Fault diagnosis and maintenance method and system for smart street lamps

Abstract

The present invention provides a fault diagnosis and maintenance method and system for a smart street lamp, the method comprising: collecting data on the operating parameters of each component node of the smart street lamp, performing noise reduction processing using wavelet transform, and generating structured data. By combining principal component analysis and sliding time window technology with Mahalanobis distance calculation, feature extraction and spatiotemporal correlation analysis are performed on the structured data to obtain key features and spatiotemporal correlation information between nodes. The above information is input into a spatiotemporal hybrid model to generate fault detection results, and a machine learning algorithm is used to locate the faulty components, predict their performance degradation trends, and generate targeted maintenance strategies. The present invention realizes intelligent management of the entire process from fault detection to maintenance, effectively improving the operating stability and management efficiency of smart street lamps.

Description

Technical Field

[0001] The present invention relates to the field of fault diagnosis technology, and in particular to a fault diagnosis and maintenance method and system for a smart street lamp. Background Art

[0002] With the rapid development of smart city construction, smart street lights, as a multifunctional infrastructure integrating lighting, monitoring, communication and environmental detection, have been widely used in urban management. However, since smart street lights are usually distributed in a wide range of urban areas and are exposed to various complex environments for a long time, their internal component nodes (such as power modules, control chips and sensors) are easily affected by the external environment or their own aging problems, resulting in performance degradation or failure. If the fault cannot be detected and located in time, it will not only affect the normal operation of the street lights, but may also cause additional operation and maintenance costs and reduce urban management efficiency.

[0003] At present, traditional street lamp fault detection methods mainly rely on regular manual inspections or simple operating parameter monitoring. This method has problems such as high labor intensity, limited detection coverage, and lack of real-time performance, which makes it difficult to meet the needs of smart street lamps for efficient fault diagnosis. In addition, traditional methods are usually based on single parameters or static analysis, which cannot fully explore the spatiotemporal correlation between the nodes of various components of street lamps, and it is also difficult to predict potential faults or component performance degradation trends. This limitation will lead to insufficient fault location accuracy, which in turn affects the scientificity and effectiveness of maintenance strategies. Summary of the invention

[0004] The main purpose of the present invention is to solve the technical problems of the existing intelligent street lamp fault detection method, which relies on manual inspection, lacks real-time performance, has low fault location accuracy, and has unscientific maintenance strategies;

[0005] A first aspect of the present invention provides a fault diagnosis and maintenance method for a smart street lamp, the fault diagnosis and maintenance method for a smart street lamp comprising:

[0006] Collect data on the operating parameters of each component node in the smart street lamp, and use wavelet transform to perform noise reduction on the operating parameters to obtain structured data after noise reduction;

[0007] The principal component analysis method is used to extract features from the structured data, and the time-space correlation analysis of each component node is performed based on the sliding time window technology and Mahalanobis distance calculation to obtain the key features of the smart street lamp and the time-space correlation information between the component nodes;

[0008] Inputting the key features of the smart street lamp and the spatiotemporal correlation information between the component nodes into a preset spatiotemporal hybrid model to obtain a fault detection result of the smart street lamp;

[0009] A machine learning algorithm is used to locate the fault according to the fault detection result, and the performance degradation trend of the component node where the fault is located is predicted, and a corresponding maintenance strategy is generated based on the fault detection result and the performance degradation trend of the component node where the fault is located.

[0010] Optionally, in a first implementation of the first aspect of the present invention, the principal component analysis method is used to extract features from the structured data, and a spatiotemporal correlation analysis is performed on each component node based on a sliding time window technique and Mahalanobis distance calculation to obtain key features of the smart street lamp and spatiotemporal correlation information between component nodes, including:

[0011] Constructing a covariance matrix for the structured data, and performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding cumulative contribution rates;

[0012] Selecting a principal component according to the cumulative contribution rate of the eigenvalue, and performing dimension reduction projection on the structured data corresponding to the principal component to obtain key features;

[0013] Applying the sliding time window technology to the key features, analyzing the feature data at each time point through a preset window size and step size, and obtaining local features in the time dimension;

[0014] Calculate the similarity between the nodes of each component according to the eigenvalue of the key feature and the inverse matrix of the covariance matrix using the Mahalanobis distance;

[0015] A spatial correlation matrix is ​​constructed according to the similarity, and the spatiotemporal correlation information is generated by combining the local features in the time dimension.

[0016] Optionally, in a second implementation of the first aspect of the present invention, constructing a spatial correlation matrix according to the similarity and generating spatiotemporal correlation information in combination with local features in the time dimension includes:

[0017] Constructing a spatial correlation matrix according to the similarity, and processing the spatial correlation matrix using a minimum spanning tree algorithm in graph theory to construct a topological structure between component nodes;

[0018] Applying a web page ranking algorithm to the topological structure to calculate the importance index of each component node;

[0019] Perform multi-dimensional multiplication operation on the local features in the time dimension and the importance index of the component node to obtain a multi-dimensional data structure of spatiotemporal features;

[0020] The spatiotemporal feature multidimensional data structure is subjected to tensor decomposition to obtain spatiotemporal correlation information.

[0021] Optionally, in a third implementation of the first aspect of the present invention, inputting the key features of the smart street lamp and the spatiotemporal correlation information between the component nodes into a preset spatiotemporal hybrid model to obtain a fault detection result of the smart street lamp includes:

[0022] Input the key features of the smart street lamp and the spatiotemporal correlation information between nodes into a preset spatiotemporal hybrid model, obtain the historical data and current data of the input key features through the spatiotemporal hybrid model, and calculate the accumulation and statistics of the key features based on the historical data and the current data;

[0023] The Mahalanobis distance statistic constructed based on the spatiotemporal correlation information is used to calculate the Mahalanobis distance between each component node and the normal state reference point;

[0024] The weighted fusion method is used to combine the cumulative sum statistic and the Mahalanobis distance statistic to obtain the comprehensive fault index. The kernel density estimation method is used to estimate the probability density of the comprehensive fault index in the historical data to obtain the fault threshold function.

[0025] The comprehensive fault index calculated in real time is compared with the fault threshold function, and the fault detection result of the smart street lamp is determined according to the comparison result.

[0026] Optionally, in a fourth implementation of the first aspect of the present invention, the key features of the smart street lamp and the spatiotemporal correlation information between nodes are input into a preset spatiotemporal hybrid model, historical data and current data of the input key features are obtained through the spatiotemporal hybrid model, and the cumulative and statistical values ​​of the key features are calculated based on the historical data and the current data, including:

[0027] Input the key features of the smart street lamp and the spatiotemporal correlation information between nodes into a preset spatiotemporal hybrid model, and obtain the historical data and current data of the input key features through the spatiotemporal hybrid model;

[0028] Calculate the mean and standard deviation of the historical data, and set the mean as a reference value for the cumulative sum algorithm;

[0029] The cumulative sum algorithm is applied to the current data of the key feature in time series, the deviation of the current data at each time point from the reference value is calculated, and the deviation is added to the accumulated deviation at the previous time point to obtain the cumulative sum statistic of the key feature.

