A remote monitoring and management system for new energy prime mover equipment based on Internet of Things technology

Abstract

The present invention relates to the field of equipment remote monitoring technology, specifically a remote monitoring and management system for new energy prime mover equipment based on Internet of Things technology, including a model update unit, an abnormality verification unit, a monitoring interface unit, a feature extraction unit, a model training unit, a state classification unit, a situation element prediction unit, a real-life image acquisition unit, an emergency processing unit, and an abnormality sorting unit, and also includes: a scene modeling unit, the scene modeling unit is used to perform 3D modeling on multiple new energy prime mover equipment in a target plant area to obtain multiple 3D models of new energy prime mover equipment, and based on the actual positions of the multiple new energy prime mover equipment in the target plant area. The present invention can perform accurate 3D modeling on multiple new energy prime mover equipment in a target plant area through a scene modeling unit, which makes the equipment layout and position relationship of the entire target plant area more intuitive and clear, and is conducive to the management and monitoring of equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment remote monitoring, and in particular to a remote monitoring and management system for new energy prime mover equipment based on Internet of Things technology. Background Art

[0002] Internet of Things technology refers to a technical system that connects various physical devices, sensors, software, networks, etc. through the Internet to achieve intelligent and automated information interaction and data transmission.

[0003] New energy motive equipment refers to equipment that uses new energy for power conversion and energy output. New energy is relative to traditional fossil energy, and mainly includes wind energy, solar energy, hydropower, biomass energy, geothermal energy, tidal energy, etc.

[0004] Most traditional systems use simple statistical methods or empirical rules for fault detection and prediction, and lack advanced machine learning and optimization algorithms. These methods have difficulty handling complex nonlinear relationships and high-dimensional data, resulting in low accuracy in fault identification and prediction. Due to reliance on manual inspections and manual analysis, traditional systems have a slow response speed after a fault occurs and are unable to achieve real-time monitoring and rapid emergency processing, which results in long equipment downtime and low production efficiency. Traditional systems usually require a lot of manual intervention and lack automated data processing and intelligent decision support. Traditional systems have limited visualization capabilities and can usually only provide simple charts and data reports. They lack intuitive 3D modeling and dynamic scene display, which makes the operating status of the equipment not intuitive enough and difficult to fully grasp the health status of the equipment. Summary of the invention

[0005] The purpose of the present invention is to address the problems existing in the background technology and to propose a remote monitoring and management system for new energy prime mover equipment based on the Internet of Things technology.

[0006] The technical solution of the present invention is a remote monitoring and management system for new energy prime mover equipment based on the Internet of Things technology, including a model updating unit, an abnormality verification unit, a monitoring interface unit, a feature extraction unit, a model training unit, a state classification unit, a situation element prediction unit, a real-scene image acquisition unit, an emergency processing unit and an abnormality sorting unit, and also includes:

[0007] A scene modeling unit, the scene modeling unit is used to perform 3D modeling on multiple new energy prime movers in a target plant area to obtain multiple 3D models of the new energy prime movers, and integrate the multiple 3D models of the new energy prime movers based on actual positions of the multiple new energy prime movers in the target plant area to obtain a scene model of the target plant area, and transmit the scene model of the target plant area to a grid division unit and a model updating unit;

[0008] A grid division unit, wherein the grid division unit receives the target plant scene model transmitted by the scene modeling unit, and grids the target plant scene model to obtain a plurality of grids, and proportionally associates the plurality of grids with the target plant to obtain a position relationship mapping table between positions of the plurality of new energy prime movers in the target plant and positions of the plurality of 3D models of the new energy prime movers in the target plant scene model, and transmits the position relationship mapping table to the abnormality verification unit;

[0009] A data monitoring unit, wherein the data monitoring unit is used to perform data monitoring on multiple new energy prime mover equipment in the target plant area to obtain monitoring data of the new energy prime mover equipment, wherein the monitoring data includes equipment operating status, environmental data, energy consumption data and operating parameters, and vibration data of multiple new energy prime mover equipment in the target plant area is collected through a vibration sensor to obtain a vibration data sequence, and the monitoring data is transmitted to a monitoring interface unit, and the vibration data sequence is transmitted to a feature extraction unit and a situation element prediction unit.

[0010] Preferably, the feature extraction unit receives the vibration data sequence transmitted by the data monitoring unit, and performs time domain analysis on the vibration data sequence to obtain time domain features, and performs fast Fourier transform on the vibration data sequence to obtain a spectrum diagram, and converts the time domain features into frequency domain features based on the spectrum diagram, and performs coarse-graining processing on the vibration data sequence to obtain vibration data sequences at different scales, and calculates sample entropy values ​​of the vibration data sequences at different scales to obtain sample entropy features, and calculates weighted permutation features of the vibration data sequence at each scale to obtain multi-scale weighted permutation entropy features, and transmits the time domain features, the frequency domain features, the sample entropy features and the multi-scale weighted permutation entropy features to the model training unit.

