Automobile retired power battery disassembly sequence optimization method and related device

Abstract

The invention provides an automobile retired power battery disassembly sequence optimization method and a related device, and relates to the technical field of data processing. According to the method, disassembling risk priori knowledge provided by a battery manufacturer is introduced, a multi-target disassembling sequence planning mathematical model integrated with risk perception is established, and a disassembling sequence planning algorithm based on a graph attention network and a constraint perception pointer network is adopted, so that end-to-end real-time planning of disassembling sequences of different types of battery packs is realized. According to the method, the prior disassembling knowledge of an official manual can be effectively utilized, the high efficiency and safety of the disassembling process are ensured, the method can adapt to a complicated battery pack structure and a dynamically changing disassembling scene, and a brand new solution is provided for disassembling the retired power battery.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method and related device for optimizing the disassembly sequence of retired automobile power batteries. Background Art

[0002] With the rapid development of the new energy vehicle industry, the recycling and disassembly of retired power batteries has become increasingly prominent. Retired power batteries contain a variety of hazardous substances, such as heavy metals and electrolytes. Improper handling can cause serious environmental pollution and waste valuable metal resources. Therefore, the efficient and safe disassembly of retired power batteries and the recycling of these resources have become crucial components of the sustainable development of the new energy vehicle industry.

[0003] However, the disassembly process of retired power batteries faces many challenges. First, the battery pack structure is complex, and different types of battery packs differ in terms of component composition, connection methods, etc., which makes disassembly sequence planning extremely complicated. Traditional disassembly methods often rely on manual experience and are difficult to adapt to the diverse battery pack structures. They are also prone to disassembly errors, resulting in low disassembly efficiency and may even cause safety accidents. Secondly, there are high risks in the disassembly process of retired power batteries. The chemical substances inside the battery may leak, burn or even explode due to improper operation, posing a serious threat to the personal safety of operators.

[0004] At present, although battery manufacturers provide disassembly task instructions and operating guidelines, relying solely on these static guidance documents is difficult to meet the individual differences of battery packs and the dynamic changes in the disassembly process. In the face of emergencies or abnormal battery pack structures, it is necessary to be able to quickly adjust the disassembly sequence to cope with complex and changing disassembly environments. In the existing technology, although some studies have attempted to plan disassembly sequences through optimization algorithms, most of these methods only focus on disassembly efficiency and ignore the risk factors in the disassembly process. Moreover, these methods often show poor adaptability and flexibility when faced with complex battery pack structures and dynamically changing disassembly scenarios. Therefore, there is an urgent need for a method that can comprehensively consider disassembly efficiency and disassembly risk and plan the disassembly sequence in real time to meet the actual needs of disassembly of retired power batteries. Summary of the Invention

[0005] In order to meet the actual needs of disassembling retired power batteries, the present application provides a method and related devices for optimizing the disassembly sequence of retired automobile power batteries.

[0006] In the first aspect, the present application provides a method for optimizing the disassembly sequence of retired vehicle power batteries using the following technical solutions: A method for optimizing the disassembly sequence of retired vehicle power batteries, comprising: According to the disassembly task instructions and operation guide provided by the manufacturer, the battery pack part numbers are used as nodes and an extended part priority graph is generated to generate a priority relationship matrix; Establish an optimization model that includes expressions for the number of disassembly direction changes and the number of disassembly tool changes; Using a pointer network model based on a graph attention network and a dynamic constrained mask attention mechanism, inputting the priority relationship matrix and the part feature matrix and obtaining an output result; The output results are subjected to an entropy weight approximation to an ideal solution sorting method to select a compromise optimal disassembly sequence.

[0007] Optionally, the step of establishing an optimization model including an expression for the number of disassembly direction changes and an expression for the number of disassembly tool changes includes: Establish the disassembly process direction change function , tool replacement function during disassembly And the disassembly process risk assessment function ; The optimization model based on the decomposition strategy is generated according to the disassembly process direction change function, the disassembly process tool replacement function and the disassembly process risk assessment function, wherein the disassembly of retired power batteries is performed by It is composed of parts. This is the disassembly sequence for retired power batteries. Each part is disassembled only once and meets the priority constraint relationship between parts. Indicates that the first Disassembled parts.

[0008] Optionally, the disassembly process direction conversion function include: Among them, one-hot encoding is used Define three directions as 、 and The operation directions during the disassembly process are up and down, left and right, and front and back.

[0009] Optionally, the disassembly process tool replacement function : Among them, one-hot encoding is used Define three tools as 、 and To represent the tools wrench, pliers and screwdriver used in the disassembly process.