[0030] Optionally, in a fifth implementation of the first aspect of the present invention, comparing the comprehensive fault index calculated in real time with the fault threshold function, and determining the fault detection result of the smart street lamp according to the comparison result includes:

[0031] Compare the comprehensive fault index calculated in real time with the fault threshold function obtained in advance, and if the comparison result is that the comprehensive fault index exceeds the fault threshold function, mark the corresponding time point and related components as potential fault points;

[0032] Perform time series analysis on the marked potential fault points to determine whether they are persistent anomalies. If so, perform pattern matching on persistent anomalies in combination with the historical fault pattern library to obtain fault type information;

[0033] The potential fault point, the persistent abnormality judgment result and the fault type information are combined into a fault detection result of the smart street lamp fault.

[0034] Optionally, in a sixth implementation manner of the first aspect of the present invention, the using of a machine learning algorithm to locate the fault according to the fault detection result, and predicting the performance degradation trend of the component node where the fault is located, and generating a corresponding maintenance strategy based on the fault detection result and the performance degradation trend of the component node where the fault is located includes:

[0035] Scan the smart street lamp using a pattern matching algorithm according to the fault type information in the fault detection result to determine the component node where the fault is located;

[0036] Applying a long short-term memory network to predict the performance degradation trend according to the operating parameters of the component node of the fault location to obtain the corresponding performance degradation trend;

[0037] Based on the fault detection results and performance degradation trends, a multi-objective optimization model is constructed, and a non-dominated sorting genetic algorithm is used to solve the multi-objective optimization model to obtain an optimal solution set, and a maintenance strategy for the smart street lamp is generated based on the optimal solution set.

[0038] A second aspect of the present invention provides a fault diagnosis and maintenance system for a smart street lamp, the fault diagnosis and maintenance system for a smart street lamp comprising:

[0039] A data denoising module is used to collect data on the operating parameters of each component node in the smart street lamp, and use wavelet transform to perform denoising on the operating parameters to obtain denoised structured data;

[0040] A feature analysis module is used to extract features from the structured data using a principal component analysis method, and to perform spatiotemporal correlation analysis on each component node based on a sliding time window technique and Mahalanobis distance calculation, to obtain key features of the smart street lamp and spatiotemporal correlation information between component nodes;

[0041] A fault detection module, used to input the key features of the smart street lamp and the time-space correlation information between the component nodes into a preset time-space hybrid model to obtain a fault detection result of the smart street lamp;

[0042] A maintenance strategy module is used to use a machine learning algorithm to locate the fault according to the fault detection result, and predict the performance degradation trend of the component node where the fault is located, and generate a corresponding maintenance strategy based on the fault detection result and the performance degradation trend of the component node where the fault is located.

[0043] The fault diagnosis and maintenance method and system of the above-mentioned smart street lamp collects data on the operating parameters of each component node of the smart street lamp, performs noise reduction processing using wavelet transform, and generates structured data. Through principal component analysis and sliding time window technology combined with Mahalanobis distance calculation, feature extraction and spatiotemporal correlation analysis are performed on the structured data to obtain key features and spatiotemporal correlation information between nodes. The above information is input into the spatiotemporal hybrid model to generate fault detection results, and the machine learning algorithm is used to locate the faulty components, predict their performance degradation trends, and generate targeted maintenance strategies. The present invention realizes intelligent management of the entire process from fault detection to maintenance, effectively improving the operating stability and management efficiency of smart street lamps.

[0044] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0045] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a schematic diagram of a first embodiment of a fault diagnosis and maintenance method for a smart street lamp according to an embodiment of the present invention;

[0047] Figure 2 Schematic diagram of an embodiment of a fault diagnosis and maintenance system for a smart street lamp in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0050] To facilitate understanding of this embodiment, a fault diagnosis and maintenance method for a smart street lamp disclosed in an embodiment of the present invention is first described in detail. Figure 1 As shown, the method comprises the following steps:

[0051] 101. Collect data on the operating parameters of each component node in the smart street lamp, and use wavelet transform to perform noise reduction on the operating parameters to obtain structured data after noise reduction;

[0052] In one embodiment of the present invention, data collection of operating parameters of each component node in a smart street lamp involves the collaborative work of multiple sensors and data acquisition devices. Each street lamp node is usually equipped with a variety of sensors, such as current sensors, voltage sensors, light sensors, and temperature sensors. These sensors collect data at a preset frequency (which may be every second, every minute, or every hour). The collected data is transmitted to the central data processing system through the built-in communication module of the street lamp using wireless network technology (such as 4G, 5G, or LoRa). After the data arrives at the central system, preliminary data cleaning is first performed to remove obvious outliers and missing data. Then, the system applies wavelet transform to the cleaned data for noise reduction. The process of wavelet transform includes selecting a suitable wavelet basis function (such as Daubechies wavelet or Haar wavelet), performing multi-scale decomposition on the signal, and obtaining wavelet coefficients at different frequencies. Then, these coefficients are processed by setting thresholds, usually using soft threshold or hard threshold methods. The processed coefficients are then reconstructed by inverse wavelet transform to obtain noise-reduced data. Finally, the system organizes the denoised data into a structured format, usually a multidimensional array or table, where rows represent different time points and columns represent different parameters and component nodes. This structured data facilitates subsequent analysis and processing.

[0053] 102. Use principal component analysis to extract features from structured data, and perform spatiotemporal correlation analysis on each component node based on sliding time window technology and Mahalanobis distance calculation to obtain key features of smart street lamps and spatiotemporal correlation information between component nodes;

[0054] In one embodiment of the present invention, the principal component analysis method is used to extract features from the structured data, and the spatiotemporal correlation analysis of each component node is performed based on the sliding time window technology and Mahalanobis distance calculation to obtain the key features of the smart street lamp and the spatiotemporal correlation information between the component nodes, including: constructing a covariance matrix for the structured data, and performing eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding cumulative contribution rates; selecting principal components according to the cumulative contribution rates of the eigenvalues, and performing dimensionality reduction projection on the structured data corresponding to the principal components to obtain key features; applying the sliding time window technology to the key features, analyzing the feature data at each time point through a preset window size and step size, and obtaining local features in the time dimension; using the Mahalanobis distance to calculate the similarity between each component node according to the eigenvalues ​​of the key features and the inverse matrix of the covariance matrix; constructing a spatial correlation matrix according to the similarity, and generating spatiotemporal correlation information in combination with the local features in the time dimension.

[0055] Specifically, after collecting data on the operating parameters of each component node in the smart street light system, the structured data obtained is first used to construct a covariance matrix. During the construction process, the covariance is calculated for each pair of variables to form a square matrix whose size depends on the number of variables. Specifically, for each pair of variables, the average value of the product of their deviations from their respective mean values ​​is calculated, and this value is their covariance. The diagonal elements of this matrix are the variances of each variable. Next, the covariance matrix is ​​subjected to eigenvalue decomposition. This step obtains the eigenvalue by solving the characteristic equation. The specific method is to subtract the product of the eigenvalue and the unit matrix from the covariance matrix, and then set its determinant to zero. Solving this equation obtains a series of eigenvalues. For each eigenvalue, a linear equation system is solved to obtain the eigenvector. The eigenvalue represents the variance of the data in the direction of the corresponding eigenvector. Then the cumulative contribution rate is calculated, that is, the sum of the first k eigenvalues ​​divided by the sum of all eigenvalues. This ratio represents the proportion of the total variance explained by the first k principal components. By gradually increasing k until the cumulative contribution rate reaches the preset threshold, the number of principal components to be retained is determined.

