LLM-based mobile energy storage system autonomous inspection method and intelligent robot

Abstract

The invention discloses an LLM-based mobile energy storage system autonomous inspection method and an intelligent robot. The method is integrated in a controller of the intelligent robot, and comprises the following steps: controlling the intelligent robot to move to a target detection position along a target motion path, and obtaining battery related data of at least one single battery in to-be-detected energy storage equipment associated with the target detection position; determining temperature abnormal information, sound pressure abnormal information and total volatile organic compound concentration abnormal information according to the battery related data, and further determining target abnormal information of the to-be-detected energy storage equipment for determining an abnormal event type; determining a target confidence coefficient of the abnormal event type according to the battery related data and the abnormal weight attribute corresponding to the abnormal event type through a data processing model; and if the target confidence meets a preset condition, generating a target broadcast text through a pre-constructed large language model based on the battery related data, the abnormal event type and the target confidence. And the efficiency of intelligent inspection of the energy storage system is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent inspection technology, and in particular to an autonomous inspection method and intelligent robot for mobile energy storage systems based on LLM. Background Technology

[0002] Large-scale energy storage stations are typically located in open-air or semi-indoor environments. Traditionally, manual inspections are conducted, but this method suffers from high workload, poor safety during nighttime or inclement weather inspections, and limitations in quickly locating and understanding faults due to reliance on flashing lights or buzzer alarms. Furthermore, handwritten maintenance records prevent real-time integration with the BMS / EMS. While purely visual inspection robots can only output abnormal thermal images without explaining the causes, their fixed reporting vocabulary prevents the generation of natural language when encountering new faults, resulting in limitations and inaccuracies in anomaly reporting and ultimately leading to poor inspection effectiveness for energy storage systems. Summary of the Invention

[0003] This invention provides an autonomous inspection method and intelligent robot for mobile energy storage systems based on LLM (Liquidity Management System) to solve the problem of poor inspection results for energy storage systems.

[0004] According to one aspect of the present invention, an autonomous inspection method for a mobile energy storage system based on LLM is provided, integrated into the controller of an intelligent robot, and the following method is executed based on the controller:

[0005] The intelligent robot is controlled to move to the target detection position according to a predetermined target motion path, and acquire battery-related data of at least one single cell in at least one energy storage device to be tested associated with the target detection position; wherein, the battery-related data includes at least temperature field data, sound pressure level data, and total volatile organic compound concentration data;

[0006] For at least one energy storage device to be tested, temperature anomaly information, sound pressure level information, and total volatile organic compound concentration anomaly information are obtained based on the temperature field data, sound pressure level data, and total volatile organic compound concentration data of at least one individual battery associated with the energy storage device to be tested.

[0007] Based on the abnormal temperature, abnormal sound pressure, and abnormal total volatile organic compound concentration information of each energy storage device to be tested, the target abnormal information corresponding to each energy storage device to be tested is determined, and the type of abnormal event is determined based on the target abnormal information.

[0008] Battery-related data and anomaly weight attributes corresponding to the types of abnormal events are input into the data processing model to determine the target confidence level corresponding to the types of abnormal events.

[0009] When the target confidence level meets the preset conditions, the target broadcast text is generated by pre-built large language model based on battery-related data, abnormal event types, and target confidence level.

[0010] Optionally, the target movement path is determined based on the following method: the intelligent robot moves within at least one floor associated with the target area, and during the movement, a regional map of the target area is constructed based on real-time localization and mapping technology; using A... The algorithm determines the target movement path based on the device location information of at least one energy storage device to be detected in the regional map.

[0011] Optionally, based on the temperature field data, sound pressure level data, and total volatile organic compound (TVOC) concentration data of at least one individual battery associated with the energy storage device under test, temperature anomaly information, sound pressure anomaly information, and TVOC concentration anomaly information are obtained. This includes: acquiring the temperature field data of all individual batteries associated with the energy storage device under test within the current time window, and averaging the temperature field data corresponding to each discrete time point within the current time window to determine the temperature difference information at each discrete time point based on the temperature field data and the corresponding average temperature data, and using the temperature difference information as temperature anomaly information; acquiring the sound pressure level data of the energy storage device under test within the current time window, and using the sound pressure level data corresponding to the peak sound pressure as sound pressure anomaly information; and determining the concentration difference value corresponding to each discrete time point based on the TVOC concentration data of the energy storage device under test at each discrete time point within the current time window and a preset benchmark concentration data, and using the concentration difference value at each discrete time point as TVOC concentration anomaly information.

[0012] Optionally, based on the temperature anomaly information, sound pressure anomaly information, and total volatile organic compound (TVOC) concentration anomaly information of each energy storage device to be tested, the target anomaly information corresponding to each energy storage device to be tested is determined, including: determining first data based on the temperature anomaly information of all discrete time points within the current time window; determining second data based on the sound pressure anomaly information of all discrete time points within the current time window; determining third data based on the TVOC concentration anomaly information of all discrete time points within the current time window; and determining the target anomaly information of the energy storage device to be tested based on the first data, second data, third data, and corresponding weight attributes.

[0013] Optionally, based on the target anomaly information, the anomaly event type is determined, including: when the target anomaly information does not meet the normal threshold range, the maximum value among the first data, the second data, and the third data is read, and the evaluation dimension corresponding to the maximum value is taken as the anomaly event type; wherein, the evaluation dimension includes the temperature dimension, the sound pressure dimension, and the total volatile organic compound concentration dimension.

[0014] Optionally, battery-related data and anomaly weight attributes corresponding to the abnormal event type are input into the data processing model to determine the target confidence level corresponding to the abnormal event type. This includes: inputting battery-related data corresponding to the abnormal event type into a pre-built knowledge graph to determine the anomaly weight attributes corresponding to the data content of the battery-related data; inputting the anomaly weight attributes and all battery-related data generated within the current time window into the data processing model so that the feature extraction layer in the data processing model can extract the feature information of the battery-related data; obtaining weight attributes based on the feature adjustment layer in the data processing model to update the features corresponding to the abnormal event type in the feature information based on the weight attributes, thereby obtaining the updated feature information; decoding the updated feature information based on the decoding layer in the data processing model, and determining the target confidence level corresponding to the decoded feature information based on the activation function.

[0015] Optionally, after obtaining the target broadcast text, the method further includes: controlling the speaker set on the intelligent robot to play the target broadcast text when the target confidence level is in a first interval range; and caching the target broadcast text based on the caching module on the intelligent robot when the target confidence level is in a second interval range.

[0016] Optionally, before controlling the speaker set on the intelligent robot to play the target broadcast text, the method further includes: inputting the target broadcast text into a semantic extraction model to obtain a second text, and using the second text as the target broadcast text to be played; wherein, the second text is a summary text of the generated target broadcast text.

[0017] According to another aspect of the present invention, an intelligent robot is provided, comprising:

[0018] The battery-related data acquisition module is used to control the intelligent robot to move to the target detection position according to the predetermined target motion path, and to acquire battery-related data of at least one single cell in at least one energy storage device to be tested associated with the target detection position; wherein, the battery-related data includes at least temperature field data, sound pressure level data and total volatile organic compound concentration data;

[0019] An anomaly information determination module is used to obtain temperature anomaly information, sound pressure anomaly information, and total volatile organic compound concentration anomaly information for at least one energy storage device under test, based on temperature field data, sound pressure level data, and total volatile organic compound concentration data of at least one individual battery associated with the energy storage device under test.

[0020] The abnormal event type determination module is used to determine the target abnormal information corresponding to each energy storage device under test based on the abnormal temperature information, abnormal sound pressure information, and abnormal total volatile organic compound concentration information of each energy storage device under test, so as to determine the abnormal event type based on the target abnormal information.

[0021] The target confidence level determination module is used to input battery-related data and anomaly weight attributes corresponding to the anomaly event type into the data processing model in order to determine the target confidence level corresponding to the anomaly event type.

[0022] The target broadcast text determination module is used to generate target broadcast text based on battery-related data, abnormal event types, and target confidence when the target confidence meets preset conditions, by using a pre-built large language model.

[0023] According to another aspect of the present invention, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the LLM-based autonomous inspection method for mobile energy storage systems according to any embodiment of the present invention.

[0024] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0025] At least one processor; and

[0026] A memory that is communicatively connected to at least one processor; wherein,

[0027] The memory stores a computer program that can be executed by at least one processor, such that the at least one processor is able to execute the LLM-based autonomous inspection method for mobile energy storage systems according to any embodiment of the present invention.

