Intelligent optimization control system and method for battery material recovery and preparation process

Abstract

The invention discloses an intelligent optimization control system and method for a battery material recovery and preparation process, and relates to the technical field of intelligent control. Training sample data and a training quality prediction model in the battery material recovery and preparation process in a test environment are collected, and then a state space, an action space and a reward function of a deep reinforcement learning model are constructed; fusing the prediction result of the quality prediction model and the monitoring data of the sensor in a test environment through a feature fusion strategy, generating a fusion process representation vector, inputting the fusion process representation vector into a deep reinforcement learning model, training the deep reinforcement learning model, and obtaining a fusion process representation vector; based on the real-time sensor data, the quality prediction model and the deep reinforcement learning model, obtaining an action output by the deep reinforcement learning model, and performing parameter regulation and control on the to-be-monitored recovery preparation process by taking the action as an optimization strategy; the production energy consumption is effectively reduced, and the resource utilization efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to an intelligent optimization control system and method for a battery material recycling preparation process. Background Art

[0002] The traditional battery material recycling and preparation process has many shortcomings, such as cumbersome process parameter control, large fluctuations in product quality, and high energy consumption. An intelligent and automated solution is urgently needed to optimize and control the entire recycling and preparation process, improve resource utilization, and reduce production costs.

[0003] At present, the battery material recycling and preparation process mainly relies on manual experience for parameter setting and process control. This traditional method has obvious defects. Due to the complexity and uncertainty of influencing factors, it is difficult for humans to accurately grasp the optimal parameter combination of each link, which often leads to unstable product quality and inefficient use of energy. In addition, manual control has poor response flexibility and is difficult to respond to sudden changes in working conditions in a timely manner, thus affecting production efficiency.

[0004] The Chinese patent with the authorization announcement number CN117810570B discloses a recycling and discharge control method for waste lithium batteries and a related device, which classifies waste lithium batteries based on feature labels generated by cluster analysis of state parameters of waste lithium batteries to obtain a number of waste lithium batteries of each category; obtains real-time images of a number of waste lithium batteries and determines the coordinates of their positive and negative pole positions; places a number of waste lithium batteries of corresponding categories in corresponding positions of a battery discharge device based on the coordinates of the positive and negative pole positions for discharge; in the process of discharging, a number of monitoring sensors are combined to obtain corresponding monitoring data, and the working parameters of the discharge device are adjusted based on the monitoring data; at the same time, the residual power of each waste lithium battery is continuously measured based on the monitoring data, and when the residual power reaches a preset interval, the discharge process is terminated; however, this method fails to achieve the control of recycling parameters to improve the quality of recycled materials;

[0005] To this end, the present invention proposes an intelligent optimization control system and method for a battery material recycling and preparation process. Summary of the invention

[0006] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes an intelligent optimization control system and method for the battery material recycling preparation process, which effectively reduces production energy consumption and improves resource utilization efficiency.

[0007] To achieve the above purpose, an intelligent optimization control method for battery material recycling preparation process is proposed, comprising the following steps:

[0008] Step 1: Use multimodal sensing technology to collect training sample data in the battery material recycling preparation process in the test environment;

[0009] Step 2: Based on the training sample data, train a quality prediction model with product quality indicators as output;

[0010] Step 3: Based on the prediction results of the quality prediction model, construct the state space of the deep reinforcement learning model, and construct the action space and reward function of the deep reinforcement learning model;

[0011] Step 4: The prediction results of the quality prediction model and the monitoring data of the sensor in the test environment are integrated through the feature fusion strategy to generate a fusion process representation vector;

[0012] Step 5: Input the fusion process representation vector into the deep reinforcement learning model to train the deep reinforcement learning model;

[0013] Step 6: For the recycling preparation process to be monitored, real-time sensor data is collected through multimodal sensing technology in real time, and the action output by the deep reinforcement learning model is obtained based on the real-time sensor data, quality prediction model and deep reinforcement learning model; this action is used as an optimization strategy to adjust the parameters of the recycling preparation process to be monitored.

[0014] The collection of the training sample data comprises the following steps:

[0015] Step 11: Prepare N batches of waste battery positive and negative electrode materials of different sources and compositions; N is the number of selected batches;

[0016] Step 12: In the test environment, the complete recycling and preparation process is run for each batch of waste battery positive and negative electrode materials, and the monitoring data of each sensor is recorded in real time, and various controllable key parameters during the operation are simultaneously recorded;

[0017] Step 13: Measure and record the index value of the quality index of the battery material obtained after extraction and recovery as a quality index label;

[0018] Step 14: The monitoring data collected in real time by each sensor of all batches of waste battery positive and negative electrode materials, the real-time controllable key parameters and quality indicator labels are used to form training sample data.