[0056] Specifically, according to the eigenvalues ​​and cumulative contribution rates obtained in the previous step, an appropriate number of principal components are selected. The specific operation is to sort the eigenvalues ​​by size and accumulate them from the largest one until the cumulative contribution rate reaches a preset threshold. For example, if the cumulative contribution rate of the first three eigenvalues ​​reaches a predetermined percentage, the three corresponding eigenvectors are selected as principal components. After the principal components are selected, a projection matrix is ​​constructed, whose columns are the selected eigenvectors. Then, the original structured data is multiplied by this projection matrix to obtain the reduced-dimensional data. This process can be understood as projecting the original data into a new feature space. In the new data matrix obtained, each column is a new feature, called the principal component score. These scores are linear combinations of the original features. For example, the first principal component may be the weighted sum of all the original features, with the weights determined by the elements of the corresponding eigenvectors. These new features are the key features of the smart street light system, which retain the most important change patterns in the original data while reducing the dimensionality of the data. Each principal component has a specific physical meaning and needs to be interpreted according to the specific situation.

[0057] Specifically, the sliding time window technique is applied to the obtained key features, and the specific implementation is as follows: First, determine the window size and step size. The window size can be hours, days, or weeks, depending on the frequency of data collection and analysis requirements. For example, for a system that collects data once a minute, sixty may be selected as the window size, representing one hour, and the step size may be selected as ten, indicating that the window is moved every ten minutes. Then, starting from the starting point of the time series, the data points of the previous window size are taken as the first window. Statistical features are calculated for these data points, including mean, variance, skewness, and kurtosis. The mean is the arithmetic average of all data points, the variance measures the degree of dispersion of the data, the skewness reflects the asymmetry of the distribution, and the kurtosis indicates the degree of sharpness of the distribution. These statistics constitute the local feature vector of the time window. Next, move the window forward by one step of data points, repeat the above calculation, and obtain the next local feature vector. This process continues until the window reaches the end of the time series. Finally, a new matrix is ​​obtained, the number of rows of which is equal to the number of window moves plus one, and the number of columns is equal to the number of statistics calculated. Each row of this matrix represents the local features of a time window, reflecting the behavioral characteristics of the system during that time period.

[0058] Specifically, the specific steps of using Mahalanobis distance to calculate the similarity between each component node are as follows: First, for each component node, collect its data points in the key feature space. Assuming that there are multiple nodes, each node has multiple observations, and each observation contains multiple key features, then the data of each node can be represented as a matrix. Then, calculate the mean vector and covariance matrix of each node data. The covariance matrix reflects the correlation between each feature. Next, for any two nodes, calculate the Mahalanobis distance between them. The calculation of Mahalanobis distance takes into account the covariance structure of the data, not only comparing the difference in mean values, but also considering the correlation between features. Specifically, it first calculates the difference between the mean vectors of the two nodes, and then multiplies this difference by the inverse of the covariance matrix, and then multiplies it by the transpose of the difference vector. If the covariance matrix is ​​close to singular, some mathematical tricks can be used to deal with it, such as adding a small regularization term. Repeat this calculation for all node pairs to obtain a distance matrix. Finally, to convert the distance into similarity, a mathematical function can be used so that the smaller the distance, the greater the similarity. The similarity matrix obtained in this way is a symmetric matrix. Each element represents the similarity between the corresponding two nodes. The larger the value, the more similar they are.

[0059] Specifically, the process of constructing a spatial correlation matrix based on the calculated similarity and combining it with the local features in the time dimension to generate spatiotemporal correlation information is as follows: First, the similarity matrix is ​​standardized so that the sum of each row is one, and the spatial correlation matrix is ​​obtained. This can be achieved by dividing each element by the sum of its row. Next, a spatial weight matrix is ​​constructed, which reflects the degree of influence of spatial relationships on each node. Then, the local feature matrix in the time dimension is combined with the spatial weight matrix. One method is to use a spatial lag model, that is, multiplying the spatial weight matrix with the local feature matrix to obtain a new feature matrix that takes into account spatial correlation. Another method is to use a spatiotemporal autoregressive model, which considers both spatial and temporal dependencies. By estimating the parameters of this model, a representation that comprehensively considers temporal and spatial correlations can be obtained. Finally, tensor decomposition techniques such as Tucker decomposition can be used to decompose the obtained three-dimensional data (node, time, feature) into core tensors and factor matrices. This decomposition can capture high-order correlations in the data and provide a compact representation of spatiotemporal correlations. The resulting spatiotemporal correlation information contains the complex interaction patterns of the nodes in the system in time and space, which can be used for subsequent fault diagnosis and predictive maintenance.

[0060] Furthermore, the method of constructing a spatial correlation matrix based on the similarity and combining the local features on the time dimension to generate the spatiotemporal correlation information includes: constructing a spatial correlation matrix based on the similarity, and processing the spatial correlation matrix using a minimum spanning tree algorithm in graph theory to construct a topological structure between component nodes; applying a web page sorting algorithm to the topological structure to calculate the importance index of each component node; performing multi-dimensional multiplication operations on the local features on the time dimension and the importance index of the component nodes to obtain a multi-dimensional data structure of spatiotemporal features; and performing tensor decomposition on the multi-dimensional data structure of spatiotemporal features to obtain the spatiotemporal correlation information.

[0061] Specifically, the process of constructing a spatial correlation matrix based on the similarity first involves standardizing the similarity values. The standardization process usually uses a minimum-maximum scaling or Z-score standardization method. Minimum-maximum scaling maps all values ​​between 0 and 1, while Z-score standardization converts the data into a distribution with a mean of 0 and a standard deviation of 1. Which method to choose depends on the specific distribution of the data and the analysis requirements. After standardization, these values ​​are used to fill the spatial correlation matrix. The matrix is ​​a symmetric matrix with a size of n×n, where n is the number of component nodes. Each element (i, j) in the matrix represents the strength of spatial association between node i and node j. The diagonal elements are usually set to 1, indicating that the node is completely correlated with itself. In addition, a threshold can be set to set the correlation values ​​below the threshold to 0 to reduce noise and computational complexity. The selection of this threshold requires a trade-off between information retention and complexity reduction.

[0062] Specifically, the minimum spanning tree algorithm in graph theory is used to process the spatial correlation matrix and construct the topological structure between component nodes. In this step, the component nodes are regarded as vertices in the graph, and the inverse of the similarity is used as the weight of the edge (because the minimum spanning tree algorithm usually finds the tree with the smallest weight sum). Commonly used algorithms include Kruskal algorithm and Prim algorithm. Kruskal algorithm first sorts all edges by weight, and then starts with the edge with the smallest weight, gradually adds edges, and checks whether a loop is formed. If adding an edge will form a loop, the edge is skipped. Prim algorithm starts from an arbitrary node, selects a minimum weight edge connected to the current tree each time, and adds the new node connected to it to the tree. This process continues until all nodes are connected. The choice of minimum spanning tree algorithm depends on the density of the graph and the convenience of implementation. The constructed minimum spanning tree reflects the strongest connection relationship between component nodes and provides a simplified but information-rich topological structure. This structure helps to understand the key connections and potential fault propagation paths in the system. In addition, the minimum spanning tree can also be used to identify key nodes or links in the system, which are usually bridge points connecting different subsystems.