[0028] According to another aspect of the present invention, a computer-readable storage medium is provided, which stores computer instructions for causing a processor to execute and implement the LLM-based autonomous inspection method for mobile energy storage systems according to any embodiment of the present invention.

[0029] The technical solution of this invention involves a controller integrated into an intelligent robot executing the following method: controlling the intelligent robot to move to a target detection position according to a predetermined target motion path, and acquiring battery-related data of at least one single cell in at least one energy storage device to be tested associated with the target detection position; wherein, the battery-related data includes at least temperature field data, sound pressure level data, and total volatile organic compound (TVOC) concentration data; for at least one energy storage device to be tested, based on the temperature field data, sound pressure level data, and TVOC concentration data of at least one single cell associated with the energy storage device to be tested, obtaining temperature anomaly information, sound pressure level data, and TVOC concentration data. The system collects abnormal information on temperature, sound pressure, and total volatile organic compound (TVOC) concentrations. Based on these abnormalities, it determines the target abnormal information for each energy storage device under test, and then identifies the type of abnormal event. Battery-related data and the corresponding abnormal event weight attributes are input into the data processing model to determine the target confidence level for each event type. When the target confidence level meets preset conditions, a target broadcast text is generated using a pre-built large language model based on the battery-related data, the event type, and the target confidence level. This system enables autonomous mobile inspection via intelligent robots, allowing for multi-dimensional data collection without human intervention, thus improving inspection efficiency and safety. It integrates three key data categories—temperature, sound pressure, and volatile organic compounds—and combines anomaly weights with data model verification confidence levels, significantly enhancing the accuracy and reliability of anomaly event judgment. The generated broadcast text intuitively presents core information, providing clear evidence for subsequent processing. It is suitable for the efficient operation and maintenance of large-scale energy storage equipment, solving the problem of poor inspection efficiency in the autonomous inspection of mobile energy storage systems and improving the overall efficiency of autonomous inspection of mobile energy storage systems.

[0030] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart of an autonomous inspection method for a mobile energy storage system based on LLM, provided in Embodiment 1 of the present invention.

[0033] Figure 2 This is a flowchart of an autonomous inspection method for a mobile energy storage system based on LLM, provided in Embodiment 2 of the present invention;

[0034] Figure 3 This is a structural schematic diagram of an intelligent robot provided in Embodiment 3 of the present invention;

[0035] Figure 4 This is a schematic diagram of the structure of an electronic device that implements the LLM-based autonomous inspection method for mobile energy storage systems according to embodiments of the present invention. Detailed Implementation

[0036] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0037] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus 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 apparatus.

[0038] Example 1

[0039] Figure 1 This is a flowchart of an LLM-based autonomous inspection method for mobile energy storage systems provided in Embodiment 1 of the present invention. This embodiment is applicable to situations involving autonomous inspection of energy storage systems. The method can be executed by an LLM-based autonomous inspection device for mobile energy storage systems, which can be implemented in hardware and / or software. This LLM-based autonomous inspection device can be configured within an intelligent robot. The autonomous inspection method is integrated into the controller of the intelligent robot, and the controller executes the autonomous inspection method, such as... Figure 1 As shown, the method includes:

[0040] S110. Control the intelligent robot to move to the target detection position according to the predetermined target motion path, and acquire battery-related data of at least one single cell in at least one energy storage device to be tested associated with the target detection position; wherein, the battery-related data includes at least temperature field data, sound pressure level data and total volatile organic compound concentration data.

[0041] Specifically, the target detection location can be understood as a pre-defined spatial point for the intelligent robot to collect data. It serves as the target endpoint for the robot's movement and the starting point for the detection work. For example, at least one location on a path accessible to the intelligent robot within the vicinity of the energy storage device under test can be set as the target detection location. The energy storage device under test can be understood as equipment deployed within the area of ​​a large energy storage station, composed of multiple individual battery cells, and is the core object of data collection. Battery-related data can be understood as a collection of key information reflecting the operating status of individual battery cells, acquired by the built-in detection module of the energy storage device under test. This data forms the basis for subsequent status assessments and includes, but is not limited to, temperature field data, sound pressure level data, and total volatile organic compound (TVOC) concentration data. Temperature field data refers to the quantitative record of the temperature distribution on the surface and surrounding area of ​​the individual battery cell, directly reflecting the battery's heating state. Sound pressure level data refers to the measurement result of the sound intensity generated by the individual battery cell during operation, which can help determine whether there are abnormalities such as looseness or wear in the battery's internal structure. TVOC concentration data refers to the detected value of the volatile organic compound content in the environment surrounding the battery, which can indirectly reflect potential problems such as electrolyte leakage.

[0042] Specifically, the intelligent robot first loads the target motion path calculated in real time based on the current driving environment information, and then moves precisely along the target path to the designated detection location, namely the target detection location of the energy storage device to be tested, through its own navigation system. After the intelligent robot arrives, it activates the onboard multi-dimensional detection module to simultaneously collect battery-related data of at least one single battery in the energy storage device to be tested at that location, including but not limited to temperature field data, sound pressure level data, and total volatile organic compound concentration data. The entire process can be automated without human intervention.

[0043] In this embodiment, the intelligent robot automatically loads the target motion path and moves to the target detection position to collect data automatically, avoiding positioning errors caused by manual operation; the simultaneous acquisition of multiple key data can comprehensively reflect the battery operating status and provide complete data support for the safety monitoring of energy storage devices; the automated process not only improves the data collection efficiency but also reduces the safety risks of manual entry into the detection scene, making it particularly suitable for batch battery monitoring scenarios in large-scale energy storage power stations.

[0044] Optionally, the target movement path is determined based on the following method: the intelligent robot moves within at least one floor associated with the target area, and during the movement, a regional map of the target area is constructed based on real-time localization and mapping technology; using A... The algorithm determines the target movement path based on the device location information of at least one energy storage device to be detected in the regional map.

[0045] The target area specifically refers to the overall range within which the intelligent robot conducts energy storage equipment inspections, covering at least one floor to be monitored and all energy storage devices deployed within that range. It serves as the basic spatial boundary for the robot to build its map and plan its path. The target movement path specifically refers to the path taken by the robot based on the target area map, using A... The algorithm plans the optimal movement route of the robot from the starting point to each target detection position by combining the location information of each energy storage device to be tested. This route guides the robot to complete the detection task accurately and efficiently.

[0046] Specifically, the intelligent robot moves within at least one floor of the target area equipped with an energy storage system. During its autonomous movement, it collects environmental information in real time using instant positioning and mapping technology, simultaneously creating a map of the area and locating itself. Then, combining this information with the specific location information of at least one energy storage device to be tested already marked on the area map, it uses AI... The algorithm quickly plans the optimal target motion path, that is, ultimately determines the optimal target motion path for the robot from the starting point to each target detection position.

[0047] In this embodiment, based on real-time localization and mapping (RTL) technology, the robot can autonomously construct maps of unknown or complex environments, adapting to unknown or complex areas without the need for pre-set maps, thus improving environmental adaptability; through A The algorithm quickly selects the optimal path, reducing robot movement time and improving detection efficiency. At the same time, it ensures that the robot accurately reaches the detection position corresponding to each energy storage device to be tested, thus guaranteeing the accuracy of subsequent data collection.

[0048] S120. For at least one energy storage device to be tested, based on the temperature field data, sound pressure level data, and total volatile organic compound concentration data of at least one individual battery associated with the energy storage device to be tested, obtain temperature anomaly information, sound pressure anomaly information, and total volatile organic compound concentration anomaly information.

[0049] Specifically, temperature anomaly information refers to abnormal temperature changes in the temperature field of a single battery cell, used to determine if the battery is malfunctioning. Sound pressure anomaly information refers to the sound pressure level data of a single battery cell during operation; for example, it may include the sound pressure peak value within a preset time period, which can help determine if there are abnormalities such as loosening, wear, or fault discharge in the battery's internal structure. Total volatile organic compound (TVOC) concentration anomaly information refers to data reflecting abnormal changes in the battery's chemical properties, and is an important indicator for predicting potential safety hazards such as electrolyte leakage.

[0050] Specifically, for each energy storage device to be tested, the temperature field data, sound pressure level data, and total volatile organic compound (TVOC) concentration data of all associated individual cells are first extracted. These data are then processed according to a preset data processing method, and integrated to form the temperature anomaly information, sound pressure anomaly information, and TVOC concentration anomaly information corresponding to the energy storage device. For example, the weights corresponding to different types of anomaly information can be obtained, and the corresponding information can be weighted separately to obtain the temperature anomaly information, sound pressure anomaly information, and TVOC concentration anomaly information.