[0019] The training of the quality prediction model with the product quality index as output based on the training sample data comprises the following steps:

[0020] Step 21: Convert the training sample data into sample input features;

[0021] The step of converting the training sample data into sample input features comprises the following steps:

[0022] Step 211: for each batch of waste battery positive and negative electrode materials, pre-process the corresponding monitoring data of the sensor at each moment, and perform feature splicing on the pre-processed monitoring data and controllable key parameters to obtain a feature splicing vector at each moment;

[0023] The feature concatenation vector is constructed as follows:

[0024] Perform vectorization operations on the monitoring data of each sensor at each moment to obtain the corresponding parameter feature vector;

[0025] For each controllable key parameter, use the respective parameter values ​​to form a controllable parameter feature vector;

[0026] The parameter feature vectors and controllable parameter feature vectors of all sensors of the positive and negative electrode materials of the waste battery at each moment are spliced ​​to obtain a feature splicing vector.

[0027] Step 212: preset a time window t; for each moment of each batch of waste battery positive and negative electrode materials during operation, collect the monitoring parameter sequence of the monitoring data of each sensor and the controllable parameter sequence of each controllable key parameter corresponding to the moment; the monitoring parameter sequence is a sequence composed of the parameter values ​​collected by the corresponding sensor within t moments before the moment, arranged in time order; the controllable parameter sequence is a parameter value sequence composed of the parameter values ​​of the corresponding controllable key parameter within t moments before the moment, arranged in time order;

[0028] Step 213: The monitoring parameter sequence and controllable parameter sequence at each moment constitute the sample input feature.

[0029] Step 22: Take the sample input features at each moment as input and the predicted value of the quality indicator at the subsequent moment t0 as output to train the quality prediction model; t0 is the preset time period.

[0030] The method of training the quality prediction model is:

[0031] The sample input features at each moment are used as the input of the quality prediction model, the quality prediction model takes the predicted value of the quality indicator at the subsequent moment t0 of the moment as the output, takes the quality indicator label corresponding to the subsequent moment t0 of the moment as the prediction target, takes the difference between the predicted value of the quality indicator and the quality indicator label as the prediction error, and takes minimizing the sum of prediction errors as the training goal; the quality prediction model is trained until the sum of prediction errors reaches convergence and the training is stopped; the quality prediction model is a time series prediction model.

[0032] The method of constructing the state space of the deep reinforcement learning model based on the prediction results of the quality prediction model, and constructing the action space and reward function of the deep reinforcement learning model is as follows:

[0033] The deep reinforcement learning model is set to an Actor-Critic model;

[0034] The state space is marked as S, and the state space S reflects the real-time state of the entire recycling preparation process. The state space is constructed as follows:

[0035] The recovery preparation process is segmented according to a fixed time window; the size of the fixed time window is t0;

[0036] For each time segment, the representation vector set composed of the fusion process representation vectors at each moment in time order is used as the state space;

[0037] The fusion process representation vector is constructed as follows:

[0038] Collect the feature splicing vectors corresponding to each moment in the recycling and preparation process of the positive and negative electrode materials of used batteries;

[0039] Input the feature concatenation vector at each moment into the quality prediction model to obtain the predicted value of the quality index at the subsequent moment t0 output by the quality prediction model;

[0040] A one-dimensional convolutional network is used to extract local feature sequences from the monitoring parameter sequences corresponding to the monitoring data of each sensor at each moment;

[0041] Input the local feature sequence into the LSTM model to obtain the global context encoding vector C_t;

[0042] The predicted value of the quality indicator and the global context encoding vector C_t are concatenated to obtain the fusion process representation vector.

[0043] The action space is marked as A, wherein the action space A includes the adjustment amount of the controllable key parameters output by the Actor model, and the parameter adjustment action taken in the next time segment is marked as a;

[0044] The reward function is marked as Q, and the reward value function Q=r(s, a) measures the effect of performing action a in the current state s;

[0045] The intelligent agent of the deep reinforcement learning model is designed to be an Actor-Critic structure based on RNN;

[0046] The input of the agent is state s;

[0047] The Critic network outputs the state value V(s);

[0048] The action probability π(a|s) is output by the Actor network.

[0049] The method of obtaining the action output by the deep reinforcement learning model based on the real-time sensor data, the quality prediction model and the deep reinforcement learning model is:

[0050] Construct a feature splicing vector through real-time sensor data and the parameter values ​​of various controllable key parameters of the current recycling preparation process;

[0051] Input the feature concatenation vector into the quality prediction model to obtain the predicted value of the quality index at the subsequent time t0 output by the quality prediction model;

[0052] Based on the predicted values ​​of the quality indicators and the monitoring data of various sensors, the corresponding fusion process representation vector is constructed;

[0053] Collect all fusion process representation vectors of the time segment at the current moment to form a corresponding representation vector set;

[0054] The representation vector set is input into the deep reinforcement learning model to obtain the action output by the Actor network of the deep reinforcement learning model.

