Waste battery recovery process parameter optimization method based on artificial intelligence

Abstract

The invention discloses a waste battery recovery process parameter optimization method based on artificial intelligence, and relates to the technical field of waste battery recovery, and the method comprises the steps: data collection: collecting aging characteristic data of a waste battery in real time by means of a sensor network; according to the method, dynamic factors such as battery aging and equipment abrasion are tracked accurately in real time by building the dynamic model, recycling process parameters can be adjusted flexibly, the stability and consistency of the recycling effect are improved greatly, the advanced algorithm is combined with the efficient data processing technology, the process parameters are optimized accurately, and the recycling efficiency is improved. According to the real-time operation state of the equipment, relevant parameters are intelligently adjusted, the equipment is prevented from operating under unreasonable working conditions, the equipment loss is reduced, the recovery cost is reduced, multiple innovative technologies are fused, the adaptability of the model is enhanced, and it is ensured that the recovery efficiency of the equipment is improved in a complex and changeable recovery scene. And all-around optimization of the waste battery recycling process can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of waste battery recycling, in particular to a waste battery recycling process parameter optimization method based on artificial intelligence. BACKGROUND

[0002] Under the background of the current emphasis on environmental protection and sustainable use of resources, the significance of waste battery recycling is increasingly prominent. Waste batteries are diverse, with common types including lead-acid batteries, lithium-ion batteries, and nickel-hydrogen batteries. These batteries are widely used in automobiles, electronic devices, energy storage systems, and other fields. With the update of technology and the end of the use cycle, the production of waste batteries is increasing exponentially. Waste batteries contain a wealth of valuable metals. Recycling and reusing them can effectively alleviate resource shortages, reduce overexploitation of primary mineral resources, and reduce ecological damage caused by mining activities. Optimizing waste battery recycling process parameters in the waste battery recycling process plays a crucial role in the entire recycling process. Precise parameter optimization can improve recycling efficiency, improve resource recycling quality, reduce energy consumption and cost in the recycling process, and reduce secondary pollution caused by improper recycling.

[0003] However, the current waste battery recycling technology still has some defects. Traditional recycling process parameters are mostly determined based on fixed experience values or experimental data under specific static conditions. However, in actual recycling processes, the aging degree of waste batteries varies, and the internal structure and chemical properties of different batteries differ significantly. At the same time, due to long-term high-intensity operation of recycling equipment, key components will inevitably wear out, and the operating efficiency and performance of the equipment will change. However, existing technologies cannot respond to these dynamic changes in a timely and effective manner, which leads to unstable resource recovery rates and significant fluctuations in the quality of recycled products during the recycling process, making it difficult to meet the strict requirements of industrial production for raw material quality. In order to maintain recycling operations, more energy and raw materials must be consumed, increasing the cost of recycling and reducing overall economic efficiency. Therefore, it is necessary to propose a waste battery recycling process parameter optimization method based on artificial intelligence to solve the problems in the existing technology. SUMMARY

[0004] The purpose of the present application is to overcome the shortcomings of the prior art and provide a waste battery recycling process parameter optimization method based on artificial intelligence. It can track dynamic changes such as battery aging and recycling equipment wear by building a dynamic model, accurately adjust recycling process parameters, ensure the stability and consistency of recycling results, improve resource recycling efficiency, and reduce energy consumption and cost.

[0005] To solve the above technical problems, the present application provides the following technical solution: a waste battery recycling process parameter optimization method based on artificial intelligence, which comprises the following steps: Data collection: Real-time collection of aging characteristic data of waste batteries through sensor network, covering battery internal resistance , open circuit voltage , cumulative charge and discharge times , and operating parameters of the recycling equipment, including the temperature , vibration frequency , torque of key components of the equipment; Dynamic model construction: Based on the self-attention spatio-temporal fusion network SSTFN, a dynamic model is constructed. This network uses a self-attention mechanism to adaptively assign weights to different time points and different feature dimensions, and fuses time series and spatial distribution features. The battery aging characteristic data, equipment operating parameters, and historical recycling process parameters are used as inputs to construct the initial model structure. Model training: The model is trained using the improved gradient boosting algorithm IGBA. During training, the loss function between the predicted value and the actual optimal process parameter is minimized , where is the predicted value and is the actual value, to enable the model to accurately predict the optimal process parameters under different working conditions. Real-time optimization: The real-time collected data is input into the trained model, and the model outputs the real-time optimized recycling process parameters, including the crushing power , the sorting electric field intensity , and the reaction kettle temperature . The recycling system automatically adjusts the equipment operating state based on these parameters. Feedback update: Based on the actual recycling effect feedback data, the model is updated online to continuously optimize the model prediction capability and adapt to the dynamic changes in the recycling process.

