A knowledge-driven electric vehicle power battery pack risk prediction method and system

Abstract

The present invention discloses a knowledge-driven electric vehicle power battery pack risk prediction method and system. First, detailed parameters of the electric vehicle power battery pack and vehicle operating environment data are collected through a distributed network. Then, a semantic model for electric vehicle batteries is constructed by analyzing domain knowledge, and a series of inference rules are generated based on the semantic model. Then, a risk prediction model is constructed by combining deep learning with domain knowledge to realize analysis and risk prediction of electric vehicle power battery pack data, thereby improving the overall safety and user experience of electric bicycles and promoting the sustainable development of the industry.

Description

Technical Field

[0001] The present invention relates to the technical field of electric vehicle power battery pack risk prediction, and in particular to a knowledge-driven electric vehicle power battery pack risk prediction method and system. Background Art

[0002] With the widespread use and popularity of electric vehicles (including electric bicycles, electric motorcycles and mopeds), batteries, as their core components, are directly related to the vehicle's range and user riding experience, and their performance and safety have received widespread attention. Battery pack failure not only affects the normal use of electric bicycles, but may also cause serious safety accidents. Therefore, the monitoring and risk prediction technology of electric vehicle power battery packs has become particularly important.

[0003] See also Figure 1 The current technology mainly predicts the risk of electric vehicle power battery packs according to the following process:

[0004] 1. Collect maintenance data and historical fault information of electric vehicle power battery packs;

[0005] 2. Perform data processing and feature extraction;

[0006] 3. Extract and establish a series of inference rules and set thresholds;

[0007] 4. Construct a risk prediction model for electric vehicle power battery packs;

[0008] 5. Use existing data to train the model, optimize and determine the model parameters;

[0009] 6. Obtain and process real-time operating data of electric vehicle power battery packs;

[0010] 7. Use the model to perform data analysis and prediction to obtain the final risk prediction results.

[0011] In traditional technical solutions, electric vehicle power battery management technology is mainly based on basic data monitoring, simple power estimation and basic fault alarm mechanism. Although these technologies have played a certain role in initially ensuring the use of batteries, they still have great deficiencies in dealing with the special needs of electric vehicle application scenarios. Although mining and analyzing the monitoring data of electric vehicle power battery packs through deep learning methods can improve the efficiency of risk prediction, the lack of effective integration and utilization of domain knowledge limits the system optimization and decision support capabilities. To this end, we propose a knowledge-driven electric vehicle power battery pack risk prediction method and system. Summary of the invention

[0012] The purpose of the present invention is to provide a knowledge-driven electric vehicle power battery pack risk prediction method, which collects monitoring data of electric vehicle power battery packs in real time through a distributed architecture, and at the same time enhances the efficiency and accuracy of the risk prediction model based on deep learning by constructing a semantic model for electric vehicle power battery packs, thereby improving the maintenance efficiency and safety of electric vehicle batteries.

[0013] To achieve the above object, the present invention provides the following technical solution: a knowledge-driven electric vehicle power battery pack risk prediction method, the method comprising the following steps:

[0014] S11. Collect data related to the vehicle power battery pack through the BMS of the electric bicycle and the Internet;

[0015] S12. Combine battery-related historical data and field literature to establish a field-related semantic graph by extracting concepts and semantic relationships between keywords; also establish a local semantic model for vehicles related to charging stations;

[0016] S13, generating a series of inference rules for electric bicycle power battery packs based on the semantic model;

[0017] S14, pre-processing the real-time input monitoring data of the electric bicycle power battery pack;

[0018] S15, classify the data based on the semantic model, extract feature information from it, and perform vectorization processing;

[0019] S16. Build a power battery pack risk prediction model based on deep learning methods, and perform model training and parameter optimization;

[0020] S17. Use the model to analyze and predict the input data, and perform semantic enhancement of risk prediction based on inference rules to optimize the prediction results;

[0021] S18. Feedback the risk prediction results to the user or monitoring platform in real time to achieve autonomous early warning and diagnosis of the safety of the power battery pack.

[0022] Preferably, in S11, data is collected through the battery management system of the electric vehicle. The data collected according to the standards of the electric vehicle include: battery current, voltage, cell voltage, battery temperature, charging current, voltage, power, status, vehicle mileage, speed, ampere-hour integration and watt-hour integration. The collected data is stored in an internal storage medium, affixed with a timestamp, and uploaded to the cloud through a mobile communication network.