[0055] An intelligent optimization control system for the battery material recycling preparation process is proposed, including a sample data collection module, a prediction model training module, an enhanced model training module, and an action decision module; wherein each module is electrically connected;

[0056] A sample data collection module, which uses multimodal sensing technology to collect training sample data in the battery material recycling preparation process in the test environment, and sends the training sample data to the prediction model training module;

[0057] The prediction model training module trains the quality prediction model with product quality indicators as output based on the training sample data, and sends the quality prediction model to the reinforcement model training module and the action decision module;

[0058] The reinforcement model training module constructs the state space of the deep reinforcement learning model based on the prediction results of the quality prediction model, and constructs the action space and reward function of the deep reinforcement learning model. The prediction results of the quality prediction model and the monitoring data of the sensor in the test environment are integrated through the feature fusion strategy to generate a fusion process representation vector, which is input into the deep reinforcement learning model to train the deep reinforcement learning model and send the deep reinforcement learning model to the action decision module.

[0059] The action decision module collects real-time sensor data through multimodal sensing technology for the recycling preparation process to be monitored, obtains the action output by the deep reinforcement learning model based on the real-time sensor data, quality prediction model and deep reinforcement learning model; uses this action as an optimization strategy to adjust the parameters of the recycling preparation process to be monitored.

[0060] An electronic device is proposed, comprising: a processor and a memory, wherein the memory stores a computer program that can be called by the processor;

[0061] The processor executes the above-mentioned intelligent optimization control method for the battery material recycling preparation process by calling the computer program stored in the memory.

[0062] A computer-readable storage medium is provided, on which a rewritable computer program is stored;

[0063] When the computer program is run on a computer device, the computer device executes the above-mentioned intelligent optimization control method for the battery material recycling preparation process.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] The present invention collects training sample data in a battery material recycling preparation process in a test environment by using multimodal sensing technology, trains a quality prediction model with product quality indicators as output based on the training sample data, constructs a state space of a deep reinforcement learning model based on the prediction result of the quality prediction model, and constructs an action space and a reward function of the deep reinforcement learning model, fuses the prediction result of the quality prediction model and the monitoring data of the sensor in the test environment through a feature fusion strategy, generates a fusion process representation vector, inputs the fusion process representation vector into the deep reinforcement learning model, trains the deep reinforcement learning model, collects real-time sensor data in real time through multimodal sensing technology for the recycling preparation process to be monitored, obtains the action output by the deep reinforcement learning model based on the real-time sensor data, the quality prediction model and the deep reinforcement learning model; uses the action as an optimization strategy to perform parameter control on the recycling preparation process to be monitored; uses big data modeling and machine learning technology to dig out the inherent laws that affect the preparation process, and then automatically generates the optimal parameter configuration, and seamlessly integrates it into the automatic control system, thereby completely solving the problems of dependence on manual experience and inefficient parameter control, effectively reducing production energy consumption and improving resource utilization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a flow chart of the intelligent optimization control method for the battery material recycling preparation process in Example 1 of the present invention;

[0067] Figure 2This is a module connection relationship diagram of the intelligent optimization control system for the battery material recovery and preparation process in Example 1 of the present invention;

[0068] Figure 3 This is a schematic diagram of the structure of an electronic device in Embodiment 3 of the present invention;

[0069] Figure 4 This is a schematic diagram of the computer-readable storage medium structure in Example 4 of the present invention. DETAILED DESCRIPTION

[0070] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0071] Example 1

[0072] like Figure 1 As shown, the intelligent optimization control method for the battery material recycling preparation process includes the following steps:

[0073] Step 1: Use multimodal sensing technology to collect training sample data in the battery material recycling preparation process in the test environment;

[0074] Step 2: Based on the training sample data, train a quality prediction model with product quality indicators as output;

[0075] Step 3: Based on the prediction results of the quality prediction model, construct the state space of the deep reinforcement learning model, and construct the action space and reward function of the deep reinforcement learning model;

[0076] Step 4: The prediction results of the quality prediction model and the monitoring data of the sensor in the test environment are integrated through the feature fusion strategy to generate a fusion process representation vector;

[0077] Step 5: Input the fusion process representation vector into the deep reinforcement learning model to train the deep reinforcement learning model;

[0078] Step 6: For the recycling preparation process to be monitored, real-time sensor data is collected through multimodal sensing technology in real time, and the action output by the deep reinforcement learning model is obtained based on the real-time sensor data, quality prediction model and deep reinforcement learning model; this action is used as an optimization strategy to adjust the parameters of the recycling preparation process to be monitored.

