Ship power device fault detection method based on state matching

Abstract

The invention provides a ship power device fault detection method based on state matching. The method comprises the steps of obtaining device working condition data of a ship power device; calling a device state model to perform state feature extraction on the device working condition data to obtain device state features, then determining device state features corresponding to each target device in the ship power device, and finally splicing the device state features with the device state features to obtain a device state feature; and a state matching model is called to identify fusion features obtained through splicing, and a fault detection result of the ship power device is obtained. The method and device are used for overcoming the defect that in the prior art, when a fault detection model based on deep learning is used for executing fault detection, the model is poor in environmental adaptability.

Description

Technical Field

[0001] The present invention relates to the technical field of ship fault detection, and in particular to a ship power plant fault detection method based on state matching. Background Art

[0002] The power plant is one of the most complex and critical systems in a ship. Highly reliable power plants are essential to ensure reliable and safe navigation. When power plant problems arise, timely and rapid fault location is a prerequisite for resolution. Traditional fault location methods often rely on the knowledge and experience of technicians. For simple and common faults, rule-based methods can often be used. However, these methods often have significant limitations and often fail to function properly when encountering new faults or after system upgrades. To ensure power system safety, it is necessary to assess system status to ensure early detection and timely resolution of faults.

[0003] Existing technologies use deep learning-based fault detection models to locate faults. However, these models often require large amounts of data for calculation, are not universal, and suffer from poor environmental adaptability. For example, when a fault detection model trained in one environment is migrated to another, it often requires retraining or fine-tuning to adapt to the environment. Summary of the Invention

[0004] The present invention provides a ship power plant fault detection method based on state matching, which is used to overcome the defect of poor environmental adaptability of the model when using a deep learning-based fault detection model to perform fault detection in the prior art.

[0005] The present invention provides a method for detecting a fault in a marine power plant based on state matching, the method comprising the following steps: Obtaining device operating condition data of ship power plants; Calling the device state model to extract state features of the device operating condition data to obtain device state features; Determine the equipment status characteristics corresponding to each target equipment in the ship power plant, wherein the equipment status characteristics are obtained by pre-calling the corresponding equipment status model to extract the equipment operating condition data of the target equipment; The device state feature is spliced with the equipment state feature, and the state matching model is called to identify the spliced fusion feature to obtain the fault detection result of the ship power plant.

[0006] In some embodiments, the device state model includes a graph convolutional network layer, a temporal convolutional network layer, and a spatiotemporal fusion network layer. The calling of the device state model to extract state features from the device operating condition data to obtain device state features includes: Determining graph structure data corresponding to the device operating condition data; Calling the graph convolutional network layer to extract node features from the graph structure data to obtain node spatial features; Calling the temporal convolutional network layer to extract temporal features of nodes in the graph structure data to obtain node temporal features; The node spatial features and the node temporal features are input into the spatiotemporal fusion network layer for fusion processing to obtain device state features.

[0007] In some embodiments, the training process of the device state model includes: Acquire operating condition data samples of a ship power plant, wherein the operating condition data samples include equipment operating condition samples corresponding to a plurality of target equipment; Randomly selecting an equipment operating condition sample of a target equipment from the operating condition data samples as a target operating condition sample, and removing the target operating condition sample from the operating condition data samples; Inputting the operating condition data sample after removing the target operating condition sample into the initial state model to generate a predicted operating condition sample; A loss function is constructed based on the predicted operating condition samples and the target operating condition samples, and the initial state model is trained using the loss function to obtain a device state model.

[0008] In some embodiments, the state matching model includes an MLP model, and calling the state matching model to identify the spliced fusion features to obtain a fault retrieval result of the ship power plant includes: The concatenated fusion features are input into the MLP model for prediction processing to obtain the corresponding state feature scalar; Processing the state characteristic scalar mapping into a probability distribution of a ship power plant fault state; The state with the highest probability score in the probability distribution is determined as the fault detection result of the ship power plant.

