Deep learning-based hydrogen energy commercial vehicle data anomaly detection method

Abstract

The invention relates to a hydrogen energy commercial vehicle data anomaly detection method based on deep learning. The method comprises the steps that U1, data information of voltage, current and temperature of a fuel cell stack is collected in real time, speed data information of a vehicle is collected in real time, and data information of pressure in a hydrogen tank and purity of hydrogen is collected in real time; and U2, on the basis of the data information of the voltage, the current and the temperature of the fuel cell stack and the speed data information of the vehicle, constructing a first related feature model of the fuel cell stack, and extracting a first related feature of the fuel cell stack to obtain data information of the first related feature of the fuel cell stack. According to the method, abnormal conditions can be accurately identified and early warned in time in a complex and changeable operation environment, safe operation of the vehicle is ensured, data safety is enhanced, and maintenance efficiency and user experience are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrogen energy commercial vehicles, and in particular to a method for detecting anomaly in hydrogen energy commercial vehicle data based on deep learning. Background Art

[0002] As the global demand for clean energy grows, hydrogen-powered commercial vehicles, as an efficient and environmentally friendly means of transportation, have gradually attracted widespread attention. However, the technology of hydrogen-powered commercial vehicles is relatively complex, especially in the use and management of hydrogen fuel cells, which require precise monitoring and maintenance. Traditional monitoring systems often rely on preset thresholds for anomaly detection. This method often shows limitations when facing complex and changeable actual operating environments, and is prone to missed reports or false alarms, and cannot effectively ensure the safety and reliability of vehicles. Therefore, a more intelligent and accurate anomaly detection method is urgently needed. Summary of the invention

[0003] In view of the above problems, the present invention provides a data anomaly detection method for hydrogen-powered commercial vehicles based on deep learning, which can not only accurately identify and timely warn of abnormal situations in complex and changeable operating environments to ensure the safe operation of vehicles, but also enhance data security, improve maintenance efficiency and user experience.

[0004] In order to achieve the above-mentioned purpose and other related purposes, the technical solution provided by the present invention is as follows: a method for detecting anomaly in hydrogen energy commercial vehicle data based on deep learning, the method comprising:

[0005] U1. Real-time data collection of voltage, current and temperature of the fuel cell stack, real-time data collection of vehicle speed, real-time data collection of pressure in the hydrogen tank and purity of hydrogen;

[0006] U2. Based on the data information of the voltage, current and temperature of the fuel cell stack and the speed data information of the vehicle, a first relevant characteristic model of the fuel cell stack is constructed, and the first relevant characteristic of the fuel cell stack is extracted to obtain data information of the first relevant characteristic of the fuel cell stack;

[0007] U3. Based on the data information of the voltage, current and temperature of the fuel cell stack and the data information of the pressure in the hydrogen tank and the purity of the hydrogen, a second relevant characteristic model of the fuel cell stack is constructed, and the second relevant characteristic of the fuel cell stack is extracted to obtain the data information of the second relevant characteristic of the fuel cell stack;

[0008] U4. Based on the data information of the second relevant feature of the fuel cell stack and the data information of the first relevant feature of the fuel cell stack, the first relevant feature and the second relevant feature are feature embedded using an improved L1 regularization algorithm with integrated random inertia weight to obtain the data information of the relevant feature of the fuel cell stack after feature embedding;

[0009] U5. Based on the data information of the relevant features of the fuel cell stack after the feature embedding, a deep learning data anomaly detection model for hydrogen energy commercial vehicles is constructed, the data of the hydrogen energy commercial vehicles is detected, and the data information of the detection results of the hydrogen energy commercial vehicles is output.

[0010] Furthermore, in step U2, constructing a first relevant feature model of the fuel cell stack and extracting the first relevant feature of the fuel cell stack includes:

[0011] U21. Based on the data information of the voltage, current and temperature of the fuel cell stack and the speed data information of the vehicle, establish the first related chaotic mapping function Q1 of the fuel cell stack,

[0012] Wherein, x1 is the data information of the voltage of the fuel cell stack, x2 is the data information of the current of the fuel cell stack, x3 is the data information of the temperature of the fuel cell stack, v is the data information of the speed of the vehicle, f1 is the correlation function of the speed of the vehicle and the voltage of the fuel cell stack, f2 is the correlation function of the speed of the vehicle and the current of the fuel cell stack, f3 is the correlation function of the speed of the vehicle and the temperature of the fuel cell stack, α1, α2 and α3 are the first relevant chaotic mapping factors of the fuel cell stack, the correlation matrix of the fuel cell stack and the speed of the vehicle is characterized, and the data information of the correlation matrix of the speed of the fuel cell stack and the vehicle is obtained;

[0013] U22. Based on the data information of the correlation matrix between the fuel cell stack and the speed of the vehicle, a first correlation feature extraction function W1 of the fuel cell stack is established.

[0014]

[0015] Wherein, y is the data information of the correlation matrix between the fuel cell stack and the speed of the vehicle, β1, β2 and β3 are the first feature extraction factors of the fuel cell stack;

[0016] U23. Based on the first relevant feature extraction function W1 of the fuel cell stack, the first relevant feature of the fuel cell stack is extracted to obtain data information of the first relevant feature of the fuel cell stack.

