技术领域Technical Field
本发明涉及充电故障模型技术领域，尤其涉及一种基于大语言模型的充电桩故障诊断模型。The present invention relates to the technical field of charging fault models, and in particular to a charging pile fault diagnosis model based on a large language model.
背景技术Background technique
近年来，电动汽车的推广，促进了充电桩的发展。充电桩不断增长的同时给运营维修工作带来了压力。一般而言，公用充电桩被布置在室外，暴露于自然环境中。因此，这些充电设备常年经历日晒雨淋，容易受到高温、高湿以及各种不易预测的人为条件影响。这些因素会到导致充电桩出现各种故障，需要及时处理。现有的充电桩故障诊断系统已经初步具有智能化水平。可以结合多种传感器，如视频、红外、烟雾、温度等，在远程端实现监控、重启和简单故障排查等操作。但是，具体的故障还是需要具有经验的维修人员现场排查，定期维护。工作人员需要检查设备外观、设备内部电气信息，甚至需要分析充电设备与电动汽车BMS之间的交互报文。In recent years, the promotion of electric vehicles has promoted the development of charging piles. The continuous growth of charging piles has brought pressure to operation and maintenance work. Generally speaking, public charging piles are arranged outdoors and exposed to the natural environment. Therefore, these charging equipment are exposed to the sun and rain all year round and are easily affected by high temperature, high humidity and various unpredictable human conditions. These factors will cause various faults in the charging piles and need to be handled in time. The existing charging pile fault diagnosis system has initially reached an intelligent level. It can combine a variety of sensors, such as video, infrared, smoke, temperature, etc., to realize monitoring, restarting and simple troubleshooting at the remote end. However, specific faults still require on-site troubleshooting by experienced maintenance personnel and regular maintenance. The staff needs to check the appearance of the equipment, the internal electrical information of the equipment, and even analyze the interactive messages between the charging equipment and the electric vehicle BMS.
现有的故障诊断手段已经初步具备智能处理的能力，包括远程诊断、远程重启、模块化诊断等等。但是，目前这些智能手段仅适用于单一故障场景。多种潜在故障并存的场景，仍然需要具有检修经验的工人处理。工人需要检查设备外观、设备内部电气信息，甚至需要分析充电设备与电动汽车BMS之间的交互报文。这种诊断方式需要耗费大量的人力，而且诊断的经验不容易推广应用。总之，需要进一步提升充电设备故障诊断系统的智能化水平。一方面需要降低人工维修成本，另一方面需要从人工维修经验中快速提取可以推广的范式。Existing fault diagnosis methods have initially acquired the ability of intelligent processing, including remote diagnosis, remote restart, modular diagnosis, etc. However, these intelligent methods are currently only applicable to single fault scenarios. Scenarios where multiple potential faults coexist still require workers with maintenance experience to handle. Workers need to check the appearance of the equipment, the internal electrical information of the equipment, and even need to analyze the interactive messages between the charging equipment and the electric vehicle BMS. This diagnostic method requires a lot of manpower, and the diagnostic experience is not easy to promote and apply. In short, the intelligence level of the charging equipment fault diagnosis system needs to be further improved. On the one hand, it is necessary to reduce the cost of manual maintenance, and on the other hand, it is necessary to quickly extract paradigms that can be promoted from manual maintenance experience.
现有技术中：申请号为201810508107.5的中国专利申请：一种充电桩故障检修系统及方法,系统包括故障采集器、故障录播仪、智能自检插座以及故障诊断云系统，故障采集器用于采集充电桩的硬件运行状态以及通讯总线的通讯数据；故障录播仪，用于对接收到的所述硬件运行状态和所述通讯数据进行在线压缩处理和数据加密处理；智能自检插座，用于为所述充电桩提供模拟负载接口:故障诊断云系统，用于接收所述故障录播仪发来的数据,并对数据进行归类分析，并根据归类分析结果通过智能自检插座对充电桩进行离线诊断和分析。In the prior art: Chinese patent application with application number 201810508107.5: A charging pile fault inspection and repair system and method, the system includes a fault collector, a fault recorder, an intelligent self-check socket and a fault diagnosis cloud system, the fault collector is used to collect the hardware operation status of the charging pile and the communication data of the communication bus; the fault recorder is used to perform online compression processing and data encryption processing on the received hardware operation status and the communication data; the intelligent self-check socket is used to provide a simulated load interface for the charging pile: the fault diagnosis cloud system is used to receive the data sent by the fault recorder, and classify and analyze the data, and perform offline diagnosis and analysis on the charging pile through the intelligent self-check socket according to the classification and analysis results.