[0010] Optionally, the disassembly process risk assessment function include: Where s represents the part risk factor.

[0011] Optionally, the optimization model is expressed as: in, , and the weight vector Each element in has been normalized and sums to 1.

[0012] Optionally, in the pointer network model based on the graph attention network and the dynamic constrained mask attention mechanism, the following is further included: A two-layer encoder is used to process part features. A gated recurrent unit network is first used to aggregate feature sequence information, and then a graph attention network is used to embed part features into a high-dimensional space. The decoder uses an attention mechanism with a dynamic mask mechanism to generate a disassembly sequence.

[0013] Optionally, before the step of inputting the priority relationship matrix and the part feature matrix and obtaining the output result, the method further includes: The model parameters are optimized through the central self-criticism reinforcement learning algorithm, and iterative learning is performed based on the policy gradient update rule.

[0014] Optionally, the step of optimizing model parameters by a central self-criticism reinforcement learning algorithm and performing iterative learning based on a policy gradient update rule includes: Standardize the notation of the multi-objective mathematical model of disassembly sequence programming into a reinforcement learning problem, is the state space, is the action space, each state is defined as The set of all parts that have been disassembled at the moment, that is ,action Defined as the selection of parts to be disassembled at the next time step t+1 ; The policy is defined as , indicating that the given A set of parts states that have been disassembled at a certain moment , the strategy returns the probability distribution of the next disassembled part, where Represents the trainable parameters of the neural network; The expected reward is defined as the negative of the subproblem objective function: in It is a space where parts are gathered. yes All possible permutations space, It is predicted by the neural network distribution of Objective function Defined as the expected reward, i.e. The gradient of the objective function is calculated as follows: in, is the batch size for training, The number of decomposition sequence solutions generated for each subproblem, For the The first sub-problem The objective function value of the disassembly sequence, As the baseline, For the Subproblem Given part feature input Next, output the disassembly sequence The probability of training is that the gradient ascent method is used to optimize the parameters : in, Is the learning rate, a positive scalar used to control the step size of parameter adjustment at each update. The learning rate determines the extent to which the algorithm advances in the gradient direction of the loss function. The selection of the policy gradient baseline adopts Central self-critic, the baseline The calculation formula is: in Indicates the reward obtained by using a greedy strategy to obtain the disassembly sequence, Indicates the adoption of the strategy The reward for randomly sampling the disassembly sequence.

[0015] In a second aspect, the present application provides a device for optimizing the disassembly sequence of retired vehicle power batteries, comprising: A matrix generation module is used to generate a priority relationship matrix by taking the battery pack part numbers as nodes and generating an extended part priority graph based on the disassembly task instructions and operation guide provided by the manufacturer; A model building module is used to establish an optimization model including an expression for the number of disassembly direction changes and the number of disassembly tool changes; A data input module is used to adopt a pointer network model based on a graph attention network and a dynamic constrained mask attention mechanism to input the priority relationship matrix and the part feature matrix and obtain an output result; The sequence determination module is used to select the compromise optimal disassembly sequence for the output results by using the entropy weight approximation ideal solution sorting method.

[0016] In summary, this application incorporates prior knowledge of disassembly risks provided by battery manufacturers, establishes a risk-aware, multi-objective disassembly sequence planning mathematical model, and employs a disassembly sequence planning algorithm based on a graph attention network and a constraint-aware pointer network to achieve end-to-end real-time planning of disassembly sequences for different battery pack categories. This approach not only effectively leverages prior disassembly knowledge from official manuals to ensure the efficiency and safety of the disassembly process, but also adapts to complex battery pack structures and dynamically changing disassembly scenarios, providing a novel solution for the disassembly of retired power batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of the first embodiment of the method for optimizing the disassembly sequence of retired vehicle power batteries of the present application; Figure 2 This is a model architecture diagram of the present application's method for optimizing the disassembly sequence of retired vehicle power batteries; Figure 3 It is the input feature vector graph of the regularization of the sub-problem of the method for optimizing the disassembly sequence of retired automobile power batteries in this application; Figure 4 This is a structural block diagram of the first embodiment of the present application's device for optimizing the disassembly sequence of retired automobile power batteries. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is further described in detail below through the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0019] The present application embodiment provides a method for optimizing the disassembly sequence of retired vehicle power batteries. Figure 1 , Figure 1 This is a flow chart of the first embodiment of the method for optimizing the disassembly sequence of retired automobile power batteries of the present application.