[0011] Preferably, the model training unit receives the time domain features, the frequency domain features, the sample entropy features and the multi-scale weighted permutation entropy features transmitted by the feature extraction unit, and constructs a state classification model based on a support vector machine algorithm, and divides the time domain features, the frequency domain features and the multi-scale weighted permutation entropy features into a training data set, and optimizes the penalty parameters and kernel function parameters of the state classification model using the gray wolf optimization algorithm based on the training data set to obtain an optimal combination of penalty parameters and kernel function parameters, and applies the optimal combination of penalty parameters and kernel function parameters to the state classification model to obtain a trained state classification model, and transmits the trained state classification model to the state classification unit.

[0012] Preferably, the situation element prediction unit receives the vibration data sequence transmitted by the data monitoring unit, and constructs the vibration data sequence into time series data through a sliding window, and inputs the time series data into a trained situation prediction model. The situation prediction model adopts a gated recurrent unit, and outputs the vibration data sequence prediction value corresponding to the vibration data sequence at the next preset moment through the trained situation prediction model, and transmits the vibration data sequence prediction value to the state classification unit.

[0013] Preferably, the state classification unit receives the trained state classification model transmitted by the model training unit and the vibration data sequence prediction value transmitted by the situation factor prediction unit, and extracts the time domain features, the frequency domain features, the sample entropy features and the multi-scale weighted permutation entropy features of the vibration data sequence prediction value, and inputs the time domain features, the frequency domain features, the sample entropy features and the multi-scale weighted permutation entropy features into the trained state classification model to obtain the future device state corresponding to the new energy prime mover device, and extracts the time domain features, the frequency domain features, the sample entropy features and the multi-scale weighted permutation entropy features of the real-time value of the vibration data sequence, and inputs the time domain features, the frequency domain features, the sample entropy features and the multi-scale weighted permutation entropy features into the trained state classification model to obtain the real-time device state corresponding to the new energy prime mover device, and transmits the future device state and the real-time device state corresponding to the new energy prime mover device to the model updating unit.

[0014] Preferably, the model updating unit receives the target plant scene model transmitted by the scene modeling unit and the future equipment status and the real-time equipment status corresponding to the new energy driving equipment transmitted by the state classification unit, and determines the abnormal new energy driving equipment according to the future equipment status and the real-time equipment status corresponding to the new energy driving equipment, and dynamically renders the 3D model of the abnormal new energy driving equipment in the target plant scene model to obtain the target plant dynamic scene model, and sets a dynamic update cycle to perform regular dynamic update on the target plant dynamic scene model, and transmits the target plant dynamic scene model to the monitoring interface unit and the abnormality verification unit.

[0015] Preferably, the real-scene image acquisition unit is used to set up multiple groups of cameras in the target factory area, acquire the real-scene images of various locations in the target factory area through the multiple groups of cameras, and transmit the real-scene images of various locations in the target factory area to the abnormality verification unit.

[0016] Preferably, the abnormality verification unit receives the position relationship mapping table transmitted by the grid division unit, the dynamic scene model of the target plant area transmitted by the model updating unit, and the real scene images of various locations of the target plant area transmitted by the real scene image acquisition unit, and judges the abnormal position of the target plant area according to the position of the grid where the 3D model of the abnormal new energy prime mover equipment in the dynamic scene model of the target plant area is located in the position relationship mapping table, and retrieves the real scene image at the abnormal position, verifies whether the abnormal position has an abnormality through the real scene image, and if an abnormality occurs, generates abnormal information, and transmits the abnormal information to the abnormal sorting unit.

[0017] Preferably, the exception sorting unit receives the exception information transmitted by the exception verification unit, and constructs an associated knowledge graph based on the exception information to determine the sequential relationship, concurrent relationship and conflict relationship of the exception information, and calculates the similarity between each of the exception information through a similarity model, and compares the obtained similarity between each of the exception information with a preset threshold to determine whether there is a correlation between the exception information. If there is a correlation, the associated exception information will be merged in the associated knowledge graph, and the corresponding exception processing level will be increased according to the number of exception information in each node in the associated knowledge graph, and the content in each node in the associated knowledge graph and the corresponding exception processing level will be transmitted to the emergency processing unit.

[0018] Preferably, the emergency processing unit receives the content of each node in the associated knowledge graph and the corresponding exception processing level transmitted by the exception sorting unit, and generates a troubleshooting list based on the content of each node in the associated knowledge graph and the corresponding exception processing level, and troubleshoots the occurred fault according to the troubleshooting list.

[0019] Preferably, the monitoring interface unit receives the monitoring data transmitted by the data monitoring unit and the target plant dynamic scene model transmitted by the model updating unit, and displays the target plant dynamic scene model on the monitoring interface of the monitoring intelligent terminal, and hides the monitoring data in the corresponding 3D model of the new energy prime mover equipment in the target plant dynamic scene model. When the user clicks on the corresponding 3D model of the new energy prime mover equipment in the target plant dynamic scene model, the monitoring data pops up in the form of a pop-up window.