[0063] Specifically, a web page ranking algorithm is applied to the topological structure to calculate the importance index of each component node. The most commonly used web page ranking algorithm is PageRank, which was originally developed by the founder of Google for web page ranking, but is now widely used in various network analyses. The core idea of ​​the PageRank algorithm is that the importance of a node depends not only on the number of nodes directly connected to it, but also on the importance of these connected nodes. The algorithm simulates a random walk process and calculates the probability that a random walker will eventually stay on each node. In specific implementation, the importance score of each node is first initialized, usually uniformly distributed. Then iterative calculation is performed, and in each iteration, the new score of the node is obtained by weighted summing the scores of all its incoming edge nodes according to the weight of the edge. This process is repeated many times until the scores of all nodes tend to be stable, that is, the convergence condition is reached. In practical applications, a damping factor (usually 0.85) is usually set to simulate the probability of a random walker randomly jumping to any node, which can ensure the convergence of the algorithm. In addition, the problem of hanging nodes (nodes without outgoing edges) needs to be dealt with. The usual practice is to evenly distribute the weights of these nodes to all other nodes.

[0064] Specifically, the local features on the time dimension are multi-dimensionally multiplied with the importance index of the component node to obtain a multi-dimensional data structure of spatiotemporal features. The purpose of this step is to combine the time features and spatial importance to form a comprehensive spatiotemporal representation. The specific operations are as follows: First, the local features on the time dimension are organized into a matrix T, where each row represents a time point and each column represents a feature. Then, the node importance index obtained by the PageRank algorithm is organized into a vector P. Next, for each time point t and each node n, its spatiotemporal feature vector S(t,n) = T(t) * P(n) is calculated, where T(t) is the tth row of the time feature matrix and P(n) is the importance index of node n. This operation can be understood as weighting the time features with the importance of the node. The result is a three-dimensional tensor whose dimensions are time, node and feature. This tensor captures the dynamic characteristics of the system over time while taking into account the importance of the node in the network. In actual implementation, computational efficiency needs to be considered, especially when the amount of data is large. Parallel computing or distributed computing technology can be used to accelerate this process. In addition, the results need to be standardized to ensure that data in different dimensions are comparable.

[0065] Specifically, tensor decomposition is performed on the multidimensional data structure of the spatiotemporal features to obtain spatiotemporal correlation information. Tensor decomposition is a technique for decomposing high-dimensional data into a series of low-dimensional factors. Common methods include CANDECOMP / PARAFAC (CP) decomposition and Tucker decomposition. CP decomposition decomposes an N-dimensional tensor into the sum of the outer products of N matrices. For example, for a three-dimensional tensor, the result of CP decomposition can be expressed as the sum of R rank-1 tensors, each rank-1 tensor is the outer product of three vectors. Tucker decomposition decomposes the original tensor into a core tensor and a factor matrix along each dimension. Compared with CP decomposition, Tucker decomposition provides a more flexible low-rank approximation. The choice of which decomposition method to use depends on the characteristics of the data and the analysis requirements. When implementing tensor decomposition, it is first necessary to select an appropriate rank or core tensor size, which is usually determined by cross-validation or an information criterion based on model complexity. Then, an optimization algorithm such as alternating least squares or gradient descent is used to solve the decomposition problem. The result of the decomposition is the final spatiotemporal correlation information, which captures the spatiotemporal dynamic characteristics of the system in a compact and information-rich form. This representation not only reduces the dimensionality of the data, facilitating subsequent analysis, but also reveals the underlying patterns and structures in the data, helping to understand the overall behavior and local characteristics of the system.

[0066] 103. Input the key features of the smart street lamp and the time-space correlation information between the component nodes into the preset time-space hybrid model to obtain the fault detection result of the smart street lamp;

[0067] In one embodiment of the present invention, the key features of the smart street lamp and the spatiotemporal correlation information between component nodes are input into a preset spatiotemporal hybrid model to obtain the fault detection result of the smart street lamp, including: inputting the key features of the smart street lamp and the spatiotemporal correlation information between nodes into a preset spatiotemporal hybrid model, obtaining the historical data and current data of the input key features through the spatiotemporal hybrid model, and calculating the cumulative sum statistics of the key features based on the historical data and the current data; calculating the Mahalanobis distance between each component node and the normal state reference point based on the Mahalanobis distance statistics constructed based on the spatiotemporal correlation information; combining the cumulative sum statistics and the Mahalanobis distance statistics by a weighted fusion method to obtain a comprehensive fault index, and using a kernel density estimation method to perform probability density estimation on the comprehensive fault index in the historical data to obtain a fault threshold function; comparing the comprehensive fault index calculated in real time with the fault threshold function, and determining the fault detection result of the smart street lamp according to the comparison result.

[0068] Specifically, inputting the key features of smart street lamps and the spatiotemporal correlation information between nodes into the preset spatiotemporal hybrid model is the starting step of the fault detection process. This preset spatiotemporal hybrid model is a complex model that comprehensively considers time and space factors, based on a combination of multiple algorithms, such as time series analysis, spatial statistics, and machine learning methods. The input data includes the key features extracted in the previous steps, which reflect the main change patterns of the system, as well as the spatiotemporal correlation information between nodes, describing the interaction between system components. The model first preprocesses these inputs, including data standardization, missing value processing, etc. Then, the model processes the information in the time and space dimensions separately. In the time dimension, the model uses sliding window technology to capture recent change trends. In the spatial dimension, the model uses graph convolutional networks or spatial autoregressive models to process the correlation between nodes. Through this model, the system can obtain historical data and current data for each key feature. Historical data is the observation value of the past period of time (such as days or weeks), while current data is the latest observation result. These data lay the foundation for subsequent statistical analysis and anomaly detection.

[0069] Specifically, based on the acquired historical data and current data, the system calculates the cumulative sum statistics of key features. Cumulative sum (CUSUM) is an effective method for detecting small shifts or gradual trends in data. The basic idea of ​​the CUSUM algorithm is to accumulate the differences between observations and reference values. In specific implementation, you first need to determine a reference value, which is usually the average of historical data. Then, for each new observation, calculate its difference from the reference value and add this difference to the cumulative sum. If the cumulative sum exceeds the preset threshold, it may indicate an anomaly. A key parameter of the CUSUM algorithm is sensitivity, which determines how quickly the algorithm reacts to changes. Higher sensitivity can detect changes more quickly, but it may also increase the risk of false positives. In practical applications, different parameter settings need to be used for different key features, because different features may have different change patterns and importance. The calculation process of the cumulative sum statistic is continuous and is updated every time a new observation is input, which enables the system to monitor the changing trend of the feature in real time.

[0070] Specifically, a Mahalanobis distance statistic is constructed based on the spatiotemporal correlation information, and the Mahalanobis distance between each component node and the normal state reference point is calculated. The Mahalanobis distance is a distance metric that takes into account the correlation between variables and is particularly suitable for anomaly detection in multivariable systems. The process of constructing the Mahalanobis distance statistic first requires determining the reference point of the normal state, which is usually determined by analyzing historical data and can be the average value vector of historical data. At the same time, it is also necessary to calculate the covariance matrix, which describes the correlation between different features. For each component node, its Mahalanobis distance is calculated as follows: First, the difference vector between the current observation vector and the reference point is calculated. Then, this difference vector is multiplied by the inverse of the covariance matrix on the left, and then multiplied by the transpose of the difference vector. Finally, the square root of this result is taken to obtain the Mahalanobis distance. This process takes into account the correlation between features, making anomaly detection more accurate. In practical applications, it may be necessary to regularly update the reference point and the covariance matrix to adapt to long-term changes in the system. In addition, for high-dimensional data, regularization techniques may be needed to deal with possible singularity problems in the covariance matrix.

