Machine learning-based punctuality system fault detection method, system and device, and storage medium
Through machine learning-based methods, the problem of punctual system failure detection is solved, and the problem of difficulty in establishing an effective mechanism model in traditional methods is achieved, and fault detection and prediction with high accuracy and real-time performance is improved, thus improving system reliability and service quality.
Patent Information
- Application Number
- CN202411968233.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-30
- Publication Date
- 2025-04-15
AI Technical Summary
It is difficult for the prior art to establish an effective mechanism model through traditional methods to detect and predict punctual system failures, which affects the reliability and service quality of the system.
Using machine learning-based methods, fault diagnosis is performed by obtaining fault sample data, building target vectors, initializing neural network models, training models, and collecting running status data in real time.
Improve the accuracy and real-timeness of punctual system fault detection, enhance fault prediction capabilities, reduce maintenance costs, and adapt to different types of punctual systems.
Smart Images
Figure CN119830076A_ABST
Abstract
Description
Background Art
[0002] The timekeeping system is the foundation and core of the timing service. It mainly uses high-performance atomic clock groups as frequency standards to generate and maintain local standard time. It also generates a large number of physical standard frequency user signals (10MHz) and standard time user signals (1PPS) to provide standard time services to the outside world. The timekeeping system needs to be continuously and stably online for 24 hours to make the time service provided reliable and trustworthy. The timekeeping system is mainly composed of atomic clock subsystems, signal generation subsystems, measurement subsystems, and time and frequency service subsystems. There are many types and quantities of equipment involved, and the amount of data is large. In the long-term uninterrupted operation process, it is under the pressure of operating failures of time and frequency equipment at all levels, and the uncertainty of the performance of atomic clocks. The quality and state of the output signal will change in different rules.
[0003] Currently, there is no research on fault detection for timekeeping systems. It is impossible to establish a mechanism model based on certain key devices or the relationship between multiple devices. The establishment of a neural network can be used to evaluate the health status of the equipment and system, realize fault detection and prediction, and then take timely measures to maintain the stability of the system.
[0004] Therefore, it is necessary to provide a timekeeping system fault detection method, system, device and storage medium based on machine learning. Summary of the invention
[0005] This specification provides a timekeeping system fault detection method, system, device and storage medium based on machine learning, which is used to solve the problem that it is difficult to establish an effective mechanism model for fault detection and prediction through traditional methods, thereby affecting the reliability and service quality of the system.
[0006] In a first aspect of the present invention, a timekeeping system fault detection method based on machine learning is proposed, the method comprising:
[0007] Step S1, obtaining fault sample data, and generating a feature vector for each fault sample data in combination with label data;
[0008] Step S2, constructing a target vector of a fault detection method according to whether there is a fault and the fault level of the fault sample data;
[0009] Step S3, randomly initializing the weight matrix and bias vector of the hidden layer in the pre-built fault detection network model;
[0010] Step S4, using each feature vector as input data of the initialized fault detection network model, calculating the activation value of each node in the hidden layer, and constructing an activation matrix based on multiple activation values;
[0011] Step S5, calculating the connection weights from the hidden layer to the output layer based on the target vector, the activation matrix, the identity matrix and the regularization coefficient; repeating steps S1 to S5, training the fault detection network model until convergence, and obtaining the optimal model;
[0012] Step S6, collect the operating status data of the timekeeping system in real time, and filter out the fault information data. When the amount of fault information data exceeds a preset threshold, construct multiple fault information data into a fault feature set, input it into the optimal model, and obtain the fault diagnosis result.
[0013] Furthermore, the fault sample data at least includes the operating environment temperature and humidity of the timekeeping system, and the current and voltage of the atomic clock.
[0014] Furthermore, the fault detection network model is constructed based on a single hidden layer feedforward neural network model.
[0015] Furthermore, the weight matrix and bias vector of the hidden layer in the pre-built fault detection network model are randomly initialized as follows:
[0016] The number of hidden layer nodes of the single hidden layer feedforward neural network model is set to h, and random feature mapping is performed based on uniform distribution to generate weight matrix and bias vector.
