Internet of Things data efficient collection method based on active sampling

Abstract

The invention provides an Internet of Things data efficient collection method based on active sampling. The method comprises the steps that the needed minimum sampling cost is determined to be r + 1; mining a deep relationship and a structure between the grids through a space-time fusion module to obtain an augmented grid chart; sampling the selected sampling grid to obtain a sampling vector h (T + 1), and combining the sampling vector h (T + 1) with a sensing matrix X (T) to form an initial T + 1 time slot sensing matrix [X (T), h (T + 1)]; inputting the augmented grid chart and the initial sensing matrix of the T + 1 time slot into a space-time fusion module to obtain space-time representation, and exporting two initial factor matrixes through neural network training; obtaining final factor matrixes U (T + 1) and V (T + 1); and the final factor matrixes U (T + 1) and V (T + 1) are utilized to calculate data of non-sampling positions, and all grid data are obtained by combining sampling data. According to the method, the cost of data collection can be remarkably reduced, and the accuracy and reliability of the reasoning result can be ensured, so that higher-quality services are provided for applications such as environment monitoring and smart cities in the Internet of Things.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet of Things data collection, and in particular to an efficient Internet of Things data collection method based on active sampling. Background Art

[0002] With the rapid development of IoT technology, data collection is of great importance as the basis for supporting various intelligent services. Traditional data collection strategies mainly focus on densely deploying various IoT devices, such as sensors and intelligent monitoring devices, in the target area to ensure that detailed and comprehensive data can be obtained, thereby providing solid data support for high-quality IoT services. However, this strategy has encountered multiple challenges in practical applications, which has seriously restricted the further popularization and optimization of IoT technology.

[0003] First, from the perspective of economic cost, with the rapid increase in the number of IoT devices, not only has the purchase cost of the equipment become a considerable expense, but the subsequent maintenance costs have also continued to rise. Especially in the scenario of large-scale deployment, the high cost often becomes an insurmountable obstacle, limiting the widespread application of IoT technology.

[0004] Secondly, even if there is sufficient economic support to achieve large-scale device deployment, environmental factors and privacy protection requirements also bring many difficulties to data collection. Environmental instability may cause device failure or data distortion, while strict privacy protection in some areas may make it difficult to fully carry out data collection. These factors work together to make comprehensive data coverage extremely difficult, which in turn affects the overall quality and effectiveness of IoT services.

[0005] Coverage is a key indicator for measuring the quality of IoT services, and its level directly reflects the integrity and effectiveness of data collection. However, in large-scale and fine-grained IoT applications, due to various limitations of device deployment, the collected data is often incomplete and sparse. This data sparsity will seriously weaken the ability of upper-layer applications to make accurate decisions, making it difficult to provide truly optimized services.

[0006] In order to deal with the high cost of data collection and incomplete coverage, the industry has begun to explore ways to infer unsampled data from the collected data. The feasibility of this idea lies in the continuity of the IoT environment and the strong spatiotemporal correlation of sensor data, which are mathematically manifested as the low rank of data. Using matrix filling technology, missing data can be effectively inferred and filled, thereby achieving efficient data collection.

[0007] Although some studies have attempted to apply matrix filling technology to the field of data collection and have achieved certain results, most of these studies have focused on optimizing the matrix filling algorithm itself to improve the accuracy of the inference results. However, this passive optimization strategy ignores the importance of the sampling point location and the amount of information it carries. In fact, data at different locations have significant differences in the amount of information. If the sampling points are not selected properly, even if the algorithm is optimized, it is difficult to avoid serious inference bias.