[0009] In some embodiments, before obtaining the device operating condition data of the ship power plant, the method further includes: Obtain a device state model corresponding to each target device in the ship power plant. The device state model is an LSTM model or a Transformer model. The training process of the device state model includes: Obtain equipment operating condition samples for each target equipment in the ship's power plant; The initial device models of the corresponding target devices are trained separately using the device operating condition samples to obtain a device state model for each target device.

[0010] In some embodiments, after obtaining the equipment state model corresponding to each target equipment in the ship propulsion system through training, the method further includes: Obtain the current equipment operating condition data of each target equipment in the ship's power plant; Calling the device state model corresponding to the target device to extract the current device state characteristics of the current device operating condition data, and storing the current device state characteristics in a database; When a new target device is replaced in a ship power plant, the corresponding device state model is called to extract the state features of the device operating data of the new target device to obtain the corresponding new device state features; The current device status characteristics corresponding to the new target device in the database are updated based on the new device status characteristics, and the database is used to perform status matching with the device status characteristics of the ship power plant to determine the fault detection result of the ship power plant.

[0011] The present invention also provides a ship power plant fault detection device based on state matching, which includes the following modules: An acquisition module, used to acquire device operating condition data of a ship power device; An extraction module, configured to call a device state model to extract state features from the device operating condition data to obtain device state features; A determination module is used to determine the equipment status characteristics corresponding to each target equipment in the ship power plant, wherein the equipment status characteristics are obtained by pre-calling the corresponding equipment status model to extract the equipment operating condition data of the target equipment; The matching module is used to splice the device state feature with the equipment state feature, and call the state matching model to identify the spliced fusion feature to obtain the fault detection result of the ship power plant.

[0012] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for detecting faults of a ship power plant based on state matching as described above is implemented.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for detecting faults of a ship power plant based on state matching as described above is implemented.

[0014] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned methods for detecting faults of a ship power plant based on state matching.

[0015] The present invention provides a method for detecting faults in a ship power plant based on state matching. The method first extracts the device state characteristics of the ship power plant, then determines the device state characteristics of each target device extracted in advance, matches the device state characteristics with the device state characteristics by state feature matching, obtains a fault detection result, and realizes fault identification of the ship power plant. Fault identification using device state characteristics can be unaffected by environmental factors and has universality. Even when fault identification is performed in different equipment environments, there is no need to migrate the model, re-fine-tune the model, or re-train it to adapt to the new equipment environment. Instead, it is only necessary to re-extract the state characteristics under the corresponding equipment environment for state matching, which has strong environmental adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced one by one below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 It is a flow chart of the ship power plant fault detection method based on state matching provided by the present invention.

[0018] Figure 2 This is an example diagram of the target equipment structure of the ship power unit provided by the present invention.

[0019] Figure 3 It is a structural schematic diagram of the ship power plant fault detection device based on state matching provided by the present invention.

[0020] Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention.

[0021] Reference numerals: 1: Feedwater heater; 2: Feedwater pump; 3: Secondary circuit system; 4: Main circulation pump. DETAILED DESCRIPTION

[0022] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0023] The state matching-based ship power plant fault detection method provided by the present invention can be applied to the control system of the ship power plant, and is specifically deployed in the server or terminal of the control system. It can obtain in real time the device operating condition data of the ship power plant collected by the control system through sensors, and then use this method to realize the fault identification of the ship power plant, and finally feed back the fault detection result to the control system or directly use the ship power plant to issue an alarm.

[0024] The following describes the ship power plant fault detection method based on state matching of the present invention with reference to the accompanying drawings. Figure 1 FIG. 1 is a flow chart of a method for detecting a fault in a marine power plant based on state matching provided by the present invention. Figure 1 As shown, the method includes the following steps 101 to 104, which are described one by one below.

[0025] Step 101: Acquire device operating condition data of a ship power device.