上述系统的数据更新依赖故障诊断结果的维修工单，但是维修工单是维修人员结果的简单记录，缺乏维修过程的详细经验，缺少智能化辅助功能。The data update of the above system relies on the maintenance work order of the fault diagnosis result, but the maintenance work order is a simple record of the maintenance personnel's results, lacking detailed experience of the maintenance process and lacking intelligent auxiliary functions.
申请号为202110544358.0，的中国专利申请：一种基于优化深度置信网络的电动汽车交流充电桩的故障诊断方法，首先选取电动汽车交流充电桩的运行状态数据,标注运行状态,建立数据集；其次将数据集划分为训练集测试集,并对其进行预处理；接着将线性受限玻尔兹曼机作为深度置信网络的输出层,构建优化深度置信网络:然后采用改进粒子群算法调整所述网络的超参数:再使用训练集数据训练优化深变置信网络，继而得到交流充电桩故障诊断模型，并使用测试集数据评估故障诊断模型的性能；最后将电动汽车交流充电桩的运行数据输入故障诊断模型,得到所述充电桩的故障诊断结果。The Chinese patent application with application number 202110544358.0 is: A fault diagnosis method for an electric vehicle AC charging pile based on an optimized deep belief network. First, the operating status data of the electric vehicle AC charging pile is selected, the operating status is marked, and a data set is established; secondly, the data set is divided into a training set and a test set, and preprocessed; then the linear restricted Boltzmann machine is used as the output layer of the deep belief network to construct an optimized deep belief network: then the improved particle swarm algorithm is used to adjust the hyperparameters of the network: the training set data is used to train the optimized deep variable belief network, and then the AC charging pile fault diagnosis model is obtained, and the test set data is used to evaluate the performance of the fault diagnosis model; finally, the operating data of the electric vehicle AC charging pile is input into the fault diagnosis model to obtain the fault diagnosis result of the charging pile.
这种方法在实际应用中，能够给检修人员一定的建议。但是作为一种离线训练的模型，在数据集中没有出现过的故障无法被识别出来。因此，这种方法不具有持续升级的能力。This method can provide some suggestions to maintenance personnel in practical applications. However, as an offline training model, faults that have not appeared in the data set cannot be identified. Therefore, this method does not have the ability to be continuously upgraded.
申请号为202111505982.6的中国专利申请一种基于深度学习的充电设施故障检测方法及系统。包括对电动汽车充电设备的运行状态数据进行分析，并进行预处理,获得数据集；将数据集划分为训练集、验证集以及测试集；建立基于自适应深度置信网络的充电设备故障诊断模型,使用训练集对其进行训练；使用验证集和测试集对故障诊断模型的性能进行评估；将充电设备的运行状态数据输入满足要求的充电设备故障诊断模型,得到诊断结果。The Chinese patent application with application number 202111505982.6 is a method and system for detecting faults in charging facilities based on deep learning. It includes analyzing the operating status data of electric vehicle charging equipment and preprocessing it to obtain a data set; dividing the data set into a training set, a validation set, and a test set; establishing a charging equipment fault diagnosis model based on an adaptive deep belief network and training it with the training set; using the validation set and the test set to evaluate the performance of the fault diagnosis model; inputting the operating status data of the charging equipment into the charging equipment fault diagnosis model that meets the requirements to obtain a diagnosis result.
上述方法虽然实现了基于大数据的充电设施运行状态规模化在线评价，但是对于一些需要现场排查的故障，除了不能提供很好的现场辅助，也不能充分地吸收人工维修的经验Although the above method realizes the large-scale online evaluation of the operating status of charging facilities based on big data, it cannot provide good on-site assistance for some faults that require on-site troubleshooting, nor can it fully absorb the experience of manual maintenance.
综上，随着充电站桩规模的增加，目前的检修方式需要大量的人力资源，维修过程费力、维修后的经验无法有效推广，不利于充电站桩的智能化运营。In summary, with the increase in the scale of charging stations, the current maintenance method requires a large amount of human resources, the maintenance process is laborious, and the experience after maintenance cannot be effectively promoted, which is not conducive to the intelligent operation of charging stations.