[0079] Specifically, the multimodal sensing technology uses multiple types of sensors to collect corresponding sensor data in different forms in real time during the battery material recycling and preparation process.

[0080] Specifically, the multimodal sensing technology includes but is not limited to:

[0081] Spectral sensors, such as UV-Vis, NIR, Raman, etc., electrochemical sensor arrays, such as pH, ORP, ISE, etc., physical sensors, such as temperature sensor arrays, pressure sensors, flow meters, etc., imaging sensors, such as CCD, thermal imaging, etc., acoustic sensors, such as ultrasound, etc.

[0082] Among them, the in-situ Raman spectrometer monitors the changes in chemical components in the solution in real time;

[0083] X-ray fluorescence (XRF) analyzer, online detection of metal element content;

[0084] Near infrared spectroscopy (NIR) is used to monitor organic content and solvent purity;

[0085] Among them, ion selective electrodes (ISE) are used to measure the concentration of specific ions (such as Li+, Co2+, Ni2+, etc.);

[0086] pH electrodes and oxidation-reduction potential (ORP) electrodes monitor the acidity and redox state of the solution;

[0087] Among them, the temperature sensor array monitors the temperature distribution at different locations of the reactor;

[0088] Pressure sensors, used to control reaction pressure and detect abnormal conditions;

[0089] Flow meter to accurately control the amount of various solutions and reagents added;

[0090] Among them, a high-speed camera captures the changes in particle morphology and distribution during the reaction process.

[0091] Thermal imaging cameras monitor equipment hot spots to prevent potential failures;

[0092] Among them, ultrasonic sensors are used to monitor the solid content and particle size distribution in the solution.

[0093] Acoustic emission sensors detect abnormal vibrations in equipment and perform predictive maintenance.

[0094] Furthermore, the test environment refers to a simulated environment of the battery material recycling and preparation process constructed under laboratory or factory conditions to test the recycling and preparation process and collect training samples.

[0095] Furthermore, the collection of the training sample data includes the following steps:

[0096] Step 11: Prepare N batches of waste battery positive and negative electrode materials of different sources and compositions; N is the number of selected batches;

[0097] Step 12: In the test environment, the complete recycling and preparation process is run for each batch of waste battery positive and negative electrode materials, and the monitoring data of each sensor is recorded in real time, and various controllable key parameters during the operation are simultaneously recorded;

[0098] Specifically, the controllable key parameters are physical parameters that can be automatically adjusted during the recovery and preparation process, including but not limited to the temperature, pressure, and concentration of the extractant during the extraction process.

[0099] Step 13: Measure and record the index value of the quality index of the battery material obtained after extraction and recovery as a quality index label;

[0100] Specifically, the quality indicator may be the purity of the extracted recovered material, and the purity of the extracted battery raw material is related to the controllable key parameters in the extraction process.

[0101] Step 14: The monitoring data collected in real time by each sensor of all batches of waste battery positive and negative electrode materials, the real-time controllable key parameters and quality indicator labels are used to form training sample data.

[0102] Furthermore, the training of the quality prediction model with the product quality index as output based on the training sample data includes the following steps:

[0103] Step 21: Convert the training sample data into sample input features;

[0104] Specifically, the converting of the training sample data into sample input features comprises the following steps:

[0105] Step 211: for each batch of waste battery positive and negative electrode materials, pre-process the corresponding monitoring data of the sensor at each moment, and perform feature splicing on the pre-processed monitoring data and controllable key parameters to obtain a feature splicing vector at each moment;

[0106] Specifically, the preprocessing includes but is not limited to operations such as noise removal and feature dimension reduction;

[0107] Preferably, the feature splicing vector is constructed as follows:

[0108] Perform vectorization operations on the monitoring data of each sensor at each moment to obtain the corresponding parameter feature vector;

[0109] Specifically, for example: the principal component analysis method is used for spectral data to obtain the spectral principal component vector, the sensor readings are used for electrochemical data to obtain the electrochemical vector, the image data is encoded using a convolutional autoencoder to obtain the image feature vector, and the Mel frequency gradient is calculated for the sound data to obtain the sound feature vector.

[0110] For each controllable key parameter, use the respective parameter values ​​to form a controllable parameter feature vector;

[0111] The parameter feature vectors and controllable parameter feature vectors of all sensors of the positive and negative electrode materials of the waste battery at each moment are spliced ​​to obtain a feature splicing vector; it can be understood that by feature vectorizing the monitoring data of each data form, the uniformity of multimodal data is guaranteed.