[0026] like Figure 2 As shown, Figure 2 This is an example diagram of a ship power plant. The ship power plant consists of multiple target devices such as a feedwater heater 1, a feedwater pump 2, a secondary circuit system 3, and a main circulation pump 4. Various sensors or components are also installed in the ship power plant to monitor the operating data of each target device in the ship power plant during operation. For example, a sound sensor collects device sound as operating data, and a temperature sensor measures device temperature as operating data. Abnormal sound or excessively high temperature can be the cause of a malfunction in the ship power plant.

[0027] The data collected by these sensors installed in the ship's power plant are uniformly uploaded as device operating data. At this time, the device operating data of the ship's power plant is obtained. It can be understood that the device operating data is collected in real time, including the current device operating data of each target device under the operating load state.

[0028] Step 102: Call the device state model to extract state features from the device operating condition data to obtain device state features.

[0029] After obtaining the current operating condition data of the ship's power plant in step 101, the plant state model is invoked to extract state features from the operating condition data to obtain plant state features. Here, the plant state model can be a temporal convolutional graph neural network (TCGNN), specifically comprising three components: a graph convolutional network layer (GCN), a temporal convolutional network layer (TCN), and a spatiotemporal fusion network layer.

[0030] When extracting state features, the first step is to determine the graph structure data corresponding to the device operating condition data. Each component (i.e., target device) in a ship propulsion system has a certain physical or logical connection relationship. Because a component failure may also cause failures in other connected components, the device operating condition data can be mapped into a graph structure data based on the connection relationships between target devices. Specifically, each target device can be treated as a node, and the physical or logical connection relationships between target devices can be used as edges between the nodes. This creates a graph structure data based on the nodes and edges.

[0031] When extracting state features, it is necessary to extract both the state features of the node itself and the temporal features between nodes. Therefore, for graph-structured data, the graph convolutional network layer is called to extract node features from the graph-structured data to obtain node spatial features. The graph convolutional network layer generates a node feature representation that includes spatial associations by aggregating information about the node neighborhood. The formula is as follows: (1) In the above formula (1), represents the node spatial features of node v generated by the k-th graph convolutional network layer in the graph convolutional network layer, It is the set of neighbor nodes of node v in the graph structure data, and u represents Any neighbor node of the middle node v, and They represent the weight parameters and bias terms of the k-th layer graph convolutional network respectively. represents the node spatial features of the neighbor node u generated by the k-1th graph convolutional network layer in the graph convolutional network layer, Represents the association parameter between node v and its neighbor node u. is the activation function of the graph convolutional network.

[0032] Next, the temporal convolutional network layer is called to extract temporal features from the nodes in the graph structure data to obtain node temporal features. The temporal convolutional network layer consists of multiple extended convolutional networks of different scales, which can comprehensively extract features of different scales and capture the dynamic features of nodes changing over time. Here, all nodes are spliced into node sequences. The temporal convolutional network layer can divide the input node sequence according to the time step t, and then process all time steps in the sequence in parallel. The calculation formula of the node temporal features is expressed as: (2) In the above formula (2), is the node temporal feature output by the temporal convolutional network layer at time step t, k represents the number of network layers in the temporal convolutional network layer, and Represent the convolution kernel and bias term of the i-th extended convolutional network, is the node data that is d time steps away from the current time step t in the node sequence, and d is the expansion factor of the expanded convolutional network.

[0033] Temporal convolution typically uses multi-scale convolution operations to more comprehensively capture features at different scales in the time series. A single-scale convolution kernel can only capture features within a fixed time range, while multi-scale convolution can capture the dynamic features of nodes at different time scales by using convolution kernels of different sizes or dilation factors at the same time.

[0034] In order to more accurately model the dynamic characteristics of the target equipment in the ship propulsion system, in the embodiment of the present invention, the temporal convolutional network layer adopts multiple layers of extended convolutional networks of different scales, and then increases the convolution kernel of each layer to increase the network receptive field, gradually expanding from short-term time dependence to long-term dependence. In other words, the convolution operation of each layer of the extended convolutional network in the temporal convolutional network layer adopts different convolution kernel sizes. and the expansion factor d, so that the shallower layers capture short-term dynamic features and the deeper layers capture long-term trend features.