[0006] Further, in the data collection step, for the battery aging characteristic data, a multi-sensor fusion technology is used to fuse the battery internal resistance, open circuit voltage, and cumulative charge and discharge times data obtained by different types of sensors. The specific fusion method is as follows: Kalman filtering algorithm is used to process the sensor data, different weights are given according to the measurement accuracy and stability of the sensors, and the fused data value is calculated. The fused battery internal resistance , where is the battery internal resistance measured by the i th sensor, is the corresponding weight, is the number of sensors.

[0007] Further, in the data acquisition step, for the operating parameters of the recycling equipment, an outlier detection algorithm is used to preprocess the collected data, and whether each data point is an outlier is judged by calculating the local outlier factor LOF of the data, and the specific calculation formula is: Wherein is a data point, is a neighborhood set, is the reachable distance to , and the LOF value of the data point is greater than the preset threshold value, which is determined as an outlier and is corrected and removed.

[0008] Further, when constructing the self-attention spatio-temporal fusion network SSTFN, a spatio-temporal position encoding module is introduced to encode the time and space position information of the input data. The time position encoding generates time series features of different frequencies using sine and cosine functions, and the space position encoding allocates weights according to the data feature dimension. In the training process of the self-attention spatio-temporal fusion network SSTFN, an adversarial training mechanism is introduced to construct an adversarial network, which is trained with the main model to confuse the prediction of the main model, and the main model improves the prediction accuracy to interfere with the adversarial network.

[0009] Further, in the model training process, when using the improved gradient boosting algorithm IGBA, an adaptive learning rate adjustment strategy is introduced, which dynamically adjusts the learning rate according to the change of the loss function in the training process. When the loss function decreases rapidly, the learning rate is increased to speed up the convergence, and when the loss function decreases slowly or fluctuates, the learning rate is reduced. The specific adjustment formula is: Wherein is the learning rate of the step, is the initial learning rate, is the adjustment factor, is the difference between the step and the step loss function.

[0010] Further, in the real-time optimization step, after the model outputs the recycling process parameters, a parameter verification mechanism based on risk assessment is used to evaluate the possible device failure risk and recycling effect deviation risk caused by the output parameters according to the historical data and real-time state of the equipment operation, and calculate the risk assessment index Wherein is the probability of the risk, is the corresponding risk loss, is the number of risk types, and the risk assessment index exceeds the preset threshold value, the model output parameter is corrected and re-predicted.

[0011] Further, in the real-time optimization step, considering the coupling relationship between multiple process parameters in the recycling process, a parameter collaborative adjustment strategy based on multi-objective optimization is adopted to establish a multi-objective optimization function , are different optimization objectives, including improving metal recovery rate and reducing energy consumption, the multi-objective optimization function is solved by a non-dominated sorting genetic algorithm to obtain a set of optimal solutions, and a process parameter combination meeting the actual demand is selected from the solution set.

[0012] Further, in the feedback updating step, an incremental learning and knowledge distillation combined method is used to update the model online, when new recycling effect feedback data is obtained, the new data and part of the historical data are combined to form an incremental data set, and an incremental learning algorithm is used to fine-tune the model, and at the same time, the knowledge of the trained large model is migrated to the lightweight small model through the knowledge distillation technology.

[0013] Further, in the feedback updating step, the model is optimized by combining the idea of reinforcement learning, the prediction process of the model is regarded as a decision-making process of an agent in the environment, and the recycling effect is fed back to the agent as a reward, and by continuously adjusting the model parameters, the agent obtains the maximum cumulative reward in the long-term decision-making process, and the reward function is defined as , is the metal recovery rate, is the purity of the recycled product, is the recycling cost, , , is a weight coefficient, and the reinforcement learning algorithm is used to continuously optimize the model parameters to maximize the reward function value.

