Automatic alarm system and method based on battery production and transportation inventory management

Abstract

The invention relates to the technical field of battery management, and discloses an automatic alarm system and method based on battery production and transportation inventory management, and the system comprises a data collection module which is used for deploying a multi-source sensor in the whole links of battery production, storage and transportation, and collecting battery state data, environment parameter data and transportation process data in real time; the data fusion module is used for carrying out fusion processing on the collected multi-source data based on a physical information neural network and a multi-modal fusion algorithm of adaptive feature weighting, and extracting target feature parameters; the risk prediction module is used for inputting the target characteristic parameters into a space-time adaptive risk prediction model, performing probability prediction on the battery safety risk, and dividing the risk into four grades according to the prediction risk probability; the intelligent alarm decision module is used for automatically triggering corresponding alarm strategies and disposal suggestions according to the risk grades and the links where the batteries are located; according to the invention, the pertinence and practicability of alarm are improved.

Description

Technical Field

[0001] This invention relates to the field of battery management technology, and specifically to an automatic alarm system and method based on battery production and transportation inventory management. Background Technology

[0002] With the rapid development of the new energy industry, batteries, as core energy storage components, directly impact the overall efficiency of the industry chain through the efficiency of their production, transportation, and inventory management. Currently, battery management systems often employ a single threshold triggering mechanism for automatic alarms, lacking the ability to integrate and analyze multi-source data, which can easily lead to false alarms or missed alarms. Traditional alarm mechanisms only focus on current environmental anomalies, lacking the ability to analyze long-term trends in battery status and predict risks. The safety risk characteristics of batteries vary greatly across different stages such as production, storage, and transportation, and existing systems have failed to establish risk assessments. Furthermore, alarm response mechanisms are simplistic, failing to implement tiered early warning and differentiated handling strategies based on risk levels. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing an automatic alarm system and method based on battery production and transportation inventory management.

[0004] The first aspect of this invention provides an automatic alarm system based on battery production and transportation inventory management, the system comprising: The data acquisition module is used to deploy multi-source sensors throughout the entire process of battery production, storage and transportation to collect real-time battery status data, environmental parameter data and transportation process data; The data fusion module is used to fuse collected multi-source data based on a physical information neural network and an adaptive feature weighting multimodal fusion algorithm to extract target feature parameters. The risk prediction module is used to input the target feature parameters into the spatiotemporal adaptive risk prediction model built on the Transformer architecture to predict the probability of battery safety risks. Based on the predicted risk probability, the risks are divided into four levels: low risk, medium risk, high risk, and emergency risk. The intelligent alarm decision-making module is used to automatically trigger corresponding alarm strategies and handling suggestions based on risk level and the battery's current stage, using dynamic decision-making and digital twin collaborative processing.

[0005] Optionally, in a first implementation of the first aspect of the present invention, the data fusion module includes: A submodule is constructed to build physical model equations for the battery electrochemical reaction as physical constraints. These physical model equations include thermodynamic equations, electrochemical reaction equations, and gas generation equations. The sub-module is used to divide the collected multi-source data into two categories according to time series and spatial distribution characteristics, resulting in time series data and spatial distribution data. The design submodule is used to design a dual-channel feature extraction network. The temporal channel uses a one-dimensional convolutional neural network to process time series data and extract the feature change patterns in the time dimension. The spatial channel uses a two-dimensional convolutional neural network to process spatially distributed data and extract the feature distribution patterns in the spatial dimension. A submodule is introduced to dynamically calculate the weight coefficients of each feature parameter based on the battery type, current environmental conditions, and the stage in which the battery is located, using an adaptive weight allocation mechanism. The fusion submodule is used to fuse the physical model equations with the extracted temporal and spatial features, and to normalize the fused feature representation to obtain the target feature parameters.

[0006] Optionally, in a second implementation of the first aspect of the present invention, the time series data includes data on the changes in battery surface temperature, internal gas concentration, and storage environment humidity over time, and the spatial distribution data includes spatial distribution information on transportation vibration index and battery state of charge.