[0020] Preferably, the penalty parameter and kernel function parameter of the state classification model are optimized using the gray wolf optimization algorithm based on the training data set to obtain an optimal combination of the penalty parameter and the kernel function parameter, comprising the following steps:

[0021] A1. Randomly initialize individuals of the gray wolf population, where the positions of the individuals of the gray wolf population represent a parameter combination of a penalty parameter and a kernel function parameter of the state classification model;

[0022] A2. Update the position of the individual of the gray wolf population to obtain a new position of the individual of the gray wolf population. The new position update formula is as follows:

[0023] ;

[0024] in, represents the position at the t+1th iteration, Indicates the location of the prey, represents the first random parameter, and , represents the nonlinear convergence factor, express The first random number in the interval, Indicates the distance between the prey and the wolf pack, , represents the position at the tth iteration, represents the second random parameter, and , express The second random number in the interval;

[0025] A3. Calculate the new fitness of the gray wolf population individual at the new position and compare it with the optimal fitness of the previous iteration. If the new fitness is greater than the optimal fitness, replace the new fitness with the optimal fitness and retain the new position of the gray wolf population individual. The fitness calculation formula is as follows:

[0026] ;

[0027] in, represents the average accuracy of K-fold cross validation, represents the accuracy calculated at the k-th fold;

[0028] A4. If the number of iterations is greater than a preset iteration threshold, the optimization is terminated to obtain the optimal combination of penalty parameters and kernel function parameters. Otherwise, steps A2-A3 are repeated.

[0029] Preferably, the vibration data sequence is subjected to coarse-graining processing to obtain vibration data sequences at different scales, and sample entropy values ​​of the vibration data sequences at different scales are calculated, comprising the following steps:

[0030] B1. The vibration data sequence is coarse-grained by a coarse-grained formula to obtain vibration data sequences at different scales. The coarse-grained formula is as follows:

[0031] ;

[0032] in, represents the vibration data sequence at scale i, represents the scale factor, represents the original vibration data sequence, that is, the vibration data sequence before coarse-graining processing;

[0033] B2. Constructing a sequence vector based on the vibration data sequences at different scales, and based on the distance between the sequence vector and other sequence vectors, the sequence vector is as follows:

[0034] ;

[0035] in, represents the sequence vector at scale i, represents the embedding dimension;

[0036] The distance formula is as follows:

[0037] ;

[0038] in, Represents the sequence vector at scale i with any other sequence vector at any scale j The distance between express Vibration data series at scale, express Vibration data series at scale;

[0039] B3. Calculate the ratio of the distance between the sequence vector and other sequence vectors less than the preset distance threshold to the number of vectors. The ratio formula is as follows:

[0040] ;

[0041] in, Represents the ratio, Indicates the preset distance threshold, Represents the number of data in the vibration data sequence at different scales. count() represents a counting function used to calculate the number of data that meet the conditions. The number of sequence vectors;

[0042] B4. Calculate the average value of the ratio of the distance between the sequence vector and other sequence vectors that is less than a preset distance threshold to the number of vectors. The average value formula is as follows:

[0043] ;

[0044] in, represents the average value of the mth iteration;

[0045] B5. Based on the average value of the ratio of the distance between the sequence vector and other sequence vectors being less than a preset distance threshold and the number of vectors, the sample entropy values ​​of the vibration data sequences at different scales are calculated. The sample entropy value formula is as follows:

[0046] ;

[0047] in, represents the sample entropy value, represents the average value of the m+1th iteration, and ln() represents the natural logarithm function.

[0048] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0049] 1. The present invention can accurately 3D model multiple new energy prime movers in the target plant area through the scene modeling unit, and integrate them into an overall target plant area scene model, which makes the equipment layout and position relationship of the entire target plant area more intuitive and clear, which is helpful for equipment management and monitoring. The target plant area scene model is gridded by the grid division unit and proportionally associated with the actual plant area to generate a position relationship mapping table, which is helpful for accurately locating the equipment and its abnormal position, and improving the efficiency of fault detection and processing. The equipment operation status, environmental data, energy consumption data, operation parameters and vibration data sequence are collected by the data monitoring unit, and the time domain and frequency domain are extracted by the feature extraction unit. Feature analysis and multi-scale weighted arrangement feature extraction can comprehensively reflect the operating status of the equipment and its changing trends. The state classification model is constructed using the support vector machine algorithm through the model training unit, and the parameters are optimized through the gray wolf optimization algorithm to obtain the optimal combination of penalty parameters and kernel function parameters. This ensures the accuracy and robustness of the state classification model and improves the accuracy of fault identification and prediction. The dynamic scene model of the target plant area and the monitoring data are integrated into the interface of the monitoring intelligent terminal through the monitoring interface unit, so that users can view detailed monitoring data in the form of pop-up windows by clicking on the 3D model of the equipment. This humanized design greatly improves the user experience and ease of operation.