[0071] Specifically, the purpose of combining the cumulative sum statistic and the Mahalanobis distance statistic by the weighted fusion method to obtain the comprehensive fault index is to integrate the information of the time dimension (cumulative sum statistic) and the space dimension (Mahalanobis distance statistic) to form a more comprehensive fault index. The specific method of weighted fusion can be a simple linear combination, that is, assigning weights to the two statistics and then weighted summing them. The selection of weights needs to be determined according to the specific application scenario and expert experience, and needs to be optimized through experiments or cross-validation. Another method is to use more complex nonlinear fusion techniques, such as fuzzy logic or neural networks. This method can capture the nonlinear relationship between statistics, but requires more training data and computing resources. After obtaining the comprehensive fault index, the next step is to use the kernel density estimation method to estimate the probability density of the comprehensive fault index in the historical data to obtain the fault threshold function. Kernel density estimation is a non-parametric method that can estimate the probability density function of any distribution. The advantage of this method is that it does not need to assume that the data obeys a specific distribution form. In the implementation process, it is necessary to select a suitable kernel function (such as a Gaussian kernel) and bandwidth parameter. The selection of bandwidth parameter has an important impact on the estimation result and can be optimized by methods such as cross-validation. Through kernel density estimation, the probability distribution of the comprehensive fault index can be obtained, so as to determine the appropriate fault threshold function. This threshold function can be a fixed value or a dynamic threshold that changes with time or other factors.

[0072] Specifically, the process of comparing the comprehensive fault index calculated in real time with the fault threshold function and determining the fault detection result of the smart street lamp based on the comparison result is carried out in real time. Whenever new data is input, the system will calculate the latest comprehensive fault index and compare it with the threshold function. The comparison method can be a simple threshold judgment, that is, if the comprehensive fault index exceeds the threshold, it is judged as a fault state. More complex decision logic can also be used, such as considering the index values ​​at multiple consecutive time points, or combining other auxiliary information to make a judgment. In practical applications, multiple threshold levels can be set, corresponding to different warning levels, such as minor abnormalities, serious faults, etc. In addition, in order to reduce false alarms and missed alarms, the system can introduce time delay or smoothing technology, that is, the alarm is triggered only when the abnormal state lasts for a certain period of time. The fault detection result includes not only a binary judgment of whether there is a fault, but also detailed information such as the type, severity, and possible causes of the fault. This information can help maintenance personnel quickly locate and solve the problem. The system can also automatically generate reports or trigger predetermined maintenance processes based on the detection results to improve the operation and maintenance efficiency of the entire smart street lamp system.

[0073] Furthermore, the step of inputting the key features of the smart street lamp and the spatiotemporal correlation information between nodes into a preset spatiotemporal hybrid model, obtaining the historical data and current data of the input key features through the spatiotemporal hybrid model, and calculating the cumulative sum statistics of the key features based on the historical data and the current data includes: inputting the key features of the smart street lamp and the spatiotemporal correlation information between nodes into a preset spatiotemporal hybrid model, obtaining the historical data and current data of the input key features through the spatiotemporal hybrid model; calculating the mean and standard deviation of the historical data, and setting the mean as a reference value of the cumulative sum algorithm; applying the cumulative sum algorithm to the current data of the key features in time series, calculating the deviation of the current data at each time point from the reference value, and adding it to the accumulated deviation at the previous time point to obtain the cumulative sum statistics of the key features.

[0074] Specifically, the key features of the smart street lamp and the spatiotemporal correlation information between nodes are input into the preset spatiotemporal hybrid model. This model is a complex data processing system that integrates a variety of advanced analysis techniques. First, the model receives two main types of input: key features and spatiotemporal correlation information. Key features may include but are not limited to parameters such as current, voltage, power factor, and light intensity of the lamp, which are extracted from the original data through the aforementioned principal component analysis method. The spatiotemporal correlation information describes the relationship between different street lamp nodes in the time and space dimensions. The model first preprocesses these input data, including data cleaning, standardization, and missing value filling. Outliers and noise are removed during data cleaning, and standardization ensures that data of different dimensions can be compared on the same scale. For missing values, interpolation or machine learning methods can be used for estimation. After preprocessing, the model performs data analysis in two dimensions, time and space. In the time dimension, the model can use sliding window technology, which allows the model to analyze data within a fixed-size time window and update the analysis results over time. In the spatial dimension, the model can use graph convolutional networks or spatial autoregressive models, which can effectively capture the spatial dependencies between street light nodes. Through these analyses, the model can extract deeper features and patterns from the input data, providing a basis for subsequent fault detection and diagnosis. Ultimately, the model outputs historical data and current data for each key feature. Historical data usually includes a sequence of observations over a period of time in the past (such as days or weeks), while current data is the latest real-time observation.

[0075] Calculate the mean and standard deviation of the historical data and set the mean as the reference value for the cumulative sum algorithm. This step is to prepare the necessary statistical parameters for the cumulative sum algorithm. First, for each key feature, the system extracts a suitable time window from the historical data. The size of this window needs to be determined according to the nature of the specific feature, which can be hours, days or weeks. Choosing a suitable time window is crucial to capturing the normal behavior pattern of the system. Within the selected time window, the system calculates the arithmetic mean of the historical data. Specifically, the values ​​of all data points in the window are added and then divided by the total number of data points. This average value will be used as a reference value for the cumulative sum algorithm and represents the expected performance of the system under normal operating conditions. At the same time, the system also calculates the standard deviation, which reflects the degree of dispersion of the data. The calculation process of the standard deviation includes calculating the difference between each data point and the mean, squaring these differences, averaging, and finally taking the square root. Although the standard deviation is not directly used as a reference value here, it may be used in subsequent anomaly detection, such as for setting alarm thresholds or standardizing data. In practical applications, these statistics can be updated regularly to adapt to the long-term trend of the system. The frequency of updates depends on the dynamic characteristics of the system, which may be once a day, week, or month. In addition, different calculation strategies can be used for different key features. For example, for features that change quickly, methods such as exponential moving average can be used to calculate a more representative mean.

[0076] The cumulative sum algorithm is applied to the current data of the key features in time series, and the deviation of the current data from the reference value at each time point is calculated, and the cumulative deviation is added to the previous time point to obtain the cumulative sum statistic of the key feature. This process is the core of the cumulative sum algorithm. First, the system needs to determine the starting point of a time series, which is usually the moment when monitoring starts or the moment when the cumulative sum statistic is last reset. Then, for each new time point, the system obtains the current data, which is output in real time through the previous spatiotemporal hybrid model. The system compares this current data with the previously calculated reference value (the mean of historical data) to calculate the deviation. The deviation can be a simple difference, that is, the current value minus the reference value; or it can be a standardized difference, that is, the simple difference divided by the standard deviation, which makes the features of different dimensions comparable. Next, the system adds this newly calculated deviation to the cumulative deviation of the previous time point to obtain the cumulative sum statistic at the current time point. This process is actually accumulating the degree of deviation of system performance from expected performance. If the system is operating normally, the cumulative sum statistic should fluctuate randomly around zero. A persistent positive deviation will cause the cumulative sum statistic to increase, indicating that the system performance may be higher than expected; a persistent negative deviation will cause the statistic to decrease, which may mean performance degradation. To prevent the cumulative sum statistic from increasing or decreasing indefinitely, a reset mechanism is usually set, such as resetting it to zero when the statistic exceeds a predetermined threshold. This method is particularly suitable for detecting gradual trends in system performance and can detect small but continuous changes that may lead to failures early.

[0077] Furthermore, the method of comparing the comprehensive fault index calculated in real time with the fault threshold function and determining the fault detection result of the smart street lamp based on the comparison result includes: comparing the comprehensive fault index calculated in real time with the fault threshold function obtained in advance, if the comparison result is that the comprehensive fault index exceeds the fault threshold function, then marking the corresponding time point and related components as potential fault points; performing time series analysis on the marked potential fault points to determine whether they are persistent abnormalities, and if so, performing pattern matching on the persistent abnormalities in combination with the historical fault mode library to obtain fault type information; combining the potential fault points, the persistent abnormality judgment results and the fault type information into the fault detection result of the smart street lamp fault.

[0078] Specifically, comparing the comprehensive fault index calculated in real time with the pre-obtained fault threshold function is an important part of the fault detection process. This step first requires obtaining the comprehensive fault index calculated in real time, which is usually obtained by fusing the status information of multiple subsystems or components. The comprehensive fault index contains information on multiple aspects such as electrical parameters, optical parameters, and temperature parameters. At the same time, the system will call the fault threshold function determined in advance through historical data analysis and expert experience. This threshold function can be a fixed value or a dynamic function that changes with time or other factors. The comparison process usually uses a simple size comparison, that is, judging whether the comprehensive fault index exceeds the value of the threshold function. If the comprehensive fault index exceeds the threshold, the system will mark the corresponding time point and the related components as potential fault points. This marking process includes recording the specific time when the fault occurs, the number or type of the components involved, the specific value of the fault index, and other information. The purpose of marking is to provide detailed context information for subsequent analysis and processing. In practical applications, in order to improve the reliability of detection, multiple threshold levels can be set, corresponding to different warning levels, such as minor abnormalities, serious faults, etc. In addition, the system can consider the rate of change of the fault index, because some faults may manifest as a sharp change in the index rather than an absolute value exceeding the limit.

[0079] Specifically, it is an important step to confirm the fault by analyzing the time series of the marked potential fault points and determining whether it is a persistent anomaly. This process first needs to start from the marked potential fault point, extend a certain time window forward and backward, and extract the fault indicator data series within this time period. The size of the time window needs to be determined according to the specific system characteristics and fault type, which can usually range from a few minutes to a few hours. Then, the time series is analyzed. Common methods include moving average, exponential smoothing, autoregressive integrated moving average (ARIMA) model, etc. These methods can help identify the changing trend and pattern of fault indicators. The judgment of whether it is a persistent anomaly is usually based on several criteria: the duration of the anomaly, the severity of the anomaly, the frequency of the anomaly, etc. For example, if the fault indicator exceeds the threshold for multiple consecutive time points, or if it does not exceed the threshold continuously but frequently crosses the threshold, it may be judged as a persistent anomaly. In addition, the system can also consider the growth rate of the anomaly, because some faults may be manifested as a continuous growth of the indicator rather than immediately exceeding a fixed threshold. If it is determined to be a persistent anomaly, the system will enter the next stage of fault type identification. If it is not a persistent anomaly, it may be just a temporary fluctuation. The system can continue to monitor but will not alarm immediately.

[0080] Specifically, if it is confirmed to be a persistent anomaly, the system will combine the historical fault mode library to perform pattern matching on this persistent anomaly to determine the specific fault type. The historical fault mode library is a database containing various known fault type features. This library is usually established through long-term data collection, expert analysis and machine learning algorithm training. The pattern matching process first needs to extract features from the persistent anomaly data, which can include the duration of the anomaly, the amplitude of the anomaly, the frequency characteristics of the anomaly, the type of parameters involved, etc. Then, the system will compare these features with various fault modes in the fault mode library. The comparison method can include the nearest neighbor algorithm, support vector machine, decision tree or deep learning network. Each method has its own characteristics. For example, the nearest neighbor algorithm is simple and intuitive but has a large amount of calculation, while the deep learning network can capture complex nonlinear relationships but requires a large amount of training data. The matching process usually gives several most likely fault types with corresponding confidence levels. If the confidence level of the matching result is not high enough, the system will mark it as "unknown fault type" and remind the human to conduct further analysis. The result of pattern matching, that is, the fault type information, will provide an important basis for subsequent maintenance decisions.

[0081] Specifically, combining the potential fault point, the result of the persistent abnormality judgment and the fault type information into the fault detection result of the smart street lamp fault is the last step of the entire fault detection process. This step first needs to integrate all the information obtained in the previous stages. The information of the potential fault point includes the specific time of the fault, the number or type of the component involved, the specific value of the fault indicator, etc. The result of the persistent abnormality judgment includes quantitative indicators such as the duration of the abnormality, the severity of the abnormality, and the frequency of the abnormality. The fault type information includes the possible fault types and their confidence obtained by pattern matching. The system organizes this information into a structured data format, which can usually be a JSON object or a specially designed data structure. This structured fault detection result not only contains the original detection data, but also can include the fault description automatically generated by the system, the recommended maintenance measures, etc. When combining this information, the system can also add some metadata, such as the generation time of the detection result, the detection algorithm version used, etc., which are very useful for subsequent fault tracking and system optimization. Finally, the system may assign a priority to this detection result according to the severity of the fault to help maintenance personnel decide the order of processing. This combined fault detection result will be stored in the system database and trigger the corresponding alarm mechanism, such as sending SMS and email notifications to relevant personnel.

[0082] 104. Use machine learning algorithms to locate faults based on fault detection results, and predict performance degradation trends of component nodes where faults are located. Generate corresponding maintenance strategies based on the fault detection results and performance degradation trends of component nodes where faults are located.

[0083] In one embodiment of the present invention, the method of using a machine learning algorithm to locate the fault according to the fault detection result, and predicting the performance degradation trend of the component node where the fault is located, and generating a corresponding maintenance strategy based on the fault detection result and the performance degradation trend of the component node where the fault is located includes: using a pattern matching algorithm to scan the smart street lamp according to the fault type information in the fault detection result to determine the component node where the fault is located; using a long short-term memory network to predict the performance degradation trend according to the operating parameters of the component node where the fault is located to obtain the corresponding performance degradation trend; constructing a multi-objective optimization model based on the fault detection result and the performance degradation trend, and using a non-dominated sorting genetic algorithm to solve the multi-objective optimization model to obtain an optimal solution set, and generating a maintenance strategy for the smart street lamp based on the optimal solution set.