[0007] Optionally, in a third implementation of the first aspect of the present invention, the risk prediction module includes: The input submodule is used to input the target feature parameters into the spatiotemporal adaptive risk prediction model built on the Transformer architecture, which contains 4 encoder layers, each of which consists of a multi-head self-attention mechanism and a feedforward neural network. The capture submodule is used to capture the long-term dependency between battery safety risks and multi-source data by calculating the attention weights at different time points and spatial locations using a multi-head self-attention mechanism, and obtain the associated features. The extraction submodule is used to perform nonlinear transformation on the correlation features output by the multi-head self-attention mechanism of the feedforward neural network to extract deep spatiotemporal features. After processing by a 4-layer encoder, a risk feature vector with fused spatiotemporal information is obtained. The output submodule is used to input the risk feature vector into the fully connected layer. The fully connected layer outputs the safety risk probability of the battery in the current spatiotemporal scenario through the sigmoid activation function, and classifies the risk level according to the preset threshold.

[0008] Optionally, in the fourth implementation of the first aspect of the present invention, when the risk probability is less than or equal to 0.2, it is determined to be low risk; when the risk probability is greater than 0.2 and less than or equal to 0.5, it is determined to be medium risk; when the risk probability is greater than 0.5 and less than or equal to 0.8, it is determined to be high risk; and when the risk probability is greater than 0.8, it is determined to be emergency risk.

[0009] Optionally, in a fifth implementation of the first aspect of the present invention, the intelligent alarm decision module includes: The optimization submodule is used to train and learn the handling strategies under different risk levels and battery stages based on historical alarm data and historical handling data, and continuously optimize the decision-making strategy through the Q-learning algorithm. The simulation submodule is used to construct a 1:1 digital twin model in the digital space based on the physical entities of battery production, storage, and transportation. The battery digital twin model is used to simulate the risk changes and potential impacts of batteries under different disposal schemes. The triggering submodule is used to match the optimal handling plan according to the stage and risk level of the battery, convert the optimal handling plan into specific operation instructions, and automatically trigger the execution of the warning operation.

[0010] Optionally, in a sixth implementation of the first aspect of the present invention, the optimization submodule includes: Acquire historical alarm data and historical handling data from the past 1-3 years to form a training dataset for the Q-learning algorithm; Define the state space, action space, and reward function, and construct the Q-table based on the state space and action space; Initially, all Q values ​​are set to 0. A sample is randomly selected from the training dataset to determine the current state and the actual action taken for the current sample. Based on the treatment results in the sample, the immediate reward corresponding to the current action is calculated in combination with the reward function. The next state after the current sample is treated is found in the training dataset. The Q values ​​of all actions corresponding to the next state are found in the Q table, and the largest Q value is selected. The current Q value is updated by combining the preset learning rate and discount factor. The update is iterated until the change in Q value in the Q table is less than the preset threshold, and then the iteration stops. After training is completed, for each state, the action with the largest Q value in that state is found in the Q table and determined as the corresponding handling strategy.

[0011] A second aspect of this invention provides an automatic alarm method based on battery production and transportation inventory management, the method comprising the following steps: Deploy multi-source sensors throughout the entire process of battery production, storage, and transportation to collect real-time battery status data, environmental parameter data, and transportation process data; A multimodal fusion algorithm based on physical information neural network and adaptive feature weighting is used to fuse collected multi-source data and extract target feature parameters. The target feature parameters are input into the spatiotemporal adaptive risk prediction model built on the Transformer architecture to predict the probability of battery safety risks. Based on the predicted risk probability, the risks are divided into four levels: low risk, medium risk, high risk and emergency risk. Based on risk classification and the stage of the battery, dynamic decision-making and digital twin collaborative processing are used to automatically trigger corresponding alarm strategies and handling suggestions.