[0020] In this embodiment, the method for optimizing the disassembly sequence of retired vehicle power batteries includes the following steps: Step S10: According to the disassembly task instructions and operation guide provided by the manufacturer, the battery pack part numbers are used as nodes and an extended part priority graph is generated to generate a priority relationship matrix.

[0021] It's important to note that for the widespread availability of retired power batteries, manufacturers typically provide an official disassembly task description and detailed operating instructions, providing a relatively safe and guiding path for manual disassembly. These instructions detail the battery pack disassembly steps and systematically number the components to ensure a standardized and safe disassembly process. This study uses these numbers as nodes in a parts priority graph, integrating the manufacturer's disassembly guidelines with actual disassembly requirements. This lays the foundation for the subsequent development of a multi-objective disassembly sequence planning model.

[0022] To address the high risk of power battery disassembly, this embodiment introduces a core dimension of risk prioritization. By directly linking the order of component disassembly with potential hazards, a "risk-driven disassembly topology" is formed. For example, when disassembling positive electrode components, by forcibly disconnecting the logical node connected to the positive electrode busbar, the transmission path of the electric shock risk is completely blocked, achieving a paradigm shift from "passive risk avoidance" to "active design safety."

[0023] Therefore, in the process of constructing the part priority relationship diagram, the physical constraint relationships between parts are first accurately identified based on CAD drawings or manual operation experience. Based on these constraint relationships and following the part numbering rules mentioned above, the mutually constrained parts are connected through directed arrows, with the arrow direction pointing from the constraint imposing party to the constrained party, thereby constructing a standard part priority relationship diagram. On this basis, a risk priority constraint mechanism is further introduced, and finally an extended part priority diagram is formed. This innovative extended relationship diagram not only ensures the logic and systematic nature of the disassembly process, but also significantly improves the safety and reliability of the disassembly process, providing more scientific guidance for subsequent disassembly planning and operations.

[0024] Step S20: establishing an optimization model including an expression for the number of disassembly direction changes and an expression for the number of disassembly tool changes.

[0025] It should be noted that the step of establishing an optimization model including the expression of the number of disassembly direction changes and the number of disassembly tool replacement expressions includes: establishing a disassembly process direction change function , tool replacement function during disassembly And the disassembly process risk assessment function ; The optimization model based on the decomposition strategy is generated according to the disassembly process direction change function, the disassembly process tool replacement function and the disassembly process risk assessment function, wherein the disassembly of retired power batteries is performed by It is composed of parts. This is the disassembly sequence for retired power batteries. Each part is disassembled only once and meets the priority constraint relationship between parts. Indicates that the first Disassembled parts.

[0026] The disassembly process direction transformation function include: Among them, one-hot encoding is used Define three directions as 、 and The operation directions during the disassembly process are up and down, left and right, and front and back.

[0027] The disassembly process tool replacement function : Among them, one-hot encoding is used Define three tools as 、 and To represent the tools wrench, pliers and screwdriver used in the disassembly process.

[0028] It should be noted that the internal structure of the power battery adopts a crisscross arrangement structure, and the adjacent components have a high coupling characteristic. In view of this, if non-continuous disassembly operations are implemented, the safety risk of the disassembly operation will be significantly increased. For this reason, battery manufacturers usually provide standardized disassembly processes as a reference. In addition, different parts have different risk levels due to differences in structure and function, so a part risk factor is specially introduced for this purpose. This indicator. Its definition rules are mainly based on potential safety accidents. The higher the risk coefficient value, the greater the safety risk that the component may bring during the disassembly process. Special attention should be paid during the disassembly process. The disassembly parts should be as consistent as possible with the official instructions. Based on the risk coefficient of the disassembled parts, in order to minimize the risk of disassembly operations, the degree of consistency between the disassembly sequence and the official manual is used as the risk assessment standard. A disassembly process risk assessment function is established. include: Where s represents the part risk factor.

[0029] Based on the decomposition strategy, a multi-objective optimization mathematical model is obtained, and weights are set for each objective. The multi-objective problem is decomposed into single-objective sub-problems with different weights. By solving the optimal solution of each sub-problem, the Pareto frontier of the problem is finally obtained by merging and non-dominated sorting. Assume that the three objective preferences of the decomposition sequence programming problem are , weight vector Each element in has been normalized and summed to 1, where larger values ​​indicate higher importance. The optimization model is expressed as: .

[0030] Step S30: adopting a pointer network model based on a graph attention network and a dynamic constrained mask attention mechanism, inputting the priority relationship matrix and the part feature matrix and obtaining an output result.

[0031] It should be noted that traditional iterative search algorithms have significant limitations when dealing with battery disassembly sequence planning problems. Not only do these algorithms struggle to meet the requirements of real-time disassembly sequence planning, but even slight changes in input features require algorithm redesign or re-iteration, resulting in poor generalization. In contrast, pointer networks, as a network model specifically designed for solving sequence outputs, have significant advantages. Once trained, pointer networks can quickly solve results in a short period of time and exhibit good generalization performance for different problems. Therefore, pointer networks are highly applicable to battery disassembly sequence planning problems.

[0032] In the specific implementation, a multi-objective optimization mathematical model is obtained based on the decomposition strategy, and weights are set for each objective. The multi-objective problem is decomposed into single-objective sub-problems with different weights. By solving the optimal solution of each sub-problem, the Pareto frontier of the problem is finally obtained by merging the non-dominated sorting. Assume that the three objective preferences of the decomposition sequence programming problem are , weight vector Each element in has been normalized and summed to 1, where larger values ​​represent higher importance. Therefore, the weight vector Next disassembly sequence The scalar objective function is expressed as follows: in, Represents the given retired power battery part feature matrix, target weight vector and part priority constraint relationship state, Indicates that in a given feature matrix Time disassembly sequence The goal of pointer network training is to learn a random strategy , given the retired power battery parts feature matrix, target weight vector and parts priority constraint relationship Next, give the disassembly sequence with smaller objective function value Output higher probability, pointer network decomposes probability using chain rule as follows: Each factor on the right side of the equation All by the pointer network The output of the layer is obtained.

[0033] It should be noted that further improvements are needed to ensure that the output of the pointer network can satisfy the priority relationship of the extended parts priority graph. To this end, this embodiment constructs a pointer network model that introduces a graph attention network and a dynamic constrained mask attention mechanism. The graph attention network can efficiently process graph structured data and can assign differentiated weights to different nodes in the neighborhood, so as to better adapt to neighborhoods of different sizes. The dynamic constrained mask attention mechanism can update the evaluation score of the attention mechanism during the decoding process to ensure that the generated disassembly sequence strictly follows the disassembly order constraints. Through these improvements, this embodiment not only improves the applicability of the pointer network in battery disassembly sequence planning, but also significantly enhances the real-time and generalization capabilities of the model. Figure 2 This is the model architecture of the embodiment.

[0034] This embodiment uses GRU-GAT dual-layer encoding to promote effective learning of part features and ensure full utilization of information. Since the risk assessment function of the disassembly sequence multi-objective programming mathematical model is related to the order of input, and GRU has similar functions to LSTM but with fewer training parameters, the GRU network is selected to aggregate input feature information and input feature sequence information, and store the part features in a vector of fixed size. Use the graph attention network (GAT) to process the part priority relationship matrix , to extract the priority relationship information between parts.

[0035] Encoder input includes part feature information (in Quantity of parts, is the feature dimension of each part) and the part priority relationship matrix .

[0036] In order to use a single model to solve the sub-problems decomposed under different weights, it is necessary to The weights of the sub-problems are then appended, and the specific definition rules for the encoder GRU network input are as follows: a. The feature vector of the part and the weight vector of the target are stitched together to form the input vector. The order of the targets corresponding to the elements in the feature vector should be the same as the order of the weight vector; b. In the feature vector, the feature dimensions associated with each target are kept consistent. If the feature dimension of one part is smaller than that of the other parts, the smaller feature dimension needs to be padded based on the maximum feature dimension. The padding value is set to 1.

[0037] like Figure 3The regularized input feature vector for the subproblem consists of the disassembly tool, disassembly direction, and risk factor. The disassembly tool and direction are three-dimensional one-hot encoded, while the risk factor is a real number, so it needs to be padded with 1 to form a three-dimensional vector. The weight vector attached to the end of the part features for the same subproblem is the same. By constructing rules, the order and position of the elements in the input vector are given practical meaning, increasing the amount of information provided to the model. Furthermore, the target weight is also included in the input vector, ensuring that the information sent to all subsequent modules includes the target weight.

[0038] The GAT network processes the part feature matrix to obtain high-dimensional features of each part As shown in the formula: in Indicates parts high-dimensional features.

[0039] After GRU processing, the fusion features of the parts are obtained As the final output of the encoder.