[0050] 2. The present invention uses a gated loop unit to construct a situation prediction model through a situation element prediction unit to predict the vibration data sequence. The state classification unit combines the predicted value and the real-time value to extract various features and input them into the state classification model, thereby obtaining the future state and real-time state of the equipment. This dual state monitoring method enhances the system's early warning capability and real-time response capability. The model update unit regularly updates the dynamic scene model of the target plant area, and combines the real-scene images obtained by multiple groups of cameras. The abnormal verification unit verifies the abnormal information, which improves the authenticity and confirmation speed of the fault information. The abnormal sorting unit constructs an associated knowledge graph based on the abnormal information, analyzes the order, concurrency and conflict relationship of the abnormal information, calculates the similarity and improves the processing level. This intelligent sorting and associated analysis method can better manage multiple faults, improve processing efficiency and accuracy, and generate a fault troubleshooting list through the emergency processing unit and troubleshoot according to this list, which helps to quickly respond to and handle faults, reduce equipment downtime, and ensure production continuity and safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of the overall system flow in an embodiment of the present invention;

[0052] Figure 2 A schematic diagram of a process flow of parameter optimization in an embodiment of the present invention;

[0053] Figure 3 The figure is a schematic diagram of a flow chart of sample entropy value calculation in an embodiment of the present invention.

[0054] Figure numerals: 1. scene modeling unit; 2. grid division unit; 3. model updating unit; 4. abnormality verification unit; 5. data monitoring unit; 6. monitoring interface unit; 7. feature extraction unit; 8. model training unit; 9. state classification unit; 10. situation element prediction unit; 11. real scene image acquisition unit; 12. emergency handling unit; 13. abnormality sorting unit. DETAILED DESCRIPTION

[0055] Embodiment 1, as Figure 1 As shown, the present invention proposes a remote monitoring and management system for new energy prime mover equipment based on the Internet of Things technology, including a model updating unit 3, an abnormality verification unit 4, a monitoring interface unit 6, a feature extraction unit 7, a model training unit 8, a state classification unit 9, a situation element prediction unit 10, a real scene image acquisition unit 11, an emergency processing unit 12 and an abnormality sorting unit 13, and also includes:

[0056] The scene modeling unit 1 is used to perform 3D modeling on multiple new energy prime movers in the target plant area to obtain multiple 3D models of the new energy prime movers, and integrate the multiple 3D models of the new energy prime movers based on the actual positions of the multiple new energy prime movers in the target plant area to obtain a scene model of the target plant area, and transmit the scene model of the target plant area to the grid division unit 2 and the model updating unit 3;

[0057] A grid division unit 2 receives the target plant scene model transmitted by the scene modeling unit 1, and performs grid division on the target plant scene model to obtain a plurality of grids, and proportionally associates the plurality of grids with the target plant to obtain a position relationship mapping table between the positions of the plurality of new energy prime movers in the target plant and the positions of the plurality of 3D models of the new energy prime movers in the target plant scene model, and transmits the position relationship mapping table to the abnormality verification unit 4;

[0058] The data monitoring unit 5 is used to perform data monitoring on multiple new energy prime mover equipment in the target plant area to obtain monitoring data of the new energy prime mover equipment, wherein the monitoring data includes equipment operating status, environmental data, energy consumption data and operating parameters, and collect vibration data of multiple new energy prime mover equipment in the target plant area through a vibration sensor to obtain a vibration data sequence, and transmit the monitoring data to the monitoring interface unit 6, and transmit the vibration data sequence to the feature extraction unit 7 and the situation factor prediction unit 10.

[0059] In an optional embodiment, the feature extraction unit 7 receives the vibration data sequence transmitted by the data monitoring unit 5, and performs time domain analysis on the vibration data sequence to obtain time domain features, and performs fast Fourier transform on the vibration data sequence to obtain a spectrum diagram, and converts the time domain features into frequency domain features based on the spectrum diagram, and performs coarse-graining processing on the vibration data sequence to obtain vibration data sequences at different scales, and calculates sample entropy values ​​of vibration data sequences at different scales to obtain sample entropy features, and calculates weighted arrangement features of the vibration data sequence at each scale to obtain multi-scale weighted arrangement entropy features, and transmits the time domain features, frequency domain features, sample entropy features and multi-scale weighted arrangement entropy features to the model training unit 8.

[0060] It should be noted that sample entropy is a feature used to measure signal complexity. For the coarse-grained vibration data sequence at each scale, its sample entropy value can be calculated. The higher the sample entropy, the higher the complexity of the sequence. Weighted permutation entropy is a feature based on sorting and arrangement, which is used to describe the complexity and irregularity of the signal. In vibration data sequences at different scales, the weighted permutation entropy features at each scale can be calculated. These features combine the information of vibration data sequences at different scales.

[0061] In an optional embodiment, the model training unit 8 receives the time domain features, frequency domain features, sample entropy features and multi-scale weighted permutation entropy features transmitted by the feature extraction unit 7, and constructs a state classification model based on the support vector machine algorithm, and divides the time domain features, frequency domain features and multi-scale weighted permutation entropy features into a training data set, and optimizes the penalty parameters and kernel function parameters of the state classification model based on the training data set using the gray wolf optimization algorithm to obtain the optimal combination of penalty parameters and kernel function parameters, and applies the optimal combination of penalty parameters and kernel function parameters to the state classification model to obtain a trained state classification model, and transmits the trained state classification model to the state classification unit 9.