[0017] Furthermore, the constraint function g1 of the first feature extraction factor of the fuel cell stack is:

[0018]

[0019] Among them, the value range of the constraint function g1 is (1,2).

[0020] Furthermore, the constraints of the first related chaotic mapping factors α1, α2 and α3 of the fuel cell stack are:

[0021]

[0022] Furthermore, the correlation function f1 between the speed of the vehicle and the voltage of the fuel cell stack is:

[0023]

[0024] The correlation function f2 between the speed of the vehicle and the current of the fuel cell stack is:

[0025]

[0026] The correlation function f3 between the speed of the vehicle and the temperature of the fuel cell stack is:

[0027]

[0028] Among them, x1 is the data information of the voltage of the fuel cell stack, x2 is the data information of the current of the fuel cell stack, x3 is the data information of the temperature of the fuel cell stack, and v is the data information of the speed of the vehicle.

[0029] Furthermore, in step U3, constructing the second relevant feature model of the fuel cell stack and extracting the second relevant feature of the fuel cell stack includes:

[0030] U31. Based on the data information of the voltage, current and temperature of the fuel cell stack and the data information of the pressure in the hydrogen tank and the purity of the hydrogen, a second related chaotic mapping function Q2 of the fuel cell stack is established.

[0031]

[0032] Among them, x1 is the data information of the voltage of the fuel cell stack, x2 is the data information of the current of the fuel cell stack, x3 is the data information of the temperature of the fuel cell stack, x4 is the data information of the pressure in the hydrogen tank, x5 is the data information of the purity of the hydrogen in the hydrogen tank, δ1, δ2 and δ3 are the second related chaotic mapping factors of the fuel cell stack, and the correlation matrix between the fuel cell stack and the hydrogen tank is characterized to obtain the data information of the correlation matrix between the fuel cell stack and the hydrogen tank;

[0033] U32. Based on the data information of the correlation matrix between the fuel cell stack and the hydrogen tank, a second correlation feature extraction function W2 of the fuel cell stack is established.

[0034]

[0035] Wherein, z is the data information of the correlation matrix between the fuel cell stack and the hydrogen tank, γ1, γ2 and γ3 are the first feature extraction factors of the fuel cell stack;

[0036] U33. Based on the second relevant feature extraction function W2 of the fuel cell stack, the second relevant feature of the fuel cell stack is extracted to obtain data information of the second relevant feature of the fuel cell stack.

[0037] Furthermore, in step U4, the step of embedding the first related feature and the second related feature using the improved L1 regularization algorithm with integrated random inertia weight includes:

[0038] U41. Based on the data information of the second relevant feature of the fuel cell stack and the data information of the first relevant feature of the fuel cell stack, establish an L1 regularized loss function R of the relevant feature of the fuel cell stack,

[0039]

[0040] Wherein, r1 is the data information of the first relevant characteristic of the fuel cell stack, r2 is the data information of the second relevant characteristic of the fuel cell stack, η1, η2 and η3 are the loss factors of the fuel cell stack, and the integrated random weights of the relevant characteristics of the fuel cell stack are characterized to obtain the data information of the integrated random weights of the relevant characteristics of the fuel cell stack;

[0041] U42. Based on the data information of the integrated random weights of the relevant characteristics of the fuel cell stack, a feature embedding function P of the fuel cell stack is established.

[0042]

[0043] Wherein, r1 is data information of a first relevant characteristic of a fuel cell stack, r2 is data information of a second relevant characteristic of a fuel cell stack, and λ is an integrated stacking weight of relevant characteristics of a fuel cell stack;

[0044] U43. Based on the feature embedding function P of the fuel cell stack, feature embedding is performed on the first related feature and the second related feature to obtain data information of the related features of the fuel cell stack after feature embedding.

[0045] Furthermore, in step U5, the deep learning data anomaly detection model for hydrogen energy commercial vehicles is constructed, and the data of hydrogen energy commercial vehicles is detected, including:

[0046] U51. Based on the data information of the relevant features of the fuel cell stack after the feature embedding, construct a data set of relevant features of the fuel cell stack after the feature embedding;

[0047] U52. Input the data set of relevant features of the fuel cell stack after the feature embedding into the data anomaly detection model of the deep learning of hydrogen energy commercial vehicles, and determine the anomaly detection function S of the hydrogen energy commercial vehicle.

[0048]

[0049] Among them, h is the data set of relevant features of the fuel cell stack after feature embedding, μ1, μ2 and μ3 are the anomaly detection factors of the fuel cell stack, and the deep learning data anomaly detection model of the trained hydrogen energy commercial vehicle is obtained;

[0050] U53. Based on the trained deep learning data anomaly detection model of the hydrogen energy commercial vehicle, the data information of the relevant features of the fuel cell stack after the feature embedding is input, the data of the hydrogen energy commercial vehicle is detected, and the data information of the detection result of the hydrogen energy commercial vehicle is output.

[0051] In order to achieve the above-mentioned objectives and other related objectives, the present invention also provides a hydrogen energy commercial vehicle data anomaly detection system based on deep learning, including a computer device, which is programmed or configured to execute any one of the steps of the hydrogen energy commercial vehicle data anomaly detection method based on deep learning.