[0012] Optionally, in the first implementation of the second aspect of the present invention, the multimodal fusion algorithm based on physical information neural network and adaptive feature weighting performs fusion processing on the collected multi-source data to extract target feature parameters, including: Physical model equations for the battery electrochemical reaction are constructed as physical constraints, including thermodynamic equations, electrochemical reaction equations, and gas generation equations. The collected multi-source data were divided into two categories according to their time series and spatial distribution characteristics, resulting in time series data and spatial distribution data. Design a dual-channel feature extraction network, in which the temporal channel uses a one-dimensional convolutional neural network to process time series data and extract the feature change patterns in the time dimension, and the spatial channel uses a two-dimensional convolutional neural network to process spatially distributed data and extract the feature distribution patterns in the spatial dimension. An adaptive weight allocation mechanism is introduced to dynamically calculate the weight coefficients of each characteristic parameter based on the battery type, current environmental conditions, and the stage in which the battery is located. The physical model equations are fused with the extracted temporal and spatial features, and the fused feature representations are normalized to obtain the target feature parameters.

[0013] Optionally, in a second implementation of the second aspect of the present invention, the step of inputting the target feature parameters into a spatiotemporal adaptive risk prediction model built on the Transformer architecture to predict the probability of battery safety risks, and classifying the risks into four levels—low risk, medium risk, high risk, and emergency risk—based on the predicted risk probability, includes: The target feature parameters are input into the spatiotemporal adaptive risk prediction model built on the Transformer architecture, which contains 4 encoder layers, each of which consists of a multi-head self-attention mechanism and a feedforward neural network. By employing a multi-head self-attention mechanism, attention weights at different time points and spatial locations are calculated to capture the long-term dependency between battery safety risks and multi-source data, thereby obtaining associated features. The feedforward neural network performs a nonlinear transformation on the correlation features output by the multi-head self-attention mechanism to extract deep spatiotemporal features. After processing by a 4-layer encoder, a risk feature vector with fused spatiotemporal information is obtained. The risk feature vector is input into the fully connected layer, which outputs the probability of the battery's safety risk in the current spatiotemporal scenario through the sigmoid activation function, and classifies the risk level according to a preset threshold.

[0014] The technical solution provided by this invention deploys multi-source sensors throughout the entire process of battery production, storage, and transportation to collect real-time battery status data, environmental parameter data, and transportation process data. A multi-modal fusion algorithm based on physical information neural networks and adaptive feature weighting is used to fuse the collected multi-source data and extract target feature parameters. These target feature parameters are then input into a spatiotemporal adaptive risk prediction model built on a Transformer architecture to predict battery safety risks probabilistically. Based on the predicted risk probabilities, the risks are classified into four levels: low risk, medium risk, high risk, and emergency risk. According to the risk classification and the stage in which the battery is located, dynamic decision-making and digital twin collaborative processing are used to automatically trigger corresponding alarm strategies and handling suggestions. This invention solves the problem of inaccurate early warnings caused by a single data source in existing technologies, effectively captures the complex nonlinear relationship between battery safety risks and multi-source data, and designs a dynamic risk classification and alarm strategy based on the stage in which the battery is located, improving the targeting and practicality of alarms. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0016] Figure 1 A schematic diagram of the structure of an automatic alarm system based on battery production and transportation inventory management provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the data fusion module provided in an embodiment of the present invention; Figure 3 A flowchart of an automatic alarm method based on battery production and transportation inventory management provided in an embodiment of the present invention. Detailed Implementation

[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0018] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 A schematic diagram of an automatic alarm system based on battery production and transportation inventory management provided in this embodiment of the invention. The system includes: The data acquisition module is used to deploy multi-source sensors throughout the entire process of battery production, storage and transportation to collect real-time battery status data, environmental parameter data and transportation process data; The data fusion module is used to fuse collected multi-source data based on a physical information neural network and an adaptive feature weighting multimodal fusion algorithm to extract target feature parameters. The risk prediction module is used to input the target feature parameters into the spatiotemporal adaptive risk prediction model built on the Transformer architecture to predict the probability of battery safety risks. Based on the predicted risk probability, the risks are divided into four levels: low risk, medium risk, high risk, and emergency risk. The intelligent alarm decision-making module is used to automatically trigger corresponding alarm strategies and handling suggestions based on risk level and the battery's current stage, using dynamic decision-making and digital twin collaborative processing.