[0008] Therefore, when faced with uneven information amounts at different locations, the passive sampling strategies of existing methods have obvious deficiencies, and it is difficult to ensure the accuracy and reliability of the inference results. This has become a bottleneck restricting the further development of IoT data collection technology, and new technical solutions are urgently needed to break through this limitation. Summary of the invention

[0009] In order to solve the problems of high cost and incomplete coverage in traditional IoT data collection methods, the purpose of the present invention is to provide an efficient IoT data collection method based on active sampling, which can accurately infer unsampled data from sparse data by actively selecting information-rich sampling locations and combining matrix filling technology. This method can not only significantly reduce the cost of data collection, but also ensure the accuracy and reliability of the inference results, thereby providing better services for environmental monitoring, smart cities and other applications in the IoT.

[0010] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0011] An efficient method for collecting IoT data based on active sampling includes the following steps:

[0012] The target area is divided into N grids, and the sensor data of each grid in the historical T time slots are obtained to form a sensor matrix X with a scale of N×T (T) ; According to the sensing matrix X (T) The rank r determines that the minimum sampling cost required for the next time slot T+1 is r+1;

[0013] Augment the grid map and mine the deep relationship and structure between grids through the spatiotemporal fusion module to obtain an augmented grid map.

[0014] The augmented grid graph is divided into r+1 clusters using the spectral clustering method, and at least one sampling grid is selected from each cluster;

[0015] Sample the selected sampling grid and obtain the sampling vector h (T+1) and compare it with the sensor matrix X (T) Combined to form the initial T+1 time slot sensing matrix [X (T) ,h (T+1) ];

[0016] The augmented grid map and the initial sensor matrix of the T+1 time slot are input into the spatiotemporal fusion module to obtain the spatiotemporal representation, and two initial factor matrices are derived through neural network training;

[0017] Iteratively update the two initial factor matrices to obtain the final factor matrix U (T+1) and V (T+1) ;

[0018] Using the final factor matrix U (T+1) and V (T+1) , the data of the unsampled positions are calculated and combined with the sampled data to obtain all the collected grid data.

[0019] According to an efficient IoT data collection method based on active sampling provided by the present invention, when determining the minimum sampling cost, the historical sensor matrix X is known. (T) , assuming its rank is r, X (T+1) (X (T+1) =[X (T) , x (T+1) ]) is at most r+1, so the minimum sampling cost is r+1 according to the basic properties of the matrix; where X (T+1) The space of is composed of r+1 linearly independent vectors.

[0020] According to an efficient IoT data collection method based on active sampling provided by the present invention, when performing augmentation processing, the deep relationship and structure between grids are mined according to the historical data of the grids, that is, the grid graph G = (V, ε) is augmented to obtain an augmented grid graph, where V = {v1, v2, …, v N} is the set of grids, ε is the set of edges; the adjacency matrix of the grid graph G is denoted by A, a ij ∈A, if v i , v j ∈V,ε ij ∈ε, then a ij =1, otherwise 0.

[0021] According to an efficient Internet of Things data collection method based on active sampling provided by the present invention, for grid v i (1≤i≤N), whose temporal representation s spans multiple time steps i It can be expressed as s i =TC (m) (b i ), where TC represents temporal convolution, m represents multiple temporal convolutions to aggregate information from all time steps, It is the intermediate product of the space-time fusion module and is expressed as the following formula:

[0022] TC(X(T) ) → B (T)

[0023] TC(SC(B (T) ,A))→Y (T)

[0024] Among them, TC represents temporal convolution, SC represents spatial convolution, and They represent different stages of representation, D is the representation dimension, T′ and T″ are the time dimensions after convolution, and after stacking multiple blocks, the final representation of the grid can be obtained.

[0025] According to an efficient IoT data collection method based on active sampling provided by the present invention, the cosine distance represented by time is used to measure the similarity between two grids, and the grid v i and grid v j The similarity between (1≤j≤N) can be expressed as the following formula:

[0026]

[0027] According to the formula, we can get a similarity matrix All diagonal elements are equal to 1, w ij The larger it is, the more similar the data patterns between the grids are, and vice versa.