[0017] Furthermore, the activation value h il , which is calculated as:
[0018] Using the infinitely differentiable sigmoid activation function, the m feature vectors x i As the input of the fault detection network model, calculate its corresponding activation value h in the hidden layer il :
[0019]
[0020] in, is the transpose of a row vector of the lth hidden layer node in the weight matrix A, b l is the bias vector of the lth hidden layer node.
[0021] Furthermore, the connection weight γ is calculated as follows:
[0022] γ=(H T H+ωI) -1 H T T;
[0023] Among them, H is the activation matrix, H T is the transpose of the activation matrix, ω is the regularization coefficient, T is the target vector, and I is the identity matrix.
[0024] Furthermore, after obtaining the fault diagnosis result, the source of the fault information is traced.
[0025] In a second aspect of the present invention, a timekeeping system fault detection system based on machine learning is proposed. Based on the timekeeping system fault detection method based on machine learning in the first aspect, the system comprises:
[0026] A data set construction module, which is configured to obtain fault sample data and generate a feature vector for each fault sample data in combination with the label data;
[0027] A target vector calculation module configured to construct a target vector of a fault detection method according to whether there is a fault and the fault level of the fault sample data;
[0028] A network model parameter initialization module configured to randomly initialize a weight matrix and a bias vector of a hidden layer in a pre-built fault detection network model;
[0029] an activation matrix building module, configured to use each feature vector as input data of the initialized fault detection network model, calculate an activation value of each node of the hidden layer, and build an activation matrix based on the multiple activation values;
[0030] A training module is configured to calculate the connection weights from the hidden layer to the output layer based on the target vector, the activation matrix, the identity matrix and the regularization coefficient; and repeatedly train the fault detection network model until convergence to obtain an optimal model.
[0031] The fault diagnosis module is configured to collect the operating status data of the timekeeping system in real time and filter out the fault information data. When the amount of fault information data exceeds a preset threshold, multiple fault information data are constructed into a fault feature set and input into the optimal model to obtain the fault diagnosis result.
[0032] According to a third aspect of the present invention, an electronic device is provided, comprising:
[0033] at least one processor; and
[0034] a memory communicatively connected to at least one of the processors; wherein,
[0035] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement a timekeeping system fault detection method based on machine learning.
[0036] In a fourth aspect of the present invention, a computer-readable storage medium is proposed, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to be executed by the computer to implement a timekeeping system fault detection method based on machine learning.
[0037] At least one of the above technical solutions adopted in the embodiments of this specification can achieve the following beneficial effects:
[0038] Improve reliability: By introducing machine learning algorithms, especially neural network models optimized for complex and high-dimensional data environments, the accuracy and real-time performance of timekeeping system fault detection are improved.
[0039] Enhanced predictive capabilities: Using historical data to train models enables the system to not only identify current problems, but also predict potential future failures, allowing preventive measures to be taken in advance.
[0040] Reduce maintenance costs: Automated fault detection reduces the need for manual monitoring, while precise location of the fault source can guide maintenance work to be carried out more efficiently.
[0041] Strong adaptability: This method does not need to rely on specific physical or mathematical models, so it can be flexibly applied to different types of timekeeping systems and can continuously self-adjust and optimize with new data. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0043] Figure 1 A schematic diagram of a fault detection network model training process of a timekeeping system fault detection method based on machine learning of the present invention;
[0044] Figure 2 The figure is a schematic diagram of the diagnostic process of a timekeeping system fault detection method based on machine learning of the present invention. DETAILED DESCRIPTION
[0045] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.
[0046] The technical solutions provided by various embodiments of the present application are described in detail below in conjunction with the accompanying drawings.
[0047] like Figure 1 and Figure 2As shown, a method for detecting faults in a timekeeping system based on machine learning of the present invention can detect faults in a timely manner when a key time-frequency device in the timekeeping system fails, providing a basis for fault handling. The extreme learning machine (ELM) algorithm used in the present invention is a machine learning algorithm used to train an artificial neural network model. The network structure of the ELM algorithm is the same as that of the single hidden layer feedforward neural network (SLFN), except that in the training stage, it is no longer the gradient-based algorithm (backward propagation) that has been tried and tested in traditional neural networks, but uses random input layer weights and deviations, and the output layer weights are calculated by generalized inverse matrix theory. After the weights and deviations on all network nodes are obtained, the training of the extreme learning machine (ELM) is completed. At this time, when the test data comes, the network output can be calculated using the output layer weights just obtained to complete the prediction of the data.
[0048] The method includes:
[0049] Step S1, obtaining fault sample data, and generating a feature vector for each fault sample data in combination with label data;
[0050] Step S2, constructing a target vector of a fault detection method according to whether there is a fault and the fault level of the fault sample data;
[0051] Step S3, randomly initializing the weight matrix and bias vector of the hidden layer in the pre-built fault detection network model;
[0052] Step S4, using each feature vector as input data of the initialized fault detection network model, calculating the activation value of each node in the hidden layer, and constructing an activation matrix based on multiple activation values;
[0053] Step S5, calculating the connection weights from the hidden layer to the output layer based on the target vector, the activation matrix, the identity matrix and the regularization coefficient; repeating steps S1 to S5, training the fault detection network model until convergence, and obtaining the optimal model;
[0054] Step S6, collect the operating status data of the timekeeping system in real time, and filter out the fault information data. When the amount of fault information data exceeds a preset threshold, construct multiple fault information data into a fault feature set, input it into the optimal model, and obtain the fault diagnosis result.
[0055] In order to more clearly illustrate the timekeeping system fault detection method based on machine learning of the present invention, the following is combined with Figure 1 Each step in the embodiment of the present invention is described in detail.
[0056] A timekeeping system fault detection method based on machine learning according to a first embodiment of the present invention includes steps S1 to S6, each of which is described in detail as follows:
[0057] Step S1, obtaining fault sample data, and generating a feature vector for each fault sample data in combination with label data;
[0058] In the present invention, structured data containing a large number of fault samples are prepared as fault sample data, and the fault sample data at least includes the operating environment temperature and humidity of the timekeeping system, and the current and voltage of the atomic clock.
[0059] Arrange m fault sample data and corresponding labels to form m feature vectors, and finally use these feature vectors to construct an m×n feature matrix X:
[0060]
[0061] The fault sample data includes fault data and non-fault data.
[0062] Step S2, constructing a target vector of a fault detection method according to whether there is a fault and the fault level of the fault sample data;
[0063] Among them, the target vector T is:
[0064]
[0065] At this stage, a target value (or label) is defined for each sample, which reflects whether the sample has a fault and the level of the fault. For a binary classification problem, it is a simple 0 (no fault) and 1 (fault). For multi-classification or multi-level fault detection, there are multiple different integer values to represent different types of faults or fault severity.
[0066] The target vector T is an m×k matrix, where m is the number of samples and k is the number of output categories (for single-output problems, k=1). Therefore, for each sample i, there is a corresponding target value ti, which together constitute the entire target vector T.
[0067] The present invention can help the model distinguish between data under normal operating conditions and different types of fault conditions by setting a target vector. The target vector not only detects whether a fault has occurred, but also grades the severity of the fault, thereby managing maintenance activities more finely. By learning from a large amount of historical data, the model can respond quickly when a new fault occurs, reducing diagnostic time.
[0068] Step S3, randomly initializing the weight matrix and bias vector of the hidden layer in the pre-built fault detection network model;
[0069] In step S3, the number of hidden layer nodes of the single hidden layer feedforward neural network model is set to h, and random feature mapping is performed based on uniform distribution, that is, the hidden layer parameters are randomly initialized to generate a weight matrix and a bias vector. Specifically:
[0070] Set the structure of the neural network model and randomly initialize the weight matrix A and bias vector B of the hidden layer. A single hidden layer feedforward neural network is used here, and the number of hidden layer nodes is set to h. The dimension of the weight matrix A is h×n (h is the number of hidden layers, n is the number of input features), and the bias vector B is a one-dimensional array of length h. These parameters are randomly drawn from a uniform distribution to ensure that the network has a certain degree of randomness in the initial state, which helps to avoid local optimal solutions.
[0071] Step S4, using each feature vector as input data of the initialized fault detection network model, calculating the activation value of each node in the hidden layer, and constructing an activation matrix based on multiple activation values;
[0072] The present invention adopts the infinitely differentiable sigmoid activation function to transform the m feature vectors x i As the input of the fault detection network model, calculate its corresponding activation value h in the hidden layer il :
[0073]
[0074] in, is the transpose of a row vector of the lth hidden layer node in the weight matrix A, b l is the bias vector of the lth hidden layer node.
[0075] When given an input feature vector x i When , the Sigmoid function is used as the activation function to calculate the activation value h of each node in the hidden layer il The Sigmoid function is an infinitely differentiable nonlinear function that can map any real value to the interval (0,1), which is very useful for processing probabilities or normalized outputs.
[0076] h in the formula il represents the activation value of the i-th sample on the l-th hidden layer node, It represents the linear combination of the input features and the weights plus the bias term. The activation value obtained after conversion by the Sigmoid function can be used to reflect the degree of "excitation" of the input features for a specific hidden layer node.
[0077] The activation values corresponding to all m samples together constitute the activation matrix H, whose dimension is m×h, that is, the number of samples multiplied by the number of hidden layers.
[0078] Step S5, calculating the connection weights from the hidden layer to the output layer based on the target vector, the activation matrix, the identity matrix and the regularization coefficient; repeating steps S1 to S5, training the fault detection network model until convergence, and obtaining the optimal model;
[0079] The connection weight γ in the present invention is calculated as follows:
[0080] γ= ( H T H+ωI ) -1 H T T;
[0081] Among them, H is the activation matrix, H T is the transpose of the activation matrix, ω is the regularization coefficient, T is the target vector, and I is the identity matrix.
[0082] The present invention determines the connection weight γ from the hidden layer to the output layer, which is accomplished by minimizing the training error. Specifically, a set of weights γ is found so that the predicted output is as close as possible to the true target vector T. The least squares method is used for optimization, and a regularization term ωI (where I is the identity matrix) is added to prevent overfitting and improve the generalization ability of the model.
[0083] formula ( H T H+ωI ) -1 H T T is the analytical solution used to calculate the optimal weight γ, H T is the transpose of the activation matrix H,
[0084] H T T is a square matrix that represents the covariance information between all sample activation values; adding the regularization coefficient ω multiplied by the unit matrix I is to ensure the reversibility of the matrix and control the complexity of the model. The final result γ is the connection weight of the output layer, which directly determines the output of the model.
[0085] Repeat the above steps until the model converges, that is, the training error no longer decreases significantly. The saved model is the best trained model and can be used for subsequent timekeeping system fault detection tasks.
[0086] Step S6, collect the operating status data of the timekeeping system in real time, and filter out the fault information data. When the amount of fault information data exceeds a preset threshold, construct multiple fault information data into a fault feature set, input it into the optimal model, and obtain the fault diagnosis result.
[0087] Since the real-time collection of the operating status data of the timing system may contain a lot of noise or irrelevant information, directly feeding all the data into the neural network may lead to a waste of computing resources and an increase in false positives. Therefore, the first step of online diagnosis is to filter out obviously irrelevant data points through a preliminary screening mechanism.
[0088] Collect real-time data from the sensors and other monitoring devices of the timekeeping system. This data should include all the key characteristics mentioned before, such as temperature, humidity, current, voltage, etc.
[0089] According to the set rules (such as threshold comparison), the information suspected of fault is selected from the collected data. This step can be based on simple statistical methods or pre-defined logical conditions to reduce the amount of data for subsequent processing while ensuring that important fault signals are not missed.
[0090] In order to further reduce the false alarm rate and optimize the computing efficiency, a reasonable activation threshold n should be set. By activating the diagnostic module, a deeper diagnostic analysis will be triggered only when the number of suspected fault information obtained after preliminary screening exceeds this threshold. This strategy can help the system avoid frequent initiation of comprehensive diagnosis due to occasional very individual error messages, thereby saving computing resources.
[0091] Once the diagnostic module is activated, the suspected fault data that has been initially screened out needs to be preprocessed in preparation for input into the neural network model. The preprocessing steps usually include:
[0092] Standardization / normalization: Make the data scale of different features consistent to ensure model performance.
[0093] Missing value processing: fill in or delete data records with missing values.
[0094] Outlier handling: Identify and handle data points that are far from the normal range to prevent them from affecting the prediction accuracy of the model.
[0095] After the above preprocessing steps, a set of fault feature vectors suitable for input to the neural network is obtained, which together constitute the fault feature set. The format of the feature set should match the feature matrix X used in the training phase, that is, each sample consists of the same number of features.
[0096] Finally, the preprocessed fault feature set is input into the optimal model saved in advance, and the model will give a diagnosis result based on the previously learned knowledge. The diagnosis result is a classification label (such as "no fault", "minor fault", "serious fault", etc.) or a probability assessment of a continuous value, indicating the probability of a specific type of fault.
[0097] Although the various steps in the above embodiment are described in the above-mentioned order, those skilled in the art can understand that in order to achieve the effect of this embodiment, different steps do not have to be executed in such an order. They can be executed simultaneously (in parallel) or in a reverse order. These simple changes are within the scope of protection of the present invention.
[0098] A timekeeping system fault detection system based on machine learning according to a second embodiment of the present invention is based on a timekeeping system fault detection method based on machine learning, and the system includes:
[0099] A data set construction module, which is configured to obtain fault sample data and generate a feature vector for each fault sample data in combination with the label data;
[0100] A target vector calculation module configured to construct a target vector of a fault detection method according to whether there is a fault and the fault level of the fault sample data;
[0101] A network model parameter initialization module configured to randomly initialize a weight matrix and a bias vector of a hidden layer in a pre-built fault detection network model;
[0102] an activation matrix building module, configured to use each feature vector as input data of the initialized fault detection network model, calculate an activation value of each node of the hidden layer, and build an activation matrix based on the multiple activation values;
[0103] A training module is configured to calculate the connection weights from the hidden layer to the output layer based on the target vector, the activation matrix, the identity matrix and the regularization coefficient; and repeatedly train the fault detection network model until convergence to obtain an optimal model.
[0104] The fault diagnosis module is configured to collect the operating status data of the timekeeping system in real time and filter out the fault information data. When the amount of fault information data exceeds a preset threshold, multiple fault information data are constructed into a fault feature set and input into the optimal model to obtain the fault diagnosis result.
[0105] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.
[0106] It should be noted that the above embodiment provides a timekeeping system fault detection system based on machine learning, which is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be decomposed or combined. For example, the modules in the above embodiments can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps, and are not regarded as improper limitations of the present invention.
[0107] An electronic device according to a third embodiment of the present invention includes:
[0108] at least one processor; and
[0109] a memory communicatively connected to at least one of the processors; wherein,
[0110] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the above-mentioned timekeeping system fault detection method based on machine learning.
[0111] A computer-readable storage medium according to a fourth embodiment of the present invention stores computer instructions, and the computer instructions are used to be executed by the computer to implement the above-mentioned timekeeping system fault detection method based on machine learning.
[0112] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related instructions of the storage device and processing device described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.
[0113] Those skilled in the art should be able to appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software or a combination of the two, and the programs corresponding to the software modules and method steps can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the technical field. In order to clearly illustrate the interchangeability of electronic hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.
[0114] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.
Claims
1. A timekeeping system fault detection method based on machine learning, characterized in that: The method includes: Step S1, obtaining fault sample data, and generating a feature vector for each fault sample data in combination with label data; Step S2, constructing a target vector of a fault detection method according to whether there is a fault and the fault level of the fault sample data; Step S3, randomly initializing the weight matrix and bias vector of the hidden layer in the pre-built fault detection network model; Step S4, using each feature vector as input data of the initialized fault detection network model, calculating the activation value of each node in the hidden layer, and constructing an activation matrix based on multiple activation values; Step S5, calculating the connection weights from the hidden layer to the output layer based on the target vector, the activation matrix, the identity matrix and the regularization coefficient; repeating steps S1 to S5, training the fault detection network model until convergence, and obtaining the optimal model; Step S6, collect the operating status data of the timekeeping system in real time, and filter out the fault information data. When the amount of fault information data exceeds a preset threshold, construct multiple fault information data into a fault feature set, input it into the optimal model, and obtain the fault diagnosis result.
2. A timekeeping system fault detection method based on machine learning according to claim 1, characterized in that: The fault sample data at least includes the operating environment temperature and humidity of the timekeeping system, and the current and voltage of the atomic clock.
3. The method for detecting timekeeping system faults based on machine learning according to claim 1, characterized in that: The fault detection network model is constructed based on a single hidden layer feedforward neural network model.
4. The method for detecting timekeeping system faults based on machine learning according to claim 1, characterized in that: The weight matrix and bias vector of the hidden layer in the pre-built fault detection network model are randomly initialized as follows: The number of hidden layer nodes of the single hidden layer feedforward neural network model is set to h, and random feature mapping is performed based on uniform distribution to generate weight matrix and bias vector.
5. The method for detecting timekeeping system faults based on machine learning according to claim 1, characterized in that: The activation value h il , which is calculated as: Using the infinitely differentiable sigmoid activation function, the m feature vectors x i As the input of the fault detection network model, calculate its corresponding activation value h in the hidden layer il : in, is the transpose of a row vector of the lth hidden layer node in the weight matrix A, b l is the bias vector of the lth hidden layer node.
6. The method for detecting timekeeping system faults based on machine learning according to claim 1, characterized in that: The connection weight γ is calculated as follows: γ=(H T H+ωI) -1 H T T; Among them, H is the activation matrix, H T is the transpose of the activation matrix, ω is the regularization coefficient, T is the target vector, and I is the identity matrix.
7. The method for detecting timekeeping system faults based on machine learning according to claim 1, characterized in that: After obtaining the fault diagnosis results, trace the source of the fault information.
8. A timekeeping system fault detection system based on machine learning, based on a timekeeping system fault detection method based on machine learning according to any one of claims 1 to 7, characterized in that: The system includes: A data set construction module, which is configured to obtain fault sample data and generate a feature vector for each fault sample data in combination with the label data; A target vector calculation module configured to construct a target vector of a fault detection method according to whether there is a fault and the fault level of the fault sample data; A network model parameter initialization module configured to randomly initialize a weight matrix and a bias vector of a hidden layer in a pre-built fault detection network model; an activation matrix building module, configured to use each feature vector as input data of the initialized fault detection network model, calculate an activation value of each node of the hidden layer, and build an activation matrix based on the multiple activation values; A training module is configured to calculate the connection weights from the hidden layer to the output layer based on the target vector, the activation matrix, the identity matrix and the regularization coefficient; and repeatedly train the fault detection network model until convergence to obtain an optimal model. The fault diagnosis module is configured to collect the operating status data of the timekeeping system in real time and filter out the fault information data. When the amount of fault information data exceeds a preset threshold, multiple fault information data are constructed into a fault feature set and input into the optimal model to obtain the fault diagnosis result.
9. An electronic device, characterized in that: include: at least one processor; as well as a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement a timekeeping system fault detection method based on machine learning as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to be executed by the computer to implement a timekeeping system fault detection method based on machine learning as described in any one of claims 1-7.