发明内容Summary of the invention
本发明的目的在于提出一种基于大语言模型，具有智能化交互能力，并且持续更新故障诊断信息的充电桩故障诊断模型。The purpose of the present invention is to propose a charging pile fault diagnosis model based on a large language model, with intelligent interaction capabilities, and continuously updating fault diagnosis information.
为达到上述目的，本发明提出一种基于大语言模型的充电桩故障诊断模型，基于充电桩故障诊断多模态输入信息和历史故障排除经验，构建可持续更新的充电桩故障诊断知识库，并且转换为充电桩故障知识图谱；通过所述充电桩故障知识图谱微调预训练大语言模型，实现持续更新的充电桩故障诊断大语言模型的建立；To achieve the above-mentioned purpose, the present invention proposes a charging pile fault diagnosis model based on a large language model. Based on the multimodal input information of the charging pile fault diagnosis and historical troubleshooting experience, a continuously updated charging pile fault diagnosis knowledge base is constructed and converted into a charging pile fault knowledge graph; the pre-trained large language model is fine-tuned through the charging pile fault knowledge graph to achieve the establishment of a continuously updated charging pile fault diagnosis large language model;
所述充电桩故障诊断大语言模型在充电桩故障排除过程中为工作人员提供交互功能，输出故障排除辅助，以及获取故障排除的人工反馈经验；将故障排除经验补充加入所述知识库，用于持续更新所述充电桩故障诊断大语言模型。The charging pile fault diagnosis large language model provides interactive functions for staff during the charging pile fault troubleshooting process, outputs troubleshooting assistance, and obtains manual feedback experience for troubleshooting; the troubleshooting experience is added to the knowledge base for continuous updating of the charging pile fault diagnosis large language model.
进一步的，所述多模态输入信息包括视频、音频、电学信号和文本信息。Furthermore, the multimodal input information includes video, audio, electrical signals and text information.
进一步的，所述充电桩故障知识图谱的构建，需要将充电桩故障诊断知识库的诊断信息进行关键词优化提取便于模型的理解，包括实体识别、关系抽取和事件抽取。Furthermore, the construction of the charging pile fault knowledge graph requires keyword optimization and extraction of the diagnostic information of the charging pile fault diagnosis knowledge base to facilitate the understanding of the model, including entity recognition, relationship extraction and event extraction.
进一步的，所述预训练大语言模型为获取开源的预训练大语言模型，以作为模型基座。Furthermore, the pre-trained large language model is an open source pre-trained large language model to serve as a model base.
进一步的，微调预训练大语言模型包括以下步骤：Furthermore, fine-tuning the pre-trained large language model includes the following steps:
S1:加载预训练模型；S1: Load the pre-trained model;
S2:冻结预训练模型的底层权重；S2: Freeze the underlying weights of the pre-trained model;
S3:定义新的任务相关层，在模型的顶部或中间添加新的层次或模块，以适应充电桩的故障诊断任务；S3: define new task-related layers, add new layers or modules at the top or in the middle of the model to adapt to the fault diagnosis task of charging piles;
S4:定义损失函数，用于衡量模型的预测与实际标签之间的差异；S4: Define the loss function to measure the difference between the model's prediction and the actual label;
S5:数据输入和批处理，训练数据准备成模型可以接受的格式，并设置合适的批处理大小；S5: Data input and batch processing: prepare the training data into a format acceptable to the model and set an appropriate batch size;
S6:训练模型，使用训练数据集，通过反向传播算法来更新模型的权重，以最小化损失函数；S6: Train the model, using the training data set to update the model weights through the back propagation algorithm to minimize the loss function;
S7:验证和监控，在每个训练周期结束后，使用验证数据集来评估模型的性能；S7: Validation and monitoring, after each training cycle, use the validation dataset to evaluate the performance of the model;
S8:超参数调整，根据验证性能需要微调超参数；S8: Hyperparameter adjustment, fine-tune hyperparameters according to the verification performance needs;
S9:保存模型，即保存训练好的模型权重，以备后续的推理和部署；S9: Save the model, that is, save the trained model weights for subsequent reasoning and deployment;
S10:重复迭代，多次重复上述步骤，以进一步提高模型性能。持续监控模型性能，并根据实际需求进行微调和改进。S10: Repeat the above steps multiple times to further improve the model performance. Continuously monitor the model performance and make fine-tuning and improvements based on actual needs.