[0112] Step 212: preset a time window t; for each moment of each batch of waste battery positive and negative electrode materials during operation, collect the monitoring parameter sequence of the monitoring data of each sensor and the controllable parameter sequence of each controllable key parameter corresponding to the moment; the monitoring parameter sequence is a sequence composed of the parameter values ​​collected by the corresponding sensor within t moments before the moment, arranged in time order; the controllable parameter sequence is a parameter value sequence composed of the parameter values ​​of the corresponding controllable key parameter within t moments before the moment, arranged in time order;

[0113] Step 213: The monitoring parameter sequence and controllable parameter sequence at each moment constitute the sample input feature.

[0114] Step 22: Take the sample input features at each moment as input and the predicted value of the quality indicator at the subsequent moment t0 as output to train the quality prediction model; t0 is the preset time period.

[0115] Specifically, the training quality prediction model is as follows:

[0116] The sample input features at each moment are used as the input of the quality prediction model, the quality prediction model takes the predicted value of the quality indicator at the subsequent moment t0 of the moment as the output, takes the quality indicator label corresponding to the subsequent moment t0 of the moment as the prediction target, takes the difference between the predicted value of the quality indicator and the quality indicator label as the prediction error, and takes minimizing the sum of prediction errors as the training goal; the quality prediction model is trained until the sum of prediction errors reaches convergence and the training is stopped; the quality prediction model is a time series prediction model; the time series prediction model is any one of an RNN model or an LSTM model.

[0117] Furthermore, the state space of the deep reinforcement learning model is constructed based on the prediction results of the quality prediction model, and the action space and reward function of the deep reinforcement learning model are constructed in the following manner:

[0118] The deep reinforcement learning model is set to an Actor-Critic model;

[0119] The state space is marked as S, and the state space S reflects the real-time state of the entire recycling preparation process. The state space is constructed as follows:

[0120] The recycling preparation process is segmented according to a fixed time window; the size of the fixed time window is t0; it can be understood that by setting the fixed time window to the same size as the preset time period, the quality index in the subsequent time segment can be predicted, so as to guide the adjustment amount of the controllable key parameters according to the subsequent quality index;

[0121] For each time segment, the representation vector set composed of the fusion process representation vectors at each moment in time order is used as the state space;

[0122] Specifically, the fusion process representation vector is constructed as follows:

[0123] Collect the feature splicing vectors corresponding to each moment in the recycling and preparation process of the positive and negative electrode materials of used batteries;

[0124] Input the feature concatenation vector at each moment into the quality prediction model to obtain the predicted value of the quality index at the subsequent moment t0 output by the quality prediction model;

[0125] A one-dimensional convolutional network is used to extract local feature sequences from the monitoring parameter sequences corresponding to the monitoring data of each sensor at each moment;

[0126] Input the local feature sequence into the LSTM model to obtain the global context encoding vector C_t;

[0127] The predicted value of the quality indicator and the global context encoding vector C_t are concatenated to obtain the fusion process representation vector.

[0128] It should be noted that the representation vector of the fusion process captures the intrinsic semantic association between different features, so that the subsequent model training process can learn the correlation and coupling mode between features, and the fused feature vector can be directly input into the subsequent reinforcement learning model. Simple splicing usually requires a complex model architecture to process, and training convergence is more difficult.

[0129] The action space is marked as A, and the action space A includes the adjustment amounts of the controllable key parameters output by the Actor model, such as the temperature adjustment amount, the pressure adjustment amount, and the extraction concentration adjustment amount, and the parameter adjustment action taken in the next time segment is marked as a;

[0130] The reward function is marked as Q, and the reward value function Q=r(s, a) measures the effect of performing action a in the current state s;

[0131] Specifically, the reward value function Q can be set to: Q = b1×|P_t-P|-b2×∑ i hi; b1 and b2 are both preset proportional coefficients;

[0132] Among them, P_t represents the actual value of the quality indicator at the current moment, and P represents the predicted value of the quality indicator, which is used to measure the improvement of the value of the quality indicator after executing action a. The greater the improvement, the greater the reward value Q;

[0133] i is the number of the controllable key parameter, hi is the specific adjustment amount of the i-th controllable key parameter. The smaller the adjustment amount, the smaller the disturbance to the recycling preparation process, the better the execution effect, and the greater the reward Q.

[0134] The intelligent agent of the deep reinforcement learning model is designed to be an Actor-Critic structure based on RNN;

[0135] The input of the agent is state s;

[0136] The Critic network outputs the state value V(s);

[0137] The action probability π(a|s) is output by the Actor network.

[0138] Furthermore, the fusion process representation vector is input into the deep reinforcement learning model, and the deep reinforcement learning model is trained in the following manner:

[0139] The characterization vector set of each time segment of the recycling preparation process to be monitored is used as the current state s; the current state s is input to the Actor model of the deep reinforcement learning model, and the Actor model outputs the action in the action space A so as to maximize the state-behavior value output by the Critic model, and outputs the action decision of the adjustment amount of each controllable key parameter in the next time segment of the recycling preparation process to be monitored;

[0140] Execute action decision to obtain the next state s' and actual reward value;

[0141] The critic network outputs the state-behavior value after executing the action decision in the current state, calculates the TD error δ of the critic network, updates the critic network parameters according to δ, and minimizes the critic loss function; the state-behavior value output by the critic model is calculated by the state-behavior value function generated by the temporal difference optimization method.