[0052] In order to achieve the above-mentioned purpose and other related purposes, the present invention also provides a computer-readable storage medium, which stores a computer program programmed or configured to execute any one of the deep learning-based hydrogen energy commercial vehicle data anomaly detection methods.

[0053] The present invention has the following positive effects:

[0054] 1. The present invention constructs a first relevant feature model of the fuel cell stack to extract the first relevant feature of the fuel cell stack, and combines it with constructing a second relevant feature model of the fuel cell stack to extract the second relevant feature of the fuel cell stack. This not only improves the consistency and availability of data and eliminates the differences between different data sources, but also improves the accuracy and reliability of data to meet the operating requirements under different working conditions.

[0055] 2. The present invention embeds the first relevant feature and the second relevant feature by adopting an improved L1 regularization algorithm with integrated random inertia weight, and combines it with building a deep learning data anomaly detection model for hydrogen energy commercial vehicles to detect the data of hydrogen energy commercial vehicles. It can not only accurately identify and timely warn of abnormal situations in complex and changeable operating environments to ensure the safe operation of vehicles, but also enhance data security, improve maintenance efficiency and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0057] Figure 2 A schematic diagram of a process for constructing a first relevant characteristic model of a fuel cell stack according to the present invention;

[0058] Figure 3 A schematic diagram of a process for constructing a second related characteristic model of a fuel cell stack according to the present invention;

[0059] Figure 4 It is a flow chart of the improved L1 regularization algorithm integrating random inertia weight of the present invention;

[0060] Figure 5 A schematic diagram of the process of constructing a deep learning data anomaly detection model for hydrogen energy commercial vehicles according to the present invention. DETAILED DESCRIPTION

[0061] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0062] Example 1: Figure 1 As shown, a method for detecting anomaly in hydrogen energy commercial vehicle data based on deep learning, the method comprising:

[0063] U1. Real-time data collection of voltage, current and temperature of the fuel cell stack, real-time data collection of vehicle speed, real-time data collection of pressure in the hydrogen tank and purity of hydrogen;

[0064] U2. Based on the data information of the voltage, current and temperature of the fuel cell stack and the speed data information of the vehicle, a first relevant characteristic model of the fuel cell stack is constructed, and the first relevant characteristic of the fuel cell stack is extracted to obtain data information of the first relevant characteristic of the fuel cell stack;

[0065] U3. Based on the data information of the voltage, current and temperature of the fuel cell stack and the data information of the pressure in the hydrogen tank and the purity of the hydrogen, a second relevant characteristic model of the fuel cell stack is constructed, and the second relevant characteristic of the fuel cell stack is extracted to obtain the data information of the second relevant characteristic of the fuel cell stack;

[0066] U4. Based on the data information of the second relevant feature of the fuel cell stack and the data information of the first relevant feature of the fuel cell stack, the first relevant feature and the second relevant feature are feature embedded using an improved L1 regularization algorithm with integrated random inertia weight to obtain the data information of the relevant feature of the fuel cell stack after feature embedding;

[0067] U5. Based on the data information of the relevant features of the fuel cell stack after the feature embedding, a deep learning data anomaly detection model for hydrogen energy commercial vehicles is constructed, the data of the hydrogen energy commercial vehicles is detected, and the data information of the detection results of the hydrogen energy commercial vehicles is output.

[0068] In this embodiment, if Figure 2 As shown, in step U2, constructing a first relevant feature model of the fuel cell stack and extracting the first relevant feature of the fuel cell stack includes:

[0069] U21. Based on the data information of the voltage, current and temperature of the fuel cell stack and the speed data information of the vehicle, establish the first related chaotic mapping function Q1 of the fuel cell stack,

[0070] Wherein, x1 is the data information of the voltage of the fuel cell stack, x2 is the data information of the current of the fuel cell stack, x3 is the data information of the temperature of the fuel cell stack, v is the data information of the speed of the vehicle, f1 is the correlation function of the speed of the vehicle and the voltage of the fuel cell stack, f2 is the correlation function of the speed of the vehicle and the current of the fuel cell stack, f3 is the correlation function of the speed of the vehicle and the temperature of the fuel cell stack, α1, α2 and α3 are the first relevant chaotic mapping factors of the fuel cell stack, the correlation matrix of the fuel cell stack and the speed of the vehicle is characterized, and the data information of the correlation matrix of the speed of the fuel cell stack and the vehicle is obtained;

[0071] U22. Based on the data information of the correlation matrix between the fuel cell stack and the speed of the vehicle, a first correlation feature extraction function W1 of the fuel cell stack is established.

[0072]

[0073] Wherein, y is the data information of the correlation matrix between the fuel cell stack and the speed of the vehicle, β1, β2 and β3 are the first feature extraction factors of the fuel cell stack;

[0074] U23. Based on the first relevant feature extraction function W1 of the fuel cell stack, the first relevant feature of the fuel cell stack is extracted to obtain data information of the first relevant feature of the fuel cell stack.

[0075] In this embodiment, the constraint function g1 of the first feature extraction factor of the fuel cell stack is:

[0076]

[0077] Among them, the value range of the constraint function g1 is (1,2).