进一步的，所述充电桩故障诊断大语言模型与工作人员的交互方式通过：将所述充电桩故障诊断大语言模型内嵌到工作人员的运维软件中，工作人员通过语音交流、文字交互或者上传图片的方式，从大语言模型获得基本的故障信息、可能故障原因以及处理方式；工作人员将新观测到的信息告诉大语言模型获得进一步的辅助信息。Furthermore, the charging pile fault diagnosis big language model interacts with the staff in the following way: the charging pile fault diagnosis big language model is embedded in the staff's operation and maintenance software, and the staff obtains basic fault information, possible fault causes and treatment methods from the big language model through voice communication, text interaction or uploading pictures; the staff tells the big language model the newly observed information to obtain further auxiliary information.
与现有技术相比，本发明的优势之处在于：Compared with the prior art, the advantages of the present invention are:
1、本发明模型采用线上持续训练和知识库更新的方式，避免离线模型不能检测新故障的缺点，提升故障诊断方法的长期有效性。1. The model of the present invention adopts the method of online continuous training and knowledge base updating to avoid the disadvantage that the offline model cannot detect new faults and improve the long-term effectiveness of the fault diagnosis method.
2、本发明采用大预言模型为维修充电桩的工作人员提供交互式辅助服务，降低了工作人员的工作难度，相比于单纯的故障类型诊断，具有更高的实际应用价值。2. The present invention adopts a large prediction model to provide interactive auxiliary services for staff who repair charging piles, which reduces the difficulty of the staff's work. Compared with simple fault type diagnosis, it has higher practical application value.
3、本发明通过大预言模型与工作人员的交互，可以实现无感自动收集故障维修经验，用于模型性能的进一步优化，有利于人工检修经验的推广。3. The present invention can realize the automatic and seamless collection of fault maintenance experience through the interaction between the large prediction model and the staff, which can be used for further optimization of the model performance and is conducive to the promotion of manual maintenance experience.
4、本发明采用多模态信息和人工经验结合的方法，结合了数据驱动方法的效率和人工检修的准确性。4. The present invention adopts a method combining multimodal information and manual experience, combining the efficiency of data-driven methods and the accuracy of manual maintenance.
附图说明BRIEF DESCRIPTION OF THE DRAWINGS
图1为本发明实施例中基于大语言模型的充电桩故障诊断模型构建流程图；FIG1 is a flow chart of building a charging pile fault diagnosis model based on a large language model in an embodiment of the present invention;
图2为本发明的实施例中基于大语言模型的充电桩故障诊断模型的应用示意图。FIG2 is a schematic diagram of an application of a charging pile fault diagnosis model based on a large language model in an embodiment of the present invention.
具体实施方式Detailed ways
为使本发明的目的、技术方案和优点更加清楚，下面将对本发明的技术方案作进一步地说明。In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be further described below.
如图1所示，本发明提出一种基于大语言模型的充电桩故障诊断模型，用于辅助对充电桩的故障诊断，所述模型的构建步骤如下：As shown in FIG1 , the present invention proposes a charging pile fault diagnosis model based on a large language model, which is used to assist in the fault diagnosis of the charging pile. The steps of constructing the model are as follows:
Step1：获取充电桩故障诊断多模态输入信息和历史故障排除经验，形成充电桩故障诊断知识库。Step 1: Obtain multi-modal input information of charging pile fault diagnosis and historical troubleshooting experience to form a charging pile fault diagnosis knowledge base.
多模态输入包括但不限于：1)视频，充电站的监控视频；2)音频，充电站的监控音频；3)电学信号，来自温度传感器、烟雾传感器、充电桩与电动汽车交互报文内包含的电压、电流、功率等；4)文本信息，记录故障发生时间、地点和故障检修记录的工单。Multimodal input includes but is not limited to: 1) video, monitoring video of the charging station; 2) audio, monitoring audio of the charging station; 3) electrical signals, such as voltage, current, power, etc. contained in the interactive messages between temperature sensors, smoke sensors, charging piles and electric vehicles; 4) text information, such as work orders that record the time and location of the fault and the fault repair record.
