge of those skilled in the art.

[0071] The multi-decoding layer Transformer model includes a Transformer model composed of a plurality of decoding layers.

[0072] The decoding layers include, in sequence, a self-attention module, a cross-attention module, and a feedforward neural network.

[0073] The self-attention module performs linear transformation on the input vector to obtain queries, keys, and values, and in a plurality of parallel heads, respectively calculates attention scores, and after splicing the outputs of all heads, performs linear mapping, and then does residual connection with the input for normalization fusion.

[0074] The cross-attention module uses the output features of the self-attention module as queries, and maps the pre-generated static content into keys and values for cross-modal attention calculation. After calculation, the output features of the cross-attention module are connected with the output features of the self-attention module, and finally normalized fusion is performed to output the output features of the cross-attention module.

[0075] The feedforward neural network is composed of two layers of linear transformation and an intermediate nonlinear activation function. After calculation, the same normalization fusion step is performed.

[0076] In order to prevent overfitting, Dropout is used after the above self-attention module, cross-attention module, and feedforward neural network.

[0077] The multi-decoding layer Transformer model constructs a loss function based on cross-entropy loss, uniqueness penalty, and ranking consistency penalty.

[0078] The cross-entropy loss includes the difference between the task ID output by the multi-decoding layer Transformer model and the preset true target sequence, which is represented by the following formula:

[0079] ;

[0080] where, represents the numerical value of cross-entropy loss, which quantifies the error between the model prediction distribution and the target real distribution. The symbol represents the batch size, i.e., the number of samples input to the model at a time, while represents the length of the task ID sequence in each sample. The index and correspond to the index of the sample within the batch and the index of the position in the sequence, respectively. The variable represents the real task label of the th sample at position , while is the probability predicted by the model that the task at this position belongs to a particular class. The natural logarithm function is used to transform the probability distribution, and the normalization factor outside the integral ensures that the loss value will not be unreasonably biased due to changes in batch size and sequence length, thus promoting equal treatment of each task ID position during model training.

[0081] The uniqueness penalty includes: penalizing repeated task IDs in the generated sequence, represented by the following formula:

[0082] ;

[0083] where, represents the numerical value of the uniqueness penalty, reflecting the model's generation of repeated tasks. The variable still represents the batch size, while is the time index in the task sequence, ranging from 1 to . Among them, represents the task ID generated by the th sample at time , which is compared with the set of previously generated tasks . The indicator function returns 1 when the condition is met (i.e., the current task ID already exists in the previously generated set), and 0 otherwise. After summing over all samples and sequence positions, the is normalized to measure the overall proportion of repeated task IDs. Through this penalty term, the likelihood of the model generating repeated tasks can be effectively reduced, thus ensuring the uniqueness of the task scheduling sequence;

[0084] The sorting consistency penalty includes: penalizing task IDs with inconsistent scheduling order, represented by the following formula:

[0085] ;

[0086] wherein, quantifies the deviation of the model-predicted task sequence from the real task sequence at each time position. The variable still represents the batch size, is the position index of the task sequence. The symbol represents the task ID predicted by the model for the th sample at position , while represents the actual real task ID at this position. By calculating the absolute error between the predicted value and the real value, this loss term can truly reflect the ordering difference between the generated sequence and the target sequence. After outer summation and normalization, this loss term constrains the model to order, thereby pushing the generated task sequence and the real order to be as consistent as possible;

[0087] The above loss terms are weighted and combined to form the loss function:

[0088] ;

[0089] wherein, and are the weight coefficients of the corresponding penalty terms.

[0090] Step 3: Based on the autoregressive generation strategy, repeatedly input the task ID obtained this time and the fused features as the input of the Transformer model to obtain the score and task ID for the next step of battery disassembly until the disassembly of the battery is completed.

[0091] The task ID autoregressive generation strategy includes initial input setting, step-by-step generation mechanism, repeated task shielding, and predicted update generation sequence.

[0092] The initial input setting includes setting a dynamic input for the multi-decoding layer Transformer model and setting the initial input of the dynamic input as a zero vector.

[0093] The step-by-step generation mechanism includes inputting the task ID output by the multi-decoding layer Transformer model at the current step as input.

[0094] The repeated task shielding includes adjusting the corresponding value of the logits vector output by the multi-decoding layer Transformer model at the current step in the corresponding generated task ID set to negative infinity for shielding processing.

[0095] The logits vector is the original output of the last linear layer of the multi-decoding layer Transformer model, which is a real number vector without normalization processing. After normalization processing, the logits vector becomes a probability distribution, and the final output task ID is obtained based on the probability distribution.

[0096] The predicted update generation sequence includes: selecting the task ID corresponding to the highest probability value logits vector in the logits vector after the repeated task masking processing as the output of the current step.