[0040] In the specific implementation, the description of the decoder is as follows: In order to minimize the consumption of training model, this embodiment adopts the RNN model decoder of gated recurrent unit (GRU), stores the node embedding information of all inputs into the decoder (i.e., the mixed feature information of all parts), and uses the node embedding information of the last disassembled part. , and the decoder hidden state at the previous time , the decoder can output a decoding vector containing information about all parts disassembled so far, that is, the hidden state at the current moment , used as The input of the attention mechanism at this moment.

[0041] It should be noted that the attention mechanism includes: Based on the fusion features and decoder hidden state at all times , output the probability vector of disassembly parts in the next time step ,The working principle of the attention mechanism with the introduction of dynamic priority constrained mask is as follows: Initialize the decoder's hidden state (usually using the encoder's last state or other initialization methods), and for each time step (From 1 to N) perform the following steps: Step 1: Based on the current decoder hidden state and the fusion features of the encoder , calculate the attention score (without mask): in, is the transpose of the query vector, is the fusion feature output by the encoder The weight of is the hidden state of the decoder output at each moment The weights of , all three are learnable parameters of the neural network.

[0042] Step 2: To satisfy the priority constraint relationship between parts in the generated extended part priority graph, obtain the dynamic mask of the current time step through the dynamic mask generation mechanism , add the mask to the attention score: Step 3: Layer gets the current time The probability distribution of selecting the next part to be disassembled.

[0043] Step 4: Based on the probability distribution Select the part to be disassembled , sampling and greedy methods are used during training, and greedy strategy is used during testing.

[0044] Step 5: Update the disassembly status and remove the parts The disassembly state is set to 1, that is, , and update the undetached collection .

[0045] Step 6: Embed the selected parts Input to the decoder GRU to update the hidden state.

[0046] Return to step 1 until all parts are disassembled.

[0047] Each time step of this embodiment The specific process of dynamic mask generation under the following conditions is: let the priority relationship matrix be ,in Indicates disassembly parts Before disassembling the parts , that is, there is a line from arrive The directed edge of yes At the beginning of disassembly, all parts are not disassembled. We maintain a disassembly state vector , initially Indicates parts Not disassembled, set to 1 after disassembly. Define the set of currently unassembled parts as , then for each part , calculate the number of its predecessor parts that have not been disassembled as shown in the following formula.

[0048] Note that the parts here It's a spare part The front parts, so when and (Component Not disassembled), parts will be added Number of pre-installed parts that have not been disassembled . Then the set of detachable parts is: Dynamic Mask is a vector of the same length as the number of parts and is defined as follows: .

[0049] It should be noted that, given the difficulty in obtaining labels for optimal solutions to battery pack disassembly sequence planning, traditional supervised learning methods, which rely on large amounts of labeled data, face the challenges of a large initial workload and are difficult to adapt to when solving unknown optimal solutions. This paper employs an unsupervised learning approach, combined with a central self-critical reinforcement learning algorithm, to train the neural network model.

[0050] The actor-critic model is developed from the policy gradient method and is designed as a hybrid model to leverage both policy-based and value-based reinforcement learning methods. Generally speaking, there are two main components in the actor-critic model. The actor is responsible for generating the policy and then deciding what action the agent should take next. The actor in this embodiment is the proposed improved pointer network, which provides a complete disassembly sequence. The critic uses a value-based approach to evaluate the actions taken by the agent. In the disassembly sequence planning problem, the actor provides the disassembly sequence, and the critic receives the sequence and provides a reward value for the sequence. This reward value can be used as an approximate reward value when calculating the loss function to update the weights of the two networks.

[0051] The self-criticism approach uses a single artificial neural network model, rather than the typical two-actor and critic model. In this architecture, the artificial neural network model acts as both an actor and a critic. When acting as an actor, the agent randomly selects the next city to visit from a probability distribution generated by the model. In this case, the model returns the generated itinerary and the probability distribution for the next city. When the model is acting as a critic, the agent no longer randomly selects the next city to visit, but instead chooses the city with the highest probability and generates a reward corresponding to the unraveling sequence.

[0052] This strategy uses two different methods to generate disassembly sequences, obtained using the same artificial neural network: one is randomly generated, and the other is the best disassembly sequence under the greedy strategy. Based on this method, the difference between the reward of the randomly sampled disassembly sequence and the central self-criticism baseline is calculated, and the gradient is calculated to update the network parameters. Compared with using the exponential moving average of the reward, using the central self-criticism baseline accelerates convergence. The specific steps are as follows: In the pointer network model based on graph attention network and dynamic constrained mask attention mechanism, it also includes: using a two-layer encoder to process part features, first using a gated recurrent unit network to aggregate feature sequence information, and then entering the graph attention network to embed part features into a high-dimensional space, and the decoder uses an attention mechanism with a dynamic mask mechanism to generate a disassembly sequence.