Step2：将知识库中的信息转化为充电桩故障诊断文本信息，并建立充电桩故障知识图谱。Step 2: Convert the information in the knowledge base into charging pile fault diagnosis text information and establish a charging pile fault knowledge graph.
通过模态转化工具(例如，BERT模型)，将视频、音频、电学信号等转化为文本信息。考虑到充电桩故障诊断文本信息含有大量的专有名词和关系，需要对文本进行关键词优化提取便于后续模型的理解。具体需要完成实体识别、关系抽取和事件抽取。1)识别实体：通过分类方法识别文本中的实体概念，包括各种电子电气元件、零件以及部件的名称，如继电器、电线和通信模块；2)关系抽取：通过分类方法识别出文本里面具有关系的实体，及它们之间的关系。如(停机-过温-风扇故障)；3)事件抽取：通过识别特殊的实体作为事件触发器，并分析事件论元。例如，将事件类型作为触发器：充电桩过温故障；事件论元，时间-2023年9月1日22点22分21秒；地点-xx地区1号站；事件主体-3号充电桩；事件客体-充电模块过温；事件原因-风扇故障。Through modal conversion tools (for example, BERT model), video, audio, electrical signals, etc. are converted into text information. Considering that the text information of charging pile fault diagnosis contains a large number of proper nouns and relationships, it is necessary to optimize the text for keyword extraction to facilitate the understanding of subsequent models. Specifically, entity recognition, relationship extraction and event extraction need to be completed. 1) Identify entities: Identify entity concepts in text through classification methods, including the names of various electronic and electrical components, parts and components, such as relays, wires and communication modules; 2) Relationship extraction: Identify entities with relationships in the text and the relationships between them through classification methods. Such as (shutdown-overtemperature-fan failure); 3) Event extraction: By identifying special entities as event triggers and analyzing event arguments. For example, using event type as a trigger: charging pile overtemperature failure; event argument, time-22:22:21 on September 1, 2023; location-station 1 in xx area; event subject-charging pile No. 3; event object-charging module overtemperature; event cause-fan failure.
Step3获取预训练大语言模型作为基座模型。Step 3: Get the pre-trained large language model as the base model.
获取开源的预训练大语言模型，包括但不限于，LLaMA2。这种预训练的大模型具有很强的理解、推理能力以及连续对话的能力，但是对专业领域内工作的处理能力不突出。Obtain open source pre-trained large language models, including but not limited to LLaMA2. This pre-trained large model has strong understanding, reasoning and continuous dialogue capabilities, but is not good at handling work in professional fields.
Step4根据所述充电桩故障诊断文本信息微调预训练大语言模型，构建充电桩故障诊断大语言模型。微调模型的步骤包括：Step 4 is to fine-tune the pre-trained large language model according to the charging pile fault diagnosis text information to construct a charging pile fault diagnosis large language model. The steps of fine-tuning the model include:
1)加载预训练模型，通过开源深度学习框架(如PyTorch)加载已经选择的预训练模型；1) Load the pre-trained model and load the selected pre-trained model through an open source deep learning framework (such as PyTorch);
2)将选择的预训练模型的底层层次的权重冻结，以保留它们在预训练原始任务中学到的特征；2) Freeze the weights of the bottom layers of the selected pre-trained model to preserve the features they learned in the original pre-training task;
3)定义新的任务相关层，在模型的顶部或中间添加新的层次或模块，以适应充电桩的故障诊断任务。例如，为了对充电桩的多种故障加以识别，添加卷积层(CNN layer)用于分类，添加循环层(RNN layer)用于时间预测，添加全连接层(FNN layer)用于输出降维等。3) Define new task-related layers, add new layers or modules at the top or in the middle of the model to adapt to the fault diagnosis task of charging piles. For example, in order to identify various faults of charging piles, add a convolutional layer (CNN layer) for classification, add a recurrent layer (RNN layer) for time prediction, and add a fully connected layer (FNN layer) for output dimensionality reduction.
4)定义损失函数，用于衡量模型的预测与实际标签之间的差异。注意，实际标签来自步骤二的知识库。例如，充电桩的故障诊断任务，使用交叉熵损失。4) Define a loss function to measure the difference between the model's prediction and the actual label. Note that the actual label comes from the knowledge base in step 2. For example, for the fault diagnosis task of charging piles, cross entropy loss is used.