[0051] In this embodiment, based on targeted comparison of three types of core data, abnormal signals in the thermal, acoustic and chemical characteristics of the battery can be accurately captured, enabling comprehensive fault monitoring. The synchronous output of multi-dimensional abnormal information can help staff quickly locate the potential fault type of the battery, providing accurate basis for timely maintenance of energy storage equipment and effectively reducing the risk of fault expansion.

[0052] Optionally, based on the temperature field data, sound pressure level data, and total volatile organic compound (TVOC) concentration data of at least one individual battery associated with the energy storage device under test, temperature anomaly information, sound pressure anomaly information, and TVOC concentration anomaly information are obtained. This includes: acquiring the temperature field data of all individual batteries associated with the energy storage device under test within the current time window, and averaging the temperature field data corresponding to each discrete time point within the current time window to determine the temperature difference information at each discrete time point based on the temperature field data and the corresponding average temperature data, and using the temperature difference information as temperature anomaly information; acquiring the sound pressure level data of the energy storage device under test within the current time window, and using the sound pressure level data corresponding to the peak sound pressure as sound pressure anomaly information; and determining the concentration difference value corresponding to each discrete time point based on the TVOC concentration data of the energy storage device under test at each discrete time point within the current time window and a preset benchmark concentration data, and using the concentration difference value at each discrete time point as TVOC concentration anomaly information.

[0053] Specifically, for the energy storage device under test, temperature field data of all individual cells within the current time window are first collected. The mean value of the temperature field data at each discrete time point is calculated. The temperature difference data is obtained by comparing the temperature field data at each discrete time point with the corresponding mean value, and the highest temperature difference data is used as temperature anomaly information. At the same time, the peak data of the sound pressure level data within the time window is extracted and directly used as sound pressure anomaly information. Then, the total volatile organic compound concentration data at each discrete time point within the current time window is combined with the preset benchmark concentration data to calculate the concentration difference at each time point, which is used as total volatile organic compound concentration anomaly information.

[0054] In this embodiment, three targeted data processing methods—average and temperature difference data processing within a time window, peak extraction, and benchmark comparison—are used to accurately capture abnormal temperature fluctuations, instantaneous sound pressure anomalies, and concentration deviations, making the extraction of abnormal information more targeted and accurate. Based on refined analysis of discrete time points, the temporal characteristics of abnormal data can be completely preserved, providing detailed support for tracing the timing of anomalies and judging their development trends. At the same time, the processing logic is simple and efficient, adapting to the needs of real-time detection scenarios.

[0055] S130. Based on the abnormal temperature information, abnormal sound pressure information, and abnormal total volatile organic compound concentration information of each energy storage device to be tested, determine the target abnormal information corresponding to each energy storage device to be tested, so as to determine the type of abnormal event based on the target abnormal information.

[0056] Specifically, the target anomaly information can be understood as the information obtained after processing the temperature anomaly information, sound pressure anomaly information, and total volatile organic compound concentration anomaly information of a certain energy storage device under test according to a preset data processing method. For example, the target anomaly information can be obtained by weighting various types of anomaly information using weighted data. The anomaly event type can be understood as a standardized classification of the abnormal state of the energy storage device determined based on the target anomaly information and according to preset mapping rules. For example, it can be a single temperature overheating anomaly, a sound pressure peak anomaly, an electrolyte leakage-induced concentration anomaly, or a composite fault type with multiple anomalies superimposed. It is a precise definition of the nature of the anomaly.

[0057] Specifically, the system first aggregates abnormal temperature, sound pressure, and total volatile organic compound (TVOC) concentration information for each energy storage device under test. Then, it integrates and filters these three types of abnormal information using pre-defined information fusion rules to extract target abnormal information that comprehensively reflects the core abnormal state of the device. These pre-defined information fusion rules include, but are not limited to, abnormal information weighting and abnormal logic judgment rules. Next, based on pre-defined abnormal event type mapping rules (e.g., a single abnormality corresponds to a specific type, and multiple abnormalities correspond to a composite type), the abnormal event type corresponding to the energy storage device is determined.

[0058] In this embodiment, the target anomaly information is obtained by integrating three types of subdivided anomaly information, avoiding the one-sidedness of single-dimensional anomaly judgment and making the presentation of the anomaly state more comprehensive. By relying on clear mapping rules to match the anomaly event type, the judgment logic is standardized and reusable, which not only improves the efficiency of anomaly event type judgment, but also ensures the consistency of judgment results between different devices, providing a clear direction for subsequent accurate anomaly handling.

[0059] Optionally, based on the temperature anomaly information, sound pressure anomaly information, and total volatile organic compound (TVOC) concentration anomaly information of each energy storage device to be tested, the target anomaly information corresponding to each energy storage device to be tested is determined, including: determining first data based on the temperature anomaly information of all discrete time points within the current time window; determining second data based on the sound pressure anomaly information of all discrete time points within the current time window; determining third data based on the TVOC concentration anomaly information of all discrete time points within the current time window; and determining the target anomaly information of the energy storage device to be tested based on the first data, second data, third data, and corresponding weight attributes.

[0060] The first data refers to the comprehensive data obtained by summarizing and processing (e.g., statistical extreme values, cumulative deviations, average values, etc.) the temperature anomaly information (i.e., temperature difference information) of all discrete time points within the current time window, and is used to centrally reflect the overall temperature anomaly situation of individual cells within this time window. The second data refers to the core data formed by integrating and processing (e.g., extracting the maximum peak value, statistically analyzing the frequency of peak occurrences, etc.) the sound pressure anomaly information (i.e., sound pressure peak data) within the current time window, focusing on reflecting the key characteristics of sound pressure anomalies within the time window. The third data refers to the comprehensive data obtained by summarizing and calculating (e.g., cumulative differences, average differences, etc.) the total volatile organic compound concentration anomaly information (i.e., concentration differences) of all discrete time points within the current time window, and is used to present the overall abnormal deviation of relevant concentrations within this time window.

[0061] Specifically, for each energy storage device to be tested, the first data is obtained by summarizing and calculating (e.g., statistical extreme values, cumulative deviations, etc.) the temperature anomaly information (such as temperature difference data) at all discrete time points within the current time window; the second data is obtained by integrating and processing the sound pressure anomaly information (i.e., sound pressure peak data) at each discrete time point within the time window (e.g., extracting the maximum peak value, statistically analyzing the frequency of peak occurrence, etc.); at the same time, the third data is obtained by summarizing the total volatile organic compound concentration anomaly information (i.e., concentration difference) at each discrete time point; finally, the target anomaly information of the energy storage device is obtained by combining the first, second, and third data and their respective preset weight attributes through weighted calculation and other methods.

[0062] For example, after summarizing the temperature anomaly information, sound pressure anomaly information, and total volatile organic compound concentration anomaly information for each energy storage device to be tested, the method for determining the temperature anomaly information includes:

[0063] Calculate the average temperature: ;

[0064] Calculate the temperature difference value: ;

[0065] in, Let N represent the temperature of the i-th individual cell, and N represent the total number of individual cells in the energy storage device to be tested; the calculated temperature difference value is determined as temperature anomaly information. This refers to the first data point. The highest sound pressure level within the current measurement window is selected as the sound pressure anomaly information. This is the second set of data, used to identify abnormal fan wear or mechanical noise. Abnormal information regarding total volatile organic compound (TVOC) concentration. The third data point is determined using the following formula:

[0066] ;

[0067] in, The current total volatile organic compound (TVOC) concentration of the energy storage device to be tested. This represents the historical stable value of the total volatile organic compound concentration of the energy storage device under test.

[0068] It should be noted that different abnormal data have varying degrees of importance in determining the current abnormal event. Therefore, the weight of each abnormal data point can be determined based on the mapping relationship between abnormal information and weighted data. Then, the corresponding abnormal data can be adjusted according to the weights to obtain the final abnormal information for temperature, sound pressure, and total volatile organic compound concentration. , and Then, you can select... , and The maximum value in the range is used to determine the corresponding event type, for example... If the value is the maximum, then the abnormal event type is a temperature abnormal event. It is also possible to obtain... , and One or more of the data are matched with the preset mapping relationship between abnormal data range and abnormal event type, so as to determine the abnormal event type based on the comprehensive data.