[0014] Compared with the prior art, the waste battery recycling process parameter optimization method based on artificial intelligence has the following beneficial effects:

[0015] The present application can flexibly adjust the recycling process parameters by constructing a dynamic model to track dynamic factors such as battery aging and equipment wear and tear in real time and accurately, and can greatly improve the stability and consistency of recycling effect, and by using advanced algorithms combined with efficient data processing technology, the potential value behind the data can be deeply mined, the process parameters can be accurately optimized, and the resource recovery rate can be significantly improved, according to the real-time running state of the equipment, the related parameters can be intelligently adjusted, the equipment can be prevented from running under unreasonable working conditions, the equipment wear and tear can be effectively reduced, the service life of the equipment can be prolonged, the recycling cost can be significantly reduced, and by integrating various innovative technologies, the adaptability of the model is enhanced, and the waste battery recycling process can be optimized in all aspects in complex and variable recycling scenarios.

[0016] Additional advantages, objects, and features of the application will be apparent to those skilled in the art upon examination of the following detailed description, it being understood that each BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0018] Figure 1 Flow chart of the artificial intelligence-based waste battery recycling process parameter optimization method. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be described clearly and completely below, obviously, the described embodiments are only some embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0020] Embodiment one

[0021] In a large waste battery recycling plant, waste batteries from different regions and different sources are received every day, including lithium ion batteries used in electric vehicles, nickel hydrogen batteries in various electronic devices, and traditional lead-acid batteries, etc. These waste batteries have significant differences in battery type, battery aging degree, initial state, etc. And the equipment in the recycling plant, such as the crusher, the sorting machine, the chemical leaching equipment, etc., is in a long-term high-load operation state, resulting in different degrees of wear of the equipment parts. In order to ensure that the waste battery recycling process can be carried out efficiently, stably and economically, it is necessary to optimize the recycling process parameters, at which time the artificial intelligence-based waste battery recycling process parameter optimization method of the present application can be applied.

[0022] At the starting point of the waste battery recycling process, a complete sensor system is arranged to comprehensively collect data of the waste batteries entering the recycling process, and special sensors are used to measure the internal resistance , open circuit voltage and cumulative charge-discharge times and other aging characteristic data of the batteries, at the same time, corresponding sensors are installed on the key components of the recycling equipment, such as the core components of the crusher, the transmission parts of the sorting machine and the key elements of the reaction kettle, to monitor the temperature of the key components of the equipment , vibration frequency , and torque and other operating parameters, these sensors continuously collect data to provide raw information for subsequent process parameter optimization.

[0023] For battery aging characteristic data, multi-sensor fusion technology is used. For multiple sensor data measuring the same attribute, such as multiple sensors measuring battery internal resistance, their data is respectively represented as , represents the th sensor, according to the accuracy and stability of each sensor, the corresponding weight is given, and the fused battery internal resistance data is calculated by formula to ensure the accuracy and reliability of the data.

[0024] For the operating parameters of the recycling equipment, an outlier detection algorithm is used for data preprocessing. For any data point in the equipment operating parameters, the local outlier factor LOF is calculated by formula , where is the neighborhood set of , and is the reachable distance from to . When the value of exceeds the preset threshold, the data point is marked as an outlier and processed accordingly to ensure data quality.

[0025] A dynamic model based on self-attention spatio-temporal fusion network SSTFN is constructed. In the construction process, a spatio-temporal position encoding module is introduced. For the input data, the time position encoding uses sine and cosine functions to generate time series features of different frequencies, and the spatial position encoding allocates weights according to the data feature dimension, so that the model can better capture the change law of data in the spatio-temporal dimension.

[0026] A large amount of waste battery recycling historical data is collected, including characteristic data of different types of waste batteries in different aging states, and operating data of recycling equipment in different wear states corresponding to them, as well as the corresponding historical recycling process parameters. These data are integrated into a training data set and divided into a training set, a validation set and a test set.