[0084] Specifically, the pattern matching algorithm is used to scan the smart street lamp according to the fault type information in the fault detection result to determine the component node of the fault location. This process first requires extracting the fault type information from the fault detection result, which usually includes the specific description of the fault, the system part or component involved, etc. Then, the system loads the pre-established fault mode library, which contains the mapping relationship between various known fault types and the components that may be affected. The pattern matching algorithm compares the current fault type information with the records in the pattern library. The comparison method can include string matching, fuzzy logic matching, or similarity calculation based on machine learning. During the matching process, the algorithm can consider multiple factors, such as keywords in the fault description, the severity of the fault, the operating environment of the system, etc. After the matching is completed, the algorithm generates a list of components that may be affected and assigns a probability score to each component. Next, the system conducts a targeted scan of the smart street lamp system based on this list. The scanning process can include checking the real-time status of these components, analyzing their historical performance data, and even triggering some specific diagnostic procedures. Through these in-depth inspections, the system can further narrow the scope of possible faults and ultimately determine the specific component nodes for fault location.

[0085] Specifically, the long short-term memory network is used to predict the performance degradation trend according to the operating parameters of the fault location component node, and the corresponding performance degradation trend is obtained. This step first requires the collection of historical operating parameter data of the fault location component node. These data may include time series data of multiple dimensions such as voltage, current, temperature, and light intensity. Data preprocessing is necessary, including removing outliers, processing missing data, and performing necessary standardization or normalization. Then, the system will build a long short-term memory (LSTM) network model. The architecture of the LSTM network needs to be designed according to the specific problem, and can include one or more LSTM layers, as well as some fully connected layers. The number of nodes in the input layer depends on the number of parameters considered, and the output layer usually predicts performance indicators in the future. The training process of the model uses historical data and continuously adjusts the network parameters through the back propagation algorithm. After the training is completed, the model will be used to predict future performance trends. When predicting, the model will consider short-term fluctuations and long-term trends, which is the advantage of the LSTM network. The prediction result is usually a time series that represents the performance change trend of the component in the future. The system can calculate the confidence interval of the prediction to reflect the uncertainty of the prediction. Finally, the predicted performance degradation trends need to be interpreted and visualized for easy understanding and use by decision makers.

[0086] Specifically, based on the fault detection results and performance degradation trends, a multi-objective optimization model is constructed, and the non-dominated sorting genetic algorithm is used to solve the model to obtain the optimal solution set, based on which the maintenance strategy of the smart street lamp is generated. The construction of the multi-objective optimization model first requires the definition of the optimization objectives, which usually include minimizing maintenance costs, maximizing system reliability, minimizing energy consumption, etc. Each objective requires a mathematical expression to quantify. Constraints also need to be clearly defined, which can include budget constraints, minimum reliability requirements, maintenance time windows, etc. The decision variables of the model can include the time point of maintenance, the type of maintenance (preventive or reactive), the replaced parts, etc. After the construction is completed, the non-dominated sorting genetic algorithm (NSGA) is used to solve this multi-objective optimization problem. NSGA first randomly generates a set of initial solutions, and then generates new solutions through selection, crossover and mutation operations. In each iteration, the algorithm performs non-dominated sorting and crowding distance calculations on the solutions to maintain the diversity of solutions. This process continues for multiple generations until the preset termination condition is reached. The algorithm will eventually obtain a set of Pareto optimal solutions, that is, the optimal solution set. Each solution represents a possible maintenance strategy that strikes a balance between different objectives. Finally, the system will analyze these optimal solutions and select one of them as the final maintenance strategy based on the decision maker's preference, or provide a decision support interface for manual selection. The selected maintenance strategy will include specific maintenance schedules, a list of parts that need maintenance, estimated costs, and expected system reliability improvements.

[0087] In this embodiment, data is collected on the operating parameters of each component node of the smart street lamp, and noise reduction is performed using wavelet transform to generate structured data. Principal component analysis and sliding time window technology are combined with Mahalanobis distance calculation to perform feature extraction and spatiotemporal correlation analysis on the structured data to obtain key features and spatiotemporal correlation information between nodes. The above information is input into the spatiotemporal hybrid model to generate fault detection results, and a machine learning algorithm is used to locate the faulty components, predict their performance degradation trends, and generate targeted maintenance strategies. The present invention realizes intelligent management of the entire process from fault detection to maintenance, effectively improving the operating stability and management efficiency of smart street lamps.

[0088] The above describes the fault diagnosis and maintenance method of the smart street lamp in the embodiment of the present invention. The following describes the fault diagnosis and maintenance system of the smart street lamp in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, a fault diagnosis and maintenance system for a smart street lamp includes:

[0089] The data denoising module 201 is used to collect data on the operating parameters of each component node in the smart street lamp, and perform denoising on the operating parameters using wavelet transform to obtain denoised structured data;

[0090] The feature analysis module 202 is used to extract features from the structured data using a principal component analysis method, and to perform spatiotemporal correlation analysis on each component node based on a sliding time window technique and Mahalanobis distance calculation to obtain key features of the smart street lamp and spatiotemporal correlation information between component nodes;

[0091] A fault detection module 203 is used to input the key features of the smart street lamp and the time-space correlation information between the component nodes into a preset time-space hybrid model to obtain a fault detection result of the smart street lamp;

[0092] The maintenance strategy module 204 is used to use a machine learning algorithm to locate the fault according to the fault detection result, and predict the performance degradation trend of the component node where the fault is located, and generate a corresponding maintenance strategy based on the fault detection result and the performance degradation trend of the component node where the fault is located.

[0093] In an embodiment of the present invention, the fault diagnosis and maintenance system of the smart street lamp runs the fault diagnosis and maintenance method of the smart street lamp. The fault diagnosis and maintenance system of the smart street lamp collects data on the operating parameters of each component node of the smart street lamp, performs noise reduction processing using wavelet transform, and generates structured data. By combining principal component analysis and sliding time window technology with Mahalanobis distance calculation, feature extraction and spatiotemporal correlation analysis are performed on the structured data to obtain key features and spatiotemporal correlation information between nodes. The above information is input into a spatiotemporal hybrid model to generate fault detection results, and a machine learning algorithm is used to locate the faulty components, predict their performance degradation trends, and generate targeted maintenance strategies. The present invention realizes intelligent management of the entire process from fault detection to maintenance, effectively improving the operating stability and management efficiency of smart street lamps.

[0094] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0095] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0096] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. 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 embodiments of the present invention.

Claims

1. A fault diagnosis and maintenance method for a smart street lamp, characterized in that: The fault diagnosis and maintenance method of the smart street lamp includes: Collect data on the operating parameters of each component node in the smart street lamp, and use wavelet transform to perform noise reduction on the operating parameters to obtain structured data after noise reduction; The principal component analysis method is used to extract features from the structured data, and the time-space correlation analysis of each component node is performed based on the sliding time window technology and Mahalanobis distance calculation to obtain the key features of the smart street lamp and the time-space correlation information between the component nodes; the principal component analysis method is used to extract features from the structured data, and the time-space correlation analysis of each component node is performed based on the sliding time window technology and Mahalanobis distance calculation to obtain the key features of the smart street lamp and the time-space correlation information between the component nodes, including: constructing a covariance matrix for the structured data, and performing eigenvalue analysis on the covariance matrix. The solution is obtained to obtain the eigenvalue and the corresponding cumulative contribution rate; the principal component is selected according to the cumulative contribution rate of the eigenvalue, and the structured data corresponding to the principal component is projected with dimensionality reduction to obtain the key feature; the sliding time window technology is applied to the key feature, and the feature data of each time point is analyzed through the preset window size and step size to obtain the local feature in the time dimension; the Mahalanobis distance is used to calculate the similarity between the nodes of each component according to the eigenvalue of the key feature and the inverse matrix of the covariance matrix; the spatial correlation matrix is ​​constructed according to the similarity, and the spatiotemporal correlation information is generated in combination with the local features in the time dimension; Inputting the key features of the smart street lamp and the spatiotemporal correlation information between the component nodes into a preset spatiotemporal hybrid model to obtain a fault detection result of the smart street lamp; A machine learning algorithm is used to locate the fault according to the fault detection result, and the performance degradation trend of the component node where the fault is located is predicted, and a corresponding maintenance strategy is generated based on the fault detection result and the performance degradation trend of the component node where the fault is located.