[0142] Specifically, the state-behavior value function generated by the temporal difference optimization method can be:

[0143] Q(s, a) = r(s, a) + γ × V(s'), γ is the discount factor, V(s') is the estimated value when in state s'; generally, the estimated value can be replaced by Q(s', a'), that is, V(s') represents the estimated value of the state-behavior value function corresponding to the next state s' and action a', so as to ensure that Q(s, a) can consider the long-term, rather than just the current state;

[0144] Then the TD error δ is δ=(r(s,a)+γ×V(s')-Q(s,a)) 2 ;

[0145] The Actor model takes the state-behavior value output by the Critic model as the output target and uses the policy gradient update method to update the parameters of the Actor network model.

[0146] Furthermore, the method of obtaining the action output by the deep reinforcement learning model based on the real-time sensor data, the quality prediction model and the deep reinforcement learning model is:

[0147] Construct a feature splicing vector through real-time sensor data and the parameter values ​​of various controllable key parameters of the current recycling preparation process;

[0148] Input the feature concatenation vector into the quality prediction model to obtain the predicted value of the quality index at the subsequent time t0 output by the quality prediction model;

[0149] Based on the predicted values ​​of the quality indicators and the monitoring data of various sensors, the corresponding fusion process representation vector is constructed;

[0150] Collect all fusion process representation vectors of the time segment at the current moment to form a corresponding representation vector set;

[0151] The representation vector set is input into the deep reinforcement learning model to obtain the action output by the Actor network of the deep reinforcement learning model.

[0152] Example 2

[0153] like Figure 2 As shown, the intelligent optimization control system for the battery material recycling preparation process includes a sample data collection module, a prediction model training module, an enhancement model training module, and an action decision module; wherein each module is electrically connected;

[0154] A sample data collection module, which uses multimodal sensing technology to collect training sample data in the battery material recycling preparation process in the test environment, and sends the training sample data to the prediction model training module;

[0155] The prediction model training module trains the quality prediction model with product quality indicators as output based on the training sample data, and sends the quality prediction model to the reinforcement model training module and the action decision module;

[0156] The reinforcement model training module constructs the state space of the deep reinforcement learning model based on the prediction results of the quality prediction model, and constructs the action space and reward function of the deep reinforcement learning model. The prediction results of the quality prediction model and the monitoring data of the sensor in the test environment are integrated through the feature fusion strategy to generate a fusion process representation vector, which is input into the deep reinforcement learning model, trained, and sent to the action decision module.

[0157] The action decision module collects real-time sensor data through multimodal sensing technology for the recycling preparation process to be monitored, obtains the action output by the deep reinforcement learning model based on the real-time sensor data, quality prediction model and deep reinforcement learning model; uses this action as an optimization strategy to adjust the parameters of the recycling preparation process to be monitored.

[0158] Example 3

[0159] Figure 3 1 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present application. Figure 3 As shown, according to another aspect of the present application, an electronic device 100 is also provided. The electronic device 100 may include one or more processors and one or more memories. The memories store computer readable codes, and when the computer readable codes are run by one or more processors, the intelligent optimization control method for the battery material recycling preparation process as described above may be executed.

[0160] The method or device according to the embodiment of the present application can also be used by Figure 3 The electronic device architecture shown in FIG. Figure 3 As shown, the electronic device 100 may include a bus 101, one or more CPUs 102, a ROM 103, a RAM 104, a communication port 105 connected to a network, an input / output component 106, a hard disk 107, etc. The storage device in the electronic device 100, such as the ROM 103 or the hard disk 107, may store the intelligent optimization control method for the battery material recycling preparation process provided in the present application.

[0161] Furthermore, the electronic device 100 may further include a user interface 108. Figure 3 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different devices. Figure 3 One or more components of an electronic device are shown.

[0162] Example 4

[0163] Figure 4 Schematic diagram of the computer-readable storage medium structure provided by an embodiment of the present application. Figure 4 As shown, it is a computer-readable storage medium 200 according to an embodiment of the present application. Computer-readable instructions are stored on the computer-readable storage medium 200. When the computer-readable instructions are executed by the processor, the intelligent optimization control method for the battery material recycling preparation process according to the embodiment of the present application described with reference to the above figures can be executed. The computer-readable storage medium 200 includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0164] In addition, according to the implementation of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium, the non-transitory machine-readable storage medium stores machine-readable instructions, the machine-readable instructions can be run by a processor to execute instructions corresponding to the method steps provided by the present application, and when the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are executed.