[0069] In this embodiment, by summarizing the three types of abnormal information within a time window, more representative first, second, and third data are extracted, providing a high-quality foundation for subsequent integration. Weight attributes are introduced to adapt to the differences in the importance of different abnormal information, making the integration of target abnormal information more in line with actual detection needs and improving the accuracy of the comprehensive abnormal status reflection. The overall processing flow is highly structured and logically clear, ensuring both the comprehensiveness of the target abnormal information and highlighting the core abnormal dimensions, providing reliable support for subsequent abnormal event type determination.

[0070] Optionally, based on the target anomaly information, the anomaly event type is determined, including: when the target anomaly information does not meet the normal threshold range, the maximum value among the first data, the second data, and the third data is read, and the evaluation dimension corresponding to the maximum value is taken as the anomaly event type; wherein, the evaluation dimension includes the temperature dimension, the sound pressure dimension, and the total volatile organic compound concentration dimension.

[0071] Specifically, first, it is determined whether the target abnormal information of the energy storage device under test is within the normal threshold range. If it is not met, the abnormal event type determination is triggered. Then, the maximum value among the first data, the second data, and the third data is extracted to determine the value to which the maximum value belongs. The evaluation dimension is directly used as the abnormal event type corresponding to the energy storage device. The evaluation dimensions include temperature dimension, sound pressure dimension, or total volatile organic compound concentration dimension.

[0072] In this embodiment, the judgment logic is simple and intuitive. By using threshold filtering and maximum value matching, the core abnormal dimension can be quickly identified, which greatly improves the efficiency of judging abnormal event types and adapts to the timeliness requirements of real-time detection scenarios. By focusing on the key evaluation dimension corresponding to the maximum value, the main abnormal factors affecting the operation of energy storage equipment can be accurately located, avoiding interference from multi-dimensional abnormal information in the core judgment and providing a clear direction for subsequent targeted handling of abnormalities.

[0073] S140. Input battery-related data and the anomaly weight attributes corresponding to the anomaly event type into the data processing model to determine the target confidence level corresponding to the anomaly event type.

[0074] Specifically, the data processing model can be understood as a pre-trained algorithmic model used to analyze anomaly-related data of energy storage devices. It can receive battery-related data and anomaly weight attributes of corresponding anomaly event types as input, and output a quantitative index reflecting the reliability of the anomaly event type determination result through built-in algorithm calculation and data fitting. The target confidence score can be understood as a numerical value that quantifies the credibility of the determined anomaly event type. The value directly reflects the reliability of the anomaly determination result and provides a key decision-making basis for whether to initiate an early warning and take handling measures. The target confidence score can be output by the data processing model.

[0075] Specifically, the battery-related data of the energy storage device to be tested is first collected, and the abnormal weight attribute (the priority or impact weight that adapts to the abnormal type) corresponding to the determined abnormal event type is retrieved. Both types of data are used as input parameters and fed into the preset data processing model. Through the algorithm calculation and data fitting of the model, the target confidence level corresponding to the abnormal event type is finally output, which is the reliability index of the judgment result.

[0076] In this embodiment, by combining the original battery-related data with targeted anomaly weight attributes, more comprehensive and focused input information is provided to the model, effectively improving the accuracy of target confidence calculation. By replacing manual judgment with a data processing model, the objectivity and consistency of confidence results are ensured, while significantly improving computational efficiency. This adapts to the batch processing needs of anomaly detection in large-scale energy storage equipment, providing a scientific and reliable basis for subsequent early warning and handling measures.

[0077] S150. When the target confidence level meets the preset conditions, the target broadcast text is generated by pre-built large language model based on battery-related data, abnormal event types and target confidence level.

[0078] Specifically, the pre-built large language model can be understood as a natural language processing model that has been pre-trained and adapted to the anomaly reporting scenario of energy storage devices. It has the ability to integrate professional data and transform it into plain language expressions. It can receive input information such as battery-related data, anomaly event types, and target confidence levels, and generate standardized and easy-to-understand text content according to preset logic. The target reporting text can be understood as a structured text that integrates the core data of the energy storage device's battery, the specific type of anomaly event, and the credibility of the judgment result (target confidence level). It contains key professional information and is presented in a concise and intuitive way. It can be directly used by operation and maintenance personnel to quickly grasp the anomaly situation and carry out subsequent processing. It can be output by the pre-built large language model after processing the input data such as battery-related data, anomaly event types, and target confidence levels.

[0079] Specifically, the process first determines whether the target confidence level output by the data processing model meets the preset conditions, i.e., whether the target confidence level is higher than the set confidence level threshold. If so, the text generation process is triggered. Battery-related data of the energy storage device to be tested, the identified abnormal event types, and the target confidence level are used as input information and fed into a pre-built large language model. The model automatically integrates and extracts the core content according to a preset broadcast format and information presentation logic, generating a clear and easy-to-understand target broadcast text. Optionally, battery-related data, abnormal event data, and the target confidence level are filled into the broadcast text prompt word template to generate corresponding broadcast text prompt words. These prompt words are then input into the pre-built large language model to generate the target broadcast text.

[0080] In this embodiment, the broadcast text is generated based on the premise that the confidence level meets the standard, so as to avoid false alarms caused by low reliability judgment results and improve the rigor of the broadcast. With the help of the natural language processing capabilities of the large language model, professional raw data, anomaly types and confidence indicators can be transformed into easy-to-understand and intuitive text, reducing the threshold for information interpretation. The automated generation process does not require manual intervention, which not only improves the broadcast efficiency, but also ensures the integrity and consistency of the text information, making it convenient for operation and maintenance personnel to quickly grasp the core situation of the anomaly and take countermeasures.

[0081] Based on the above embodiments, after obtaining the target broadcast text, the method further includes: controlling the speaker set on the intelligent robot to play the target broadcast text when the target confidence level is in a first interval range; and caching the target broadcast text based on the caching module on the intelligent robot when the target confidence level is in a second interval range.

[0082] The first interval can be understood as a confidence level range set to determine whether to play the target broadcast text. It is a preset high-confidence range, corresponding to a high level of credibility for the target anomaly information and target broadcast text (e.g., confidence level ≥ 80%). Information within this interval is deemed sufficiently reliable by the system and can directly trigger the intelligent robot's speaker broadcast function without additional verification, ensuring the rapid transmission of key and effective information. The second interval can be understood as a preset low-to-medium confidence range, referring to a range where the target confidence level is below the high threshold but above the set low threshold (or only below the high threshold). This represents a range where the reliability of the anomaly determination result needs further verification, corresponding to scenarios where temporarily stored information is not broadcast immediately. Both intervals together constitute the basis for determining whether to differentiate the broadcast text based on confidence level. The first interval is larger than the second interval.

[0083] Specifically, after generating the target broadcast text, the range of the target confidence level is first determined. If it is in the first range (high confidence range), the speaker module of the intelligent robot is triggered to directly play the target broadcast text to immediately convey the abnormal information. If it is in the second range (medium to low confidence range), the target broadcast text is stored and retained through the caching module on the robot and is not actively broadcast for the time being.

[0084] In this embodiment, the broadcast text is processed differently according to the confidence level range. When the confidence level is high, the text is broadcast instantly via voice, which enables maintenance personnel to respond quickly to critical anomalies. When the confidence level is low, the text is cached and retained to avoid false alarms, thus balancing the timeliness of response and the adaptability of the scenario. The voice broadcast is intuitive and efficient, and the caching function can retain data for subsequent review and traceability. This not only meets the needs of real-time maintenance, but also provides a basis for subsequent data analysis and anomaly investigation, improving the flexibility and practicality of the overall processing flow.

[0085] Optionally, before controlling the speaker set on the intelligent robot to play the target broadcast text, the method further includes: inputting the target broadcast text into a semantic extraction model to obtain a second text, and using the second text as the target broadcast text to be played; wherein, the second text is a summary text of the generated target broadcast text.

[0086] Specifically, the semantic extraction model can be understood as a pre-trained natural language processing model focused on extracting core information. It possesses the ability to identify key text content and remove redundant information. For target broadcast texts related to energy storage device anomalies, it can accurately extract and condense key information such as anomaly type, core data, and target confidence level. The second text can be understood as a concise and simplified version of the original target broadcast text. It retains the core information of the anomaly event while removing irrelevant and redundant content, resulting in more concise language and shorter length. This adapts to the needs of intelligent robot speaker voice broadcasts, facilitating maintenance personnel to quickly capture core anomaly information.