[0028] According to an efficient IoT data collection method based on active sampling provided by the present invention, the adjacency relationship between grids is re-evaluated according to the similarity matrix W, and the grid v i and grid v j The edge between ij The probability of adding or removing follows the Bernoulli distribution with probability p ij , expressed as the following formula:

[0029]

[0030] Among them, max(W·(1-I)) represents the maximum value of the matrix W except the diagonal elements, represents their average value.

[0031] According to an efficient IoT data collection method based on active sampling provided by the present invention, if the grid v i and grid v j Physical locations are adjacent, with probability ρ ij Delete edge ε ij , ρ ij With the similarity w ij decreases with the increase of; if the grid v iand grid v j If the physical locations are not adjacent, then 1-ρ ij The probability of adding an edge between the two; the probability of adding an edge is related to the similarity w ij The trend remains consistent;

[0032] Get the augmented grid map

[0033] According to an efficient IoT data collection method based on active sampling provided by the present invention, when the augmented grid graph is divided using the spectral clustering method, the division is based on the Laplace matrix L, which is derived from the similarity matrix W, that is, L = DW, and D is obtained by the following formula:

[0034]

[0035] After obtaining the Laplace matrix L, first decompose its eigenvalues ​​to obtain a new matrix composed of its eigenvectors; then use the K-Means algorithm to divide the grid into r+1 clusters according to the new matrix.

[0036] According to an efficient IoT data collection method based on active sampling provided by the present invention, the following steps are performed for each cluster to select a cluster center as a sampling grid:

[0037] Calculate node degree: For each node in the cluster, calculate its degree, that is, count the number of edges directly connected to the node, which is used as a measure of node degree centrality;

[0038] Sorting nodes: All nodes in the cluster are sorted in descending order according to the calculated node degree. Nodes with higher degrees are ranked higher, indicating that they have stronger connectivity in the cluster and higher importance.

[0039] Select cluster centers: From the sorted node list, select one or more nodes with the highest degree as the cluster centers of the cluster and as representative points of the sampling grid;

[0040] Forming a sampling grid: The selected cluster centers in each cluster are combined to form a sampling grid for the entire data set. The sampling grid can reflect the structural characteristics of the data set and is used for subsequent data analysis, sampling, or visualization of clustering results.

[0041] According to an efficient Internet of Things data collection method based on active sampling provided by the present invention, the augmented grid map and the initial sensing matrix [X (T) ,h (T+1) ] Input the spatiotemporal fusion module and output the spatiotemporal representation

[0042] Through neural network training, from spatiotemporal representation Derive two initial factor matrices and

[0043] The two initial factor matrices are trained to obtain the final factor matrix. The process is expressed as the following formula:

[0044]

[0045] in,‖‖ F represents the F norm, Z (T+1) is [1,1,…,1,z (T+1) ] is a simple expression, z (T+1) Indicates whether the grid is sampled in the T+1 time slot. If it is sampled, it is marked as 1, otherwise it is marked as 0;

[0046] After getting the final factor matrix U (T+1) and V (T+1) After that, through U (T+1) V (T+1)T The data at the unsampled locations are inferred and combined with the sampled data to form all the collected grid data.

[0047] It can be seen that compared with the IoT applications of the prior art, efficient data collection is crucial for scenarios such as environmental monitoring and smart cities. However, traditional data collection methods face challenges such as high cost and incomplete data coverage. In order to solve these problems, the present invention proposes an efficient IoT data collection method based on active sampling. This method actively selects information-rich areas for sampling, and combines a matrix filling algorithm to accurately infer the perception data of unsampled areas from sparse samples. This technology can not only ensure the accuracy of the data, but also significantly reduce the cost of IoT data collection. Therefore, the present invention has the following

[0048] Beneficial effects:

[0049] 1. Minimum data collection cost: The present invention conducts in-depth analysis of historical data to determine the minimum sampling requirements required for accurate inference in the next time slot, thereby reducing unnecessary sampling work and achieving the lowest data collection cost.

[0050] 2. Improve the accuracy of data reasoning: The present invention actively selects sampling locations within the sampling cost range, fully considers the information differences in different regions, avoids reasoning deviations caused by insufficient information at the sampling points, and thus significantly improves the accuracy and reliability of data reasoning.

[0051] 3. Flexible response to complex environments: The active sampling method of the present invention is highly flexible and can dynamically adjust the sampling strategy according to actual environmental changes. It has strong adaptability and can maintain efficient data collection capabilities in complex and dynamic IoT environments.

[0052] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a flow chart of an embodiment of a method for efficiently collecting IoT data based on active sampling of the present invention.

[0054] Figure 2 It is a schematic diagram of an embodiment of an efficient method for collecting IoT data based on active sampling of the present invention.

[0055] Figure 3 It is an application principle diagram of the space-time fusion module in an embodiment of an efficient collection method for Internet of Things data based on active sampling of the present invention.

[0056] Figure 4 It is a schematic diagram of the steps of obtaining a factor matrix in an embodiment of an efficient method for collecting IoT data based on active sampling of the present invention. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. 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.

[0058] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0059] See also Figure 1 and Figure 2 This embodiment provides an efficient method for collecting IoT data based on active sampling, and the method includes the following steps:

[0060] Step S1: Divide the target area into N grids, and obtain the sensor data of each grid in the historical T time slots to form a sensor matrix X with a scale of N×T. (T) ;

[0061] Step S2: According to the sensing matrix X (T) The rank r determines that the minimum sampling cost required for the next time slot T+1 is r+1;

[0062] Step S3, augmenting the grid map, mining the deep relationship and structure between grids through the spatiotemporal fusion module to obtain an augmented grid map;

[0063] Step S4, using a spectral clustering method to divide the augmented grid graph into r+1 clusters, and selecting at least one sampling grid from each cluster;

[0064] Step S5, sampling the selected sampling grid to obtain a sampling vector h (T+1) and compare it with the sensor matrix X (T) Combined to form the initial T+1 time slot sensing matrix [X (T) ,h (T+1) ];

[0065] Step S6, inputting the augmented grid map and the initial sensor matrix of the T+1 time slot into the space-time fusion module to obtain the space-time representation, and deriving two initial factor matrices through neural network training;

[0066] Step S7, iteratively update the two initial factor matrices to obtain the final factor matrix U (T+1) and V (T+1) ;

[0067] Step S8, using the final factor matrix U (T+1) and V (T+1) , the data of the unsampled positions are calculated and combined with the sampled data to obtain all the collected grid data.

[0068] In this embodiment, for a given target area, it can be divided into N grids, and the sensing data of the N grids in the historical T time slots can be expressed as a sensing matrix X with a scale of N×T: (T) ={x (1) ,x (2) ,…,x (T)}, where x (1) ,x (2) ,…,x (T) Respectively represent the sensor data collected in N grids in the 1st, 2nd, …, Tth time slots in history. For the future T+1 time slot, the goal is to select a small number of grids for sampling and use the sensor matrix X (T) The small amount of sampled data can be used to accurately infer the sensor data of other unsampled locations, that is, the sensor matrix x at the T+1 time slot is obtained. (T+1) .

[0069] It can be seen that this embodiment provides an efficient IoT data collection method based on active sampling, which mainly includes three parts: 1) determining the minimum sampling cost that can ensure accurate reasoning; 2) selecting key grids for sampling without exceeding the sampling cost; 3) using matrix filling to infer data of other unsampled locations.

[0070] In the above step S1, when determining the minimum sampling cost, the historical sensing matrix X is known. (T) , assuming its rank is r, X (T+1) (X (T+1) =[X (T) , x (T+1) ]) is at most r+1, so the minimum sampling cost is r+1 according to the basic properties of the matrix; where X (T+1) The space of is composed of r+1 linearly independent vectors.