[0023] Preferably, data related to the electric vehicle battery field is collected through multiple data sources, including relevant literature in the field, and combined with the historical monitoring data collected by S11, the key field knowledge contained in these data is identified and extracted to establish a global semantic model for electric vehicle battery monitoring. , semantic model It can be further expressed as:

[0024] ;

[0025] in, Indicates the battery component, Indicates battery performance. Indicates the cause of the fault. Indicates symptoms, Represents failure mode; multiple levels of concepts are further established under each core concept to form a conceptual hierarchy.

[0026] Preferably, for A set of concepts, and further establishes a set of semantic relationships between concepts based on domain knowledge and experience , based on semantic model and its relationship set , generate a series of inference rules, and record the rule set as , any rule The form is as follows:

[0027] ;

[0028] in, for The concept of entity in for The concept of entity in .

[0029] Preferably, the distributed network architecture of "cloud-charging station-vehicle terminal" is used to collect monitoring data, and the monitoring data is collected for a certain period of time. All vehicles that are relatively stably connected to a charging station within the range can further build a local semantic model ; For each charging station in the area Internal energy consumption and pattern characteristics, and establish a corresponding local semantic model based on the ontology representation language , the model includes meta-information related to the energy consumption data of charging stations and the relationships between energy consumption related entities .

[0030] Preferably, in S12, the data is segmented according to the battery operating conditions and time intervals, each charging and discharging segment is a data segment, and the data is filtered according to the model excitation conditions; at the same time, the data is normalized to the [0,1] interval.

[0031] Preferably, the semantic model is used in S13 The data is classified based on the classification system, including remaining battery energy, battery power, battery efficiency, battery life and battery safety status. On this basis, two types of risk assessment parameters are formed: fault indicators and aging / health indicators.

[0032] Preferably, in S14, a power battery pack risk prediction model based on deep learning is constructed, feature extraction and vectorization processing are performed on the data through a feature extraction network, and then the vectorized data is classified using a classification network, and the network structure and training parameters are optimized at the same time.

[0033] Preferably, the trained and optimized model is used to analyze the battery performance data input later. Perform analysis and prediction to obtain the corresponding symptom manifestation result set ;

[0034] Will and Perform pre-order similarity matching on the rules in the table to find the rule with the highest matching degree for the conditional part of the rule ; The matching steps are as follows:

[0035] S21, yes Any rule in , extract its conditional part ;

[0036] S22, will and Perform similarity matching. If the matching result exceeds the preset empirical threshold, ,Will Mark as a candidate rule;

[0037] S23, traversal All the rules in the loop execute the operations of S21 and S22 to obtain a rule subset ;

[0038] S24, will The rules in step S22 are arranged in descending order according to their similarity values ​​in step S22, and the top Rules;

[0039] By matching rules, we can get the corresponding reasoning results. If the vehicle transmits data directly to the cloud, the result is used as the final battery risk prediction result; if the vehicle transmits data through a charging station, while performing the above prediction and reasoning, it is further based on the local semantic model corresponding to the charging station. , and the inference results Perform semantic-based pattern matching to obtain the final battery risk prediction results .

[0040] A knowledge-driven electric vehicle power battery pack risk prediction system encapsulates the above-mentioned knowledge-driven electric vehicle power battery pack risk prediction method into an application system and deploys it to a cloud environment.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] The present invention first collects detailed parameters of electric vehicle power battery packs and vehicle operating environment data through a distributed network; then, a semantic model for electric vehicle batteries is constructed by analyzing domain knowledge, and a series of reasoning and blame are generated based on the semantic model; then, a risk prediction model is constructed by combining deep learning with domain knowledge to realize the analysis and risk prediction of electric vehicle power battery pack data, thereby improving the overall safety and user experience of electric bicycles and promoting the sustainable development of the industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Flowchart for risk prediction of electric vehicle power battery packs for current technology;

[0044] Figure 2 A flow chart of risk prediction of electric vehicle power battery packs driven by the knowledge of the present invention. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] Embodiment 1