[0087] Specifically, before triggering the intelligent robot speaker to play the target broadcast text, the generated target broadcast text is first input into the pre-trained semantic extraction model. The model condenses and simplifies the original broadcast text through the logic of extracting core information and deleting redundant content, and outputs a summary text that retains the core information of the anomaly (such as anomaly type, key data, and confidence level), which is the second text. Finally, the second text is used as the target broadcast text that the speaker actually plays.

[0088] In this embodiment, the original text is summarized by a semantic extraction model, which greatly shortens the playback time, allowing maintenance personnel to quickly grasp the core abnormal information and improve the efficiency of information transmission. The simplified text language is more concise and highlights the key points, avoiding the omission of key information due to lengthy descriptions. At the same time, it is adapted to the characteristics of speaker voice broadcasting, making the broadcast content easier to understand and remember quickly.

[0089] The technical solution of this embodiment executes the following method through a controller integrated into an intelligent robot: controlling the intelligent robot to move to a target detection position according to a predetermined target motion path, and acquiring battery-related data of at least one single cell in at least one energy storage device to be tested associated with the target detection position; wherein, the battery-related data includes at least temperature field data, sound pressure level data, and total volatile organic compound (TVOC) concentration data; for at least one energy storage device to be tested, based on the temperature field data, sound pressure level data, and TVOC concentration data of at least one single cell associated with the energy storage device to be tested, obtaining temperature anomaly information and sound pressure level data. Anomaly information and total volatile organic compound (TVOC) concentration anomaly information; based on the temperature anomaly information, sound pressure anomaly information, and TVOC concentration anomaly information of each energy storage device to be tested, determine the target anomaly information corresponding to each energy storage device to be tested, so as to determine the anomaly event type based on the target anomaly information; input battery-related data and anomaly weight attributes corresponding to the anomaly event type into the data processing model to determine the target confidence level corresponding to the anomaly event type; when the target confidence level meets the preset conditions, generate the target broadcast text based on the battery-related data, the anomaly event type, and the target confidence level through a pre-built large language model. This system enables autonomous mobile inspection via intelligent robots, allowing for multi-dimensional data collection without human intervention, thus improving inspection efficiency and safety. It integrates three key data categories—temperature, sound pressure, and volatile organic compounds—and combines anomaly weights with data model verification confidence levels, significantly enhancing the accuracy and reliability of anomaly event judgment. The generated broadcast text intuitively presents core information, providing clear evidence for subsequent processing. It is suitable for the efficient operation and maintenance of large-scale energy storage equipment, solving the problem of poor inspection efficiency in the autonomous inspection of mobile energy storage systems and improving the overall efficiency of autonomous inspection of mobile energy storage systems.

[0090] Example 2

[0091] Figure 2This is a flowchart of an LLM-based autonomous inspection method for mobile energy storage systems provided in Embodiment 2 of the present invention. The method in this embodiment is a further optimization of the method in the above embodiments. Optionally, the battery-related data corresponding to the abnormal event type is input into a pre-constructed knowledge graph to determine the abnormal weight attribute corresponding to the data content of the battery-related data; the abnormal weight attribute and all battery-related data generated within the current time window are input into a data processing model so that the feature extraction layer in the data processing model extracts the feature information of the battery-related data; the weight attribute is obtained based on the feature adjustment layer in the data processing model, and the features corresponding to the abnormal event type in the feature information are updated based on the weight attribute to obtain the updated feature information; the updated feature information is decoded by the decoding layer in the data processing model, and the target confidence level corresponding to the decoded feature information is determined based on the activation function. Figure 2 As shown, the method includes:

[0092] S210. Control the intelligent robot to move to the target detection position according to the predetermined target motion path, and acquire battery-related data of at least one single cell in at least one energy storage device to be tested associated with the target detection position; wherein, the battery-related data includes at least temperature field data, sound pressure level data and total volatile organic compound concentration data.

[0093] S220. For at least one energy storage device to be tested, based on the temperature field data, sound pressure level data, and total volatile organic compound concentration data of at least one individual battery associated with the energy storage device to be tested, obtain temperature anomaly information, sound pressure anomaly information, and total volatile organic compound concentration anomaly information.

[0094] S230. Based on the abnormal temperature information, abnormal sound pressure information, and abnormal total volatile organic compound concentration information of each energy storage device to be tested, determine the target abnormal information corresponding to each energy storage device to be tested, so as to determine the type of abnormal event based on the target abnormal information.

[0095] S240. Input the battery-related data corresponding to the abnormal event type into the pre-built knowledge graph to determine the abnormal weight attribute corresponding to the data content of the battery-related data.

[0096] Specifically, the pre-built knowledge graph can be understood as a structured semantic network that integrates information such as various abnormal event types of energy storage devices, corresponding battery-related data, and weight configuration rules. By sorting out the relationship between data and weights under different abnormal scenarios, a standardized mapping logic is formed, providing structured support for rapid weight matching. Optionally, the knowledge graph construction method includes: obtaining all abnormal types and corresponding historical battery data, and obtaining calibrated parameter weight attributes, and constructing the knowledge graph based on the above information; the root node of the knowledge graph corresponds to the weight, the child nodes correspond to the data content of historical battery data, and the leaf nodes correspond to entities. The abnormal weight attribute is a quantitative indicator adapted to a specific abnormal event type, reflecting the degree or priority of the influence of the corresponding battery-related data on the abnormality judgment. Its value is determined by the knowledge graph based on the input battery-related data, and is used to improve the accuracy of subsequent target confidence calculation.

[0097] Specifically, the identified abnormal event types are first determined, and the corresponding battery-related data is extracted. Then, this battery-related data is input into a pre-built knowledge graph of energy storage device anomalies. Through the graph's built-in association mapping rules and data matching logic, the abnormal weight attributes corresponding to the data content of the battery-related data and adapted to the current anomaly type are retrieved and determined.

[0098] For example, the pre-built knowledge graph is a pre-built Energy Storage O&M Knowledge Graph, which includes: various abnormal event nodes (temperature consistency, fan anomaly, TVOC anomaly, etc.); equipment nodes (battery clusters, fans, combiner boxes, etc.); and the relationship between anomaly patterns and sensing quantities (e.g., "TVOC anomaly → electrolyte evaporation risk"). The information recorded in the graph includes: temperature difference contributes significantly to temperature-related anomalies; peak sound pressure contributes significantly to fan failures; and TVOC changes contribute the most to leakage-related anomalies. Based on this, three weights are assigned. The initial value is given by the spectral relationship, for example: Different initial values ​​are assigned based on different event types. In actual deployment, the system adopts an online update strategy for the above three types of weights: when an inspection event is confirmed by maintenance personnel, it is used as a labeled sample to update the knowledge graph; lightweight gradient updates (such as Online SGD) can also be used to adjust the weights. This allows it to gradually approach the true abnormal contribution level, thus achieving self-learning.

[0099] In this embodiment, based on the structured association capabilities of the knowledge graph, a precise correspondence between battery-related data and abnormal weight attributes can be quickly established, avoiding the subjectivity of weight allocation and improving the accuracy of weight attribute determination. The knowledge graph can integrate massive abnormal scenarios and weight configuration experience to adapt to the differentiated needs of different abnormal event types, making the weight attributes more consistent with the actual abnormal situation, providing a reliable weight basis for subsequent target confidence calculation. At the same time, the entire process is completed automatically, improving the efficiency of the abnormal handling process.

[0100] S250. Input the abnormal weight attribute and all battery-related data generated within the current time window into the data processing model so that the feature extraction layer in the data processing model can extract the feature information of the battery-related data.

[0101] Specifically, the data processing model can be understood as a structured algorithm model designed for anomaly detection of energy storage devices. It includes core modules such as feature extraction layer, feature adjustment layer, and decoding layer. It can receive battery-related data and anomaly weight attributes as input. First, the feature extraction layer mines key features of the data. Then, the feature adjustment layer combines the weight attributes to optimize the core anomaly features. Finally, the decoding layer transforms the features and maps them through an activation function to output the target confidence level that quantifies the reliability of anomaly judgment.

[0102] Specifically, the abnormal weight attribute that is appropriate for the current abnormal event type is first obtained, as well as all battery-related data generated within the current time window. Both types of data are then input into the preset data processing model. The feature extraction layer in the model automatically mines and extracts key feature information related to anomaly determination from the battery-related data through built-in algorithms, including but not limited to data fluctuation patterns, extreme value features, and deviation trends, providing core feature support for subsequent confidence calculation.