[0035] Finally, the node spatial features and node temporal features are input into the spatiotemporal fusion network layer for fusion processing to obtain the device state features, which are recorded as Here, the spatiotemporal fusion network layer is used to fuse temporal and spatial features, integrating the feature information of both spatial and temporal dimensions to generate a more accurate representation of the system state: (3) In the above formula (3), represents the final calculated device status characteristics, is the node spatial feature output by the graph convolutional network layer, It is the node temporal feature output by the temporal convolutional network layer. Fusion represents the calculation function of feature fusion. The calculation function can be to set corresponding weights for the two respectively, and then sum the weighted results, or to directly splice the feature vectors, or to calculate through the self-attention mechanism.

[0036] In this way, the present invention simultaneously considers the time-varying operating state of each target device and the operating states between target devices, extracting and fusing the corresponding temporal and spatial state features, thereby improving the accuracy of the representation of the ship propulsion system state. Furthermore, a temporal convolutional network is used to extract temporal state features. Due to the stability of the temporal convolutional network, the stability of temporal feature modeling is improved, achieving efficient temporal feature extraction.

[0037] The device status model is a deep learning model. The model needs to be pre-trained before it can extract features from device operating condition data. The following describes the training process of the device status model.

[0038] First, we obtain the operating condition data samples of the ship's power plant. These data samples include equipment operating condition samples corresponding to multiple target equipment. These operating condition data samples include positive samples and negative samples. Positive samples are equipment operating condition data of the ship's power plant in a normal state, while negative samples are equipment operating condition data of the ship's power plant in a faulty state. We then randomly select an equipment operating condition sample of a target equipment from the operating condition data samples as the target operating condition sample. This target operating condition sample is then removed from the operating condition data samples. Next, the operating condition data samples after removing the target operating condition sample are input into the initial state model to generate a predicted operating condition sample.

[0039] Here, a target device, such as a main circulation pump or feedwater heater, is randomly selected from the operating data samples of all target devices in the ship's power plant. The operating data for this target device is then removed, for example, by removing the valve opening information for the main circulation pump or the speed information for the feedwater heater. This removed operating data sample is then re-input into the initial state model, which predicts the missing values and regenerates the data based on the predicted state features to recover the removed data, such as the valve opening information for the main circulation pump or the speed information for feedwater heater 1.

[0040] Finally, a loss function is constructed based on the predicted operating condition samples and the target operating condition samples. This loss function is then used to train the initial state model, resulting in the device state model. During initial state model training, to verify the accuracy of the model's extraction of device state features, the regenerated predicted operating condition samples are compared with the original, but removed, target operating condition samples. A corresponding loss function, such as a cross-entropy loss function or a mean squared error loss function, is constructed. The initial state model is then trained using this loss function. Training is terminated when the preset number of training rounds is reached or when the loss function begins to converge, resulting in the device state model.

[0041] In an embodiment of the present invention, by training a device state model, when it is necessary to perform fault detection on a ship power plant, the pre-trained device state model can efficiently and accurately extract features from the device operating condition data of the ship power plant, thereby ensuring the correctness of subsequent state matching and improving the accuracy of fault prediction.

[0042] Step 103: Determine the device state characteristics corresponding to each target device in the ship power plant. The device state characteristics are obtained by pre-calling the corresponding device state model to extract the state characteristics of the device operating data of the target device.