[0019] In this embodiment, please refer to Figure 2 The data fusion module includes: A submodule is constructed to build physical model equations for the battery electrochemical reaction as physical constraints. These physical model equations include thermodynamic equations, electrochemical reaction equations, and gas generation equations. The sub-module is used to divide the collected multi-source data into two categories according to time series and spatial distribution characteristics, resulting in time series data and spatial distribution data. The time series data includes data on the changes of battery surface temperature, internal gas concentration and storage environment humidity over time, and the spatial distribution data includes spatial distribution information of transportation vibration index and battery state of charge. The design submodule is used to design a dual-channel feature extraction network. The temporal channel uses a one-dimensional convolutional neural network to process time series data and extract the feature change patterns in the time dimension. The spatial channel uses a two-dimensional convolutional neural network to process spatially distributed data and extract the feature distribution patterns in the spatial dimension. A submodule is introduced to dynamically calculate the weight coefficients of each feature parameter based on the battery type, current environmental conditions, and the stage in which the battery is located, using an adaptive weight allocation mechanism. The fusion submodule is used to fuse the physical model equations with the extracted temporal and spatial features, and to normalize the fused feature representation to obtain the target feature parameters.

[0020] In this embodiment, the risk prediction module includes: The input submodule is used to input the target feature parameters into the spatiotemporal adaptive risk prediction model built on the Transformer architecture, which contains 4 encoder layers, each of which consists of a multi-head self-attention mechanism and a feedforward neural network. The capture submodule is used to capture the long-term dependency between battery safety risks and multi-source data by calculating the attention weights at different time points and spatial locations using a multi-head self-attention mechanism, and obtain the associated features. The extraction submodule is used to perform nonlinear transformation on the correlation features output by the multi-head self-attention mechanism of the feedforward neural network to extract deep spatiotemporal features. After processing by a 4-layer encoder, a risk feature vector with fused spatiotemporal information is obtained. The output submodule is used to input the risk feature vector into the fully connected layer. The fully connected layer outputs the safety risk probability of the battery in the current spatiotemporal scenario through the sigmoid activation function, and classifies the risk level according to the preset threshold. When the risk probability is less than or equal to 0.2, it is judged as low risk; when the risk probability is greater than 0.2 and less than or equal to 0.5, it is judged as medium risk; when the risk probability is greater than 0.5 and less than or equal to 0.8, it is judged as high risk; and when the risk probability is greater than 0.8, it is judged as emergency risk.

[0021] In this embodiment, the intelligent alarm decision module includes: The optimization submodule is used to train and learn the handling strategies under different risk levels and battery stages based on historical alarm data and historical handling data, and continuously optimize the decision-making strategy through the Q-learning algorithm. The simulation submodule is used to construct a 1:1 digital twin model in the digital space based on the physical entities of battery production, storage, and transportation. The battery digital twin model is used to simulate the risk changes and potential impacts of batteries under different disposal schemes. The triggering submodule is used to match the optimal handling plan according to the stage and risk level of the battery, convert the optimal handling plan into specific operation instructions, and automatically trigger the execution of the warning operation.