[0078] In this embodiment, the constraints of the first relevant chaotic mapping factors α1, α2 and α3 of the fuel cell stack are:

[0079]

[0080] In this embodiment, the correlation function f1 between the speed of the vehicle and the voltage of the fuel cell stack is:

[0081]

[0082] The correlation function f2 between the speed of the vehicle and the current of the fuel cell stack is:

[0083]

[0084] The correlation function f3 between the speed of the vehicle and the temperature of the fuel cell stack is:

[0085]

[0086] Among them, x1 is the data information of the voltage of the fuel cell stack, x2 is the data information of the current of the fuel cell stack, x3 is the data information of the temperature of the fuel cell stack, and v is the data information of the speed of the vehicle.

[0087] In this embodiment, if Figure 3 As shown, in step U3, the second relevant feature model of the fuel cell stack is constructed, and the second relevant feature of the fuel cell stack is extracted, including:

[0088] U31. Based on the data information of the voltage, current and temperature of the fuel cell stack and the data information of the pressure in the hydrogen tank and the purity of the hydrogen, a second related chaotic mapping function Q2 of the fuel cell stack is established.

[0089]

[0090] Among them, x1 is the data information of the voltage of the fuel cell stack, x2 is the data information of the current of the fuel cell stack, x3 is the data information of the temperature of the fuel cell stack, x4 is the data information of the pressure in the hydrogen tank, x5 is the data information of the purity of the hydrogen in the hydrogen tank, δ1, δ2 and δ3 are the second related chaotic mapping factors of the fuel cell stack, and the correlation matrix between the fuel cell stack and the hydrogen tank is characterized to obtain the data information of the correlation matrix between the fuel cell stack and the hydrogen tank;

[0091] U32. Based on the data information of the correlation matrix between the fuel cell stack and the hydrogen tank, a second correlation feature extraction function W2 of the fuel cell stack is established.

[0092]

[0093] Wherein, z is the data information of the correlation matrix between the fuel cell stack and the hydrogen tank, γ1, γ2 and γ3 are the first feature extraction factors of the fuel cell stack;

[0094] U33. Based on the second relevant feature extraction function W2 of the fuel cell stack, the second relevant feature of the fuel cell stack is extracted to obtain data information of the second relevant feature of the fuel cell stack.

[0095] Example 2: Based on the deep learning-based hydrogen energy commercial vehicle data anomaly detection method in Example 1, the present invention is further illustrated and described below.

[0096] like Figure 1 As shown, a method for detecting anomaly in hydrogen energy commercial vehicle data based on deep learning, the method comprising:

[0097] U1. Real-time data collection of voltage, current and temperature of the fuel cell stack, real-time data collection of vehicle speed, real-time data collection of pressure in the hydrogen tank and purity of hydrogen;

[0098] U2. Based on the data information of the voltage, current and temperature of the fuel cell stack and the speed data information of the vehicle, a first relevant characteristic model of the fuel cell stack is constructed, and the first relevant characteristic of the fuel cell stack is extracted to obtain data information of the first relevant characteristic of the fuel cell stack;

[0099] U3. Based on the data information of the voltage, current and temperature of the fuel cell stack and the data information of the pressure in the hydrogen tank and the purity of the hydrogen, a second relevant characteristic model of the fuel cell stack is constructed, and the second relevant characteristic of the fuel cell stack is extracted to obtain the data information of the second relevant characteristic of the fuel cell stack;

[0100] U4. Based on the data information of the second relevant feature of the fuel cell stack and the data information of the first relevant feature of the fuel cell stack, the first relevant feature and the second relevant feature are feature embedded using an improved L1 regularization algorithm with integrated random inertia weight to obtain the data information of the relevant feature of the fuel cell stack after feature embedding;

[0101] U5. Based on the data information of the relevant features of the fuel cell stack after the feature embedding, a deep learning data anomaly detection model for hydrogen energy commercial vehicles is constructed, the data of the hydrogen energy commercial vehicles is detected, and the data information of the detection results of the hydrogen energy commercial vehicles is output.

[0102] In this embodiment, if Figure 4 As shown, in step U4, the improved L1 regularization algorithm with integrated random inertia weight is used to embed the first related feature and the second related feature, including:

[0103] U41. Based on the data information of the second relevant feature of the fuel cell stack and the data information of the first relevant feature of the fuel cell stack, establish an L1 regularized loss function R of the relevant feature of the fuel cell stack,

[0104]

[0105] Wherein, r1 is the data information of the first relevant characteristic of the fuel cell stack, r2 is the data information of the second relevant characteristic of the fuel cell stack, η1, η2 and η3 are the loss factors of the fuel cell stack, and the integrated random weights of the relevant characteristics of the fuel cell stack are characterized to obtain the data information of the integrated random weights of the relevant characteristics of the fuel cell stack;

[0106] U42. Based on the data information of the integrated random weights of the relevant characteristics of the fuel cell stack, a feature embedding function P of the fuel cell stack is established.