2. The fault diagnosis and maintenance method of the smart street lamp according to claim 1 is characterized in that: The step of constructing a spatial correlation matrix according to the similarity and generating spatiotemporal correlation information by combining local features in the time dimension includes: Constructing a spatial correlation matrix according to the similarity, and processing the spatial correlation matrix using a minimum spanning tree algorithm in graph theory to construct a topological structure between component nodes; Applying a web page ranking algorithm to the topological structure to calculate the importance index of each component node; Perform multi-dimensional multiplication operation on the local features in the time dimension and the importance index of the component node to obtain a multi-dimensional data structure of spatiotemporal features; The spatiotemporal feature multidimensional data structure is subjected to tensor decomposition to obtain spatiotemporal correlation information.

3. The fault diagnosis and maintenance method of the smart street lamp according to claim 1 is characterized in that: The step of inputting the key features of the smart street lamp and the time-space correlation information between the component nodes into a preset time-space hybrid model to obtain the fault detection result of the smart street lamp includes: Input the key features of the smart street lamp and the spatiotemporal correlation information between nodes into a preset spatiotemporal hybrid model, obtain the historical data and current data of the input key features through the spatiotemporal hybrid model, and calculate the accumulation and statistics of the key features based on the historical data and the current data; The Mahalanobis distance statistic constructed based on the spatiotemporal correlation information is used to calculate the Mahalanobis distance between each component node and the normal state reference point; The weighted fusion method is used to combine the cumulative sum statistic and the Mahalanobis distance statistic to obtain the comprehensive fault index. The kernel density estimation method is used to estimate the probability density of the comprehensive fault index in the historical data to obtain the fault threshold function. The comprehensive fault index calculated in real time is compared with the fault threshold function, and the fault detection result of the smart street lamp is determined according to the comparison result.

4. The fault diagnosis and maintenance method of the smart street lamp according to claim 3 is characterized in that: The step of inputting the key features of the smart street lamp and the spatiotemporal correlation information between nodes into a preset spatiotemporal hybrid model, obtaining historical data and current data of the input key features through the spatiotemporal hybrid model, and calculating the accumulation and statistics of the key features based on the historical data and the current data includes: Input the key features of the smart street lamp and the spatiotemporal correlation information between nodes into a preset spatiotemporal hybrid model, and obtain the historical data and current data of the input key features through the spatiotemporal hybrid model; Calculate the mean and standard deviation of the historical data, and set the mean as a reference value for the cumulative sum algorithm; The cumulative sum algorithm is applied to the current data of the key feature in time series, the deviation of the current data at each time point from the reference value is calculated, and the deviation is added to the accumulated deviation at the previous time point to obtain the cumulative sum statistic of the key feature.

5. The fault diagnosis and maintenance method of the smart street lamp according to claim 3 is characterized in that: The step of comparing the comprehensive fault index calculated in real time with the fault threshold function and determining the fault detection result of the smart street lamp according to the comparison result includes: Compare the comprehensive fault index calculated in real time with the fault threshold function obtained in advance, and if the comparison result is that the comprehensive fault index exceeds the fault threshold function, mark the corresponding time point and related components as potential fault points; Perform time series analysis on the marked potential fault points to determine whether they are persistent anomalies. If so, perform pattern matching on persistent anomalies in combination with the historical fault pattern library to obtain fault type information; The potential fault point, the persistent abnormality judgment result and the fault type information are combined into a fault detection result of the smart street lamp fault.

6. The fault diagnosis and maintenance method of the smart street lamp according to claim 1, characterized in that: The method of using a machine learning algorithm to locate the fault according to the fault detection result and predicting the performance degradation trend of the component node where the fault is located, and generating a corresponding maintenance strategy based on the fault detection result and the performance degradation trend of the component node where the fault is located includes: Scan the smart street lamp using a pattern matching algorithm according to the fault type information in the fault detection result to determine the component node where the fault is located; Applying a long short-term memory network to predict the performance degradation trend according to the operating parameters of the component node of the fault location to obtain the corresponding performance degradation trend; Based on the fault detection results and performance degradation trends, a multi-objective optimization model is constructed, and a non-dominated sorting genetic algorithm is used to solve the multi-objective optimization model to obtain an optimal solution set, and a maintenance strategy for the smart street lamp is generated based on the optimal solution set.

7. A fault diagnosis and maintenance system for smart street lamps, characterized in that: The fault diagnosis and maintenance system of the smart street lamp is applied to the fault diagnosis and maintenance method of the smart street lamp according to any one of claims 1 to 6, and the fault diagnosis and maintenance system of the smart street lamp comprises: A data denoising module is used to collect data on the operating parameters of each component node in the smart street lamp, and use wavelet transform to perform denoising on the operating parameters to obtain denoised structured data; The feature analysis module is used to extract features from the structured data using the principal component analysis method, and to perform spatiotemporal correlation analysis on each component node based on the sliding time window technology and Mahalanobis distance calculation to obtain key features of the smart street lamp and spatiotemporal correlation information between component nodes; the feature extraction from the structured data using the principal component analysis method, and to perform spatiotemporal correlation analysis on each component node based on the sliding time window technology and Mahalanobis distance calculation to obtain key features of the smart street lamp and spatiotemporal correlation information between component nodes include: constructing a covariance matrix for the structured data, and performing Decompose the eigenvalues ​​to obtain the eigenvalues ​​and the corresponding cumulative contribution rates; select the principal components according to the cumulative contribution rates of the eigenvalues, perform dimension reduction projection on the structured data corresponding to the principal components, and obtain the key features; apply the sliding time window technology to the key features, analyze the feature data at each time point through the preset window size and step size, and obtain the local features in the time dimension; use the Mahalanobis distance to calculate the similarity between the nodes of each component according to the eigenvalues ​​of the key features and the inverse matrix of the covariance matrix; construct a spatial correlation matrix according to the similarity, and generate spatiotemporal correlation information in combination with the local features in the time dimension; A fault detection module, used to input the key features of the smart street lamp and the time-space correlation information between the component nodes into a preset time-space hybrid model to obtain a fault detection result of the smart street lamp; A maintenance strategy module is used to use a machine learning algorithm to locate the fault according to the fault detection result, and predict the performance degradation trend of the component node where the fault is located, and generate a corresponding maintenance strategy based on the fault detection result and the performance degradation trend of the component node where the fault is located.

Images