[0043] Through step 102, the device status characteristics of the current device operating condition data of the ship power device are extracted Finally, the device status characteristics corresponding to each target device are required Matching is performed to determine whether there are differences, thereby determining whether there is a fault in the device. However, this requires determining the device status characteristics corresponding to each target device in the ship's power plant. This device status characteristic The state characteristics of the equipment condition data of the target equipment are extracted by calling the corresponding equipment condition model in advance. The equipment condition data of the target equipment are all the condition data of the ship power unit in normal state. Therefore, the extracted equipment condition characteristics It can be used as a criterion for determining whether a fault occurs during status matching.

[0044] Before acquiring device operating condition data for a ship power plant, embodiments of the present invention obtain a device state model corresponding to each target device in the ship power plant. Specifically, before performing fault detection, a device state model corresponding to each target device in the ship power plant is pre-selected and trained. Specifically, a corresponding device state model is trained for each target device.

[0045] Here, the device state model can use the long short-term memory network (LSTM) model or the Transformer model. Taking the LSTM model as an example, when extracting device state features When the equipment working condition data at different times Input into the LSTM model, and then process the input data through the forget gate, input gate and output gate of the model respectively. The calculation formulas of the forget gate and input gate are as follows: (4) (5) In the above formulas (4) and (5), and Represent the activation values of the forget gate and input gate respectively, represents the equipment operating condition data input into the model at time step t, represents the hidden state features corresponding to the equipment operating condition data input into the model at the last time step t-1, and Represent the weight matrix and bias term of the hidden layer in the forget gate, and They represent the weight matrix and bias term of the hidden layer in the input gate respectively. is the sigmoid activation function.

[0046] In the LSTM model, the core is the cell state, which is like an information conveyor belt that runs through the entire time series. The cell state can maintain relatively stable information transmission in each time step t of the sequence, so that the network can remember long-distance information. At each time step t, the cell state The information will be updated and transmitted according to the control of the input gate, forget gate and output gate. In the input gate of time step t, the candidate memory cell in the cell state The calculation formula is as follows: (6) In the above formula (6), and It also represents the relevant weight parameters and bias terms in the input gate. represents the equipment operating condition data input into the model at time step t, Represents the hidden state features corresponding to the equipment operating condition data input into the model at time step t-1. represents the hyperbolic tangent function.

[0047] Then update the activation values of the input gate and forget gate to the cell state. The formula is as follows: (7) In the above formula (7), represents the cell state corresponding to time step t, represents the cell state corresponding to the last time step t-1, and Represent the activation values of the forget gate and input gate respectively.

[0048] When the LSTM model outputs, it is necessary to update the activation value of the output gate according to the cell state of the corresponding time step, and then use the updated activation value as the hidden state output of the model at the current time step t. The calculation formula of the output gate activation value is: (8) In the above formula (8), represents the activation value of the output gate, represents the equipment operating condition data input into the model at time step t, represents the hidden state features corresponding to the equipment operating condition data input into the model at the last time step t-1, and denote the weight matrix and bias term of the hidden layer in the output gate, respectively. is the sigmoid activation function.

[0049] Finally, through the cell state Update the activation value of the output gate to obtain the hidden state output of the model at the current time step t , expressed as: (9) In the above formula (9), represents the hyperbolic tangent function.

[0050] In this way, at the last time step t, the hidden state output obtained by the LSTM model can be used as the device state feature of the target device.

[0051] The following describes the training process of the device state model.

[0052] First, obtain the equipment operating condition samples of each target equipment in the ship's power plant. Training the model still requires data samples, so here we can obtain the equipment operating condition samples of each target equipment in the ship's power plant under normal conditions without any failure, such as the valve opening information of the main circulation pump under normal operating conditions, the speed information of the feed water heater, etc.

[0053] The initial device model for each target device is then trained using the device condition samples to generate a device state model for each target device. This means that the initial device model is trained independently using the device condition samples of the target device. The training process is independent and does not interfere with each other. Alternatively, the training method involves removing some information from the device condition samples of the target device, such as removing the target condition data for a specific time period. The device condition samples without the target condition data are then fed into the initial device model for regeneration, recovering the target condition data. The specific training process is similar to that for the device state model and will not be detailed here.