[0027] The improved gradient boosting algorithm IGBA is used to train the constructed dynamic model. During the training process, the loss function is calculated according to the difference between the predicted value and the actual value , and the loss function is calculated by formula The learning rate is adaptively adjusted, where For the first The learning rate of the step. The initial learning rate, To adjust the factor, For the first Step and the first The difference in the step loss function is used to achieve efficient training and optimization of the model.

[0028] During the training process, an adversarial training mechanism is also introduced, in which the constructed adversarial network is trained against the main model. The adversarial network aims to confuse the predictions of the main model by generating interference data, while the main model continuously optimizes its own parameters to improve its resistance to interference, thereby enhancing the robustness and generalization ability of the model.

[0029] In the actual process of recycling used batteries, real-time battery status data and equipment status data are input into a pre-trained dynamic model.

[0030] The model will output real-time optimized recycling process parameters, including but not limited to crushing power. Sorting electric field strength and reactor temperature For these parameters, considering the coupling relationships between them, a multi-objective optimization function is constructed. The function is solved by a non-dominated sorting genetic algorithm to obtain a set of optimal solutions. The combination of process parameters that best fits the current recycling conditions and equipment status is selected from this solution set.

[0031] A risk assessment-based parameter validation mechanism is adopted, and a risk assessment index is calculated for each predicted combination of process parameters. ,in For the first The probability of such risks occurring To mitigate the corresponding risk of loss, if the risk assessment indicators exceed the preset threshold, the combination of process parameters will be adjusted or re-predicted to ensure that the recycling equipment operates under safe and stable conditions.

[0032] After completing a batch of waste battery recycling, feedback data is collected based on the actual recycling effect, such as metal recovery rate and purity of recycled products. This data is used as a new information source and combined with some historical data to form an incremental dataset.

[0033] We employ a combination of incremental learning and knowledge distillation to update the model online. Incremental learning is used to fine-tune the model, while knowledge distillation is used to transfer knowledge from the trained large model to a lightweight small model. This reduces computational costs and storage requirements while ensuring model performance, enabling the model to better adapt to dynamic changes during the recycling process.

[0034] At the same time, the prediction process of the model is combined with reinforcement learning, and the recovery effect is fed back to the model as a reward by defining a reward function , wherein is the metal recovery rate, is the purity of the recovered product, is the recovery cost, , , is the weight coefficient, and the model parameters are constantly adjusted to make the model tend to select the process parameter combination that can bring the maximum cumulative reward in the long-term decision-making process.

[0035] The effect brought by the embodiment: Through the embodiment, the waste battery recycling plant realizes the intelligentization and dynamic optimization of the waste battery recycling process. First, due to the use of multi-sensor fusion and outlier detection technology in the data collection stage, the quality of the input data is guaranteed, providing a reliable data foundation for subsequent optimization. The innovative techniques in the dynamic model construction and training process, such as adaptive learning rate adjustment and adversarial training, enable the model to better learn the complex dynamic relationships in the waste battery recycling process, and have stronger adaptability and prediction accuracy for different types and states of batteries and different degrees of equipment wear. In the real-time optimization stage, the parameter verification mechanism based on risk assessment ensures the safe and stable operation of the recycling equipment, avoiding equipment failure and poor recovery effect caused by unreasonable process parameters. The multi-objective optimization function and the corresponding solving algorithm ensure that the recycling process can consider multiple important indicators simultaneously, achieving comprehensive optimization of resource recovery rate, energy consumption, and cost. Through incremental learning, knowledge distillation, and reinforcement learning in the feedback update stage, the model can continuously update and optimize itself, continuously improving the recovery effect, reducing costs, and extending the service life of equipment, ultimately achieving efficient, stable, environmentally friendly, and economic comprehensive benefits of waste battery recycling, and improving the sustainability and market competitiveness of the entire recycling process.