5)数据输入和批处理，训练数据准备成模型可以接受的格式，并设置合适的批处理大小，批处理大小由训练效率决定。例如，训练数据是前文所述的知识图谱数据，批处理大小为32时，训练速度太慢，就提升为64或者128。5) Data input and batch processing: prepare the training data into a format acceptable to the model and set an appropriate batch size, which is determined by the training efficiency. For example, if the training data is the knowledge graph data mentioned above, the training speed is too slow when the batch size is 32, so it is increased to 64 or 128.
6)训练模型，使用训练数据集，通过反向传播算法来更新模型的权重，以最小化损失函数。通常需要进行多个训练周期(epochs)，每个周期包括多个批次的训练。例如，采用Adam、SGD等优化算法。6) Train the model, using the training data set to update the model weights through the back propagation algorithm to minimize the loss function. Usually multiple training cycles (epochs) are required, each cycle includes multiple batches of training. For example, optimization algorithms such as Adam and SGD are used.
7)验证和监控，在每个训练周期结束后，使用验证数据集来评估模型的性能。这有助于监测模型是否过拟合或欠拟合，以及是否需要进行调整。7) Validation and monitoring: After each training cycle, use the validation dataset to evaluate the performance of the model. This helps monitor whether the model is overfitting or underfitting and whether adjustments are needed.
8)超参数调整，根据验证性能需要微调超参数，如学习率、正则化参数等，以提高模型性能。8) Hyperparameter adjustment: fine-tune hyperparameters such as learning rate, regularization parameters, etc. according to the verification performance to improve model performance.
9)保存模型，一旦模型在验证集上表现良好，即保存训练好的模型权重，以备后续的推理和部署。9) Save the model. Once the model performs well on the validation set, save the trained model weights for subsequent reasoning and deployment.
10)重复迭代，多次重复上述步骤，以进一步提高模型性能。持续监控模型性能，并根据实际需求进行微调和改进。10) Repeat the above steps multiple times to further improve the model performance. Continuously monitor the model performance and make fine-tuning and improvements based on actual needs.
Step5：所述充电桩故障诊断大语言模型，在充电桩故障排除过程中为工作人员提供交互式提示服务。具体的，在使用过程中，如图2所示，该大语言模型会内嵌到工作人员的运维软件中，工作人员可以通过语音交流、文字交互或者上传图片的方式，从大语言模型获得基本的故障信息、可能故障原因以及处理方式。工作人员可以将新观测到的信息告诉大语言模型获得进一步的辅助信息，比如描述某个部件出现裂痕，或者上传某一个部件出现裂痕的照片，大语言模型将返回检修注意事项。Step 5: The charging pile fault diagnosis large language model provides interactive prompt services to the staff during the charging pile troubleshooting process. Specifically, during use, as shown in Figure 2, the large language model will be embedded in the staff's operation and maintenance software. The staff can obtain basic fault information, possible fault causes and treatment methods from the large language model through voice communication, text interaction or uploading pictures. The staff can tell the newly observed information to the large language model to obtain further auxiliary information, such as describing a crack in a certain component, or uploading a photo of a crack in a certain component. The large language model will return maintenance precautions.
Step6：所述充电桩故障诊断大语言模型，在充电桩故障排除后与工作人员交互，获取故障排除反馈经验。例如，任务完成之后，首先让工作人员确认故障是否被排除，然后大预言模型生成工作日志，包括故障检修的时间、地点、原因和维修方法，让工作人员确认日志是否属实，并且对大预言模型的表现打分。Step 6: The charging pile fault diagnosis large language model interacts with the staff after the charging pile fault is eliminated to obtain feedback on the troubleshooting experience. For example, after the task is completed, the staff is first asked to confirm whether the fault has been eliminated, and then the large prediction model generates a work log, including the time, location, cause and repair method of the fault repair, and the staff is asked to confirm whether the log is true and score the performance of the large prediction model.