[0047] See also Figure 2 The present invention provides a technical solution: a knowledge-driven electric vehicle power battery pack risk prediction method, the method comprising the following steps:

[0048] S11. Collect data related to the vehicle power battery pack through the BMS of the electric bicycle and the Internet;

[0049] S12. Combine battery-related historical data and field literature to establish a field-related semantic graph by extracting concepts and semantic relationships between keywords; also establish a local semantic model for vehicles related to charging stations;

[0050] S13, generating a series of inference rules for electric bicycle power battery packs based on the semantic model;

[0051] S14, pre-processing the real-time input monitoring data of the electric bicycle power battery pack;

[0052] S15, classify the data based on the semantic model, extract feature information from it, and perform vectorization processing;

[0053] S16. Build a power battery pack risk prediction model based on deep learning methods, and perform model training and parameter optimization;

[0054] S17. Use the model to analyze and predict the input data, and perform semantic enhancement of risk prediction based on inference rules to optimize the prediction results;

[0055] S18. Feedback the risk prediction results to the user or monitoring platform in real time to achieve autonomous early warning and diagnosis of the safety of the power battery pack.

[0056] As a preferred implementation, data is collected through the battery management system of the electric vehicle in S11. The data collected according to the standards of the electric vehicle include: battery current, voltage, cell voltage, battery temperature, charging current, voltage, power, status, vehicle mileage, speed, ampere-hour integration and watt-hour integration. The collected data is stored in an internal storage medium, affixed with a timestamp, and uploaded to the cloud through a mobile communication network.

[0057] As a preferred implementation, data related to the electric vehicle battery field is collected through multiple data sources, including relevant literature in the field, and combined with historical monitoring data collected by S11, the key domain knowledge contained in these data is identified and extracted to establish a global semantic model for electric vehicle battery monitoring. , semantic model It can be further expressed as:

[0058] ;

[0059] in, Indicates the battery component, Indicates battery performance. Indicates the cause of the fault. Indicates symptoms, Represents failure mode; multiple levels of concepts are further established under each core concept to form a conceptual hierarchy.

[0060] As a preferred embodiment, A set of concepts, and further establishes a set of semantic relationships between concepts based on domain knowledge and experience , based on semantic model and its relationship set , generate a series of inference rules, and record the rule set as , any rule The form is as follows:

[0061] ;

[0062] in, for The concept of entity in for The concept of entity in .

[0063] As a preferred implementation, monitoring data collection is performed based on a distributed network architecture of "cloud-charging station-vehicle terminal" to collect data within a certain period of time. All vehicles that are relatively stably connected to a charging station within the range can further build a local semantic model ; For each charging station in the area Internal energy consumption and pattern characteristics, and establish a corresponding local semantic model based on the ontology representation language , the model includes meta-information related to the energy consumption data of charging stations and the relationships between energy consumption related entities .

[0064] As a preferred implementation, in S12, the data is segmented according to the battery operating conditions and time intervals, each charging and discharging segment is a data segment, and the data is filtered according to the model excitation conditions; at the same time, the data is normalized to the [0,1] interval.

[0065] As a preferred implementation, the semantic model is used in S13 The data is classified based on the classification system, including remaining battery energy, battery power, battery efficiency, battery life and battery safety status. On this basis, two types of risk assessment parameters are formed: fault indicators and aging / health indicators.

[0066] As a preferred implementation, a power battery pack risk prediction model based on deep learning is constructed in S14, feature extraction and vectorization of data are performed through a feature extraction network, and then the vectorized data is classified using a classification network, while the network structure and training parameters are optimized.

[0067] As a preferred embodiment, the trained and optimized model is used to analyze the battery performance data input subsequently. Perform analysis and prediction to obtain the corresponding symptom manifestation result set ;

[0068] Will and Perform pre-order similarity matching on the rules in the table to find the rule with the highest matching degree for the conditional part of the rule ; The matching steps are as follows:

[0069] S21, yes Any rule in , extract its conditional part ;

[0070] S22, will and Perform similarity matching. If the matching result exceeds the preset empirical threshold, ,Will Mark as a candidate rule;

[0071] S23, traversal All the rules in the loop execute the operations of S21 and S22 to obtain a rule subset ;

[0072] S24, will The rules in step S22 are arranged in descending order according to their similarity values ​​in step S22, and the top Rules;

[0073] By matching rules, we can get the corresponding reasoning results. If the vehicle transmits data directly to the cloud, the result is used as the final battery risk prediction result; if the vehicle transmits data through a charging station, while performing the above prediction and reasoning, it is further based on the local semantic model corresponding to the charging station. , and the inference results Perform semantic-based pattern matching to obtain the final battery risk prediction results .