[0036] Embodiment Two

[0037] In a medium-sized mobile device waste battery recycling processing center, mainly responsible for processing waste lithium ion batteries from mobile phones, tablet computers, notebook computers and other small electronic devices. Since these waste batteries come from a wide range of sources, their battery aging conditions differ greatly, and the equipment in this recycling center is relatively compact, requiring more precise adjustments to process parameters. At the same time, considering the pressure of market competition, the recycling center needs to reduce costs as much as possible while ensuring high recovery rate and high-quality recovered products, and improve the utilization efficiency of equipment. Therefore, the waste battery recycling process parameter optimization method based on artificial intelligence of the present application is adopted to achieve its goal.

[0038] At the entrance of the recycling center, specially designed miniaturized, high-precision sensors are installed to accurately measure the aging characteristics of the incoming used lithium-ion batteries, including their internal resistance. Open circuit voltage and cumulative charge / discharge cycles Meanwhile, to ensure data comprehensiveness, multiple sensors will be installed in different parts of the battery to obtain more accurate information on its internal state. For recycling equipment, due to space constraints, miniature sensors are used to monitor the temperature of key components. Vibration frequency and torque These sensors can collect data at a high frequency based on operating parameters, providing detailed information for subsequent processing.

[0039] For multi-sensor fusion of battery aging characteristic data, considering the measurement characteristics and environmental adaptability of different sensors, different weights are assigned to each sensor data based on its reliability and accuracy. Through formula This is used to calculate the fused battery internal resistance data, ensuring the accuracy and consistency of the collected battery aging data.

[0040] For the operating parameters of the recycling equipment, outlier detection algorithms are used to preprocess the data. In this small recycling center, because the equipment has a low tolerance for anomalies, stricter outlier detection standards are adopted for the data points in the equipment operating parameters. Using formula Calculate the Local Outlier Factor (LOF), once If the value exceeds a preset lower threshold, the data point will be processed immediately to prevent abnormal data from affecting subsequent operations.

[0041] A dynamic model is constructed based on the Self-Attention Spatiotemporal Fusion Network (SSTFN). In this model, the characteristics of the waste batteries processed by the center and the compactness of the equipment are taken into account. In the spatiotemporal location coding module, the parameters of time location coding and spatial location coding are precisely adjusted according to the time and space characteristics of the recycling process, so that the model can better capture the spatiotemporal variation patterns of the data in this specific scenario.

[0042] Historical waste battery recycling data from the center were collected, including aging data of waste lithium-ion batteries from different batches and sources, data of equipment under different operating conditions, and corresponding historical recycling process parameters. A dedicated training dataset was constructed, which was then divided into training, validation, and test sets according to a certain ratio to meet the needs of model training and evaluation.

[0043] An improved gradient boosting algorithm, IGBA, is used to train the dynamic model. During training, the model is trained based on the predicted values. and actual values calculate the loss function In order to complete the model training quickly under limited computing resources, a more aggressive adjustment strategy is adopted during adaptive learning rate adjustment, i.e. adjusting the learning rate according to the formula while closely monitoring the changes in the loss function, ensuring that the model can converge to a better performance level in a shorter time.

[0044] In the adversarial training phase, according to the actual data characteristics of the center, a targeted adversarial network is designed to generate interference that is more in line with actual situations, and the main model is trained in an adversarial manner to enhance the robustness and generalization ability of the model in handling small electronic device waste batteries.

[0045] In daily waste battery recycling operations, real-time collected battery status data and device status data are input into the trained dynamic model.

[0046] The model outputs real-time optimized recycling process parameters, such as crushing power , sorting electric field intensity and reaction kettle temperature , etc. Considering the sensitivity of the center's equipment to process parameter changes, a more detailed multi-objective optimization function is constructed, where the optimization objectives include more accurate metal recovery rate, purer recovered product quality and lower energy consumption, etc. Non-dominated sorting genetic algorithm is used to solve this function to obtain the optimal solution that meets the center's needs.

[0047] For the process parameters output by the model, a risk assessment-based parameter verification mechanism is adopted to calculate the risk assessment index Due to the compactness of the center's equipment, the risk tolerance is relatively weak, and a more conservative risk threshold is set. Once the risk assessment index approaches or exceeds the threshold, the parameter adjustment mechanism is immediately activated to ensure that the equipment operates within a safe and reliable range, avoiding any possible equipment damage and production interruption.