Step7：所述充电桩故障诊断大语言模型，自动收集处理该故障的多模态状态信息和故障维修经验，补充加入知识库。具体的，在诊断过后，大语言模型要求工作人员确认故障真实原因和处理结果。例如，如果故障在大预言模型的指导下顺利被排除，故障检修的时间、地点、原因、维修方法与故障排除的正标签会反馈到Step1知识库中。反之，如果故障在大预言模型的指导下没有顺利被排除，故障检修的时间、地点、原因、维修方法与故障未被排除的负标签会被反馈到Step4中重新微调模型。需要注意的是，成功和失败的经验都会被用于下一阶段的模型微调。二者的区别在于，成功经验和其他的多模态输入被放到知识库中，当这些知识积累到一定的数量，比如10000条，会被再次用于大模型的微调。失败的经验会即时地反馈给大预言模型，记录类似的错误经验，避免类似错误经验的在短时间内再次输出。Step 7: The charging pile fault diagnosis large language model automatically collects and processes the multimodal state information and fault repair experience of the fault, and adds them to the knowledge base. Specifically, after the diagnosis, the large language model requires the staff to confirm the real cause of the fault and the handling result. For example, if the fault is successfully eliminated under the guidance of the large prediction model, the time, location, cause, repair method of the fault repair and the positive label of the fault elimination will be fed back to the Step 1 knowledge base. On the contrary, if the fault is not successfully eliminated under the guidance of the large prediction model, the time, location, cause, repair method of the fault repair and the negative label of the fault not being eliminated will be fed back to Step 4 to re-fine the model. It should be noted that both successful and failed experiences will be used for model fine-tuning in the next stage. The difference between the two is that successful experiences and other multimodal inputs are put into the knowledge base. When these knowledge accumulates to a certain number, such as 10,000, they will be used again for fine-tuning the large model. Failed experiences will be fed back to the large prediction model in real time, and similar wrong experiences will be recorded to avoid similar wrong experiences being output again in a short time.
如图2所示，本发明把来自多种传感器的数据和工作人员的工作经验集成给大预言模型进行训练，得到专用于充电桩故障诊断辅助方法及装置。具体地，通过传感器和人收集充电桩故障诊断的数据和经验，利用本地计算设备进行模式转换，接着将数据传入存储设备来构建数据库，然后在云计算设备微调大模型，最后形成移动端软件为工作人员提供服务。As shown in Figure 2, the present invention integrates data from multiple sensors and the work experience of staff into a large prediction model for training, and obtains a method and device dedicated to auxiliary diagnosis of charging pile faults. Specifically, the data and experience of charging pile fault diagnosis are collected through sensors and people, and the mode is converted using a local computing device. Then the data is transferred to a storage device to build a database, and then the large model is fine-tuned on a cloud computing device, and finally a mobile terminal software is formed to provide services for the staff.
具体的，多种传感器包括视频传感器、图像传感器、温度传感器和电气传感器等；人指的是工作人员。Specifically, the various sensors include video sensors, image sensors, temperature sensors, electrical sensors, etc.; people refer to staff.
上述仅为本发明的优选实施例而已，并不对本发明起到任何限制作用。任何所属技术领域的技术人员，在不脱离本发明的技术方案的范围内，对本发明揭露的技术方案和技术内容做任何形式的等同替换或修改等变动，均属未脱离本发明的技术方案的内容，仍属于本发明的保护范围之内。The above is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any technician in the relevant technical field, without departing from the scope of the technical solution of the present invention, makes any form of equivalent replacement or modification to the technical solution and technical content disclosed in the present invention, which does not depart from the content of the technical solution of the present invention and still falls within the protection scope of the present invention.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410028520.7ACN117829296A (en) | 2024-01-09 | 2024-01-09 | Charging pile fault diagnosis model based on large language model |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410028520.7ACN117829296A (en) | 2024-01-09 | 2024-01-09 | Charging pile fault diagnosis model based on large language model |
| Publication Number | Publication Date |
|---|---|
| CN117829296Atrue CN117829296A (en) | 2024-04-05 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410028520.7APendingCN117829296A (en) | 2024-01-09 | 2024-01-09 | Charging pile fault diagnosis model based on large language model |
| Country | Link |
|---|---|
| CN (1) | CN117829296A (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118261583A (en)* | 2024-04-15 | 2024-06-28 | 上海复数时空科技有限公司 | A factory equipment maintenance system based on FMEA and large language model |
| CN118396606A (en)* | 2024-07-01 | 2024-07-26 | 南方科技大学 | Manufacturing industry product production optimization method based on multi-mode large model |