[0062] It should be noted that support vector machine is a supervised learning algorithm used for classification and regression tasks. In classification problems, SVM separates data points of different categories by finding an optimal hyperplane. It can handle linear and nonlinear classification problems and effectively classify in high-dimensional space. The gray wolf optimization algorithm is a heuristic optimization algorithm inspired by the social behavior of gray wolf groups. The algorithm simulates the collaboration between leaders, followers and peripheral members in the gray wolf group and is used to find the global optimal solution. In this case, GWO is used to optimize the penalty parameters and kernel function parameters of the SVM model to improve the classification performance of the model. The penalty parameter C controls the balance between the soft interval and the interval boundary of the classifier. A larger C value will cause the model to classify the training data more strictly, which may lead to overfitting, while a smaller C value will be more tolerant of misclassification of some training data, which may lead to underfitting. The kernel function is used to map the input space to a higher-dimensional feature space in order to better segment the data. The kernel function parameters affect the representation of the data in the feature space. Different kernel function parameters can affect the classification effect of the model.

[0063] In an optional embodiment, the situation element prediction unit 10 receives the vibration data sequence transmitted by the data monitoring unit 5, and constructs the vibration data sequence into time series data through a sliding window, and inputs the time series data into a trained situation prediction model. The situation prediction model adopts a gated loop unit, and outputs the vibration data sequence prediction value corresponding to the next preset moment through the trained situation prediction model, and transmits the vibration data sequence prediction value to the state classification unit 9.

[0064] It should be noted that the gated recurrent unit (GRU) is a variant of the recurrent neural network (RNN) that is specifically used to process sequence data. GRU introduces a gate mechanism to control the flow of information, solving the gradient vanishing problem of the standard RNN in long sequences. Its structure enables GRU to capture long-term dependencies while having high computational efficiency. The situation prediction model is a machine learning model used to predict future states or trends. This model is based on GRU and is used to predict future vibration data sequences. This type of model can help predict the operating status of equipment in practical applications.

[0065] In an optional embodiment, the state classification unit 9 receives the trained state classification model transmitted by the model training unit 8 and the vibration data sequence prediction value transmitted by the situation factor prediction unit 10, and extracts the time domain characteristics, frequency domain characteristics, sample entropy characteristics and multi-scale weighted permutation entropy characteristics of the vibration data sequence prediction value, and inputs the time domain characteristics, frequency domain characteristics, sample entropy characteristics and multi-scale weighted permutation entropy characteristics into the trained state classification model to obtain the future device state corresponding to the new energy driving device, and extracts the time domain characteristics, frequency domain characteristics, sample entropy characteristics and multi-scale weighted permutation entropy characteristics of the real-time value of the vibration data sequence, and inputs the time domain characteristics, frequency domain characteristics, sample entropy characteristics and multi-scale weighted permutation entropy characteristics into the trained state classification model to obtain the real-time device state corresponding to the new energy driving device, and transmits the future device state and the real-time device state corresponding to the new energy driving device to the model updating unit 3.

[0066] In an optional embodiment, the model updating unit 3 receives the target plant scene model transmitted by the scene modeling unit 1 and the future equipment status and real-time equipment status corresponding to the new energy driving equipment transmitted by the state classification unit 9, and determines the abnormal new energy driving equipment according to the future equipment status and real-time equipment status corresponding to the new energy driving equipment, and dynamically renders the 3D model of the abnormal new energy driving equipment in the target plant scene model to obtain the target plant dynamic scene model, and sets a dynamic update cycle to perform regular dynamic updates on the target plant dynamic scene model, and transmits the target plant dynamic scene model to the monitoring interface unit 6 and the abnormality verification unit 4.

[0067] In an optional embodiment, the real-scene image acquisition unit 11 is used to set up multiple groups of cameras in the target factory area, acquire the real-scene images of various locations in the target factory area through the multiple groups of cameras, and transmit the real-scene images of various locations in the target factory area to the abnormality verification unit 4.

[0068] In an optional embodiment, the abnormality verification unit 4 receives the position relationship mapping table transmitted by the grid division unit 2, the dynamic scene model of the target plant area transmitted by the model updating unit 3, and the real scene images of various locations in the target plant area transmitted by the real scene image acquisition unit 11, and judges the abnormal position of the target plant area according to the position of the grid where the 3D model of the new energy prime mover equipment with an abnormality in the dynamic scene model of the target plant area is located in the position relationship mapping table, and retrieves the real scene image at the abnormal position, verifies whether an abnormality occurs at the abnormal position through the real scene image, and if an abnormality occurs, generates abnormal information, and transmits the abnormal information to the abnormal sorting unit 13.

[0069] In an optional embodiment, the exception sorting unit 13 receives the exception information transmitted by the exception verification unit 4, and constructs an associated knowledge graph based on the exception information to determine the sequential relationship, concurrent relationship and conflict relationship of the exception information, and calculates the similarity between each exception information through a similarity model, and compares the obtained similarity between each exception information with a preset threshold to determine whether there is a correlation between the exception information. If there is a correlation, the associated exception information will be merged in the associated knowledge graph, and the corresponding exception processing level will be increased according to the number of exception information in each node in the associated knowledge graph, and the content in each node in the associated knowledge graph and the corresponding exception processing level will be transmitted to the emergency processing unit 12.