[0107]

[0108] Wherein, r1 is data information of a first relevant characteristic of a fuel cell stack, r2 is data information of a second relevant characteristic of a fuel cell stack, and λ is an integrated stacking weight of relevant characteristics of a fuel cell stack;

[0109] U43. Based on the feature embedding function P of the fuel cell stack, feature embedding is performed on the first related feature and the second related feature to obtain data information of the related features of the fuel cell stack after feature embedding.

[0110] In this embodiment, if Figure 5 As shown, in step U5, the deep learning data anomaly detection model for hydrogen energy commercial vehicles is constructed, and the data of hydrogen energy commercial vehicles is detected, including:

[0111] U51. Based on the data information of the relevant features of the fuel cell stack after the feature embedding, construct a data set of relevant features of the fuel cell stack after the feature embedding;

[0112] U52. Input the data set of relevant features of the fuel cell stack after the feature embedding into the data anomaly detection model of the deep learning of hydrogen energy commercial vehicles, and determine the anomaly detection function S of the hydrogen energy commercial vehicle.

[0113]

[0114] Among them, h is the data set of relevant features of the fuel cell stack after feature embedding, μ1, μ2 and μ3 are the anomaly detection factors of the fuel cell stack, and the deep learning data anomaly detection model of the trained hydrogen energy commercial vehicle is obtained;

[0115] U53. Based on the trained deep learning data anomaly detection model of the hydrogen energy commercial vehicle, the data information of the relevant features of the fuel cell stack after the feature embedding is input, the data of the hydrogen energy commercial vehicle is detected, and the data information of the detection result of the hydrogen energy commercial vehicle is output.

[0116] In this embodiment, the present invention provides a hydrogen energy commercial vehicle data anomaly detection system based on deep learning, including a computer device, which is programmed or configured to execute any one of the steps of the hydrogen energy commercial vehicle data anomaly detection method based on deep learning.

[0117] In this embodiment, the present invention provides a computer-readable storage medium, on which is stored a computer program programmed or configured to execute any one of the deep learning-based hydrogen energy commercial vehicle data anomaly detection methods.

[0118] 1) Data collection

[0119] Vehicle operation data, including fuel cell status, temperature, pressure, hydrogen purity and speed, are collected in real time through high-precision sensors and IoT devices.

[0120] 1. Sensor selection: High-precision sensors are installed at various key locations of hydrogen energy commercial vehicles to monitor key parameters such as the voltage, current, temperature of the fuel cell stack, the speed of the vehicle, and the pressure and purity of the hydrogen tank.

[0121] Fuel cell monitoring sensor: monitors the voltage, current and temperature of the fuel cell stack.

[0122] Temperature sensors: used to monitor the temperature of fuel cells and other critical components to prevent overheating and fire risks.

[0123] Pressure sensor: monitors the pressure in the hydrogen tank and fuel cell to ensure the normal operation and safety of the system.

[0124] Hydrogen Purity Sensor: Measures the purity of hydrogen to ensure efficient operation of fuel cells.

[0125] Speed ​​sensor: monitors the real-time speed of the vehicle and analyzes the dynamic performance of the vehicle.

[0126] GPS module: obtains the vehicle's location information and provides geo-referenced data for energy management and anomaly detection.

[0127] 2. Data acquisition hardware design: Design hardware modules that integrate multiple sensors to ensure the accuracy and reliability of data acquisition.

[0128] Data acquisition module: A data acquisition unit that integrates multiple sensors and is capable of collecting and processing multiple types of data in real time.

[0129] Embedded system: Use high-performance embedded system for real-time data processing and transmission.

[0130] Data storage device: used to temporarily store collected data to ensure that data is not lost due to transmission problems.

[0131] 3. Data collection protocol: Use appropriate data collection protocols to ensure the accuracy and real-time nature of data transmission.

[0132] CAN bus protocol: used for data transmission between various electronic control units (ECUs) of the vehicle, with high real-time performance and reliability.

[0133] Modbus protocol: supports a variety of sensors and data acquisition devices, with good compatibility and scalability.

[0134] MQTT protocol: A lightweight messaging protocol suitable for data transmission between IoT devices.

[0135] Data transmission: The collected data is transmitted to the central server or cloud platform through wired and wireless methods.

[0136] Wired transmission: high-bandwidth data transmission is achieved through interfaces such as USB and Ethernet.

[0137] Wireless transmission: Use wireless communication technologies such as WiFi, 4G / 5G, LoRa, etc. to transmit data to a remote server or cloud platform.

[0138] 2) Data preprocessing

[0139] The collected data are cleaned, denoised and normalized to improve data quality.

[0140] Data cleaning: Use advanced filtering algorithms to remove noise and outliers to ensure data accuracy and consistency.

[0141] Data normalization: Normalize data to improve data consistency and availability and eliminate differences between different data sources.

[0142] Adaptive filtering: Apply adaptive filtering technology to further improve data accuracy and reliability to meet operating requirements under different working conditions.

[0143] 3) Feature extraction and embedding

[0144] Feature extraction and embedding are key steps in data processing. Deep learning networks are used to extract important features from data and generate feature embeddings to facilitate subsequent anomaly detection.

[0145] 1. Variational Autoencoder (VAE): Variational Autoencoder is a generative model that learns the latent distribution of data by maximizing the likelihood function of the data. We use VAE to extract temporal features and complex relationships of sensor data.