[0165] The methods, apparatuses, and devices of the present application may be implemented in many ways. For example, the methods, apparatuses, and devices of the present application may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present application are not limited to the order specifically described above unless otherwise specified. In addition, in some embodiments, the present application may also be implemented as programs recorded in a recording medium, which include machine-readable instructions for implementing the method according to the present application. Therefore, the present application also covers recording media storing programs for executing the method according to the present application.

[0166] In addition, the parts of the above-mentioned technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0167] The specific implementation modes as described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation mode of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0168] The above preset parameters or preset thresholds are all set by technicians in this field according to actual conditions or obtained through large-scale data simulation.

[0169] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. Intelligent optimization control method for battery material recycling and preparation process, characterized in that: The following steps are involved: Step 1: Use multimodal sensing technology to collect training sample data in the battery material recycling preparation process in the test environment; Step 2: Based on the training sample data, train a quality prediction model with product quality indicators as output; Step 3: Based on the prediction results of the quality prediction model, construct the state space of the deep reinforcement learning model, and construct the action space and reward function of the deep reinforcement learning model; Step 4: The prediction results of the quality prediction model and the monitoring data of the sensor in the test environment are integrated through the feature fusion strategy to generate a fusion process representation vector; Step 5: Input the fusion process representation vector into the deep reinforcement learning model to train the deep reinforcement learning model; Step 6: For the recycling preparation process to be monitored, real-time sensor data is collected through multimodal sensing technology in real time, and the action output by the deep reinforcement learning model is obtained based on the real-time sensor data, quality prediction model and deep reinforcement learning model; this action is used as an optimization strategy to adjust the parameters of the recycling preparation process to be monitored.

2. The intelligent optimization control method for battery material recycling and preparation process according to claim 1 is characterized in that: The collection of the training sample data comprises the following steps: Step 11: Prepare N batches of waste battery positive and negative electrode materials of different sources and compositions; N is the number of selected batches; Step 12: In the test environment, the complete recycling and preparation process is run for each batch of waste battery positive and negative electrode materials, and the monitoring data of each sensor is recorded in real time, and various controllable key parameters during the operation are recorded synchronously; Step 13: Measure and record the index value of the quality index of the battery material obtained after extraction and recovery as a quality index label; Step 14: The monitoring data collected in real time by each sensor of all batches of waste battery positive and negative electrode materials, the real-time controllable key parameters and quality indicator labels are used to form training sample data.

3. The intelligent optimization control method for battery material recycling and preparation process according to claim 2 is characterized in that: The training of the quality prediction model with the product quality index as output based on the training sample data comprises the following steps: Step 21: Convert the training sample data into sample input features; Step 22: Take the sample input features at each moment as input and the predicted value of the quality indicator at the subsequent moment t0 as output to train the quality prediction model; t0 is the preset time period.

4. The intelligent optimization control method for battery material recycling and preparation process according to claim 3 is characterized in that: The step of converting the training sample data into sample input features comprises the following steps: Step 211: for each batch of waste battery positive and negative electrode materials, pre-process the corresponding monitoring data of the sensor at each moment, and perform feature splicing on the pre-processed monitoring data and controllable key parameters to obtain a feature splicing vector at each moment; Step 212: preset a time window t; for each moment of each batch of waste battery positive and negative electrode materials during operation, collect the monitoring parameter sequence of the monitoring data of each sensor and the controllable parameter sequence of each controllable key parameter corresponding to the moment; the monitoring parameter sequence is a sequence composed of the parameter values ​​collected by the corresponding sensor within t moments before the moment, arranged in time order; the controllable parameter sequence is a parameter value sequence composed of the parameter values ​​of the corresponding controllable key parameter within t moments before the moment, arranged in time order; Step 213: The monitoring parameter sequence and controllable parameter sequence at each moment constitute the sample input feature.

5. The intelligent optimization control method for battery material recycling and preparation process according to claim 4 is characterized in that: The method of training the quality prediction model is: The sample input features at each moment are used as the input of the quality prediction model, the quality prediction model takes the predicted value of the quality indicator at the subsequent moment t0 of the moment as the output, takes the quality indicator label corresponding to the subsequent moment t0 of the moment as the prediction target, takes the difference between the predicted value of the quality indicator and the quality indicator label as the prediction error, and takes minimizing the sum of prediction errors as the training goal; the quality prediction model is trained until the sum of prediction errors reaches convergence and the training is stopped; the quality prediction model is a time series prediction model.