[0070] It should be noted that the associative knowledge graph is a graph structure used to store and represent the associative relationships between knowledge. It consists of nodes and edges. Nodes represent different entities or concepts, and edges represent the relationships between these entities or concepts.

[0071] In an optional embodiment, the emergency processing unit 12 receives the content and corresponding exception handling level of each node in the associated knowledge graph transmitted by the exception sorting unit 13, and generates a troubleshooting list based on the content and corresponding exception handling level of each node in the associated knowledge graph, and troubleshoots the occurred fault according to the troubleshooting list.

[0072] In an optional embodiment, the monitoring interface unit 6 receives the monitoring data transmitted by the data monitoring unit 5 and the target plant dynamic scene model transmitted by the model updating unit 3, and displays the target plant dynamic scene model on the monitoring interface of the monitoring intelligent terminal, and hides the monitoring data in the corresponding new energy prime mover equipment 3D model in the target plant dynamic scene model. When the user clicks on the corresponding new energy prime mover equipment 3D model in the target plant dynamic scene model, the monitoring data pops up in the form of a pop-up window.

[0073] Embodiment 2, as Figure 2-Figure 3As shown, the remote monitoring and management system of new energy prime mover equipment based on the Internet of Things technology proposed by the present invention, compared with the first embodiment, this embodiment also includes: optimizing the penalty parameters and kernel function parameters of the state classification model based on the training data set using the gray wolf optimization algorithm to obtain the optimal combination of the penalty parameters and kernel function parameters, including the following steps:

[0074] A1. Randomly initialize individuals of the gray wolf population. The positions of the individuals of the gray wolf population represent the parameter combination of the penalty parameter and the kernel function parameter of the state classification model.

[0075] A2. Update the position of the individual gray wolf population to obtain the new position of the individual gray wolf population. The new position update formula is as follows:

[0076] ;

[0077] in, represents the position at the t+1th iteration, Indicates the location of the prey, represents the first random parameter, and , represents the nonlinear convergence factor, express The first random number in the interval, Indicates the distance between the prey and the wolf pack, , represents the position at the tth iteration, represents the second random parameter, and , express The second random number in the interval;

[0078] A3. Calculate the new fitness of the gray wolf population individual in the new position and compare it with the optimal fitness of the previous iteration. If the new fitness is greater than the optimal fitness, replace the new fitness with the optimal fitness and retain the new position of the gray wolf population individual. The fitness calculation formula is as follows:

[0079] ;

[0080] in, represents the average accuracy of K-fold cross validation, represents the accuracy calculated at the k-th fold;

[0081] A4. If the number of iterations is greater than a preset iteration threshold, the optimization is terminated to obtain the optimal combination of penalty parameters and kernel function parameters. Otherwise, steps A2-A3 are repeated.

[0082] In an optional embodiment, the vibration data sequence is coarse-grained to obtain vibration data sequences at different scales, and sample entropy values ​​of the vibration data sequences at different scales are calculated, including the following steps:

[0083] B1. The vibration data sequence is coarse-grained by a coarse-grained formula to obtain vibration data sequences at different scales. The coarse-grained formula is as follows:

[0084] ;

[0085] in, represents the vibration data sequence at scale i, represents the scale factor, represents the original vibration data sequence, that is, the vibration data sequence before coarse-graining processing;

[0086] B2. Constructing a sequence vector based on the vibration data sequences at different scales, and based on the distance between the sequence vector and other sequence vectors, the sequence vector is as follows:

[0087] ;

[0088] in, represents the sequence vector at scale i, represents the embedding dimension;

[0089] The distance formula is as follows:

[0090] ;

[0091] in, Represents the sequence vector at scale i with any other sequence vector at any scale j The distance between express Vibration data series at scale, express Vibration data series at scale;

[0092] B3. Calculate the ratio of the distance between the sequence vector and other sequence vectors less than the preset distance threshold to the number of vectors. The ratio formula is as follows:

[0093] ;

[0094] in, Represents the ratio, Indicates the preset distance threshold, Represents the number of data in the vibration data sequence at different scales. count() represents a counting function used to calculate the number of data that meet the conditions. The number of sequence vectors;

[0095] B4. Calculate the average value of the ratio of the distance between the sequence vector and other sequence vectors that is less than a preset distance threshold to the number of vectors. The average value formula is as follows:

[0096] ;

[0097] in, represents the average value of the mth iteration;

[0098] B5. Based on the average value of the ratio of the distance between the sequence vector and other sequence vectors being less than a preset distance threshold and the number of vectors, the sample entropy values ​​of the vibration data sequences at different scales are calculated. The sample entropy value formula is as follows:

[0099] ;

[0100] in, represents the sample entropy value, represents the average value of the m+1th iteration, and ln() represents the natural logarithm function.