[0022] In this embodiment, the optimization submodule includes: Historical data from the past 1-3 years is extracted from the system database. The data includes historical alarm records with the risk level at the time of the alarm, such as low risk, medium risk, etc.; the stage of the battery's operation, such as production, storage, and transportation; alarm trigger time and associated characteristic parameters, such as temperature deviation values ​​in the production stage and vibration acceleration values ​​in the transportation stage, along with corresponding handling data including the handling strategies adopted, such as adjusting equipment temperature for medium-risk production and temporary stop inspection for high-risk transportation; and the results of the handling, such as whether the risk has been reduced, handling time, and resource consumption, such as energy consumption changes after temperature adjustment and transportation delays caused by stop inspections. Invalid data, such as alarm data without handling results and records with more than 3 missing data fields, are removed. The valid data is then categorized and stored according to the relationship between alarm information, handling strategy, and handling result to form the training dataset for the Q-learning algorithm. Define a state space, setting states as combinations of risk level and battery stage, such as low risk, production stage, medium risk, transportation stage, high risk, and storage stage, covering all combinations of low, medium, high, and emergency risk levels with the three battery stages, forming a clear set of states; define an action space, with each state corresponding to multiple possible actions, such as adjusting equipment operating parameters (fine-tuning temperature), increasing inspection frequency, pausing the process for 5 minutes for inspection, etc., for a medium risk state; define a reward function, setting reward values ​​based on the action results: if the risk level decreases after action (e.g., from medium to low risk) and there are no additional losses (e.g., no production delays, no transportation timeouts), the reward value is set to +10; if the risk level remains unchanged after action but there are no additional losses, the reward value is set to 0; if the risk level increases after action or additional losses occur (e.g., production delays exceeding 2 hours, transportation timeouts exceeding 1 hour), the reward value is set to negative (e.g., -5, -10), with larger losses resulting in larger negative values. Based on the defined state space and action space, a Q-table is constructed. The rows of the Q-table represent states, and the columns represent the actions corresponding to each state. The values ​​in the table are the values ​​corresponding to the state and actions. Initially, all Q values ​​are set to 0. The dimensions of the Q-table are determined by the number of states and the number of actions for each state. For example, if there are 12 states, 4 risk levels × 3 steps, and each state corresponds to 3 actions, then the Q-table is an initial table with 12 rows and 3 columns. The Q values ​​will be gradually updated through training. Randomly select a sample from the training dataset to determine the current state and the actual action taken for that sample. Based on the action result in the sample, calculate the immediate reward corresponding to the action using the reward function. For example, if the risk is reduced to low risk and there is no delay after the action, the reward is +10. Find the next state after the action of the sample in the dataset, find the Q-values ​​of all actions corresponding to the next state in the Q-table, and select the largest Q-value. Update the Q-value of the current state-action according to the Q-learning update logic, combined with a preset learning rate to control the magnitude of each update (e.g., set to 0.1) and a discount factor to control the impact of future rewards (e.g., set to 0.8). Each time, select different samples from the training dataset and update the corresponding state-action Q-values ​​sequentially. The number of iterations is set to 5000-10000 times until the change magnitude of the Q-values ​​of each state-action in the Q-table is less than a preset threshold (e.g., 0.01), then stop the iteration and complete the initial training. After training, for each state, the action with the largest Q value in the Q-table is found and determined as the optimal handling strategy for that state. For example, if the action with the largest Q value in a state is to move the battery in the target area, then this action will be used as the default handling strategy for that state. During actual system operation, whenever a new alarm is generated and a handling strategy is executed, the handling result is recorded and the immediate reward R is calculated. The next state after handling is obtained, and the Q-table is updated in real time with the newly generated data according to the Q-value update logic in step four. The newly added historical alarm and handling data are periodically added to the training dataset, and the iterative training process in step four is repeated to perform secondary optimization of the Q-table to ensure that the decision-making strategy is continuously adjusted according to changes in the actual scenario.

[0023] Please see Figure 3 The flowchart of the automatic alarm method based on battery production and transportation inventory management provided in this embodiment of the invention includes the following steps: Step 301: Deploy multi-source sensors throughout the entire process of battery production, storage, and transportation to collect battery status data, environmental parameter data, and transportation process data in real time; Step 302: The multi-source data collected is fused based on the physical information neural network and the adaptive feature weighting multimodal fusion algorithm to extract target feature parameters; Step 303: Input the target feature parameters into the spatiotemporal adaptive risk prediction model built on the Transformer architecture to predict the probability of battery safety risks. Based on the predicted risk probability, the risks are divided into four levels: low risk, medium risk, high risk and emergency risk. Step 304: Based on the risk classification and the stage of the battery, adopt dynamic decision-making and digital twin collaborative processing to automatically trigger the corresponding alarm strategies and handling suggestions.