[0103] In this embodiment, by combining the anomaly weight attribute with complete time window battery correlation data, feature extraction can be more focused on core influencing factors, improving the relevance of feature information. The feature extraction layer automatically completes feature mining, avoiding the subjectivity and inefficiency of manual extraction, while capturing deep data features that are difficult for humans to detect, laying a solid foundation for the accurate calculation of subsequent target confidence. The overall process is adapted to the needs of batch and real-time anomaly detection, improving processing efficiency and judgment reliability.

[0104] S260. Obtain weight attributes based on the feature adjustment layer in the data processing model, and update the features in the feature information corresponding to the abnormal event type based on the weight attributes to obtain the updated feature information.

[0105] Specifically, the feature adjustment layer of the data processing model first obtains the determined anomaly weight attributes, and then, in combination with the influence degree and priority of the corresponding attributes, updates the feature information output by the feature extraction layer in a targeted manner, focusing on strengthening the feature weights related to the current anomaly event type and weakening the influence of irrelevant features, and finally obtaining updated feature information that focuses on the core anomaly.

[0106] In this embodiment, feature updates are guided by weight attributes, making the feature information more consistent with the current abnormal event type and improving the accuracy of subsequent confidence calculations. The targeted optimization of the feature adjustment layer avoids interference from redundant features, highlights the role of core abnormal features, and enables the data processing model to adapt to the judgment requirements of different types of abnormalities, thereby enhancing the flexibility and adaptability of the model.

[0107] S270. Based on the decoding layer in the data processing model, the updated feature information is decoded and processed, and the target confidence corresponding to the decoded feature information is determined based on the activation function.

[0108] Specifically, the decoding layer of the data processing model receives the updated feature information output by the feature adjustment layer, and transforms the high-dimensional features into interpretable low-dimensional data using a built-in decoding algorithm. Then, it calls a preset activation function (such as Sigmoid, Softmax, etc.) to perform calculations on the decoded feature data, mapping the results to the numerical range corresponding to the confidence level, and finally outputting the target confidence level that quantifies the reliability of the anomaly determination. The model expression of the feature adjustment layer in the data processing model is shown below:

[0109] ;

[0110] in, denoted by 'abnormal weight attribute', and z represents the feature information output by the feature extraction layer.

[0111] In this embodiment, the decoding layer effectively transforms high-dimensional features, making the feature information more suitable for the needs of confidence calculation; the activation function accurately completes the numerical mapping, ensuring that the target confidence value meets the preset range and improving the standardization of the results; the two-step processing flow is logically rigorous and computationally efficient, ensuring both the accuracy of confidence calculation and the ability to quickly output results, thus meeting the timeliness requirements of real-time anomaly detection.

[0112] S280. When the target confidence level meets the preset conditions, the target broadcast text is generated by pre-built large language model based on battery-related data, abnormal event types and target confidence level.

[0113] An example of how event types and attribute values ​​are constructed: Event Instance It must include structured attribute values ​​related to the event type. For example, for a high-temperature event (temp_high), the attributes include the highest temperature, average temperature, and temperature difference. For temperature inconsistency anomalies (temp_imbalance), the core attribute is the temperature difference. For acoustical anomalous events (acoustic_abnormal), the attribute is peak sound pressure level. For the TVOC exception event (tvoc_high), the attribute is... These structured attributes will serve as key fields in the Prompt, providing them to large language models for natural language generation.

[0114] For example, a "rack" field can be set in the event label of the characterization type. "Rack" identifies the location of the device to which the detected object belongs. In an energy storage system, "rack" refers to the energy storage device under inspection managed by the BMS system, such as a battery rack. It should be noted that a rack consists of multiple individual cells (dozens to hundreds of cells); the BMS is responsible for collecting temperature, voltage, equalization, and alarm data on this rack; the inspection robot stops at the observation point corresponding to each battery rack and collects multimodal data. Therefore, "rack="C3" indicates that the current anomaly corresponds to the third battery rack (area C) within the BMS's management scope.

[0115] For example, after determining the event type, the system automatically constructs a Prompt. Taking a high-temperature event as an example, <event> {"type":"temp_high","deltaT":7,"rack":"C3"}< / event> <style>zh_informal_v1< / style> Where: type corresponds to the event type; deltaT is the temperature difference, used to explain high temperature or consistency anomalies; rack indicates which battery rack the current event belongs to. The LLM then generates a broadcast text, such as: "Attention! Battery pack C3 temperature is 7°C higher than average. It is recommended to check for fan blockage or coolant level."

[0116] It should be noted that the text generation is not executed in isolation, but is triggered when the following conditions are met: That is, only when the confidence level reaches the broadcast threshold will the LLM generate semantic broadcast and hand it over to the TTS for speech output.

[0117] For example, the confidence interval in this embodiment is divided into three levels: , , The corresponding strategy is: : Invoke TTS (Text-To-Speech) to broadcast the message and push the work order; Push notifications will be sent in text format only; no voice announcements will be provided. Recorded locally only, not pushed externally. It is important to emphasize that this embodiment does not unconditionally "generate the broadcast text first, then decide whether to broadcast it," but rather adopts a two-level process of "first determining confidence level, then deciding whether to generate text and whether to execute TTS," to avoid wasting computing resources on low-confidence events. The execution order is as follows:

[0118] For identified abnormal events First, determine the multimodal feature vector z based on the battery correlation data, and then calculate the confidence level:

[0119] ;

[0120] according to The preset interval determines whether to call LLM to generate the broadcast text y:

[0121] like The event is deemed unreliable, so it is only recorded in the local log, without calling LLM to generate text or triggering TTS.

[0122] like : Call LLM to generate a Chinese broadcast text However, it is only used as a text alert or work order description push in the App / platform and is not used for speech synthesis;

[0123] like The LLM module generates text y, which is then processed by the TTS module to synthesize y into speech. This speech is then broadcast on-site via the robot's loudspeaker, while simultaneously pushing the corresponding work order.

[0124] Therefore, there is a clear sequential relationship between text generation and voice broadcasting:

[0125] for The event: neither generates text nor broadcasts it;

[0126] for The event: First, generate text y, then decide whether to hand it over to TTS for broadcast based on the confidence interval. Among these, Text push only Then text push and voice broadcast will be carried out simultaneously.

[0127] Based on the above embodiments, the system can also perform self-learning. The system's self-learning relies on the "confirmation feedback" of inspection events. That is, when maintenance personnel manually verify the events reported by the robot and complete the repairs, the system will use the structured label of the event as a training sample to update the feature encoder. The LoRA–LLM model for generating broadcast text. Its tag format is as follows: TTS (Text-to-Speech) only indicates that the event has reached the high confidence interval. The system deems the event highly credible and requires immediate alerting of on-site personnel. However, true "confirmed repair" must be manually marked by maintenance personnel in the work order system, such as "reviewed," "processed," or "anomaly confirmed." Only after manual confirmation will the event be included in the case accumulation process. Therefore: TTS is invoked to generate voice broadcast information of the target broadcast text. The robot broadcasts the information based on the voice broadcast information. After confirmation by maintenance personnel, the event becomes a valid tag for self-learning. The "root-cause" in the tag refers to the true root cause category after manual confirmation. In energy storage inspection scenarios, root causes include, but are not limited to: blocked heat dissipation ducts, individual unit performance degradation, poor contact of busbars or connecting pieces, fan failure or wear, minor electrolyte leakage (corresponding to TVOC abnormality), environmental temperature influence, and false alarms (noise-related false alarms).