[0054] After training the initial device model corresponding to each target device, the corresponding device state model can be called to extract the state features of the device operating data of the target device to obtain the corresponding device state features. .

[0055] Prior to fault detection, this embodiment of the present invention pre-collects equipment condition data for each target device during normal operation of the ship's propulsion system. This data is then used to train a device state model for the target device. The corresponding device state features are then extracted from the device state model and used as the criteria for fault detection. Therefore, even in different device environments, subsequent state matching can be performed simply by pre-calculating the device state features within the corresponding device environment. This eliminates the need to migrate the fault detection model, retrain the model, or fine-tune it to adapt to the new device environment, thereby improving environmental adaptability and fault detection efficiency.

[0056] Step 104: splice the device state features with the equipment state features, and call the state matching model to identify the spliced fusion features to obtain the fault detection results of the ship power plant.

[0057] The device status characteristics of the target device in the normal state are calculated by step 103 Then, the device status characteristics Device Status Characteristics The splicing is performed, and the state matching model is called to identify the spliced fusion features to obtain the fault detection results of the ship power unit.

[0058] Specifically, the state matching model includes a multi-layer perceptron (MLP) model. Of course, the state matching model can also be a machine learning classifier or a fully connected neural network, which is not limited here. Device Status Characteristics After splicing, the fusion features are obtained , the splicing can be done directly by vector splicing. Then the spliced fusion feature Input into the MLP model for prediction processing to obtain the corresponding state feature scalar. The prediction processing can be calculated and The feature similarity of As a key value, As a query, a cross-attention calculation is then performed. The predicted state feature scalar is one-dimensional and can be directly mapped to the corresponding probability through an activation function. Here, the activation function of the MLP model can be used to map the state feature scalar to a probability distribution of the fault state of the ship power plant. Finally, the state with the highest probability score in the probability distribution is determined as the fault detection result of the ship power plant. If the state with the highest probability score is a normal state, it means that the ship power plant is not faulty. If the state with the highest probability score is a faulty state, it means that at least one target device in the ship power plant is faulty.

[0059] In an embodiment of the present invention, the device state features of the ship's power plant are first extracted. The pre-extracted device state features of each target device are then determined. The device state features are then matched with the device state features through state feature matching to obtain a fault detection result, thereby enabling fault identification of the ship's power plant. Fault identification using device state features is unaffected by environmental factors. Even when performing fault identification in a changing device environment, there is no need to migrate the model, fine-tune the model, or retrain it for the new device environment. Instead, the state features in the corresponding device environment need only be re-extracted for state matching. This method is versatile and highly adaptable to environmental factors.

[0060] In some embodiments, there may be a situation where the ship's power plant is replaced with a new target device, such as Figure 2 In the case of a failure of the secondary circuit system or the feed water pump, a new secondary circuit system or a new feed water pump is replaced. Then the device status characteristics of the new target device will also change. When detecting a failure in the ship power plant, it is necessary to recalculate the device status characteristics of the new target device for state matching. If the new target device is replaced frequently, in order to facilitate real-time state matching. After pre-training the device status model corresponding to each target device in the ship power plant, the embodiment of the present invention obtains the current equipment operating condition data of each target device in the ship power plant, and then calls the device status model corresponding to the target device to extract the current equipment status characteristics of the current equipment operating condition data, and stores the current equipment status characteristics in the database. This database is used to perform state matching with the device status characteristics of the ship power plant to determine the fault detection result of the ship power plant. Storing the current equipment status characteristics in the database facilitates the updating and maintenance of the feature data and performs state matching in real time.

[0061] When a new target device is replaced in a ship's power plant, the corresponding device status model is simply called to extract the state features of the new target device's equipment condition data. This generates the corresponding new device status features. The database then updates the current device status features of the new target device based on the new device status features, enabling real-time maintenance of the feature data in the database. When fault detection is required for the ship's power plant, the extracted device status features are directly matched against the database.

[0062] In this embodiment of the present invention, a device state model for each target device is pre-trained. The current device state features corresponding to each target device in the ship's propulsion system are then extracted, stored in a database, and maintained in real time. When a new target device is replaced in the ship's propulsion system, there's no need to retrain or fine-tune the device state model to adapt it to the new target device. Instead, the existing device state model is used to extract the new device state features, and the corresponding current device state features are then updated in the database. This allows for real-time state matching in a changing, dynamic environment to perform fault detection, further improving fault detection efficiency.

[0063] The following describes the ship power plant fault detection device based on state matching provided by the present invention. The ship power plant fault detection device based on state matching described below and the ship power plant fault detection method based on state matching described above can refer to each other.

[0064] like Figure 3 As shown, the state matching-based marine power plant fault detection device includes: an acquisition module 301, an extraction module 302, a determination module 303, and a matching module 304. The acquisition module 301 is used to acquire the device operating condition data of the marine power plant; the extraction module 302 is used to call the device state model to extract state features from the device operating condition data to obtain device state features; the determination module 303 is used to determine the device state features corresponding to each target device in the marine power plant. The device state features are obtained by pre-calling the corresponding device state model to extract state features from the device operating condition data of the target device; and the matching module 304 is used to splice the device state features with the device state features, and call the state matching model to identify the spliced fusion features to obtain the marine power plant fault detection result.

[0065] It should be noted that the beneficial effects of the ship power plant fault detection device based on state matching here and the ship power plant fault detection method based on state matching mentioned above can correspond to each other, so the beneficial effects of the ship power plant fault detection device based on state matching will not be repeated here.

[0066] Figure 4An example of a physical structure diagram of an electronic device is shown below. Figure 4 As shown, the electronic device may include: a processor 410, a communications interface 420, a memory 430, and a communications bus 440, wherein the processor 410, the communications interface 420, and the memory 430 communicate with each other via the communications bus 440. The processor 410 may call the logic instructions in the memory 430 to execute a ship power plant fault detection method based on state matching, the method comprising: obtaining device operating condition data of the ship power plant; calling a device state model to extract state features of the device operating condition data to obtain device state features; determining the device state features corresponding to each target device in the ship power plant, the device state features being obtained by calling the corresponding device state model in advance to extract state features of the device operating condition data of the target device; The device state feature is spliced with the equipment state feature, and the state matching model is called to identify the spliced fusion feature to obtain the fault detection result of the ship power plant.

[0067] Furthermore, the logic instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0068] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the ship power plant fault detection method based on state matching provided by the above methods, and the method includes: obtaining the device operating condition data of the ship power plant; calling the device state model to extract the state features of the device operating condition data to obtain the device state features; determining the device state features corresponding to each target device in the ship power plant, and the device state features are obtained by pre-calling the corresponding device state model to extract the state features of the device operating condition data of the target device; splicing the device state features with the equipment state features, and calling the state matching model to identify the spliced fusion features to obtain the fault detection results of the ship power plant.

[0069] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the ship power plant fault detection method based on state matching provided by the above-mentioned methods, the method comprising: obtaining device operating condition data of the ship power plant; calling a device state model to extract state features of the device operating condition data to obtain device state features; determining the device state features corresponding to each target device in the ship power plant, the device state features being obtained by pre-calling the corresponding device state model to extract state features of the device operating condition data of the target device; splicing the device state features with the equipment state features, and calling a state matching model to identify the spliced fusion features to obtain a fault detection result of the ship power plant.

[0070] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0071] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for detecting faults in a ship power plant based on state matching, characterized in that: The method comprises: Obtaining device operating condition data of ship power plants; Calling the device state model to extract state features of the device operating condition data to obtain device state features; Determine the equipment status characteristics corresponding to each target equipment in the ship power plant, wherein the equipment status characteristics are obtained by pre-calling the corresponding equipment status model to extract the equipment operating condition data of the target equipment; The device state feature is spliced with the equipment state feature, and the state matching model is called to identify the spliced fusion feature to obtain the fault detection result of the ship power plant.

2. The ship power plant fault detection method based on state matching according to claim 1 is characterized in that: The device state model includes a graph convolutional network layer, a time series convolutional network layer, and a spatiotemporal fusion network layer. The device state model is called to extract state features from the device operating condition data to obtain device state features, including: Determining graph structure data corresponding to the device operating condition data; Calling the graph convolutional network layer to extract node features from the graph structure data to obtain node spatial features; Calling the temporal convolutional network layer to extract temporal features of nodes in the graph structure data to obtain node temporal features; The node spatial features and the node temporal features are input into the spatiotemporal fusion network layer for fusion processing to obtain device state features.

3. The ship power plant fault detection method based on state matching according to claim 1, characterized in that: The training process of the device state model includes: Acquire operating condition data samples of a ship power plant, wherein the operating condition data samples include equipment operating condition samples corresponding to a plurality of target equipment; Randomly selecting an equipment operating condition sample of a target equipment from the operating condition data samples as a target operating condition sample, and removing the target operating condition sample from the operating condition data samples; Inputting the operating condition data sample after removing the target operating condition sample into the initial state model to generate a predicted operating condition sample; A loss function is constructed based on the predicted operating condition samples and the target operating condition samples, and the initial state model is trained using the loss function to obtain a device state model.

4. The method for detecting faults in a ship power plant based on state matching according to claim 1, characterized in that: The state matching model includes an MLP model. The calling of the state matching model to identify the spliced fusion features to obtain the fault retrieval results of the ship power plant includes: The concatenated fusion features are input into the MLP model for prediction processing to obtain the corresponding state feature scalar; Processing the state characteristic scalar mapping into a probability distribution of a ship power plant fault state; The state with the highest probability score in the probability distribution is determined as the fault detection result of the ship power plant.

5. The method for detecting faults in a marine power plant based on state matching according to claim 1, characterized in that: Before obtaining the device operating condition data of the ship power plant, the method further includes: Obtain a device state model corresponding to each target device in the ship power plant. The device state model is an LSTM model or a Transformer model. The training process of the device state model includes: Obtain equipment operating condition samples for each target equipment in the ship's power plant; The initial device models of the corresponding target devices are trained separately using the device operating condition samples to obtain a device state model for each target device.

6. The method for detecting faults in a marine power plant based on state matching according to claim 5, characterized in that: After the training obtains the device state model corresponding to each target device in the ship power plant, the method further includes: Obtain the current equipment operating condition data of each target equipment in the ship's power plant; Calling the device state model corresponding to the target device to extract the current device state characteristics of the current device operating condition data, and storing the current device state characteristics in a database; When a new target device is replaced in a ship power plant, the corresponding device state model is called to extract the state features of the device operating data of the new target device to obtain the corresponding new device state features; The current device status characteristics corresponding to the new target device in the database are updated based on the new device status characteristics, and the database is used to perform status matching with the device status characteristics of the ship power plant to determine the fault detection result of the ship power plant.

7. A ship power plant fault detection device based on state matching, characterized in that: The device comprises: An acquisition module, used to acquire device operating condition data of a ship power device; An extraction module, configured to call a device state model to extract state features from the device operating condition data to obtain device state features; A determination module is used to determine the equipment status characteristics corresponding to each target equipment in the ship power plant, wherein the equipment status characteristics are obtained by pre-calling the corresponding equipment status model to extract the equipment operating condition data of the target equipment; The matching module is used to splice the device state feature with the equipment state feature, and call the state matching model to identify the spliced fusion feature to obtain the fault detection result of the ship power plant.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the ship power plant fault detection method based on state matching as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for detecting faults of a ship power plant based on state matching as claimed in any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method for detecting faults of a ship power plant based on state matching as claimed in any one of claims 1 to 6 is implemented.

Images