[0024] In this embodiment, when deploying multi-source sensors throughout the entire process of battery production, storage, and transportation, it is necessary to accurately match the sensor type and installation location according to the characteristics of each stage: In the production stage, high-precision purity sensors, such as laser spectral sensors for detecting lithium salt purity and X-ray fluorescence sensors for lithium nickel cobalt manganese oxide materials, are installed at the raw material conveying pipeline of the electrode preparation process. Contact temperature sensors are deployed in the drying section of the cell coating equipment to monitor baking temperature fluctuations, and pressure sensors are installed at the cell packaging station to collect pressure values ​​during the packaging process. At the same time, voltage and current sensors and internal resistance testers are deployed on the formation testing line to capture real-time data on the cell charging and discharging status. In the storage stage, weight sensors are installed under the shelf panels according to warehouse location zones to indirectly count inventory balances. Temperature and humidity sensors are deployed in different areas of the warehouse, including high-temperature storage areas, normal-temperature storage areas, and low-temperature storage areas, to record changes in ambient temperature and humidity. RFID tags are affixed to the outside of each battery tray, in conjunction with RFID readers at the warehouse entrance and exit. The reader collects battery entry time, batch information, and exit records. During transportation, multiple distributed temperature sensors are installed on the top of the cargo compartment of each transport vehicle to cover the front, back, left, and right areas. A triaxial vibration acceleration sensor is installed on the bottom of the cargo compartment to monitor the vibration intensity and direction during transportation. At the same time, the vehicles are equipped with Beidou / GPS dual-mode locators to obtain the vehicle's location and speed in real time, and door magnetic sensors are installed at the cargo compartment doors to record the opening and closing status of the cargo compartment doors. All sensors are connected to the system data center through industrial Ethernet, warehouse LAN, and 4G / 5G wireless network to achieve millisecond-level collection and real-time transmission of battery status data such as cell voltage during production, remaining shelf life of batteries during storage, and battery appearance integrity feedback during transportation; environmental parameter data such as cleanliness of the production workshop, ambient temperature and humidity of the warehouse, temperature and humidity inside the cargo compartment during transportation; and transportation process data such as real-time vehicle location, speed, vibration acceleration, and estimated arrival time, ensuring the continuity and timeliness of data throughout the entire process.

[0025] In this embodiment, a physical model equation for the battery electrochemical reaction is constructed as a physical constraint, which includes thermodynamic equations, electrochemical reaction equations, and gas generation equations. The collected multi-source data is divided into two categories based on time series and spatial distribution characteristics, resulting in time series data and spatial distribution data. A dual-channel feature extraction network is designed, where the time series channel uses a one-dimensional convolutional neural network to process the time series data and extract the feature change patterns in the time dimension, and the spatial channel uses a two-dimensional convolutional neural network to process the spatial distribution data and extract the feature distribution patterns in the spatial dimension. An adaptive weight allocation mechanism is introduced to dynamically calculate the weight coefficients of each feature parameter based on the battery type, current environmental conditions, and the stage in which the battery is located. The physical model equations are fused with the extracted time series and spatial features, and the fused feature representation is normalized to obtain the target feature parameters.

[0026] In this embodiment, the target feature parameters are input into a spatiotemporal adaptive risk prediction model built on the Transformer architecture. The Transformer architecture contains four encoder layers, each consisting of a multi-head self-attention mechanism and a feedforward neural network. The multi-head self-attention mechanism calculates the attention weights at different time points and spatial locations to capture the long-term dependency between battery safety risks and multi-source data, thus obtaining associated features. The feedforward neural network performs a nonlinear transformation on the associated features output by the multi-head self-attention mechanism to extract deep spatiotemporal features. After processing by the four encoder layers, a risk feature vector integrating spatiotemporal information is obtained. The risk feature vector is input into a fully connected layer, which outputs the battery's safety risk probability in the current spatiotemporal scenario through a sigmoid activation function and classifies the risk level according to a preset threshold.