[0101] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A remote monitoring and management system for new energy prime mover equipment based on the Internet of Things technology, comprising a model updating unit (3), an abnormality verification unit (4), a monitoring interface unit (6), a feature extraction unit (7), a model training unit (8), a state classification unit (9), a situation element prediction unit (10), a real-scene image acquisition unit (11), an emergency processing unit (12) and an abnormality sorting unit (13), characterized in that: A scene modeling unit (1), the scene modeling unit (1) being used to perform 3D modeling on a plurality of new energy prime movers in a target plant area to obtain a plurality of 3D models of the new energy prime movers, and to integrate the plurality of 3D models of the new energy prime movers based on actual positions of the plurality of new energy prime movers in the target plant area to obtain a scene model of the target plant area, and to transmit the scene model of the target plant area to a grid division unit (2) and a model updating unit (3); A grid division unit (2), the grid division unit (2) receiving the target plant scene model transmitted by the scene modeling unit (1), and gridding the target plant scene model to obtain a plurality of grids, and proportionally associating the plurality of grids with the target plant to obtain a position relationship mapping table between the positions of the plurality of new energy prime movers in the target plant and the positions of the 3D models of the plurality of new energy prime movers in the target plant scene model, and transmitting the position relationship mapping table to the abnormality verification unit (4); A data monitoring unit (5), the data monitoring unit (5) being used to perform data monitoring on a plurality of new energy prime movers in the target plant area to obtain monitoring data of the new energy prime movers, the monitoring data including equipment operating status, environmental data, energy consumption data and operating parameters, and to collect vibration data of the plurality of new energy prime movers in the target plant area through a vibration sensor to obtain a vibration data sequence, and to transmit the monitoring data to a monitoring interface unit (6), and to transmit the vibration data sequence to a feature extraction unit (7) and a situation element prediction unit (10); The feature extraction unit (7) receives the vibration data sequence transmitted by the data monitoring unit (5), performs time domain analysis on the vibration data sequence to obtain time domain features, performs fast Fourier transform on the vibration data sequence to obtain a frequency spectrum, converts the time domain features into frequency domain features based on the frequency spectrum, performs coarse-graining processing on the vibration data sequence to obtain vibration data sequences at different scales, calculates sample entropy values ​​of the vibration data sequences at different scales to obtain sample entropy features, calculates weighted permutation features of the vibration data sequence at each scale to obtain multi-scale weighted permutation entropy features, and transmits the time domain features, the frequency domain features, the sample entropy features, and the multi-scale weighted permutation entropy features to the model training unit (8); The vibration data sequence is subjected to coarse-graining processing to obtain vibration data sequences at different scales, and sample entropy values ​​of the vibration data sequences at different scales are calculated, including the following steps: B1. The vibration data sequence is coarse-grained by a coarse-grained formula to obtain vibration data sequences at different scales. The coarse-grained formula is as follows: ； in, represents the vibration data sequence at scale i, represents the scale factor, represents the original vibration data sequence, that is, the vibration data sequence before coarse-graining processing; B2. Constructing a sequence vector based on the vibration data sequences at different scales, and based on the distance between the sequence vector and other sequence vectors, the sequence vector is as follows: ； in, represents the sequence vector at scale i, represents the embedding dimension; The distance formula is as follows: ； in, Represents the sequence vector at scale i With any other sequence vector at any scale j The distance between express Vibration data series at scale, express Vibration data series at scale; B3. Calculate the ratio of the distance between the sequence vector and other sequence vectors less than the preset distance threshold to the number of vectors. The ratio formula is as follows: ； in, Represents the ratio, Indicates the preset distance threshold, Represents the number of data in the vibration data sequence at different scales. count() represents a counting function used to calculate the number of data that meet the conditions. The number of sequence vectors; B4. Calculate the average value of the ratio of the distance between the sequence vector and other sequence vectors that is less than a preset distance threshold to the number of vectors. The average value formula is as follows: ； in, represents the average value of the mth iteration; B5. Based on the average value of the ratio of the distance between the sequence vector and other sequence vectors being less than a preset distance threshold and the number of vectors, the sample entropy values ​​of the vibration data sequences at different scales are calculated. The sample entropy value formula is as follows: ； in, Represents the sample entropy value; En in stands for "Entropy", i.e. sample entropy; represents the average value of the m+1th iteration; ln() represents the natural logarithm function; The model training unit (8) receives the time domain features, the frequency domain features, the sample entropy features and the multi-scale weighted permutation entropy features transmitted by the feature extraction unit (7), and constructs a state classification model based on a support vector machine algorithm, and divides the time domain features, the frequency domain features and the multi-scale weighted permutation entropy features into a training data set, and optimizes the penalty parameters and kernel function parameters of the state classification model using a gray wolf optimization algorithm based on the training data set to obtain an optimal combination of the penalty parameters and the kernel function parameters, and applies the optimal combination of the penalty parameters and the kernel function parameters to the state classification model to obtain a trained state classification model, and transmits the trained state classification model to the state classification unit (9); Based on the training data set, the penalty parameter and the kernel function parameter of the state classification model are optimized by using the gray wolf optimization algorithm to obtain the optimal combination of the penalty parameter and the kernel function parameter, including the following steps: A1. Randomly initialize individuals of the gray wolf population, where the positions of the individuals of the gray wolf population represent a parameter combination of a penalty parameter and a kernel function parameter of the state classification model; A2. Update the position of the individual of the gray wolf population to obtain a new position of the individual of the gray