| CN118819941A (en)* | 2024-09-19 | 2024-10-22 | 杭州亚信软件有限公司 | Fault diagnosis method, device, equipment, storage medium and program product |
| CN119027092A (en)* | 2024-08-13 | 2024-11-26 | 内蒙古电力(集团)有限责任公司电力调度控制分公司 | A network source coordination system status diagnosis method and system based on knowledge generation |
| CN119293232A (en)* | 2024-09-14 | 2025-01-10 | 杭州电子科技大学上虞科学与工程研究院有限公司 | Fault diagnosis method for power generation equipment based on large language model |
| CN119397326A (en)* | 2024-09-27 | 2025-02-07 | 天津大学 | An internal combustion engine fault diagnosis and defect detection intelligent agent system and construction method |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118261583A (en)* | 2024-04-15 | 2024-06-28 | 上海复数时空科技有限公司 | A factory equipment maintenance system based on FMEA and large language model |
| CN118396606A (en)* | 2024-07-01 | 2024-07-26 | 南方科技大学 | Manufacturing industry product production optimization method based on multi-mode large model |
| CN119027092A (en)* | 2024-08-13 | 2024-11-26 | 内蒙古电力(集团)有限责任公司电力调度控制分公司 | A network source coordination system status diagnosis method and system based on knowledge generation |
| CN119293232A (en)* | 2024-09-14 | 2025-01-10 | 杭州电子科技大学上虞科学与工程研究院有限公司 | Fault diagnosis method for power generation equipment based on large language model |
| CN118819941A (en)* | 2024-09-19 | 2024-10-22 | 杭州亚信软件有限公司 | Fault diagnosis method, device, equipment, storage medium and program product |
| CN119397326A (en)* | 2024-09-27 | 2025-02-07 | 天津大学 | An internal combustion engine fault diagnosis and defect detection intelligent agent system and construction method |
| Publication | Publication Date | Title |
|---|---|---|
| CN117829296A (en) | Charging pile fault diagnosis model based on large language model | |
| US20220137612A1 (en) | Transformer fault diagnosis and positioning system based on digital twin | |
| CN118394561B (en) | Method and device for diagnosing server fault, storage medium and electronic equipment | |
| CN109977624A (en) | Photovoltaic plant soft fault monitoring method based on deep neural network | |
| CN114330495A (en) | Wind power prediction method and device based on deep learning fusion model | |
| CN116541196A (en) | Intelligent inspection diagnosis method, system, device and medium for hydropower plant | |
| CN116448161B (en) | Artificial intelligence-based environment monitoring equipment fault diagnosis method | |
| CN112329937B (en) | GIS fault diagnosis method based on case and fault reasoning | |
| CN115221776A (en) | Training method and device for intelligent sensing diagnosis model of novel power distribution network | |
| CN115309736B (en) | Time sequence data anomaly detection method based on self-supervision learning multi-head attention network | |
| CN118277767A (en) | A sub-seasonal precipitation forecast method based on LA-UNET-LSTM | |
| CN116106856B (en) | Thunderstorm and Gale Identification Model Establishment Method, Identification Method and Calculation Equipment | |
| CN118868418A (en) | Integrated line intelligent monitoring system and method | |
| CN119625953B (en) | An abnormal early warning system for power equipment used in mobile terminals | |
| CN119420038A (en) | Distribution automation terminal operation and maintenance method based on 104 protocol | |
| CN119720070A (en) | Transmission tower status monitoring method and device in emergency scenarios | |
| CN119009253A (en) | On-line management system for internal temperature of battery | |
| CN118864386A (en) | Real-time identification method and device for front-end defects of transmission line drone inspection | |
| CN118886586A (en) | Photovoltaic power station interactive management system, photovoltaic power station interactive management method | |
| CN115334121B (en) | An automobile engine operating status monitoring system based on industrial Internet of Things | |
| CN112700425B (en) | A method for determining the quality of X-ray images of power equipment | |
| CN112016240B (en) | Prediction method for residual stable service life of incomplete degradation equipment with similar evidence | |
| CN111476383B (en) | A dynamic decision-making method for condition-based maintenance of pumping station units | |
| CN113807704A (en) | An intelligent algorithm platform construction method for urban rail transit data | |
| Sun et al. | Threat Target Detection System for UAV Equipment on the End Side of Transmission Lines |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |