A smart manhole cover monitoring and early warning method and system based on AI sampling

Abstract

The present invention discloses a smart manhole cover monitoring and early warning method and system based on AI sampling, including: S1: collecting original data of smart manhole covers and preprocessing to obtain preprocessed smart manhole cover data; S2: constructing original node features, then combining the manhole cover position to obtain manhole cover rasterization features, and finally extracting node fusion features; S3: calculating the betweenness, proximity, harmony and eccentricity centrality of each node, then combining the node fusion features to construct a node correlation matrix and a node influence range matrix; S4: calculating the comprehensive importance index of each node, then sorting all nodes in descending order and extracting important nodes; S5: extracting important node features, and calculating the first deleted node and the second deleted node; then deleting nodes from all nodes to obtain the remaining node numbers, and collecting smart manhole cover data accordingly. The present invention can solve the problem of insufficient battery life in traditional smart manhole cover monitoring methods.

Description

Technical Field

[0001] The present invention relates to the field of smart manhole cover monitoring technology, and in particular to a smart manhole cover monitoring and early warning method and system based on AI sampling. Background Art

[0002] In modern urban and park management, smart manhole cover monitoring, as an innovative technical means, is gradually becoming an important part of improving the intelligence level of park drainage systems and ensuring the safety of park drainage environments. Smart manhole cover monitoring provides strong data support for flood control and drainage in parks by real-time monitoring of key data such as manhole cover status, rainfall, and groundwater level.

[0003] The data collection efficiency of smart manhole covers in traditional parks is low, and data needs to be collected for each manhole cover one by one, which not only means cumbersome power on and off operations, but also is limited by the battery power mode. Frequent data collection seriously weakens the battery life of the manhole cover. Especially in large parks, due to the wide distribution of manhole covers and numerous monitoring points, how to efficiently manage these monitoring devices and ensure their continuous and stable operation under limited energy conditions has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] In view of this, the present invention aims to provide a smart manhole cover monitoring and early warning method and system based on AI sampling to solve the problem of insufficient battery life of traditional smart manhole cover monitoring methods.

[0005] A smart manhole cover monitoring and early warning method based on AI sampling, comprising:

[0006] S1: Collect and preprocess the original data of the smart manhole cover to obtain the preprocessed smart manhole cover data;

[0007] S2: According to the manhole cover position, the Euclidean distance between each manhole cover is calculated and the original node feature is constructed; then, a new blank map is created, and the rasterized manhole cover map is obtained by combining the manhole cover position, and then spatial feature extraction is performed to obtain the rasterized manhole cover feature; finally, the node fusion feature is extracted by combining the original node feature with the rasterized manhole cover feature;

[0008] S3: Based on the Euclidean distance between each manhole cover, the betweenness centrality, closeness centrality, harmonic centrality and eccentricity centrality of each node are calculated, and then the node correlation matrix and node influence range matrix are constructed by combining the node fusion features;

[0009] S4: Combine the node correlation matrix and the node influence range matrix to calculate the comprehensive importance index of each node, then sort all nodes in descending order and extract important nodes, including: when calculating the number of important nodes, use the gated recurrent unit to extract features from the node importance sequence, and then combine the multi-layer perceptron and the softmax function to calculate the number of important nodes in a classified manner;

[0010] S5: Combine the node fusion features of important nodes with the node influence range matrix, extract the important node features, and calculate the first deleted node and the second deleted node; then delete the nodes from all nodes according to the node numbers of the first deleted node and the second deleted node, obtain the remaining node numbers, and collect smart manhole cover data based on them.

[0011] Furthermore, the S1 step further includes:

[0012] S11: Collecting original data of smart manhole covers, wherein the original data of smart manhole covers includes: manhole cover number, data collection date, manhole cover location, adjacent manhole cover numbers, manhole cover switch status, and water level data;

[0013] S12: For the original data of the smart manhole cover, the isolation forest algorithm is used to detect outliers in the water level data, and outliers in the water level data are eliminated to obtain the water level data after outliers are eliminated;

[0014] S13: For the water level data after the outliers are removed, the maximum and minimum value method is used to normalize the data to obtain the normalized water level data; then the normalized water level data, the manhole cover number, the data collection date, the manhole cover location, the adjacent manhole cover number, and the manhole cover switch status are combined to obtain the pre-processed smart manhole cover data.

[0015] Furthermore, the S2 step further includes:

[0016] S21: according to the manhole cover position, the Euclidean distance between each manhole cover is calculated, and then the manhole cover number is used as a node and the Euclidean distance between each manhole cover is used as an edge to construct the original node feature, wherein the original node feature includes the manhole cover switch state, water level data and the manhole cover position;

[0017] S22: Create a new blank map, mark the location of the manhole cover in the blank map, and then rasterize the map in units of 10 meters to obtain a rasterized manhole cover map;

[0018] S23: Combined with the convolutional neural network, the spatial features of the rasterized manhole cover map are extracted to obtain the rasterized features of the manhole cover. The calculation method is:

[0019]

[0020]

[0021]

[0022]

[0023] in, is the preliminary spatial feature vector of the i-th node, i is the first node index, is the multilayer perceptron of the i-th node, is a convolutional neural network, This is the rasterized manhole cover map. is the preliminary spatial eigenvector, is the set of all nodes, To take the absolute value, is the number of nodes, is the adaptive convolution kernel, is a two-dimensional reshape operation, is a multi-layer perceptron, is the rasterized feature of the manhole cover. is the vector concatenation operation, It is a convolutional neural network using adaptive convolution kernel K;

[0024] S24: For each node, the original node features and the manhole cover rasterization features are combined to extract the node fusion features. The calculation method is:

[0025]

[0026]

[0027]

[0028]

[0029] in, is the first node feature of the i-th node, is the original node feature, is the ReLU function, is the first multi-layer perceptron, is the second node feature of the i-th node, is the sigmoid function, is the second multi-layer perceptron, is the third node feature of the i-th node, is the element-wise product, is the node fusion feature of the i-th node, It is a graph convolutional network.

[0030] Step S23 of the present invention first captures the relative position information between manhole covers in the manhole cover map through a convolution operation, then calculates the average value of the preliminary spatial feature vectors of all nodes, fuses the local features of each node, and forms a global understanding of the spatial distribution of the entire map; then, through a two-dimensional reshaping operation, an adaptive convolution kernel is generated using the global spatial features. The adaptive convolution kernel can be dynamically adjusted according to the specific spatial features of the map to improve the ability to capture spatial features; most traditional methods use fixed convolution kernels for spatial feature extraction, which is difficult to fully reflect the specific spatial distribution of the map and the complex relationship between manhole covers, thereby affecting the overall performance of the smart manhole cover monitoring and early warning system.

[0031] In step S24 of the present invention, a first node feature is obtained by using a first multilayer perceptron activated by a ReLU function. The ReLU function can introduce nonlinear changes, so that the model can capture the nonlinear relationship between features; then, the same splicing feature is processed by a second multilayer perceptron activated by a sigmoid function to obtain a second node feature. The sigmoid function can map the feature value to between 0 and 1, providing a gating mechanism for subsequent feature fusion; finally, the first node feature and the second node feature are combined by element-by-element multiplication to obtain a third node feature; this fusion method not only retains the information of the original feature, but also screens and enhances the feature through a gating mechanism, so that the model can more accurately identify key nodes in the data collection and monitoring tasks of the smart manhole covers in the park, thereby improving the efficiency and accuracy of data collection.

[0032] Furthermore, the S3 step also includes:

[0033] S31: Calculate the betweenness centrality of each node based on the Euclidean distance between each manhole cover , closeness centrality , reconciliation centrality Eccentricity ;

[0034] S32: Combine the betweenness centrality, closeness centrality, harmonic centrality, eccentricity centrality and node fusion characteristics of each node to calculate the node correlation and obtain the node correlation matrix. The calculation method is:

[0035]

[0036]

[0037] in, The deep fusion feature of the i-th node, is the node correlation between the i-th node and the j-th node, j is the second node index, i≠j, is the cosine similarity, is the Pearson correlation coefficient, is the deep fusion feature of the jth node, and the calculation method is the same as Same, only Node i in the calculation formula can be replaced by node j; the calculation results are Then, i is the row and j is the column. For elements, construct the node correlation matrix SIM;

[0038] S33: Combine node correlation and node fusion features to calculate the probability that a node will affect other nodes, and construct a node influence range matrix. The calculation method is:

[0039]

[0040] in, is the probability that node i will affect node j. When p≥0.5, it is considered that node i will affect node j. is the node fusion feature of the jth node; calculated Then, i is the row and j is the column. As elements, construct the node influence range matrix P.

[0041] In step S3 of the present invention, the centrality index of each node is first calculated according to the Euclidean distance between each manhole cover, which reflects the position, connectivity and influence of the node in the network; then, these centrality indexes are combined with the fusion features of the nodes, and nonlinear transformation is performed through a multi-layer perceptron to obtain the deep fusion features of the nodes; compared with traditional methods, the traditional methods only consider the direct connection between nodes or simple feature similarity, but ignore the global position and influence of the nodes in the network. The present invention combines the deep fusion features of the nodes calculated by the centrality index, and can calculate more accurate node correlation matrix and node influence range matrix; through the node correlation matrix, the model can identify which nodes have strong correlation and similarity, so that when deleting nodes, redundant or secondary nodes in these nodes are given priority; and through the node influence range matrix, the model can learn which nodes have a greater influence on other nodes, so that when deleting nodes, these important nodes are given priority to be retained.

[0042] Furthermore, the S4 step further includes:

[0043] S41: Combine the node correlation matrix and the node influence range matrix to calculate the comprehensive importance index of each node. The calculation method is:

[0044]

[0045] in, is the comprehensive importance index of the i-th node, is the i-th row of the node correlation matrix, is the i-th row of the node influence range matrix;

[0046] S42: sorting all nodes in descending order according to the comprehensive importance index to obtain a node importance sequence Q;

[0047] S43: Calculate the number of important nodes n according to the node importance sequence and the comprehensive importance index of each node. The calculation method is:

[0048]

[0049] in, is the probability distribution of the number of important nodes, the number of important nodes n is a discrete value in the range [4,7], and the selection The discrete value with the highest probability is taken as the number of important nodes. is the softmax function, is a gated recurrent unit;

[0050] S44: Select the first n nodes from the node importance sequence as important nodes.

[0051] The present invention combines the node correlation matrix with the node influence range matrix, and uses a multi-layer perceptron and a convolutional neural network to deeply fuse these two types of information. It not only considers the direct relationship between nodes, but also considers the global influence of nodes in the network, thereby improving the effect of the node comprehensive importance index. In addition, when calculating the number of important nodes, the traditional regression-based method will face the problem of predicted values ​​exceeding the expected range. The classification method can limit the number of important nodes to a clear range, thereby avoiding this uncertainty, so that the number of important nodes selected each time is within an appropriate range, which will neither cause excessive calculations nor cause insufficient number of important nodes, affecting the effect of subsequent node deletion.

[0052] Furthermore, the step S5 further includes:

[0053] S51: Combine the node fusion features of the important nodes with the node influence range matrix, extract the important node features, and calculate the first deletion node. The calculation method is:

[0054]

[0055]

[0056] in, is an important node feature. is the probability distribution of the first deleted node, take Highest probability Node is the first node to be deleted. If it cannot be divided evenly, round down. is the node fusion feature of the kth important node, For the The node fusion features of important nodes, is the row corresponding to the kth important node in the node influence range matrix, is the node influence range matrix The rows corresponding to the important nodes;

[0057] S52: Combine the manhole cover rasterization features, the node fusion features of the first deleted node, and the node influence range to extract the deleted features, and calculate the second deleted node based on the deleted features and the important node features. The calculation method is:

[0058]

[0059]

[0060] in, To delete features, is the node fusion feature of the mth first deleted node, m is the index of the node to be deleted, is the row corresponding to the mth first deleted node in the node influence range matrix, For the The node fusion features of the first deleted node, is the node influence range matrix The row corresponding to the first deleted node, is the probability distribution of the second deleted node, take Highest probability Node is used as the second deletion node. If If it cannot be divided evenly, round down;

[0061] S53: Deleting nodes from all nodes according to the node numbers of the first deleted node and the second deleted node to obtain remaining node numbers;

[0062] S54: According to the remaining node numbers, the shortest path algorithm is used to calculate the data collection order and perform smart manhole cover data collection.

[0063] In step S5, the present invention combines the node fusion features of important nodes with the node influence range matrix through a graph convolutional network to extract important node features. This step takes into account the attributes of the node itself and also incorporates the position of the node in the network and its influence on other nodes, so that the important node features can more comprehensively evaluate the importance of the node; in addition, the present invention calculates the first deleted node based on the important node features, and then calculates the second deleted node based on the features of the deleted node and the features of the important nodes. This staged method can more carefully evaluate the importance of the node, while reducing the difficulty of model calculation and the risk of mistakenly deleting important nodes; traditional node selection methods are mostly based on fixed rules or thresholds and lack flexibility, while the present invention automatically calculates the probability distribution of node deletion through a multi-layer perceptron and a softmax function, and deletes in stages according to the probability distribution, which is more intelligent and flexible and has a wider range of adaptability.

[0064] The present invention also discloses a smart manhole cover monitoring and early warning system based on AI sampling, comprising:

[0065] Smart manhole cover raw data collection and preprocessing module: collects raw data of smart manhole covers and preprocesses them to obtain preprocessed smart manhole cover data;

[0066] Node fusion feature extraction module: According to the manhole cover position, the Euclidean distance between each manhole cover is calculated and the original node feature is constructed; then, a new blank map is created, and the rasterized manhole cover map is obtained by combining the manhole cover position, and then spatial feature extraction is performed to obtain the rasterized manhole cover feature; finally, the node fusion feature is extracted by combining the original node feature with the rasterized manhole cover feature;

[0067] Node correlation and influence range matrix construction module: Based on the Euclidean distance between each manhole cover, the betweenness centrality, closeness centrality, harmonic centrality and eccentricity centrality of each node are calculated, and then the node correlation matrix and node influence range matrix are constructed by combining the node fusion features;

[0068] Important node extraction module: Combine the node correlation matrix and the node influence range matrix to calculate the comprehensive importance index of each node, then sort all nodes in descending order and extract important nodes;

[0069] Remaining node number calculation module: Combine the node fusion features of important nodes with the node influence range matrix, extract important node features, and calculate the first deleted node and the second deleted node; then delete nodes from all nodes according to the node numbers of the first deleted node and the second deleted node, obtain the remaining node numbers, and collect smart manhole cover data based on them.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] (1) In order to solve the problem of insufficient battery life in the traditional smart manhole cover monitoring method, the present invention proposes an intelligent data collection strategy. By calculating the importance index of the node, the key nodes are intelligently selected for data collection, thereby avoiding the huge energy consumption caused by the comprehensive collection of data from all manhole covers and improving the battery utilization efficiency. By collecting data by turning on one of the smart manhole covers, the data of other surrounding manhole covers are inferred to determine whether other surrounding manhole covers need to be turned on. That is, during normal monitoring, one smart manhole cover is turned on, and the surrounding manhole covers do not need to be turned on. The next time, another one is turned on. This method is used successively. Compared with the traditional method of turning on all manhole covers to collect data, the present invention prolongs the battery life of the smart manhole covers by several times, and greatly reduces the data processing volume and operating costs.

[0072] (2) The present invention innovatively combines the original node features of the manhole cover with the spatial features after rasterization to extract more comprehensive node fusion features, thus solving the problem of insufficient accuracy caused by traditional methods that rely only on a single feature for node analysis. At the same time, the present invention generates an adaptive convolution kernel using global spatial features through a two-dimensional reshaping operation, which can be dynamically adjusted according to the specific spatial features of the map, thereby improving the ability to capture the spatial distribution characteristics of manhole covers and providing a solid foundation for subsequent importance evaluation.

[0073] (3) The present invention proposes a comprehensive node importance evaluation method that comprehensively considers multiple centrality indicators and node fusion characteristics. By calculating the betweenness centrality, closeness centrality, harmonic centrality and eccentricity centrality of the node, and combining the node correlation matrix and the node influence range matrix, the present invention can comprehensively evaluate the importance of each node in the network, solving the problem of inaccurate and incomplete node importance evaluation in traditional methods, and improving the accuracy and reliability of monitoring and early warning.

[0074] (4) The present invention calculates the first deletion node based on the characteristics of the important nodes, and then calculates the second deletion node based on the characteristics of the deleted node and the characteristics of the important nodes. This phased method can more carefully evaluate the importance of the nodes, while reducing the difficulty of model calculation and the risk of mistakenly deleting important nodes. Through this method, the present invention realizes efficient and accurate monitoring of the status of the smart manhole cover under the condition of limited battery life. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 A schematic diagram of the process of the intelligent manhole cover monitoring and early warning method based on AI sampling provided by the present invention;

[0076] Figure 2 A schematic diagram of the algorithm flow of the adaptive convolution kernel provided by the present invention. DETAILED DESCRIPTION

[0077] The present invention is further described below in conjunction with the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention belong to the protection scope of the present invention.

[0078] Example 1: A smart manhole cover monitoring and early warning method based on AI sampling, such as Figure 1 As shown, the following steps are included:

[0079] S1: Collect and preprocess the original data of the smart manhole cover to obtain the preprocessed smart manhole cover data;

[0080] S11: Collecting original data of smart manhole covers, wherein the original data of smart manhole covers includes: manhole cover number, data collection date, manhole cover location, adjacent manhole cover numbers, manhole cover switch status, and water level data;

[0081] S12: For the original data of the smart manhole cover, the isolation forest algorithm is used to detect outliers in the water level data, and outliers in the water level data are eliminated to obtain the water level data after outliers are eliminated;

[0082] S13: For the water level data after the outliers are removed, the maximum and minimum value method is used to normalize the data to obtain the normalized water level data; then the normalized water level data, the manhole cover number, the data collection date, the manhole cover location, the adjacent manhole cover number, and the manhole cover switch status are combined to obtain the pre-processed smart manhole cover data.

[0083] S2: According to the manhole cover position, the Euclidean distance between each manhole cover is calculated and the original node feature is constructed; then, a new blank map is created, and the rasterized manhole cover map is obtained by combining the manhole cover position, and then spatial feature extraction is performed to obtain the manhole cover rasterized feature; finally, the node fusion feature is extracted by combining the original node feature with the manhole cover rasterized feature, such as Figure 2 As shown;

[0084] S21: according to the manhole cover position, the Euclidean distance between each manhole cover is calculated, and then the manhole cover number is used as a node and the Euclidean distance between each manhole cover is used as an edge to construct the original node feature, wherein the original node feature includes the manhole cover switch state, water level data and the manhole cover position;

[0085] S22: Create a new blank map, mark the location of the manhole cover in the blank map, and then rasterize the map in units of 10 meters to obtain a rasterized manhole cover map;

[0086] S23: Combined with the convolutional neural network, the spatial features of the rasterized manhole cover map are extracted to obtain the rasterized features of the manhole cover. The calculation method is:

[0087]

[0088]

[0089]

[0090]

[0091] in, is the preliminary spatial feature vector of the i-th node, i is the first node index, is the multilayer perceptron of the i-th node, is a convolutional neural network, This is the rasterized manhole cover map. is the preliminary spatial eigenvector, is the set of all nodes, To take the absolute value, is the number of nodes, is the adaptive convolution kernel, is a two-dimensional reshape operation, is a multi-layer perceptron, is the rasterized feature of the manhole cover. is the vector concatenation operation, It is a convolutional neural network using adaptive convolution kernel K;

[0092] S24: For each node, the original node features and the manhole cover rasterization features are combined to extract the node fusion features. The calculation method is:

[0093]

[0094]

[0095]

[0096]

[0097] in, is the first node feature of the i-th node, is the original node feature, is the ReLU function, is the first multi-layer perceptron, is the second node feature of the i-th node, is the sigmoid function, is the second multi-layer perceptron, is the third node feature of the i-th node, is the element-wise product, is the node fusion feature of the i-th node, It is a graph convolutional network.

[0098] Step S23 of the present invention first captures the relative position information between manhole covers in the manhole cover map through a convolution operation, then calculates the average value of the preliminary spatial feature vectors of all nodes, fuses the local features of each node, and forms a global understanding of the spatial distribution of the entire map; then, through a two-dimensional reshaping operation, an adaptive convolution kernel is generated using the global spatial features. The adaptive convolution kernel can be dynamically adjusted according to the specific spatial features of the map to improve the ability to capture spatial features; most traditional methods use fixed convolution kernels for spatial feature extraction, which is difficult to fully reflect the specific spatial distribution of the map and the complex relationship between manhole covers, thereby affecting the overall performance of the smart manhole cover monitoring and early warning system.

[0099] In step S24 of the present invention, a first node feature is obtained by using a first multilayer perceptron activated by a ReLU function. The ReLU function can introduce nonlinear changes, so that the model can capture the nonlinear relationship between features; then, the same splicing feature is processed by a second multilayer perceptron activated by a sigmoid function to obtain a second node feature. The sigmoid function can map the feature value to between 0 and 1, providing a gating mechanism for subsequent feature fusion; finally, the first node feature and the second node feature are combined by element-by-element multiplication to obtain a third node feature; this fusion method not only retains the information of the original feature, but also screens and enhances the feature through a gating mechanism, so that the model can more accurately identify key nodes in the data collection and monitoring tasks of the smart manhole covers in the park, thereby improving the efficiency and accuracy of data collection.

[0100] S3: Based on the Euclidean distance between each manhole cover, the betweenness centrality, closeness centrality, harmonic centrality and eccentricity centrality of each node are calculated, and then the node correlation matrix and node influence range matrix are constructed by combining the node fusion features;

[0101] S31: Calculate the betweenness centrality of each node based on the Euclidean distance between each manhole cover , closeness centrality , reconciliation centrality Eccentricity ;

[0102] S32: Combine the betweenness centrality, closeness centrality, harmonic centrality, eccentricity centrality and node fusion characteristics of each node to calculate the node correlation and obtain the node correlation matrix. The calculation method is:

[0103]

[0104]

[0105] in, The deep fusion feature of the i-th node, is the node correlation between the i-th node and the j-th node, j is the second node index, i≠j, is the cosine similarity, is the Pearson correlation coefficient, is the deep fusion feature of the jth node, and the calculation method is the same as Same, only Node i in the calculation formula can be replaced by node j; the calculation results are Then, i is the row and j is the column. For elements, construct the node correlation matrix SIM;

[0106] S33: Combine node correlation and node fusion features to calculate the probability that a node will affect other nodes, and construct a node influence range matrix. The calculation method is:

[0107]

[0108] in, is the probability that node i will affect node j. When p≥0.5, it is considered that node i will affect node j. is the node fusion feature of the jth node; calculated Then, i is the row and j is the column. As elements, construct the node influence range matrix P.

[0109] In step S3 of the present invention, the centrality index of each node is first calculated according to the Euclidean distance between each manhole cover, which reflects the position, connectivity and influence of the node in the network; then, these centrality indexes are combined with the fusion features of the nodes, and nonlinear transformation is performed through a multi-layer perceptron to obtain the deep fusion features of the nodes; compared with traditional methods, the traditional methods only consider the direct connection between nodes or simple feature similarity, but ignore the global position and influence of the nodes in the network. The present invention combines the deep fusion features of the nodes calculated by the centrality index, and can calculate more accurate node correlation matrix and node influence range matrix; through the node correlation matrix, the model can identify which nodes have strong correlation and similarity, so that when deleting nodes, redundant or secondary nodes in these nodes are given priority; and through the node influence range matrix, the model can learn which nodes have a greater influence on other nodes, so that when deleting nodes, these important nodes are given priority to be retained.

[0110] S4: Combine the node correlation matrix and the node influence range matrix to calculate the comprehensive importance index of each node, then sort all nodes in descending order and extract important nodes, including: when calculating the number of important nodes, use the gated recurrent unit to extract features from the node importance sequence, and then combine the multi-layer perceptron and the softmax function to calculate the number of important nodes in a classified manner;

[0111] S41: Combine the node correlation matrix and the node influence range matrix to calculate the comprehensive importance index of each node. The calculation method is:

[0112]

[0113] in, is the comprehensive importance index of the i-th node, is the i-th row of the node correlation matrix, is the i-th row of the node influence range matrix;

[0114] S42: sorting all nodes in descending order according to the comprehensive importance index to obtain a node importance sequence Q;

[0115] S43: Calculate the number of important nodes n according to the node importance sequence and the comprehensive importance index of each node. The calculation method is:

[0116]

[0117] in, is the probability distribution of the number of important nodes, the number of important nodes n is a discrete value in the range [4,7], and the selection The discrete value with the highest probability is taken as the number of important nodes. is the softmax function, is a gated recurrent unit;

[0118] S44: Select the first n nodes from the node importance sequence as important nodes.

[0119] The present invention combines the node correlation matrix with the node influence range matrix, and uses a multi-layer perceptron and a convolutional neural network to deeply fuse these two types of information. It not only considers the direct relationship between nodes, but also considers the global influence of nodes in the network, thereby improving the effect of the node comprehensive importance index. In addition, when calculating the number of important nodes, the traditional regression-based method will face the problem of predicted values ​​exceeding the expected range. The classification method can limit the number of important nodes to a clear range, thereby avoiding this uncertainty, so that the number of important nodes selected each time is within an appropriate range, which will neither cause excessive calculations nor cause insufficient number of important nodes, affecting the effect of subsequent node deletion.

[0120] In particular, after each calculation, the record of important nodes is kept; except for the first calculation, if the important nodes of this calculation are the same as those of the previous calculation, the last important node is selected. Nodes are replaced to update the important nodes this time. The calculation process is:

[0121] First, calculate Get the number of replacement nodes, if If it cannot be divided evenly, round up;

[0122] Secondly, for the number of replacement nodes, random sampling is used to sample replacement nodes in the node importance sequence;

[0123] Finally, the important nodes are updated according to the replacement nodes to obtain updated important nodes.

[0124] This replacement strategy introduces a dynamic update mechanism in the process of node importance evaluation and data collection, which maintains continuous attention on important nodes while ensuring that data collection is not limited to a fixed set of nodes, thereby capturing data changes on more different nodes and enhancing the system's adaptability to dynamic changes in the network or system.

[0125] S5: Combine the node fusion features of important nodes with the node influence range matrix, extract the important node features, and calculate the first deleted node and the second deleted node; then delete nodes from all nodes according to the node numbers of the first deleted node and the second deleted node, obtain the remaining node numbers, and collect smart manhole cover data based on them;

[0126] S51: Combine the node fusion features of the important nodes with the node influence range matrix, extract the important node features, and calculate the first deletion node. The calculation method is:

[0127]

[0128]

[0129] in, is an important node feature. is the probability distribution of the first deleted node, take Highest probability Node is the first node to be deleted. If it cannot be divided evenly, round down. is the node fusion feature of the kth important node, For the The node fusion features of important nodes, is the row corresponding to the kth important node in the node influence range matrix, is the node influence range matrix The rows corresponding to the important nodes;

[0130] S52: Combine the manhole cover rasterization features, the node fusion features of the first deleted node, and the node influence range to extract the deleted features, and calculate the second deleted node based on the deleted features and the important node features. The calculation method is:

[0131]

[0132]

[0133] in, To delete features, is the node fusion feature of the mth first deleted node, m is the index of the node to be deleted, is the row corresponding to the mth first deleted node in the node influence range matrix, For the The node fusion features of the first deleted node, is the node influence range matrix The row corresponding to the first deleted node, if If it cannot be divided evenly, round down. is the probability distribution of the second deleted node, take Highest probability Node is used as the second deletion node. If If it cannot be divided evenly, round down;

[0134] S53: Deleting nodes from all nodes according to the node numbers of the first deleted node and the second deleted node to obtain remaining node numbers;

[0135] S54: According to the remaining node numbers, the shortest path algorithm is used to calculate the data collection order and perform smart manhole cover data collection.

[0136] In step S5, the present invention combines the node fusion features of important nodes with the node influence range matrix through a graph convolutional network to extract important node features. This step takes into account the attributes of the node itself and also incorporates the position of the node in the network and its influence on other nodes, so that the important node features can more comprehensively evaluate the importance of the node; in addition, the present invention calculates the first deleted node based on the important node features, and then calculates the second deleted node based on the features of the deleted node and the features of the important nodes. This staged method can more carefully evaluate the importance of the node, while reducing the difficulty of model calculation and the risk of mistakenly deleting important nodes; traditional node selection methods are mostly based on fixed rules or thresholds and lack flexibility, while the present invention automatically calculates the probability distribution of node deletion through a multi-layer perceptron and a softmax function, and deletes in stages according to the probability distribution, which is more intelligent and flexible and has a wider range of adaptability.

[0137] For example, there are 20 smart manhole covers in the park, that is, there are 12 nodes, numbered from 1 to 20.

[0138] The number of important nodes is n=6, numbered 1, 2, 3, 4, 5, and 6;

[0139] Calculate the probability distribution of the first deleted node Then, according to the probability distribution, select the one with the highest probability Node is the first node to be deleted. , that is, select 3 nodes as the first deletion nodes, numbered 10, 13, and 17 respectively;

[0140] Then, the graph neural network is used to encode the fusion features and influence range of each first deleted node, and then the probability distribution of the second deleted node is calculated by combining the multi-layer perceptron and the important node features. ;

[0141] Calculate the probability distribution of the second deleted node Then, according to the probability distribution, select the one with the highest probability Node as the second deletion node, , that is, select 3 nodes as the second deletion nodes, numbered 12, 18, and 20 respectively;

[0142] Then, the nodes numbered 10, 13, 17, 10, 13, 17 are deleted from all nodes, and the remaining node numbers are 1, 2, 3, 4, 5, 6, 7, 8, 9, 11, 14, 15, 16, 19, a total of 14 nodes; that is, data collection is performed on these 14 nodes, saving the battery power of 6 smart manhole covers;

[0143] The node position relationship is then input into the Dijkstra algorithm to calculate the data collection order; according to the data collection order, data is collected for the smart manhole covers in sequence.

[0144] Embodiment 2: The present invention also discloses a smart manhole cover monitoring and early warning system based on AI sampling, comprising:

[0145] Smart manhole cover raw data collection and preprocessing module: collects raw data of smart manhole covers and preprocesses them to obtain preprocessed smart manhole cover data;

[0146] Node fusion feature extraction module: According to the manhole cover position, the Euclidean distance between each manhole cover is calculated and the original node feature is constructed; then, a new blank map is created, and the rasterized manhole cover map is obtained by combining the manhole cover position, and then spatial feature extraction is performed to obtain the rasterized manhole cover feature; finally, the node fusion feature is extracted by combining the original node feature with the rasterized manhole cover feature;

[0147] Node correlation and influence range matrix construction module: Based on the Euclidean distance between each manhole cover, the betweenness centrality, closeness centrality, harmonic centrality and eccentricity centrality of each node are calculated, and then the node correlation matrix and node influence range matrix are constructed by combining the node fusion features;

[0148] Important node extraction module: Combine the node correlation matrix and the node influence range matrix to calculate the comprehensive importance index of each node, then sort all nodes in descending order and extract important nodes;

[0149] Remaining node number calculation module: Combine the node fusion features of important nodes with the node influence range matrix, extract important node features, and calculate the first deleted node and the second deleted node; then delete nodes from all nodes according to the node numbers of the first deleted node and the second deleted node, obtain the remaining node numbers, and collect smart manhole cover data accordingly. This system can run the method of the above embodiment very well, and will not be repeated here.

[0150] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0151] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in multiple embodiments of the present invention.

[0152] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A smart manhole cover monitoring and early warning method based on AI sampling, characterized in that: The following steps are involved: S1: Collect and preprocess the original data of the smart manhole cover to obtain the preprocessed smart manhole cover data; The step S1 includes S11: collecting original data of the smart manhole cover, wherein the original data of the smart manhole cover includes: manhole cover number, data collection date, manhole cover location, adjacent manhole cover number, manhole cover switch status, and water level data; S2: According to the location of the manhole covers, the Euclidean distance between each manhole cover is calculated and the original node features are constructed; Then, a new blank map is created, and the rasterized manhole cover map is obtained by combining the manhole cover positions, and then spatial feature extraction is performed to obtain the rasterized features of the manhole cover; Finally, the original node features and the manhole cover rasterization features are combined to extract the node fusion features; S3: Based on the Euclidean distance between each manhole cover, the betweenness centrality, closeness centrality, harmonic centrality and eccentricity centrality of each node are calculated, and then the node correlation matrix and node influence range matrix are constructed by combining the node fusion features; S4: Combine the node correlation matrix and the node influence range matrix to calculate the comprehensive importance index of each node, then sort all nodes in descending order and extract important nodes, including: when calculating the number of important nodes, use the gated recurrent unit to extract features from the node importance sequence, and then combine the multi-layer perceptron and the softmax function to calculate the number of important nodes in a classified manner; S5: Combine the node fusion features of important nodes with the node influence range matrix, extract the important node features, and calculate the first deleted node and the second deleted node; then delete the nodes from all nodes according to the node numbers of the first deleted node and the second deleted node, obtain the remaining node numbers, and collect smart manhole cover data based on them.

2. The intelligent manhole cover monitoring and early warning method based on AI sampling according to claim 1 is characterized in that: The S1 step includes: S12: For the original data of the smart manhole cover, the isolation forest algorithm is used to detect outliers in the water level data, and outliers in the water level data are eliminated to obtain the water level data after outliers are eliminated; S13: For the water level data after the outliers are removed, the maximum and minimum value method is used to normalize the data to obtain the normalized water level data; then the normalized water level data, the manhole cover number, the data collection date, the manhole cover location, the adjacent manhole cover number, and the manhole cover switch status are combined to obtain the pre-processed smart manhole cover data.

3. The intelligent manhole cover monitoring and early warning method based on AI sampling according to claim 1 is characterized in that: The S2 step includes: After obtaining the rasterized manhole cover map, a multi-layer perceptron is assigned to each node, and an adaptive convolution kernel is constructed. Combined with the convolutional neural network, the spatial features of the rasterized manhole cover map are extracted to obtain the rasterized features of the manhole cover. For each node, the original node features and the rasterized features of the manhole cover are combined, and two multi-layer perceptrons are used to combine the ReLU function with the sigmoid function and the graph convolutional network to extract the node fusion features.

4. The intelligent manhole cover monitoring and early warning method based on AI sampling according to claim 3 is characterized in that: The S2 step includes: S21: according to the manhole cover position, the Euclidean distance between each manhole cover is calculated, and then the manhole cover number is used as a node and the Euclidean distance between each manhole cover is used as an edge to construct the original node feature, wherein the original node feature includes the manhole cover switch state, water level data and the manhole cover position; S22: Create a new blank map, mark the location of the manhole cover in the blank map, and then rasterize the map in units of 10 meters to obtain a rasterized manhole cover map; S23: Combined with the convolutional neural network, the spatial features of the rasterized manhole cover map are extracted to obtain the rasterized features of the manhole cover. The calculation method is: in, is the preliminary spatial feature vector of the i-th node, i is the first node index, is the multilayer perceptron of the i-th node, is a convolutional neural network, This is the rasterized manhole cover map. is the preliminary spatial eigenvector, is the set of all nodes, To take the absolute value, is the number of nodes, is the adaptive convolution kernel, is a two-dimensional reshape operation, is a multi-layer perceptron, is the rasterized feature of the manhole cover. is the vector concatenation operation, It is a convolutional neural network using adaptive convolution kernel K; S24: For each node, the original node features and the manhole cover rasterization features are combined to extract the node fusion features. The calculation method is: in, is the first node feature of the i-th node, is the ReLU function, is the first multi-layer perceptron, is the second node feature of the i-th node, is the sigmoid function, is the second multi-layer perceptron, is the third node feature of the i-th node, is the element-wise product, is the node fusion feature of the i-th node, It is a graph convolutional network.

5. The intelligent manhole cover monitoring and early warning method based on AI sampling according to claim 4 is characterized in that: The S3 step includes: S31: Calculate the betweenness centrality of each node based on the Euclidean distance between each manhole cover , closeness centrality , reconciliation centrality Eccentricity ; S32: Combine the betweenness centrality, closeness centrality, harmonic centrality, eccentricity centrality and node fusion characteristics of each node to calculate the node correlation and obtain the node correlation matrix. The calculation method is: in, The deep fusion feature of the i-th node, is the node correlation between the i-th node and the j-th node, j is the second node index, i≠j, is the cosine similarity, is the Pearson correlation coefficient, is the deep fusion feature of the jth node; calculated Then, i is the row and j is the column. For elements, construct the node correlation matrix SIM; S33: Combine node correlation and node fusion features to calculate the probability that a node will affect other nodes, and construct a node influence range matrix. The calculation method is: in, is the probability that node i will affect node j. When p≥0.5, it is considered that node i will affect node j. is the node fusion feature of the jth node; calculated Then, i is the row and j is the column. As elements, construct the node influence range matrix P.

6. The intelligent manhole cover monitoring and early warning method based on AI sampling according to claim 5 is characterized in that: The S4 step includes: S41: Combine the node correlation matrix and the node influence range matrix to calculate the comprehensive importance index of each node. The calculation method is: in, is the comprehensive importance index of the i-th node, is the i-th row of the node correlation matrix, is the i-th row of the node influence range matrix; S42: sorting all nodes in descending order according to the comprehensive importance index to obtain a node importance sequence Q; S43: Calculate the number of important nodes n according to the node importance sequence and the comprehensive importance index of each node. The calculation method is: in, is the probability distribution of the number of important nodes, the number of important nodes n is a discrete value in the range [4,7], and the selection The discrete value with the highest probability is taken as the number of important nodes. is the softmax function, is a gated recurrent unit; S44: Select the first n nodes from the node importance sequence as important nodes.

7. The intelligent manhole cover monitoring and early warning method based on AI sampling according to claim 6 is characterized in that: The S5 step includes: S51: Combine the node fusion features of the important nodes with the node influence range matrix, extract the important node features, and calculate the first deletion node. The calculation method is: in, is an important node feature. is the probability distribution of the first deleted node, take Highest probability Node is the first node to be deleted. If it cannot be divided evenly, round down. is the node fusion feature of the kth important node, For the The node fusion features of important nodes, is the row corresponding to the kth important node in the node influence range matrix, is the node influence range matrix The rows corresponding to the important nodes; S52: Combine the manhole cover rasterization features, the node fusion features of the first deleted node, and the node influence range to extract the deleted features, and calculate the second deleted node based on the deleted features and the important node features. The calculation method is: in, To delete features, is the node fusion feature of the mth first deleted node, m is the index of the node to be deleted, is the row corresponding to the mth first deleted node in the node influence range matrix, For the The node fusion features of the first deleted node, is the node influence range matrix The row corresponding to the first deleted node, is the probability distribution of the second deleted node, take Highest probability Node is used as the second deletion node. If If it cannot be divided evenly, round down; S53: Deleting nodes from all nodes according to the node numbers of the first deleted node and the second deleted node to obtain remaining node numbers; S54: According to the remaining node numbers, the shortest path algorithm is used to calculate the data collection order and perform smart manhole cover data collection.

8. A smart manhole cover monitoring and early warning system based on AI sampling, comprising: Smart manhole cover raw data collection and preprocessing module: collects raw data of smart manhole covers and preprocesses them to obtain preprocessed smart manhole cover data; Node fusion feature extraction module: According to the manhole cover position, the Euclidean distance between each manhole cover is calculated and the original node feature is constructed; then, a new blank map is created, and the rasterized manhole cover map is obtained by combining the manhole cover position, and then spatial feature extraction is performed to obtain the rasterized manhole cover feature; finally, the node fusion feature is extracted by combining the original node feature with the rasterized manhole cover feature; Node correlation and influence range matrix construction module: Based on the Euclidean distance between each manhole cover, the betweenness centrality, closeness centrality, harmonic centrality and eccentricity centrality of each node are calculated, and then the node correlation matrix and node influence range matrix are constructed by combining the node fusion features; Important node extraction module: Combine the node correlation matrix and the node influence range matrix to calculate the comprehensive importance index of each node, then sort all nodes in descending order and extract important nodes; Remaining node number calculation module: Combine the node fusion characteristics of important nodes and the node influence range matrix, extract important node characteristics, and calculate the first deleted node and the second deleted node; then delete nodes from all nodes according to the node numbers of the first deleted node and the second deleted node, obtain the remaining node numbers, and collect smart manhole cover data based on this, so as to realize the smart manhole cover monitoring and early warning method based on AI sampling as described in any one of claims 1-7.

Images