[0128] The technical solution of this embodiment involves controlling an intelligent robot to move to a target detection position according to a predetermined target motion path, and acquiring battery-related data of at least one single cell in at least one energy storage device to be tested associated with the target detection position. The battery-related data includes at least temperature field data, sound pressure level data, and total volatile organic compound (TVOC) concentration data. For at least one energy storage device to be tested, based on the temperature field data, sound pressure level data, and TVOC concentration data of at least one single cell associated with the energy storage device, temperature anomaly information, sound pressure anomaly information, and TVOC concentration anomaly information are obtained. Based on the temperature anomaly information, sound pressure anomaly information, and TVOC concentration anomaly information of each energy storage device to be tested, target anomaly information corresponding to each energy storage device to be tested is determined, so as to determine anomalies based on the target anomaly information. The process involves several steps: First, identifying common event types. Second, inputting battery-related data corresponding to abnormal event types into a pre-built knowledge graph to determine the abnormal weight attributes corresponding to the data content of the battery-related data. Third, inputting the abnormal weight attributes and all battery-related data generated within the current time window into the data processing model, enabling the feature extraction layer in the model to extract feature information from the battery-related data. Fourth, obtaining weight attributes based on the feature adjustment layer in the data processing model, updating the features corresponding to the abnormal event types in the feature information based on the weight attributes, and obtaining updated feature information. Fifth, decoding the updated feature information based on the decoding layer in the data processing model, and determining the target confidence level corresponding to the decoded feature information based on the activation function. Sixth, when the target confidence level meets preset conditions, generating the target broadcast text based on the battery-related data, abnormal event types, and target confidence level through a pre-built large language model. It achieves a closed loop of fully automated process from equipment movement detection, data collection, anomaly analysis to text generation, significantly reducing manual intervention and improving detection efficiency. Through multi-dimensional data collection, knowledge graph matching weights, and model hierarchical feature processing, it progressively optimizes the accuracy of anomaly judgment and reduces the false alarm rate. It integrates a large language model to generate broadcast text, transforming professional data into easily understandable information and lowering the operation and maintenance threshold. The logical connections between each link are tight and highly adaptable, which can meet the real-time anomaly detection needs of large-scale energy storage equipment and provide comprehensive and reliable support for operation and maintenance decisions.

[0129] Example 3

[0130] Figure 3 This is a structural schematic diagram of an intelligent robot provided in Embodiment 3 of the present invention. Figure 3 As shown, the intelligent robot includes:

[0131] The battery-related data acquisition module 310 is used to control the intelligent robot to move to the target detection position according to the predetermined target motion path, and to acquire battery-related data of at least one single cell in at least one energy storage device to be tested associated with the target detection position; wherein, the battery-related data includes at least temperature field data, sound pressure level data and total volatile organic compound concentration data;

[0132] The anomaly information determination module 320 is used to obtain temperature anomaly information, sound pressure anomaly information, and total volatile organic compound concentration anomaly information for at least one energy storage device to be tested, based on the temperature field data, sound pressure level data, and total volatile organic compound concentration data of at least one single cell associated with the energy storage device to be tested.

[0133] The abnormal event type determination module 330 is used to determine the target abnormal information corresponding to each energy storage device under test based on the abnormal temperature information, abnormal sound pressure information, and abnormal total volatile organic compound concentration information of each energy storage device under test, so as to determine the abnormal event type based on the target abnormal information.

[0134] The target confidence level determination module 340 is used to input battery-related data and anomaly weight attributes corresponding to the anomaly event type into the data processing model in order to determine the target confidence level corresponding to the anomaly event type.

[0135] The target broadcast text determination module 350 is used to generate target broadcast text based on battery-related data, abnormal event types, and target confidence when the target confidence meets preset conditions, by using a pre-built large language model.

[0136] The technical solution of this embodiment involves a battery-related data acquisition module controlling an intelligent robot to move to a target detection position according to a predetermined target motion path, and acquiring battery-related data of at least one single cell in at least one energy storage device to be tested associated with the target detection position. The battery-related data includes at least temperature field data, sound pressure level data, and total volatile organic compound (TVOC) concentration data. For each energy storage device to be tested, the anomaly information determination module, based on the temperature field data, sound pressure level data, and TVOC concentration data of at least one single cell associated with the device, obtains temperature anomaly information, sound pressure anomaly information, and TVOC concentration information. The system analyzes the following modules: Concentration Anomaly Information; Anomaly Event Type Determination Module: Based on the temperature anomaly information, sound pressure anomaly information, and total volatile organic compound concentration anomaly information of each energy storage device under test, determine the target anomaly information corresponding to each device, and then determine the anomaly event type based on this target anomaly information; Target Confidence Determination Module: Input battery-related data and anomaly weight attributes corresponding to the anomaly event type into the data processing model to determine the target confidence level corresponding to the anomaly event type; Target Broadcast Text Determination Module: When the target confidence level meets preset conditions, based on battery-related data, anomaly event type, and target confidence level, generate the target broadcast text through a pre-built large language model. This system enables autonomous mobile inspection via intelligent robots, allowing for multi-dimensional data collection without human intervention, thus improving inspection efficiency and safety. It integrates three key data categories—temperature, sound pressure, and volatile organic compounds—and combines anomaly weights with data model verification confidence levels, significantly enhancing the accuracy and reliability of anomaly event judgment. The generated broadcast text intuitively presents core information, providing clear evidence for subsequent processing. It is suitable for the efficient operation and maintenance of large-scale energy storage equipment, solving the problem of poor inspection efficiency in the autonomous inspection of mobile energy storage systems and improving the overall efficiency of autonomous inspection of mobile energy storage systems.

[0137] Based on the above embodiments, optionally, the battery-related data acquisition module 310 is specifically used to construct a regional map of the target area based on the movement of the intelligent robot in at least one floor associated with the target area, and during the movement, based on real-time positioning and mapping technology; using A The algorithm determines the target movement path based on the device location information of at least one energy storage device to be detected in the regional map.

[0138] Optionally, the anomaly information determination module 320 is specifically used to acquire the temperature field data of all individual batteries associated with the energy storage device under test within the current time window, and to perform mean processing on the temperature field data corresponding to each discrete time point within the current time window, so as to determine the temperature difference information of each discrete time point based on the temperature field data and the corresponding mean temperature data, and to use the temperature difference information as temperature anomaly information; acquire the sound pressure level data of the energy storage device under test within the current time window, and use the sound pressure level data corresponding to the sound pressure peak as sound pressure anomaly information; and determine the concentration difference value corresponding to each discrete time point based on the total volatile organic compound concentration data of the energy storage device under test at each discrete time point within the current time window, and the preset benchmark concentration data, and use the concentration difference value of each discrete time point as total volatile organic compound concentration anomaly information.

[0139] Optionally, the abnormal event type determination module 330 is specifically used to determine the first data based on the temperature anomaly information of all discrete time points within the current time window; determine the second data based on the sound pressure anomaly information of all discrete time points within the current time window; determine the third data based on the total volatile organic compound concentration anomaly information of all discrete time points within the current time window; and determine the target anomaly information of the energy storage device to be tested based on the first data, the second data, the third data, and the corresponding weight attributes.

[0140] Optionally, the abnormal event type determination module 330 is further specifically used to read the maximum value among the first data, the second data, and the third data when the target abnormal information does not meet the normal threshold range, and to take the evaluation dimension corresponding to the maximum value as the abnormal event type; wherein, the evaluation dimension includes the temperature dimension, the sound pressure dimension, and the total volatile organic compound concentration dimension.

[0141] Optionally, the target confidence determination module 340 is specifically used to input the battery-related data corresponding to the abnormal event type into a pre-built knowledge graph to determine the abnormal weight attribute corresponding to the data content of the battery-related data; input the abnormal weight attribute and all battery-related data generated within the current time window into the data processing model so that the feature extraction layer in the data processing model can extract the feature information of the battery-related data; obtain the weight attribute based on the feature adjustment layer in the data processing model to update the features in the feature information corresponding to the abnormal event type based on the weight attribute, and obtain the updated feature information; decode the updated feature information based on the decoding layer in the data processing model, and determine the target confidence corresponding to the decoded feature information based on the activation function.

[0142] Optionally, after obtaining the target broadcast text, the device is also used to control the speaker set on the intelligent robot to play the target broadcast text when the target confidence level is in the first interval range; and to cache the target broadcast text based on the cache module on the intelligent robot when the target confidence level is in the second interval range.

[0143] Optionally, before controlling the speaker set on the intelligent robot to play the target broadcast text, the device is also used to input the target broadcast text into the semantic extraction model to obtain a second text, and use the second text as the target broadcast text to be played; wherein, the second text is a summary text of the generated target broadcast text.

[0144] The LLM-based autonomous inspection device for mobile energy storage systems provided in this invention can execute the LLM-based autonomous inspection method for mobile energy storage systems provided in any embodiment of this invention, and has the corresponding functional modules and beneficial effects of the method.

[0145] Example 4

[0146] Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. The electronic device 10 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0147] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded into the RAM 13 from storage unit 18. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0148] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0149] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the LLM-based autonomous inspection method for mobile energy storage systems.

[0150] In some embodiments, the LLM-based autonomous inspection method for mobile energy storage systems can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the LLM-based autonomous inspection method for mobile energy storage systems described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the LLM-based autonomous inspection method for mobile energy storage systems by any other suitable means (e.g., by means of firmware).