wolf population. The new position update formula is as follows: ； in, represents the position at the t+1th iteration, Indicates the location of the prey. represents the first random parameter, and , represents the nonlinear convergence factor, r1 represents the first random number in the interval [0,1], D represents the distance between the prey and the wolf pack, , represents the position at the tth iteration, C represents the second random parameter, and , r2 represents the second random number in the interval [0,1]; A3. Calculate the new fitness of the gray wolf population individual at the new position and compare it with the optimal fitness of the previous iteration. If the new fitness is greater than the optimal fitness, replace the new fitness with the optimal fitness and retain the new position of the gray wolf population individual. The fitness calculation formula is as follows: ； in, represents the average accuracy of K-fold cross validation, represents the accuracy calculated at the k-th fold; A4. If the number of iterations is greater than a preset iteration number threshold, the optimization is terminated to obtain the optimal combination of penalty parameters and kernel function parameters. Otherwise, steps A2-A3 are repeated. The situation element prediction unit (10) receives the vibration data sequence transmitted by the data monitoring unit (5), constructs the vibration data sequence into time series data through a sliding window, and inputs the time series data into a trained situation prediction model, the situation prediction model adopts a gated recurrent unit, outputs a vibration data sequence prediction value corresponding to the vibration data sequence at the next preset time through the trained situation prediction model, and transmits the vibration data sequence prediction value to the state classification unit (9); The state classification unit (9) receives the trained state classification model transmitted by the model training unit (8) and the vibration data sequence prediction value transmitted by the situation factor prediction unit (10), and extracts the time domain features, the frequency domain features, the sample entropy features and the multi-scale weighted permutation entropy features of the vibration data sequence prediction value, and inputs the time domain features, the frequency domain features, the sample entropy features and the multi-scale weighted permutation entropy features of the vibration data sequence prediction value into the trained state classification model to obtain the future device state corresponding to the new energy driving device, and extracts the time domain features, the frequency domain features, the sample entropy features and the multi-scale weighted permutation entropy features of the real-time value of the vibration data sequence, and inputs the time domain features, the frequency domain features, the sample entropy features and the multi-scale weighted permutation entropy features of the real-time value of the vibration data sequence into the trained state classification model to obtain the real-time device state corresponding to the new energy driving device, and transmits the future device state and the real-time device state corresponding to the new energy driving device to the model updating unit (3); The model updating unit (3) receives the target plant scene model transmitted by the scene modeling unit (1) and the future device state and real-time device state corresponding to the new energy prime mover device transmitted by the state classification unit (9), determines the new energy prime mover device with abnormality according to the future device state and real-time device state corresponding to the new energy prime mover device, dynamically renders the 3D model of the new energy prime mover device with abnormality in the target plant scene model to obtain a target plant dynamic scene model, sets a dynamic update cycle to dynamically update the target plant dynamic scene model at regular intervals, and transmits the target plant dynamic scene model to the monitoring interface unit (6) and the abnormality verification unit (4); The real scene image acquisition unit (11) is used to set up multiple groups of cameras in the target factory area, acquire the real scene images of various locations in the target factory area through the multiple groups of cameras, and transmit the real scene images of various locations in the target factory area to the abnormality verification unit (4); The abnormality verification unit (4) receives the position relationship mapping table transmitted by the grid division unit (2), the target plant area dynamic scene model transmitted by the model updating unit (3), and the real scene images of various locations in the target plant area transmitted by the real scene image acquisition unit (11), and determines the abnormal location of the target plant area according to the position of the grid where the 3D model of the new energy prime mover equipment with an abnormality in the target plant area dynamic scene model is located in the position relationship mapping table, and retrieves the real scene image at the abnormal location, verifies whether an abnormality occurs at the abnormal location through the real scene image, and generates abnormality information if an abnormality occurs, and transmits the abnormality information to the abnormality sorting unit (13); The exception sorting unit (13) receives the exception information transmitted by the exception verification unit (4), and constructs an associated knowledge graph based on the exception information to determine the order relationship, concurrency relationship and conflict relationship of the exception information, and calculates the similarity between each of the exception information through a similarity model, and compares the obtained similarity between each of the exception information with a preset threshold value to determine whether there is a correlation between the exception information. If there is a correlation, the associated exception information is merged in the associated knowledge graph, and the corresponding exception processing level is increased according to the number of exception information in each node in the associated knowledge graph, and the content in each node in the associated knowledge graph and the corresponding exception processing level are transmitted to the emergency processing unit (12); The emergency processing unit (12) receives the content of each node in the associated knowledge graph and the corresponding exception processing level transmitted by the exception sorting unit (13), generates a troubleshooting list according to the content of each node in the associated knowledge graph and the corresponding exception processing level, and troubleshoots the occurred fault according to the troubleshooting list; The monitoring interface unit (6) receives the monitoring data transmitted by the data monitoring unit (5) and the target plant area dynamic scene model transmitted by the model updating unit (3), and displays the target plant area dynamic scene model on the monitoring interface of the monitoring intelligent terminal, and hides the monitoring data in the corresponding 3D model of the new energy prime mover in the target plant area dynamic scene model. When the user clicks on the corresponding 3D model of the new energy prime mover in the target plant area dynamic scene model, the monitoring data pops up in the form of a pop-up window.

Images