[0074] A knowledge-driven electric vehicle power battery pack risk prediction system encapsulates the above-mentioned knowledge-driven electric vehicle power battery pack risk prediction method into an application system and deploys it to a cloud environment.

[0075] Taking electric bicycles that support Internet data transmission as an example, traditional technical solutions still have great deficiencies when dealing with the special needs of electric bicycle application scenarios. Although mining and analyzing the monitoring data of electric bicycle power battery packs through deep learning methods can improve the efficiency of risk prediction, the lack of effective integration and utilization of domain knowledge leads to limited system optimization and decision-making support capabilities. This embodiment proposes a knowledge-driven distributed electric bicycle power battery pack risk prediction method, which improves the efficiency and accuracy of risk prediction and further improves the overall safety and user experience of electric bicycles.

[0076] Embodiment 2

[0077] Based on the method described in Example 1, this embodiment provides further specific steps as follows:

[0078] (1) First, collect relevant information in the field of electric vehicle batteries from multiple data sources, including but not limited to industry standard documents, manufacturer specifications, user reviews, etc. Then, clean and preprocess these unstructured text data, such as removing noise and unifying the format.

[0079] (2) Use the high-frequency word statistics method in natural language processing (NLP) technology to extract characteristic words related to the characteristics of electric vehicle batteries. At the same time, apply the named entity recognition (NER) algorithm to automatically extract key concepts and establish semantic relationships between them. For example, from the national standard "Safety Technical Specifications for Lithium-ion Batteries for Electric Bicycles", key concepts such as battery packs, coding labels, high-temperature resistant materials, and semantic relationships such as has_property can be extracted.

[0080] (3) Based on the information extracted above, two levels of semantic models can be further constructed: (a) Global semantic model: covers widely applicable concepts and their associations, suitable for general analysis of the entire electric vehicle battery industry. (b) Local semantic model: reflects a more detailed and specific knowledge structure for specific regions or application scenarios (such as the range of charging stations). These two models not only contain rich semantic information, but also can effectively solve the problem of inconsistent terminology between data from different sources, map similar concepts into a unified representation, and thus facilitate subsequent data fusion and classification tasks. In addition, the semantic model is also used to generate a series of rules, for example, battery pack coding identification → high temperature resistant material.

[0081] (4) Using the established semantic model, battery-related data from different units can be effectively integrated and classified, even if there are certain differences in their original descriptions. For example, expressions such as "battery pack", "power battery pack", and "electric vehicle battery pack" actually express the same or similar concepts in actual applications. In this case, these entities can be uniformly mapped to the concepts of the global or local semantic model to facilitate subsequent integration processing. For example, the battery data provided by the application unit includes the following data fields: voltage, current, temperature, SOH, SOC, mileage, number of cycles, battery status, etc. The concepts corresponding to these fields have been pre-defined in the semantic model, so the data fields can be uniformly mapped to the semantic model.

[0082] (5) On this basis, the risk prediction model can be further used to analyze and predict battery-related data, including the changes in battery voltage, current, SOC, etc. over time, detect abnormal conditions hidden in the data and give early warnings.

[0083] (6) For abnormal conditions in the risk prediction process, they are further matched with the rules generated in step (3) to obtain the final risk prediction result.