[0097] The multi-decoding layer Transformer model combines sine and cosine functions to encode the fusion features to obtain the position information of each data point in the multi-source data.

[0098] The position encoding of the fusion features combined with the sine and cosine functions includes: generating an encoding formula using the sine and cosine functions, respectively, and using the sine encoding formula to encode the position of each position and even index of the embedding dimension in the fusion features, and using the cosine encoding formula to encode the position of each position and odd index of the embedding dimension.

[0099] The sine encoding formula is represented by the following formula:

[0100] ;

[0101] wherein, represents the fusion features encoded by the sine encoding formula, represents the position, represents the even index of the embedding dimension, represents the dimension of the input vector.

[0102] The cosine encoding formula is represented by the following formula:

[0103] ;

[0104] wherein, represents the fusion features encoded by the cosine encoding formula, represents the odd index of the embedding dimension.

[0105] Based on the above steps, the battery disassembly path optimization task is trained and tested by combining the traditional genetic algorithm, particle swarm optimization algorithm, and Transformer model with the present application. The goal is to minimize the total time and cost in the disassembly process while following the task order constraints. Please refer to Table 1.

[0106] Table 1: Comparison results of the present application and various models

[0107]

[0108] Specifically, the proposed method significantly reduces the total time required for disassembly, reducing about 18%, 15%, and 11% compared to genetic algorithm, particle swarm optimization, and basic Transformer, respectively. This result shows that by optimizing task scheduling order through autoregressive generation model, the proposed method effectively reduces the waiting time and time waste between tasks. In terms of disassembly cost, the proposed method reduces about 20%, 17%, and 15% compared to genetic algorithm, particle swarm optimization, and basic Transformer, respectively. This result shows that the proposed method can effectively reduce equipment usage and resource consumption during disassembly, thereby improving the economy of disassembly. Most importantly, the total objective function value of the proposed method is 378, which is about 20%, 17%, and 14% lower than genetic algorithm, particle swarm optimization, and basic Transformer, respectively, showing its superiority in considering disassembly time and cost comprehensively.

[0109] The advantages of the proposed method mainly lie in several aspects. First, the autoregressive generation strategy is one of the core features of the proposed method. By generating task IDs step by step, the generated tasks are dynamically shielded at each generation process, avoiding repeated generation of tasks and ensuring the uniqueness of the task sequence. This strategy significantly improves the quality of task scheduling and avoids the problem of repeated tasks that may occur in traditional methods. Second, the proposed method fully integrates static disassembly data and dynamic task scheduling data through deep feature fusion and uses the Transformer autoregressive generation model for task order prediction. Unlike traditional methods that only optimize the path through heuristic search, deep learning models can automatically learn the complex relationships between tasks and optimize the task order. Third, cross-modal feature fusion is a highlight of the proposed method. By effectively combining information from static and dynamic data sources, the proposed method not only retains key information from each data source but also enhances the model's understanding of the complexity of the disassembly path, thereby improving the effectiveness of the scheduling strategy. Finally, the proposed method introduces fine optimization of the objective function, combining cross-entropy loss, uniqueness penalty, and ranking consistency penalty to ensure the uniqueness, reasonableness, and accuracy of the task sequence, thereby further improving the quality of task scheduling.

[0110] In summary, the proposed power battery disassembly path optimization system based on autoregressive generation model exhibits significant advantages in disassembly path optimization compared to traditional genetic algorithm, particle swarm optimization, and basic Transformer methods. The autoregressive generation strategy, deep feature fusion, cross-modal feature processing, and fine objective function design enable the proposed method to effectively generate task ID sequences with high efficiency, uniqueness, and reasonableness, thereby promoting the overall optimization of battery disassembly paths. Therefore, the proposed method provides a new and efficient optimization means for the battery disassembly and recycling field, with high application prospects.

[0111] The preferred embodiments of the present application have been described above in detail. It should be understood that modifications and variations to the present application can be affected by those skilled in the art without departing from the scope of the application. Accordingly, it is intended that all of the subject matter of the above description and the claims be interpreted to encompass all such modifications and changes.