6. The intelligent optimization control method for battery material recycling and preparation process according to claim 5, characterized in that: The feature concatenation vector is constructed as follows: Perform vectorization operations on the monitoring data of each sensor at each moment to obtain the corresponding parameter feature vector; For each key controllable parameter, use the respective parameter values ​​to form a controllable parameter feature vector; The parameter feature vectors and controllable parameter feature vectors of all sensors of the positive and negative electrode materials of the waste battery at each moment are spliced ​​to obtain a feature splicing vector.

7. The intelligent optimization control method for battery material recycling and preparation process according to claim 6, characterized in that: The method of constructing the state space of the deep reinforcement learning model based on the prediction results of the quality prediction model, and constructing the action space and reward function of the deep reinforcement learning model is as follows: The deep reinforcement learning model is set to an Actor-Critic model; The state space is marked as S, and the state space S reflects the real-time state of the entire recycling preparation process. The state space is constructed as follows: The recovery preparation process is segmented according to a fixed time window; the size of the fixed time window is t0; For each time segment, the representation vector set composed of the fusion process representation vectors at each moment in time order is used as the state space; The action space is marked as A, wherein the action space A includes the adjustment amount of the controllable key parameters output by the Actor model, and the parameter adjustment action taken in the next time segment is marked as a; The reward function is marked as Q, and the reward value function Q=r(s, a) measures the effect of performing action a in the current state s; The intelligent agent of the deep reinforcement learning model is designed to be an Actor-Critic structure based on RNN; The input of the agent is state s; The Critic network outputs the state value V(s); The action probability π(a|s) is output by the Actor network.

8. The intelligent optimization control method for battery material recycling and preparation process according to claim 7, characterized in that: The fusion process representation vector is constructed as follows: Collect the feature splicing vectors corresponding to each moment in the recycling and preparation process of the positive and negative electrode materials of used batteries; Input the feature concatenation vector at each moment into the quality prediction model to obtain the predicted value of the quality index at the subsequent moment t0 output by the quality prediction model; A one-dimensional convolutional network is used to extract local feature sequences from the monitoring parameter sequences corresponding to the monitoring data of each sensor at each moment; Input the local feature sequence into the LSTM model to obtain the global context encoding vector C_t; The predicted value of the quality indicator and the global context encoding vector C_t are concatenated to obtain the fusion process representation vector.

9. The intelligent optimization control method for battery material recycling and preparation process according to claim 8, characterized in that: The method of obtaining the action output by the deep reinforcement learning model based on the real-time sensor data, the quality prediction model and the deep reinforcement learning model is: Construct a feature splicing vector through real-time sensor data and the parameter values ​​of various controllable key parameters of the current recycling preparation process; Input the feature concatenation vector into the quality prediction model to obtain the predicted value of the quality index at the subsequent time t0 output by the quality prediction model; Based on the predicted values ​​of the quality indicators and the monitoring data of various sensors, the corresponding fusion process representation vector is constructed; Collect all fusion process representation vectors of the time segment at the current moment to form a corresponding representation vector set; The representation vector set is input into the deep reinforcement learning model to obtain the action output by the Actor network of the deep reinforcement learning model.

10. An intelligent optimization control system for a battery material recycling preparation process, which is used to implement the intelligent optimization control method for a battery material recycling preparation process according to any one of claims 1 to 9, characterized in that: It includes a sample data collection module, a prediction model training module, an enhanced model training module and an action decision module; wherein each module is electrically connected; A sample data collection module, which uses multimodal sensing technology to collect training sample data in the battery material recycling preparation process in the test environment, and sends the training sample data to the prediction model training module; The prediction model training module trains the quality prediction model with product quality indicators as output based on the training sample data, and sends the quality prediction model to the reinforcement model training module and the action decision module; The reinforcement model training module constructs the state space of the deep reinforcement learning model based on the prediction results of the quality prediction model, and constructs the action space and reward function of the deep reinforcement learning model. The prediction results of the quality prediction model and the monitoring data of the sensor in the test environment are integrated through the feature fusion strategy to generate a fusion process representation vector, which is input into the deep reinforcement learning model, trained, and sent to the action decision module. The action decision module collects real-time sensor data through multimodal sensing technology for the recycling preparation process to be monitored, obtains the action output by the deep reinforcement learning model based on the real-time sensor data, quality prediction model and deep reinforcement learning model; uses this action as an optimization strategy to adjust the parameters of the recycling preparation process to be monitored.

11. An electronic device, characterized in that: include: processor and memory, wherein The memory stores a computer program that can be called by the processor; The processor executes the intelligent optimization control method for the battery material recycling preparation process described in any one of claims 1 to 9 in the background by calling the computer program stored in the memory.

12. A computer-readable storage medium, characterized in that: A rewritable computer program is stored thereon; When the computer program is run on a computer device, the computer device executes the intelligent optimization control method for the battery material recycling preparation process described in any one of claims 1 to 9 in the background.

Images