[0151] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0152] The computer program used to implement the LLM-based autonomous inspection method for mobile energy storage systems of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0153] Example 5

[0154] Embodiment 5 of the present invention also provides a computer-readable storage medium storing computer instructions for causing a processor to execute an autonomous inspection method for a mobile energy storage system based on LLM, integrated in the controller of an intelligent robot, and executing the following method based on the controller:

[0155] The intelligent robot is controlled to move to the target detection position according to a predetermined target motion path, and acquire battery-related data of at least one single cell in at least one energy storage device to be tested associated with the target detection position; wherein, the battery-related data includes at least temperature field data, sound pressure level data, and total volatile organic compound concentration data;

[0156] For at least one energy storage device to be tested, temperature anomaly information, sound pressure level information, and total volatile organic compound concentration anomaly information are obtained based on the temperature field data, sound pressure level data, and total volatile organic compound concentration data of at least one individual battery associated with the energy storage device to be tested.

[0157] Based on the abnormal temperature, abnormal sound pressure, and abnormal total volatile organic compound concentration information of each energy storage device to be tested, the target abnormal information corresponding to each energy storage device to be tested is determined, and the type of abnormal event is determined based on the target abnormal information.

[0158] Battery-related data and anomaly weight attributes corresponding to the types of abnormal events are input into the data processing model to determine the target confidence level corresponding to the types of abnormal events.

[0159] When the target confidence level meets the preset conditions, the target broadcast text is generated by pre-built large language model based on battery-related data, abnormal event types, and target confidence level.

[0160] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0161] To provide interaction with an object, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the object; and a keyboard and pointing device (e.g., a mouse or trackball) through which the object provides input to the electronic device. Other types of devices can also be used to provide interaction with the object; for example, feedback provided to the object can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the object can be received in any form (including sound input, voice input, or tactile input).

[0162] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., a computer with a graphical user interface or web browser through which an item can interact with the implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0163] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0164] Example 6

[0165] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the LLM-based autonomous inspection method for mobile energy storage systems according to any embodiment of this invention.

[0166] In the implementation of a computer program product, computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​as well as conventional procedural programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0167] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0168] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. An autonomous inspection method for a mobile energy storage system based on LLM, characterized in that, Integrated in a controller of a smart robot, a method is executed based on the controller as follows: The smart robot is controlled to move to a target detection position according to a predetermined target motion path, and battery-related data of at least one single battery in at least one to-be-detected energy storage device associated with the target detection position is acquired; wherein the battery-related data at least includes temperature field data, sound pressure level data and total volatile organic compound concentration data; For the at least one to-be-detected energy storage device, temperature anomaly information, sound pressure anomaly information and total volatile organic compound concentration anomaly information are obtained according to the temperature field data, sound pressure level data and total volatile organic compound concentration data of at least one single battery associated with the to-be-detected energy storage device; According to the temperature anomaly information, the sound pressure anomaly information and the total volatile organic compound concentration anomaly information of each to-be-detected energy storage device, target anomaly information corresponding to each to-be-detected energy storage device is determined, so as to determine an abnormal event type based on the target anomaly information; The battery-related data and an abnormal weight attribute corresponding to the abnormal event type are input into a data processing model to determine a target confidence degree corresponding to the abnormal event type; When the target confidence degree meets a preset condition, a target broadcast text is generated by a pre-constructed large language model based on the battery-related data, the abnormal event type and the target confidence degree.

2. The method of claim 1, wherein, The target motion path is determined based on the following manner: The smart robot moves in at least one floor associated with a target area, and a regional map of the target area is constructed based on real-time positioning and mapping technology during the movement; Adopting A An algorithm is used to determine a target motion path based on the device location information of at least one energy storage device to be detected in the area map.

3. The method of claim 1, wherein, The temperature anomaly information, the sound pressure anomaly information and the total volatile organic compound concentration anomaly information are obtained according to the temperature field data, the sound pressure level data and the total volatile organic compound concentration data of at least one single battery associated with the to-be-detected energy storage device, including: Temperature field data of all single batteries associated with the to-be-detected energy storage device within a current time window is acquired, and the temperature field data corresponding to each discrete time point within the current time window is processed by mean value to determine temperature difference information of each discrete time point according to the temperature field data of each discrete time point and the corresponding temperature mean value data, and the temperature difference information is taken as the temperature anomaly information; The sound pressure level data of the to-be-detected energy storage device within the current time window is acquired, and the sound pressure level data corresponding to the sound pressure peak is taken as the sound pressure anomaly information; According to the total volatile organic compound concentration data of each discrete time point within the current time window of the to-be-detected energy storage device and preset reference concentration data, a concentration difference value corresponding to each discrete time point is determined, and the concentration difference value of each discrete time point is taken as the total volatile organic compound concentration anomaly information.

4. The method of claim 1, wherein, The target anomaly information corresponding to each to-be-detected energy storage device is determined according to the temperature anomaly information, the sound pressure anomaly information and the total volatile organic compound concentration anomaly information of each to-be-detected energy storage device, including: Determine first data according to temperature anomaly information of all discrete time points in the current time window; Determine second data according to the sound pressure anomaly information of all discrete time points in the current time window; Determine third data according to total volatile organic compound concentration anomaly information of all discrete time points in the current time window; Determine target anomaly information of the to-be-detected energy storage device according to the first data, the second data, the third data, and corresponding weight attributes.

5. The method of claim 1, wherein, Determine an abnormal event type based on the target anomaly information, including: When the target anomaly information does not meet the normal threshold range, read the maximum value in the first data, the second data, and the third data, and take the evaluation dimension corresponding to the maximum value as the abnormal event type; The evaluation dimension includes temperature dimension, sound pressure dimension, and total volatile organic compound concentration dimension.

6. The method of claim 1, wherein, Input the battery-related data and the abnormal weight attribute corresponding to the abnormal event type into the data processing model to determine the target confidence corresponding to the abnormal event type, including: Input the battery-related data corresponding to the abnormal event type into the pre-constructed knowledge graph to determine the abnormal weight attribute corresponding to the data content of the battery-related data; Input the abnormal weight attribute and all generated battery-related data in the current time window into the data processing model to enable the feature extraction layer in the data processing model to extract feature information of the battery-related data; Based on the feature adjustment layer in the data processing model, obtain the weight attribute to update the features corresponding to the abnormal event type in the feature information based on the weight attribute to obtain updated feature information; Based on the decoding layer in the data processing model, decode the updated feature information, and based on the activation function, determine the target confidence corresponding to the decoded feature information.

7. The method of claim 1, wherein, After obtaining the target broadcast text, the method further includes: When the target confidence is in a first interval range, control the speaker arranged on the intelligent robot to play the target broadcast text; When the target confidence is in a second interval range, cache the target broadcast text based on the cache module on the intelligent robot.

8. The method of claim 7, wherein, Before controlling the speaker arranged on the intelligent robot to play the target broadcast text, further including: Input the target broadcast text into a semantic extraction model to obtain a second text, and take the second text as the target broadcast text for playing; The second text is a summary text of the generated target broadcast text.

9. An intelligent robot, characterized in that, Including: A battery-related data acquisition module is configured to control an intelligent robot to move to a target detection position according to a pre-determined target motion path, and acquire battery-related data of at least one single battery in at least one to-be-detected energy storage device associated with the target detection position; wherein the battery-related data at least includes temperature field data, sound pressure level data, and total volatile organic compound concentration data. An abnormal information determination module is configured to obtain temperature abnormal information, sound pressure abnormal information, and total volatile organic compound concentration abnormal information according to the temperature field data, the sound pressure level data, and the total volatile organic compound concentration data of the at least one single battery associated with the at least one to-be-detected energy storage device; An abnormal event type determination module is configured to determine target abnormal information corresponding to each to-be-detected energy storage device according to the temperature abnormal information, the sound pressure abnormal information, and the total volatile organic compound concentration abnormal information of each to-be-detected energy storage device, so as to determine an abnormal event type based on the target abnormal information; A target confidence determination module is configured to input the battery-related data and an abnormal weight attribute corresponding to the abnormal event type into a data processing model to determine a target confidence corresponding to the abnormal event type; A target broadcast text determination module is configured to generate a target broadcast text through a pre-constructed large language model based on the battery-related data, the abnormal event type, and the target confidence when the target confidence satisfies a preset condition.

10. A computer program product comprising a computer program, characterized in that, The computer program, when executed by a processor, implements the LLM-based mobile energy storage system autonomous inspection method of any one of claims 1-8.

Images