[0027] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An automatic alarm system based on battery production and transportation inventory management, characterized in that, The system includes: The data acquisition module is used to deploy multi-source sensors throughout the entire process of battery production, storage and transportation to collect real-time battery status data, environmental parameter data and transportation process data; The data fusion module is used to fuse collected multi-source data based on a physical information neural network and an adaptive feature weighting multimodal fusion algorithm to extract target feature parameters. The risk prediction module is used to input the target feature parameters into the spatiotemporal adaptive risk prediction model built on the Transformer architecture to predict the probability of battery safety risks. Based on the predicted risk probability, the risks are divided into four levels: low risk, medium risk, high risk, and emergency risk. The intelligent alarm decision-making module is used to automatically trigger corresponding alarm strategies and handling suggestions based on risk level and the battery's current stage, using dynamic decision-making and digital twin collaborative processing.

2. The automatic alarm system based on battery production and transportation inventory management as described in claim 1, characterized in that, The data fusion module includes: A submodule is constructed to build physical model equations for the battery electrochemical reaction as physical constraints. These physical model equations include thermodynamic equations, electrochemical reaction equations, and gas generation equations. The sub-module is used to divide the collected multi-source data into two categories according to time series and spatial distribution characteristics, resulting in time series data and spatial distribution data. The design submodule is used to design a dual-channel feature extraction network. The temporal channel uses a one-dimensional convolutional neural network to process time series data and extract the feature change patterns in the time dimension. The spatial channel uses a two-dimensional convolutional neural network to process spatially distributed data and extract the feature distribution patterns in the spatial dimension. A submodule is introduced to dynamically calculate the weight coefficients of each feature parameter based on the battery type, current environmental conditions, and the stage in which the battery is located, using an adaptive weight allocation mechanism. The fusion submodule is used to fuse the physical model equations with the extracted temporal and spatial features, and to normalize the fused feature representation to obtain the target feature parameters.

3. The automatic alarm system based on battery production and transportation inventory management as described in claim 2, characterized in that, The time-series data includes data on the changes in battery surface temperature, internal gas concentration, and storage environment humidity over time, while the spatial distribution data includes spatial distribution information on transportation vibration index and battery state of charge.

4. The automatic alarm system based on battery production and transportation inventory management as described in claim 1, characterized in that, The risk prediction module includes: The input submodule is used to input the target feature parameters into the spatiotemporal adaptive risk prediction model built on the Transformer architecture, which contains 4 encoder layers, each of which consists of a multi-head self-attention mechanism and a feedforward neural network. The capture submodule is used to capture the long-term dependency between battery safety risks and multi-source data by calculating the attention weights at different time points and spatial locations using a multi-head self-attention mechanism, and obtain the associated features. The extraction submodule is used to perform nonlinear transformation on the correlation features output by the multi-head self-attention mechanism of the feedforward neural network to extract deep spatiotemporal features. After processing by a 4-layer encoder, a risk feature vector with fused spatiotemporal information is obtained. The output submodule is used to input the risk feature vector into the fully connected layer. The fully connected layer outputs the safety risk probability of the battery in the current spatiotemporal scenario through the sigmoid activation function, and classifies the risk level according to the preset threshold.

5. An automatic alarm system based on battery production and transportation inventory management as described in claim 4, characterized in that, When the risk probability is less than or equal to 0.2, it is judged as low risk; when the risk probability is greater than 0.2 and less than or equal to 0.5, it is judged as medium risk; when the risk probability is greater than 0.5 and less than or equal to 0.8, it is judged as high risk; and when the risk probability is greater than 0.8, it is judged as emergency risk.

6. The automatic alarm system based on battery production and transportation inventory management as described in claim 1, characterized in that, The intelligent alarm decision module includes: The optimization submodule is used to train and learn the handling strategies under different risk levels and battery stages based on historical alarm data and historical handling data, and continuously optimize the decision-making strategy through the Q-learning algorithm. The simulation submodule is used to construct a 1:1 digital twin model in the digital space based on the physical entities of battery production, storage, and transportation. The battery digital twin model is used to simulate the risk changes and potential impacts of batteries under different disposal schemes. The triggering submodule is used to match the optimal handling plan according to the stage and risk level of the battery, convert the optimal handling plan into specific operation instructions, and automatically trigger the execution of the warning operation.