Claims

1. A power battery disassembly path optimization method based on autoregressive generation strategy, characterized in that: The steps include: Step 1: Obtain multi-source data of the battery to be disassembled, and normalize the multi-source data to obtain battery data; Step 2: Input the battery data into the multi-decoding layer Transformer model, perform dimensionality reduction processing on the battery data through the multi-decoding layer Transformer model, obtain multiple copies of the reduced dimensionality data, and then splice and fuse the multiple copies of the reduced dimensionality data to obtain fused features. The multi-decoding layer Transformer model obtains the task ID and score of the battery to be disassembled based on the fused features, and completes the next step of battery disassembly based on the score and task ID; Step 3: Based on the autoregressive generation strategy, the task ID and fused features obtained this time are repeatedly used as inputs to the multi-decoding layer Transformer model to obtain the score and task ID for the next battery disassembly until the battery disassembly is completed; The multi-decoding layer Transformer model combines sine and cosine functions to perform position encoding on the fusion features to obtain the position information of each data point in the multi-source data and gradually generate the task ID; The position encoding of the fused features by combining the sine and cosine functions includes: using the sine and cosine functions to generate encoding formulas respectively, using the sine encoding formula to position encode the even index of each position and embedding dimension in the fused features, and using the cosine encoding formula to position encode the odd index of each position and embedding dimension; The sinusoidal encoding formula is expressed by the following formula: ； in, represents the fusion feature after being encoded by the sinusoidal coding formula, Indicates location, represents an even index of the embedding dimension, Represents the dimension of the input vector; The cosine encoding formula is expressed by the following formula: ； in, Represents the fusion feature after encoding by the cosine coding formula, Odd-numbered indices representing embedding dimensions.

2. The power battery disassembly path optimization method based on autoregressive generation strategy according to claim 1 is characterized in that: In step 1, the obtaining of multi-source data of the battery to be disassembled includes: obtaining multi-source data of the battery to be disassembled from a designated folder; The multi-source data is composed of data contents in a specified folder, and the data contents include fromto data in the fromto folder and machining data in the machining folder. The fromto data includes the positions of the starting module and the ending module, and the machining data includes module ID, equipment ID, disassembly time and disposal cost.

3. The power battery disassembly path optimization method based on autoregressive generation strategy according to claim 2 is characterized in that: The normalization process includes alignment and MinMax normalization process, wherein the alignment is based on the number of data rows and performs alignment process on the fromto data and the machining data; The MinMax normalization process is expressed by the following formula: ； in, is the normalized multi-source data, is any input variable to be processed by MinMax normalization in multi-source data. is the minimum value in the column data where the input variable is located. It is the maximum value in the column data where the input variable is located.

4. The power battery disassembly path optimization method based on autoregressive generation strategy according to claim 2 is characterized in that: In step 2, the dimensionality reduction process includes Emberdding processing for fromto data and fully connected linear mapping for machining data; The dimensionality reduction process of the fromto data includes: mapping the fromto data to a continuous vector space to obtain an embedding vector, and mapping the embedding vector to a feature space with half the dimension of the model's latent vector through a fully connected layer; The dimensionality reduction process of machining data includes linear mapping of machining data through a fully connected layer.

5. The power battery disassembly path optimization method based on autoregressive generation strategy according to claim 4 is characterized in that: The splicing and fusing of multiple dimensionality reduction data to obtain fusion features includes: splicing the fromto data after dimensionality reduction with the machining data on the dimensional features, and then further mapping through a fully connected layer to obtain fusion features.

6. The power battery disassembly path optimization method based on autoregressive generation strategy according to claim 1 is characterized in that: The multi-decoding layer Transformer model includes: a Transformer model composed of several decoding layers; The decoding layer includes a self-attention module, a cross-attention module and a feedforward neural network in sequence, and a Dropout operation is added after each sub-module of the self-attention module, the cross-attention module and the feedforward neural network.

7. The power battery disassembly path optimization method based on autoregressive generation strategy according to claim 6, characterized in that: The multi-decoding layer Transformer model constructs a loss function based on cross entropy loss, uniqueness penalty, and sort consistency penalty; The cross entropy loss includes: the gap between the task ID output by the multi-decoding layer Transformer model and the preset true target sequence; The uniqueness penalty includes: penalizing duplicate task IDs in the generated sequence; The sort consistency penalty includes: punishing task IDs with inconsistent scheduling orders.

8. The power battery disassembly path optimization method based on the autoregressive generation strategy according to any one of claims 1 to 7, characterized in that: The autoregressive generation strategy includes: initial input setting, step-by-step generation mechanism, repeated task shielding, and prediction update generation sequence; The initial input setting includes: setting dynamic input for the multi-decoding layer Transformer model, and setting the initial input of the dynamic input to a zero vector; The step-by-step generation mechanism includes: taking as input the task ID output by the current step of the multi-decoding layer Transformer model; The repeated task shielding includes: adjusting the corresponding value of the logits vector output by the current step of the multi-decoding layer Transformer model in the corresponding generated task ID set to negative infinity for shielding; The prediction update generation sequence includes: selecting the task ID corresponding to the logits vector with the highest probability value from the logits vector after repeated task masking as the output of the current step.

Images