[0048] After completing a batch of waste battery recycling, the recovered products are subjected to detailed quality testing, and detailed recycling effect data including metal recovery rate, recovered product purity, etc. are collected. These data are combined with part of the historical data to form an incremental data set.

[0049] Incremental learning and knowledge distillation are used to update the model. In the incremental learning process, the step size and frequency of incremental learning are adjusted according to the actual data volume and update frequency of the center to ensure that the model can quickly adapt to new data patterns. At the same time, through knowledge distillation technology, the knowledge of complex models is transferred to lightweight small models that are more suitable for the center's computing resources, to improve the update efficiency.

[0050] The model is optimized by reinforcement learning, and the reward function is wherein is the metal recovery rate, is the purity of the recovered product, is the recovery cost, , , is a weight coefficient, and the model parameters are adjusted according to the recovery effect at each feedback update, so that the model can continuously optimize the process parameters in the direction of maximizing the reward when processing waste batteries in the center.

[0051] Effects brought by the embodiment: By applying the embodiment in the medium-scale mobile device waste battery recycling processing center, a series of significant effects are achieved. First, the fine data collection ensures accurate information input for small electronic device waste batteries, providing strong support for optimization. In the model construction and training stage, through optimization for the scene, the model can quickly and accurately converge under limited computing resources, and has stronger adaptability when facing waste batteries and devices of different sources and states and subtle changes. The fine multi-objective optimization and conservative risk assessment mechanism in the real-time optimization stage ensures fine adjustment of the equipment under the premise of ensuring quality and safety, improves the operation efficiency and product quality of the equipment, and the personalized adjustment in the feedback update stage enables the model to continuously optimize and adapt to new recycling conditions. Overall, the center significantly improves the recovery rate of metals in waste batteries, reduces energy consumption in the recycling process, and reduces costs caused by equipment failure or low efficiency. In the fierce market competition, it occupies an advantage, and also makes a positive contribution to environmental protection and resource recycling, realizing efficient, economic, environmentally friendly and sustainable development of waste battery recycling.

[0052] It is apparent to those skilled in the art that the present application is not limited to the details of the foregoing exemplary embodiments, and that the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. Therefore, the embodiments should be considered in all respects as illustrative and not restrictive, and the scope of the present application should be defined by the appended claims rather than the above description, and it is intended to encompass all changes falling within the meaning and range of equivalents of the technical features of the claims. Any reference signs in the claims should not be considered as limiting the claims to which they belong.

Claims

1. An artificial intelligence based process parameter optimization method for recycling of spent batteries, characterized in that, The method comprises the following steps: Data collection: With the help of sensor network, real-time collection of aging characteristic data of waste batteries, covering battery internal resistance , open circuit voltage , cumulative charge and discharge times , and operating parameters of recycling equipment, including the temperature , vibration frequency , torque ; Dynamic model construction: a dynamic model is constructed based on a self-attention spatio-temporal fusion network SSTFN, the network uses a self-attention mechanism to adaptively assign weights of different time points and different feature dimensions, fuses time series and spatial distribution features, and takes battery aging feature data, equipment operation parameters and historical recycling process parameters as inputs to construct an initial model structure; Model training: the model is trained by using the improved gradient boosting algorithm IGBA, and in the training, the loss function between the predicted value and the actual optimal process parameter is minimized continuously adjust the model parameters, wherein the predicted value is the actual value, so that the model accurately predicts the optimal process parameter under different working conditions; Real-time optimization: input real-time collected data into the trained model, the model outputs real-time optimized recovery process parameters, including crushing power , sorting electric field intensity , reaction kettle temperature , the recovery system automatically adjusts the equipment operating state according to these parameters; Feedback updating: the model is updated online according to actual recycling effect feedback data, the prediction ability of the model is continuously optimized, and the dynamic changes in the recycling process are adapted.