7. An automatic alarm system based on battery production and transportation inventory management as described in claim 6, characterized in that, The optimization submodule includes: Acquire historical alarm data and historical handling data from the past 1-3 years to form a training dataset for the Q-learning algorithm; Define the state space, action space, and reward function, and construct the Q-table based on the state space and action space; Initially, all Q values ​​are set to 0. A sample is randomly selected from the training dataset to determine the current state and the actual action taken for the current sample. Based on the treatment results in the sample, the immediate reward corresponding to the current action is calculated in combination with the reward function. The next state after the current sample is treated is found in the training dataset. The Q values ​​of all actions corresponding to the next state are found in the Q table, and the largest Q value is selected. The current Q value is updated by combining the preset learning rate and discount factor. The update is iterated until the change in Q value in the Q table is less than the preset threshold, and then the iteration stops. After training is completed, for each state, the action with the largest Q value in that state is found in the Q table and determined as the corresponding handling strategy.

8. An automatic alarm method based on battery production and transportation inventory management, characterized in that, The method includes the following steps: Deploy multi-source sensors throughout the entire process of battery production, storage, and transportation to collect real-time battery status data, environmental parameter data, and transportation process data; A multimodal fusion algorithm based on physical information neural network and adaptive feature weighting is used to fuse collected multi-source data and extract target feature parameters. The target feature parameters are input into the spatiotemporal adaptive risk prediction model built on the Transformer architecture to predict the probability of battery safety risks. Based on the predicted risk probability, the risks are divided into four levels: low risk, medium risk, high risk and emergency risk. Based on risk classification and the stage of the battery, dynamic decision-making and digital twin collaborative processing are used to automatically trigger corresponding alarm strategies and handling suggestions.

9. The automatic alarm method based on battery production and transportation inventory management as described in claim 8, characterized in that, The multimodal fusion algorithm based on physical information neural network and adaptive feature weighting fuses collected multi-source data to extract target feature parameters, including: Physical model equations for the battery electrochemical reaction are constructed as physical constraints, including thermodynamic equations, electrochemical reaction equations, and gas generation equations. The collected multi-source data were divided into two categories according to their time series and spatial distribution characteristics, resulting in time series data and spatial distribution data. Design a dual-channel feature extraction network, in which the temporal channel uses a one-dimensional convolutional neural network to process time series data and extract the feature change patterns in the time dimension, and the spatial channel uses a two-dimensional convolutional neural network to process spatially distributed data and extract the feature distribution patterns in the spatial dimension. An adaptive weight allocation mechanism is introduced to dynamically calculate the weight coefficients of each characteristic parameter based on the battery type, current environmental conditions, and the stage in which the battery is located. The physical model equations are fused with the extracted temporal and spatial features, and the fused feature representations are normalized to obtain the target feature parameters.

10. The automatic alarm method based on battery production and transportation inventory management as described in claim 8, characterized in that, The target feature parameters are input into a spatiotemporal adaptive risk prediction model built on the Transformer architecture to predict battery safety risks probabilistically. Based on the predicted risk probabilities, the risks are divided into four levels: low risk, medium risk, high risk, and emergency risk. The target feature parameters are input into a spatiotemporal adaptive risk prediction model built on the Transformer architecture, which consists of four encoder layers, each composed of a multi-head self-attention mechanism and a feedforward neural network. By employing a multi-head self-attention mechanism, attention weights at different time points and spatial locations are calculated to capture the long-term dependency between battery safety risks and multi-source data, thereby obtaining associated features. The feedforward neural network performs a nonlinear transformation on the correlation features output by the multi-head self-attention mechanism to extract deep spatiotemporal features. After processing by a 4-layer encoder, a risk feature vector with fused spatiotemporal information is obtained. The risk feature vector is input into the fully connected layer, which outputs the probability of the battery's safety risk in the current spatiotemporal scenario through the sigmoid activation function, and classifies the risk level according to a preset threshold.

Images