[0146] Input data: including multimodal data such as fuel cell status, temperature, pressure, hydrogen purity and speed.

[0147] Encoder: compresses the input data into a low-dimensional latent space and extracts the main features.

[0148] Decoder: Reconstructs the original data from the low-dimensional latent space to ensure the effectiveness of feature extraction.

[0149] Loss function: includes reconstruction loss and KL divergence to ensure that the extracted features are representative.

[0150] 2. Generative Adversarial Network (GAN): GAN generates and discriminates data through two adversarial models (generator and discriminator), and can learn the complex distribution of data.

[0151] Generator: Generates fake data with similar distribution to real data.

[0152] Discriminator: Determines whether the input data is real data or generated data.

[0153] Feature extraction: Through adversarial training, the generator gradually learns the potential distribution of the input data and thus extracts important features.

[0154] 3. Graph Neural Network (GNN): Processes complex sensor network data and generates more accurate feature embeddings.

[0155] Input graph structure data: Represent various sensor data as graph structures.

[0156] Graph convolution: Perform convolution operations on graph structured data to capture the relationship between nodes.

[0157] Feature embedding generation: Through a multi-layer graph convolutional network (GCN), feature embedding is generated for subsequent detection models.

[0158] 4) Model training

[0159] Model training aims to use historical data to train deep learning models so that they can identify normal and abnormal operating conditions and have good generalization capabilities.

[0160] 1. Deep Reinforcement Learning (DRL): Deep reinforcement learning learns through trial and error, enabling the model to make optimal decisions in complex environments.

[0161] Environmental simulation: Build the operating environment of hydrogen-powered commercial vehicles and generate a large amount of training data through simulation.

[0162] Reward function design: Design a reasonable reward function to guide the model to learn beneficial behaviors, such as detecting anomalies and issuing timely warnings.

[0163] Policy network: The policy network learns the best actions to take in different states.

[0164] Value network: The value of each state is evaluated through the value network to assist the policy network in making decisions.

[0165] 2. Graph Convolutional Network (GCN): Graph Convolutional Network learns the complex relationships between nodes by performing convolution operations on graph structured data.

[0166] Input graph data: Input graph-structured sensor data, including the features of each node and the weights of the edges.

[0167] Graph convolution layer: Through multiple layers of graph convolution, feature extraction and transformation are performed on each node to capture the relationship in the graph structure.

[0168] Feature fusion: The features after graph convolution are fused to generate node embeddings, which are applied to the anomaly detection model.

[0169] Policy network: The policy network learns the best actions to take in different states.

[0170] 3. Transfer learning: In order to improve the generalization ability and adaptability of the model, the transfer learning method is adopted.

[0171] Pre-trained model: Use large-scale public datasets to pre-train basic models.

[0172] Fine-tuning: Fine-tuning is performed on a dedicated dataset for hydrogen-powered commercial vehicles to make the model better suited to specific application scenarios.

[0173] 4. Adversarial training: Adversarial training is performed through generative adversarial networks (GANs) to improve the stability of the model under extreme conditions.

[0174] Generate adversarial data: Generate fake data in some extreme cases.

[0175] Adversarial training: By mixing the generated data with the real data, the training model can still maintain a stable detection effect when facing different data distributions.

[0176] 5. Real-time detection: Input real-time data into the trained model for anomaly detection, identification and warning of abnormal situations.

[0177] Real-time analysis: Use the trained deep learning model to analyze the sensor data collected in real time and accurately identify abnormal situations.

[0178] Edge computing: By introducing edge computing technology, some data processing tasks can be completed locally in the vehicle, reducing data transmission delays and improving detection response speed.

[0179] Feedback mechanism: The real-time detection system will provide immediate feedback to the energy management and alarm modules based on the detection results to ensure that the system can respond to abnormal situations in a timely manner.

[0180] 6. Energy management optimization: Dynamically adjust the working state and power output of the hydrogen fuel cell according to the detection results to optimize energy management.

[0181] Dynamic adjustment: After the real-time detection system identifies an abnormality, the energy management module will dynamically adjust the operating status and power output of the fuel cell to optimize the vehicle's energy utilization.

[0182] Adaptive optimization: Through deep learning algorithms, the energy management module can adaptively adjust the management strategy and optimize according to the actual operating conditions to improve operating efficiency and extend the life of the fuel cell.

[0183] Historical data reference: Energy management optimization not only relies on real-time detection data, but also combines historical operation data for comprehensive analysis to provide more accurate optimization solutions.

[0184] 7. Hydrogen leak detection: Real-time monitoring and location of hydrogen leaks are performed through deep learning algorithms to ensure safety.

[0185] Real-time monitoring: Using hydrogen sensor data combined with deep learning models to monitor hydrogen concentration in real time and quickly identify leaks.

[0186] Virtual sensor technology: Introducing virtual sensor technology to improve the sensitivity and accuracy of hydrogen leak detection through modeling and simulation data.

[0187] Positioning and early warning: When a hydrogen leak is detected, the system can accurately locate the leak and send out an early warning signal through the alarm module to promptly notify relevant personnel to take measures.