[0053] In a specific implementation, before the step of inputting the priority relationship matrix and the part feature matrix and obtaining the output result, it also includes: optimizing the model parameters through the central self-criticism reinforcement learning algorithm, and performing iterative learning based on the policy gradient update rule.

[0054] The steps of optimizing model parameters by using a central self-criticism reinforcement learning algorithm and performing iterative learning based on a policy gradient update rule include: standardizing the multi-objective mathematical model of disassembling sequence planning into a symbol of a reinforcement learning problem, is the state space, is the action space, each state is defined as The set of all parts that have been disassembled at the moment, that is ,action Defined as the selection of parts to be disassembled at the next time step t+1 ; The policy is defined as , indicating that the given A set of parts states that have been disassembled at a certain moment , the strategy returns the probability distribution of the next disassembled part, where Represents the trainable parameters of the neural network; The expected reward is defined as the negative of the subproblem objective function: in It is a space where parts are gathered. yes All possible permutations space, It is predicted by the neural network distribution of Objective function Defined as the expected reward, i.e. The gradient of the objective function is calculated as follows: in, is the batch size for training, The number of decomposition sequence solutions generated for each subproblem, For the The first sub-problem The objective function value of the disassembly sequence, As the baseline, For the Subproblem Given part feature input Next, output the disassembly sequence The probability of training is that the gradient ascent method is used to optimize the parameters : in, Is the learning rate, a positive scalar used to control the step size of parameter adjustment at each update. The learning rate determines the extent to which the algorithm advances in the gradient direction of the loss function. The selection of the policy gradient baseline adopts Central self-critic, the baseline The calculation formula is: in Indicates the reward obtained by using a greedy strategy to obtain the disassembly sequence, Indicates the adoption of the strategy The reward for randomly sampling the disassembly sequence.

[0055] Step S40: selecting a compromise optimal disassembly sequence for the output results using an entropy weight approximation to an ideal solution sorting method.

[0056] In the specific implementation, the trained model can obtain the Pareto front optimal solution in a short time after being given the input part input feature information and the part priority constraint matrix. For this purpose, the entropy weight TOPSIS evaluation and decision method is used to select a compromise optimal solution as the final disassembly sequence plan.

[0057] Each solution in the Pareto front has three indicator values, namely the number of disassembly operation direction changes, the number of disassembly tool changes, and the risk level. To eliminate the dimensional differences between different indicators, this study uses the range standardization method to normalize the data so that the data of each indicator are unified to the same scale to ensure the accuracy and reliability of data analysis.

[0058] in, Indicates the The Pareto front solution corresponds to the An indicator value.

[0059] The entropy weight method is a highly objective indicator weight assignment method that can calculate the weight of each indicator by analyzing the observed values ​​of each indicator. The steps are as follows: Step 1: Assume the battery pack has Pareto front solutions, Item part index, according to the above index quantification method to obtain the original data matrix ; Step 2: Calculate the contribution matrix of each part status indicator , map the value of each indicator to between 0 and 1: Step 3: Calculate the The information entropy of an indicator represents the total contribution of the indicator: Step 4: Calculate the normalized coefficient of variation: Step 5: Calculate indicator weights: Step 6: To extract the best solution, continue to use the TOPSIS model to comprehensively score the fitness value of the Pareto frontier solution, and select the solution with the highest score as the best solution. The distance calculation between the fitness of a solution and the positive ideal solution is shown in the formula, and the distance calculation between the fitness of a solution and the negative ideal solution is shown below.

[0060] in, and They are The minimum and maximum values ​​of the indicators (in the normalized matrix).

[0061] Step 7: Calculate relative proximity, i.e., part recycling efficiency And sort them as shown below. The greater the relative closeness, the closer the part is to the ideal solution and the farther it is from the negative ideal solution, so the recycling benefit is higher. Sort the recycling benefits of each part by size and determine the order of recycling parts.

[0062] .

[0063] It should be noted that this embodiment addresses the increased disassembly risk caused by the complex internal structure and high degree of coupling of power batteries, as well as the discontinuous nature of disassembly operations. By proposing a multi-objective disassembly sequence planning method with integrated risk awareness, this method addresses the increased disassembly risk caused by the complex internal structure and high degree of coupling of power batteries, as well as the discontinuous nature of disassembly operations. Given that battery packs are hazardous recycling items, the disassembly process must not only maximize efficiency but also ensure safety. This embodiment innovatively constructs a multi-objective disassembly sequence planning mathematical model with integrated risk awareness. Furthermore, addressing the potential demand for real-time disassembly sequence planning in the actual recycling and disassembly of retired power batteries, a real-time battery disassembly sequence planning algorithm based on a graph attention network and a constraint-aware pointer network was designed.

[0064] Compared with traditional disassembly sequence planning techniques, this method not only introduces disassembly risk priority constraints and a multi-objective optimization model, effectively reducing the safety risks caused by improper disassembly order, but also ensures disassembly efficiency and emphasizes the advantage of real-time solution. Through an algorithm design based on a graph attention network (GAT) and a constraint-aware pointer network, this method can achieve end-to-end real-time planning and quickly output non-dominated disassembly sequence solutions. This real-time solution capability ensures that in dynamic disassembly scenarios, the model can adjust and optimize the disassembly sequence in real time based on the current disassembly state and constraints, thereby better adapting to complex disassembly environments and improving the flexibility and responsiveness of the disassembly process.

[0065] This example incorporates prior knowledge of disassembly risks provided by battery manufacturers to establish a risk-aware, multi-objective disassembly sequence planning mathematical model. This model, using a disassembly sequence planning algorithm based on a graph attention network and a constraint-aware pointer network, achieves end-to-end, real-time planning of disassembly sequences for different battery pack categories. This model not only effectively leverages prior disassembly knowledge from official manuals to ensure efficient and safe disassembly, but also adapts to complex battery pack structures and dynamically changing disassembly scenarios, providing a novel solution for the disassembly of retired power batteries.

[0066] Reference Figure 4 , Figure 4 This is a structural block diagram of the first embodiment of the device for optimizing the disassembly sequence of retired automobile power batteries of the present application.

[0067] like Figure 4 As shown, the device for optimizing the disassembly sequence of retired automobile power batteries proposed in the embodiment of the present application includes: The matrix generation module 10 is used to generate a priority relationship matrix by using the battery pack part numbers as nodes and generating an extended part priority graph according to the disassembly task instructions and operation guide provided by the manufacturer; A model building module 20 is used to establish an optimization model including an expression for the number of disassembly direction changes and an expression for the number of disassembly tool changes; A data input module 30 is configured to use a pointer network model based on a graph attention network and a dynamic constrained mask attention mechanism to input the priority relationship matrix and the part feature matrix and obtain an output result; The sequence determination module 40 is used to select the optimal compromise disassembly sequence for the output results by using an entropy weight approximation ideal solution sorting method.

[0068] It should be understood that the above is only an example and does not constitute any limitation to the technical solution of the present application. In specific applications, technicians in this field can make settings as needed, and the present application does not impose any restrictions on this.

[0069] This example incorporates prior knowledge of disassembly risks provided by battery manufacturers to establish a risk-aware, multi-objective disassembly sequence planning mathematical model. This model, using a disassembly sequence planning algorithm based on a graph attention network and a constraint-aware pointer network, achieves end-to-end, real-time planning of disassembly sequences for different battery pack categories. This model not only effectively leverages prior disassembly knowledge from official manuals to ensure efficient and safe disassembly, but also adapts to complex battery pack structures and dynamically changing disassembly scenarios, providing a novel solution for the disassembly of retired power batteries.

[0070] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this application. In actual applications, technicians in this field can select part or all of it according to actual needs to achieve the purpose of this embodiment scheme, and no restrictions are imposed here.

[0071] In addition, for technical details not fully described in this embodiment, please refer to the method for optimizing the disassembly sequence of retired automobile power batteries provided in any embodiment of the present application, and will not be repeated here.

[0072] In addition, it should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or system comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or system. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or system comprising the element.

[0073] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0074] Through the above description of the embodiments, those skilled in the art will clearly understand that the above-mentioned embodiments and methods can be implemented using software plus the necessary general-purpose hardware platform. Of course, hardware can also be used, but in many cases the former is a more preferred embodiment. Based on this understanding, the technical solution of this application, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / RAM, a magnetic disk, or an optical disk) and includes several instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of this application. The above are only preferred embodiments of this application and do not limit the scope of the patent application. Any equivalent structure or equivalent process transformation made using the contents of this application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the scope of patent protection of this application.

Claims

1. A method for optimizing the disassembly sequence of retired vehicle power batteries, characterized in that: include: According to the disassembly task instructions and operation guide provided by the manufacturer, the battery pack part numbers are used as nodes and an extended part priority graph is generated to generate a priority relationship matrix; Establish an optimization model that includes expressions for the number of disassembly direction changes and the number of disassembly tool changes; Using a pointer network model based on a graph attention network and a dynamic constrained mask attention mechanism, inputting the priority relationship matrix and the part feature matrix and obtaining an output result; The output results are subjected to an entropy weight approximation to an ideal solution sorting method to select a compromise optimal disassembly sequence.

2. The method according to claim 1, characterized in that The step of establishing an optimization model including an expression for the number of disassembly direction changes and an expression for the number of disassembly tool changes includes: Establish the disassembly process direction change function , tool replacement function during disassembly And the disassembly process risk assessment function ; The optimization model based on the decomposition strategy is generated according to the disassembly process direction change function, the disassembly process tool replacement function and the disassembly process risk assessment function, wherein the disassembly of retired power batteries is performed by It is composed of parts. The disassembly sequence of retired power batteries is as follows: Each part is disassembled only once and meets the priority constraint relationship between parts. Indicates that the first Disassembled parts.

3. The method according to claim 2, characterized in that The disassembly process direction transformation function include: Among them, one-hot encoding is used Define three directions as 、 and The operation directions during the disassembly process are up and down, left and right, and front and back.

4. The method according to claim 2, characterized in that The disassembly process tool replacement function : Among them, one-hot encoding is used Define three tools as 、 and To represent the tools wrench, pliers and screwdriver used in the disassembly process.

5. The method according to claim 1, characterized in that Disassembly process risk assessment function include: Where s represents the part risk factor.

6. The method according to claim 2, characterized in that The optimization model is expressed as: in, , and the weight vector Each element in has been normalized and sums to 1.

7. The method according to claim 1, characterized in that In the pointer network model based on graph attention network and dynamic constrained mask attention mechanism, it also includes: A two-layer encoder is used to process part features. A gated recurrent unit network is first used to aggregate feature sequence information, and then a graph attention network is used to embed part features into a high-dimensional space. The decoder uses an attention mechanism with a dynamic mask mechanism to generate a disassembly sequence.

8. The method according to claim 1, characterized in that Before the step of inputting the priority relationship matrix and the part feature matrix and obtaining the output result, the method further includes: The model parameters are optimized through the central self-criticism reinforcement learning algorithm, and iterative learning is performed based on the policy gradient update rule.

9. The method according to claim 8, characterized in that The steps of optimizing model parameters through the central self-criticism reinforcement learning algorithm and performing iterative learning based on the policy gradient update rule include: Standardize the notation of the multi-objective mathematical model of disassembly sequence programming into a reinforcement learning problem, is the state space, is the action space, each state is defined as The set of all parts that have been disassembled at the moment, that is ,action Defined as the selection of parts to be disassembled at the next time step t+1 ; The policy is defined as , indicating that the given A set of parts states that have been disassembled at a certain moment , the strategy returns the probability distribution of the next disassembled part, where Represents the trainable parameters of the neural network; The expected reward is defined as the negative of the subproblem objective function: in It is a space where parts are gathered. yes All possible permutations space, It is predicted by the neural network distribution of Objective function Defined as the expected reward, i.e. The gradient of the objective function is calculated as follows: in, is the training batch size, The number of decomposition sequence solutions generated for each subproblem, For the The first sub-problem The objective function value of the disassembly sequence, As the baseline, For the Subproblem Given part feature input Next, output the disassembly sequence The probability of training is that the gradient ascent method is used to optimize the parameters : in, Is the learning rate, a positive scalar used to control the step size of parameter adjustment at each update. The learning rate determines the extent to which the algorithm advances in the gradient direction of the loss function. The selection of the policy gradient baseline adopts Central self-critic, the baseline The calculation formula is: in Indicates the reward obtained by using a greedy strategy to obtain the disassembly sequence, Indicates the adoption of the strategy The reward for randomly sampling the disassembly sequence.

10. A disassembly sequence optimization device for retired vehicle power batteries, characterized in that: include: A matrix generation module is used to generate a priority relationship matrix by taking the battery pack part numbers as nodes and generating an extended part priority graph based on the disassembly task instructions and operation guide provided by the manufacturer; A model building module is used to establish an optimization model including an expression for the number of disassembly direction changes and the number of disassembly tool changes; A data input module is used to adopt a pointer network model based on a graph attention network and a dynamic constrained mask attention mechanism to input the priority relationship matrix and the part feature matrix and obtain an output result; The sequence determination module is used to select the compromise optimal disassembly sequence for the output results by using the entropy weight approximation ideal solution sorting method.

Images