2. The artificial intelligence based process parameter optimization method for recycling of waste batteries as claimed in claim 1 wherein, In the data acquisition step, for the battery aging characteristic data, a multi-sensor fusion technology is adopted to fuse the battery internal resistance, open-circuit voltage and cumulative charge-discharge times data obtained by different types of sensors. The specific fusion method is as follows: the Kalman filtering algorithm is used to process the sensor data, different weights are given according to the measurement accuracy and stability of the sensors, the fused data value is calculated, and the fused battery internal resistance , wherein is the battery internal resistance measured by the first sensor, is the corresponding weight, is the number of sensors. 3.The artificial intelligence-based process parameter optimization method for waste battery recycling according to claim 1, wherein, In the data acquisition step, for the operating parameters of the recycling equipment, the collected data is preprocessed by using an outlier detection algorithm, and whether each data point is an outlier is judged by calculating the local outlier factor LOF of the data, and the specific calculation formula is: Wherein is the data point, is the neighbor set of , is the reachable distance from , and the LOF value of the data point is greater than the preset threshold value, which is determined as an outlier and is corrected and removed. 4.The artificial intelligence-based process parameter optimization method for waste battery recycling according to claim 1, wherein, When the self-attention spatio-temporal fusion network SSTFN is constructed, a time-space position coding module is introduced to code the time and space position information of the input data, the time position coding generates time series features of different frequencies by using sine and cosine functions, and the space position coding allocates weights according to the feature dimension of the data. 5.The artificial intelligence-based process parameter optimization method for waste battery recycling according to claim 1, wherein, In the model training process, when the improved gradient boosting algorithm IGBA is used, an adaptive learning rate adjustment strategy is introduced. According to the change of the loss function in the training process, the learning rate is dynamically adjusted. When the loss function decreases rapidly, the learning rate is increased to speed up the convergence speed. When the loss function decreases slowly and fluctuates, the learning rate is reduced. The specific adjustment formula is: Wherein is the learning rate of the th step, is the initial learning rate, is the adjustment factor, is the difference between the loss function of the th step and the th step. 6.The artificial intelligence-based process parameter optimization method for waste battery recycling according to claim 1, wherein, In the real-time optimization step, after the model outputs the recovery process parameters, a parameter verification mechanism based on risk assessment is adopted to evaluate the device failure risk and recovery effect deviation risk that the output parameters may cause according to the historical data and real-time state of the device operation, and to calculate a risk assessment index wherein is the probability of the occurrence of the risk, is the corresponding risk loss, is the number of risk types, and if the risk assessment index exceeds a preset threshold, the model output parameters are modified and re-predicted. 7.The artificial intelligence-based process parameter optimization method for waste battery recycling according to claim 1, wherein, In the real-time optimization step, considering the coupling relationship between multiple process parameters in the recycling process, a parameter collaborative adjustment strategy based on multi-objective optimization is adopted to establish a multi-objective optimization function wherein are different optimization objectives, including improving metal recovery rate and reducing energy consumption, the multi-objective optimization function is solved by a non-dominated sorting genetic algorithm to obtain a set of optimal solutions, and a process parameter combination meeting the actual demand is selected from the solution set. 8.The artificial intelligence-based process parameter optimization method for waste battery recycling according to claim 1, wherein, In the training process of the self-attention spatio-temporal fusion network SSTFN, an adversarial training mechanism is introduced to construct an adversarial network, the main model is subjected to adversarial training to confuse the prediction of the main model, and the main model improves the prediction accuracy to interfere with the adversarial network. In the feedback updating step, the model is updated online in a combination mode of incremental learning and knowledge distillation, when there is new recycling effect feedback data, the new data and part of the historical data are combined to form an incremental data set, the model is fine-tuned by using an incremental learning algorithm, meanwhile, the knowledge of the trained large model is transferred to a light small model by using the knowledge distillation technology. 9.The artificial intelligence-based process parameter optimization method for waste battery recycling according to claim 1, wherein, In the feedback updating step, the model is optimized by combining the idea of reinforcement learning, the prediction process of the model is regarded as the decision-making process of an agent in the environment, and the recovery effect is regarded as the reward feedback to the agent. By continuously adjusting the model parameters, the agent obtains the maximum cumulative reward in the long-term decision-making process, and the reward function is defined wherein is the metal recovery rate, is the purity of the recovered product, is the recovery cost, , , is the weight coefficient, and the reinforcement learning algorithm is used to continuously optimize the model parameters to maximize the reward function value.

Images