[0188] 8. Multimodal data fusion: Fuse data from different sensors to improve the accuracy of anomaly detection.

[0189] Multimodal fusion: Use a multimodal convolutional neural network (Multimodal CNN) to comprehensively analyze the data from internal and external sensors of the vehicle to improve the robustness and accuracy of detection.

[0190] Bayesian network: The Bayesian network is introduced to optimize the fusion effect of multimodal data through probabilistic inference and ensure collaborative analysis between different data sources.

[0191] Data consistency check: During the data fusion process, the system will perform consistency checks on data from different sources, eliminate outliers and erroneous data, and ensure the quality of the fused data.

[0192] 9. Alarm processing: When an abnormal situation is detected, a warning signal is issued through the alarm module to remind relevant personnel to handle it.

[0193] Timely alarm: When the system detects an abnormal situation, such as hydrogen leakage or energy management abnormality, the alarm module will immediately send out an audible and visual alarm signal to ensure that relevant personnel can respond in time.

[0194] Adaptive alarm: The system can adaptively adjust the alarm threshold according to the severity of the abnormal situation and historical data, reduce false alarms and missed alarms, and improve the effectiveness of the alarm.

[0195] Multi-channel notification: In addition to on-site alarms, the system can also send notifications via the network, such as SMS and email, to ensure that remote managers can also obtain abnormal information in a timely manner.

[0196] Any reference to memory, storage, database or other media used in the embodiments provided in the present application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM) and memory bus dynamic RAM (RDRAM).

[0197] In summary, the present invention can not only accurately identify and timely warn of abnormal situations in a complex and changeable operating environment to ensure the safe operation of the vehicle, but also enhance data security, improve maintenance efficiency and user experience.

[0198] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for detecting anomaly in hydrogen energy commercial vehicle data based on deep learning, characterized in that: The method comprises: U1. Real-time data collection of voltage, current and temperature of the fuel cell stack, real-time data collection of vehicle speed, real-time data collection of pressure in the hydrogen tank and purity of hydrogen; U2. Based on the data information of the voltage, current and temperature of the fuel cell stack and the speed data information of the vehicle, a first relevant characteristic model of the fuel cell stack is constructed, and the first relevant characteristic of the fuel cell stack is extracted to obtain data information of the first relevant characteristic of the fuel cell stack; U3. Based on the data information of the voltage, current and temperature of the fuel cell stack and the data information of the pressure in the hydrogen tank and the purity of the hydrogen, a second relevant characteristic model of the fuel cell stack is constructed, and the second relevant characteristic of the fuel cell stack is extracted to obtain the data information of the second relevant characteristic of the fuel cell stack; U4. Based on the data information of the second relevant feature of the fuel cell stack and the data information of the first relevant feature of the fuel cell stack, the first relevant feature and the second relevant feature are feature embedded using an improved L1 regularization algorithm with integrated random inertia weight to obtain the data information of the relevant feature of the fuel cell stack after feature embedding; U5. Based on the data information of the relevant features of the fuel cell stack after the feature embedding, a deep learning data anomaly detection model for hydrogen energy commercial vehicles is constructed, the data of the hydrogen energy commercial vehicles is detected, and the data information of the detection results of the hydrogen energy commercial vehicles is output.

2. The method for detecting anomaly of hydrogen energy commercial vehicle data based on deep learning according to claim 1 is characterized in that: In step U2, constructing a first relevant feature model of the fuel cell stack and extracting the first relevant feature of the fuel cell stack includes: U21. Based on the data information of the voltage, current and temperature of the fuel cell stack and the speed data information of the vehicle, establish the first related chaotic mapping function Q1 of the fuel cell stack, Wherein, x1 is the data information of the voltage of the fuel cell stack, x2 is the data information of the current of the fuel cell stack, x3 is the data information of the temperature of the fuel cell stack, v is the data information of the speed of the vehicle, f1 is the correlation function of the speed of the vehicle and the voltage of the fuel cell stack, f2 is the correlation function of the speed of the vehicle and the current of the fuel cell stack, f3 is the correlation function of the speed of the vehicle and the temperature of the fuel cell stack, α1, α2 and α3 are the first relevant chaotic mapping factors of the fuel cell stack, the correlation matrix of the fuel cell stack and the speed of the vehicle is characterized, and the data information of the correlation matrix of the speed of the fuel cell stack and the vehicle is obtained; U22. Based on the data information of the correlation matrix between the fuel cell stack and the speed of the vehicle, a first correlation feature extraction function W1 of the fuel cell stack is established. Wherein, y is the data information of the correlation matrix between the fuel cell stack and the speed of the vehicle, β1, β2 and β3 are the first feature extraction factors of the fuel cell stack; U23. Based on the first relevant feature extraction function W1 of the fuel cell stack, the first relevant feature of the fuel cell stack is extracted to obtain data information of the first relevant feature of the fuel cell stack.

3. The method for detecting anomaly in hydrogen energy commercial vehicle data based on deep learning according to claim 2 is characterized in that: The constraint function g1 of the first feature extraction factor of the fuel cell stack is: Among them, the value range of the constraint function g1 is (1,2).