[0084] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0085] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A knowledge-driven vehicle power battery pack risk prediction method, characterized in that: The method comprises the following steps: S11. Collect historical data related to the vehicle power battery pack through the vehicle's BMS and the Internet; In the S11, data is collected through the vehicle's BMS. The data collected according to the vehicle's standards include: battery current, voltage, cell voltage, battery temperature, charging current, voltage, power, status, vehicle mileage, speed, ampere-hour integration and watt-hour integration. The collected data is stored in an internal storage medium, timestamped, and uploaded to the cloud through a mobile communication network. S12, combining historical data related to the power battery pack and field literature, extracting concepts and semantic relationships between keywords to establish a field-related semantic map; and also establishing a local semantic model for vehicles related to the charging station. Collect data related to the electric vehicle battery field through multiple data sources, including literature in the field, and combine with historical data collected by S11 to identify and extract data related to the field and key domain knowledge contained in the historical data, and establish a global semantic model for vehicle power battery pack monitoring , semantic model Further expressed as: ； in, Indicates the battery component, Indicates battery performance. Indicates the cause of the fault. Indicates symptoms, Represents the failure mode; further establishes multiple levels of concepts under each core concept to form a conceptual hierarchical relationship; S13, generates a series of inference rules for vehicle power battery packs based on the global semantic model K; against A set of concepts, and further establishes a set of semantic relationships between concepts based on domain knowledge and experience , based on semantic model and its relationship set , generate a series of inference rules, and record the rule set as , any rule The form is as follows: ； in, for The concept of entity in for The concept of entity in S14, pre-processing the vehicle power battery pack monitoring data input in real time; S15, classifying the historical data based on the global semantic model, extracting feature information therefrom, and performing vectorization processing; S16. Build a power battery pack risk prediction model based on deep learning methods, and perform model training and parameter optimization; S17, using the model to analyze and predict the monitoring data, and performing semantic enhancement of risk prediction based on inference rules to optimize risk prediction results; S18. Feedback the risk prediction results to the user or monitoring platform in real time to achieve autonomous early warning and diagnosis of the safety of the power battery pack.

2. The knowledge-driven vehicle power battery pack risk prediction method according to claim 1 is characterized in that: Data collection is carried out based on the distributed network architecture of "cloud-charging station-vehicle terminal", targeting at All vehicles that are relatively stably connected to a charging station within the range can further build a local semantic model ; For each charging station in the area Internal energy consumption and pattern characteristics, and establish a corresponding local semantic model based on the ontology representation language , the model includes meta-information related to the energy consumption data of charging stations and the relationships between energy consumption related entities .

3. The knowledge-driven vehicle power battery pack risk prediction method according to claim 2 is characterized in that: According to the operating conditions and time intervals of the power battery pack, the data is segmented, with each charging and discharging segment as a data segment, and the data is filtered according to the model excitation conditions; at the same time, the data is normalized to the [0,1] interval.

4. The knowledge-driven vehicle power battery pack risk prediction method according to claim 3 is characterized in that: Leveraging semantic models A classification system is used to classify the historical data, and the data is classified from the dimensions of remaining battery energy, battery power, battery efficiency, battery life, and battery safety status. On this basis, two types of risk assessment parameters are formed: fault indicators and aging / health indicators.

5. The knowledge-driven vehicle power battery pack risk prediction method according to claim 4 is characterized in that: A power battery pack risk prediction model is built based on deep learning. The feature extraction network is used to extract and vectorize the data. The vectorized data is then classified using the classification network. The network structure and parameters are trained and optimized.

6. The knowledge-driven vehicle power battery pack risk prediction method according to claim 5 is characterized in that: Use the trained and optimized model to monitor the subsequent real-time input of power battery pack monitoring data Perform analysis and prediction to obtain the corresponding symptom manifestation result set ; Will and Perform pre-order similarity matching on the rules in the table to find the rule with the highest matching degree for the conditional part of the rule ; The matching steps are as follows: S21, yes Any rule in , extract its conditional part ; S22, will and Perform similarity matching. If the matching result exceeds the preset empirical threshold, ,Will Mark as a candidate rule; S23, traversal All the rules in the loop execute the operations of S21 and S22 to obtain a rule subset ; S24, will The rules in S22 are sorted in descending order according to their similarity values ​​in S22, and the top Rules; By matching rules, we can get the corresponding reasoning results. If the vehicle transmits data directly to the cloud, the result As the final power battery pack risk prediction result; if the vehicle transmits data through a charging station, while performing the above prediction and reasoning, further based on the local semantic model corresponding to the charging station , and the inference results Perform semantic-based pattern matching to obtain the final power battery pack risk prediction results .

7. A knowledge-driven vehicle power battery pack risk prediction system, characterized in that: The knowledge-driven vehicle power battery pack risk prediction method described in any one of claims 1 to 6 is encapsulated as an application system and deployed to a cloud environment.

Images