4. The method for detecting anomaly in hydrogen energy commercial vehicle data based on deep learning according to claim 2 is characterized in that: The constraints of the first related chaotic mapping factors α1, α2 and α3 of the fuel cell stack are:

5. The method for detecting anomaly in hydrogen energy commercial vehicle data based on deep learning according to claim 2 is characterized in that: The correlation function f1 between the speed of the vehicle and the voltage of the fuel cell stack is: The correlation function f2 between the speed of the vehicle and the current of the fuel cell stack is: The correlation function f3 between the speed of the vehicle and the temperature of the fuel cell stack is: Among them, x1 is the data information of the voltage of the fuel cell stack, x2 is the data information of the current of the fuel cell stack, x3 is the data information of the temperature of the fuel cell stack, and v is the data information of the speed of the vehicle.

6. The method for detecting anomaly of hydrogen energy commercial vehicle data based on deep learning according to claim 2 is characterized in that: In step U3, constructing a second relevant feature model of the fuel cell stack and extracting the second relevant feature of the fuel cell stack includes: U31. Based on the data information of the voltage, current and temperature of the fuel cell stack and the data information of the pressure in the hydrogen tank and the purity of the hydrogen, a second related chaotic mapping function Q2 of the fuel cell stack is established. Among them, x1 is the data information of the voltage of the fuel cell stack, x2 is the data information of the current of the fuel cell stack, x3 is the data information of the temperature of the fuel cell stack, x4 is the data information of the pressure in the hydrogen tank, x5 is the data information of the purity of the hydrogen in the hydrogen tank, δ1, δ2 and δ3 are the second related chaotic mapping factors of the fuel cell stack, and the correlation matrix between the fuel cell stack and the hydrogen tank is characterized to obtain the data information of the correlation matrix between the fuel cell stack and the hydrogen tank; U32. Based on the data information of the correlation matrix between the fuel cell stack and the hydrogen tank, a second correlation feature extraction function W2 of the fuel cell stack is established. Wherein, z is the data information of the correlation matrix between the fuel cell stack and the hydrogen tank, γ1, γ2 and γ3 are the first feature extraction factors of the fuel cell stack; U33. Based on the second relevant feature extraction function W2 of the fuel cell stack, the second relevant feature of the fuel cell stack is extracted to obtain data information of the second relevant feature of the fuel cell stack.

7. The method for detecting anomaly of hydrogen energy commercial vehicle data based on deep learning according to claim 1 is characterized in that: In step U4, the step of embedding the first related feature and the second related feature using the improved L1 regularization algorithm with integrated random inertia weight includes: U41. Based on the data information of the second relevant feature of the fuel cell stack and the data information of the first relevant feature of the fuel cell stack, establish an L1 regularized loss function R of the relevant feature of the fuel cell stack, Wherein, r1 is the data information of the first relevant characteristic of the fuel cell stack, r2 is the data information of the second relevant characteristic of the fuel cell stack, η1, η2 and η3 are the loss factors of the fuel cell stack, and the integrated random weights of the relevant characteristics of the fuel cell stack are characterized to obtain the data information of the integrated random weights of the relevant characteristics of the fuel cell stack; U42. Based on the data information of the integrated random weights of the relevant characteristics of the fuel cell stack, a feature embedding function P of the fuel cell stack is established. Wherein, r1 is data information of a first relevant characteristic of a fuel cell stack, r2 is data information of a second relevant characteristic of a fuel cell stack, and λ is an integrated stacking weight of relevant characteristics of a fuel cell stack; U43. Based on the feature embedding function P of the fuel cell stack, feature embedding is performed on the first related feature and the second related feature to obtain data information of the related features of the fuel cell stack after feature embedding.

8. The method for detecting anomaly of hydrogen energy commercial vehicle data based on deep learning according to claim 1 is characterized in that: In step U5, the deep learning data anomaly detection model for hydrogen energy commercial vehicles is constructed, and the data of hydrogen energy commercial vehicles is detected, including: U51. Based on the data information of the relevant features of the fuel cell stack after the feature embedding, construct a data set of relevant features of the fuel cell stack after the feature embedding; U52. Input the data set of relevant features of the fuel cell stack after the feature embedding into the data anomaly detection model of the deep learning of hydrogen energy commercial vehicles, and determine the anomaly detection function S of the hydrogen energy commercial vehicle. Among them, h is the data set of relevant features of the fuel cell stack after feature embedding, μ1, μ2 and μ3 are the anomaly detection factors of the fuel cell stack, and the deep learning data anomaly detection model of the trained hydrogen energy commercial vehicle is obtained; U53. Based on the trained deep learning data anomaly detection model of the hydrogen energy commercial vehicle, the data information of the relevant features of the fuel cell stack after the feature embedding is input, the data of the hydrogen energy commercial vehicle is detected, and the data information of the detection result of the hydrogen energy commercial vehicle is output.

9. A hydrogen energy commercial vehicle data anomaly detection system based on deep learning, including a computer device, characterized in that: The computer device is programmed or configured to execute the steps of the hydrogen energy commercial vehicle data anomaly detection method based on deep learning as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program that is programmed or configured to execute the deep learning-based hydrogen energy commercial vehicle data anomaly detection method as described in any one of claims 1 to 8.

Images