[0071] In this embodiment, when selecting key grids for sampling, the deep relationship and structure between grids are mined according to the historical data of the grids, that is, the grid graph G = (V, ε) is augmented to obtain an augmented grid graph, where V = {v1, v2, …, v N} is the set of grids, ε is the set of edges; the adjacency matrix of the grid graph G is denoted by A, a ij ∈A, if v i , v j ∈V,ε ij ∈ε, then a ij =1, otherwise 0.

[0072] For the grid v i (1≤i≤N), whose temporal representation s spans multiple time steps i It can be expressed as s i =TC (m) (b i ), where TC represents temporal convolution, m represents multiple temporal convolutions to aggregate information from all time steps, is the intermediate product of the space-time fusion module. The complete process is as follows Figure 3 As shown, it can be expressed as the following formula:

[0073] TC(X (T) ) → B (T)

[0074] TC(SC(B (T) ,A))→Y (T)

[0075] Among them, TC represents temporal convolution, SC represents spatial convolution, and They represent different stages of representation, D is the representation dimension, T′ and T″ are the time dimensions after convolution, and after stacking multiple blocks, the final representation of the grid can be obtained.

[0076] The cosine distance of the temporal representation is used to measure the similarity between two meshes, mesh v i and grid v j The similarity between (1≤j≤N) can be expressed as the following formula:

[0077]

[0078] So we can get a similarity matrix All diagonal elements are equal to 1, w ij The larger the value, the more similar the data patterns between the grids are, and vice versa. The adjacency relationship between the grids is re-evaluated according to the similarity matrix W. Specifically, the grid v i and grid v j The edge between ij The probability of adding or removing follows the Bernoulli distribution with probability p ij , expressed as the following formula:

[0079]

[0080] Where max(W·(1-I)) represents the maximum value of the matrix W except the diagonal elements (where 1 represents an N×N matrix whose elements are all 1, and I represents the unit matrix of the same size). represents their average value. If the grid v i and grid v j The physical locations are adjacent. In this embodiment, the probability ρ ij Delete edge ε ij , ρ ij With the similarity w ij decreases with the increase of; if the grid v i and grid v j If the physical locations are not adjacent, then 1-ρ ij The probability of adding an edge between the two; the probability of adding an edge is related to the similarity w ij After the above augmentation process, the augmented grid graph can be obtained.

[0081] The above process obtains an augmented graph in which the grids with similar data patterns are closer in "distance", making them easier to be classified into the same cluster. When the augmented grid graph is divided using the spectral clustering method, the division is based on the Laplace matrix L, which is derived from the similarity matrix W, that is, L = DW, and D is obtained by the following formula:

[0082]

[0083] After obtaining the Laplace matrix L, first decompose its eigenvalues ​​to obtain a new matrix composed of its eigenvectors; then use the K-Means algorithm to divide the grid into r+1 clusters according to the new matrix. Compared with other methods, spectral clustering provides an effective clustering solution by reducing computational complexity and better adapting to data distribution characteristics. For each cluster, several simple and effective methods can be used to select the sampling grid (i.e., cluster center), such as degree centrality, eigenvector centrality, etc. This embodiment mainly uses degree centrality to achieve this purpose. Degree centrality is one of the simplest centrality measures, which evaluates the importance of a node by calculating its degree (i.e., the number of edges directly connected to it).

[0084] Preferably, this embodiment performs the following steps for each cluster to select a cluster center as a sampling grid, which may include the following steps:

[0085] Calculate node degree: For each node in the cluster, calculate its degree, that is, count the number of edges directly connected to the node, and use this as a measure of the node degree centrality; Sort nodes: According to the calculated node degree, all nodes in the cluster are sorted in descending order. Nodes with higher degrees are ranked at the front, indicating that they have stronger connectivity and higher importance in the cluster; Select cluster center: From the sorted node list, select one or more nodes with the highest degree as the cluster center of the cluster and as the representative point of the sampling grid; Form a sampling grid: Combine the selected cluster centers in each cluster to form a sampling grid for the entire data set. The sampling grid can reflect the structural characteristics of the data set and be used for subsequent data analysis, sampling or visualization of clustering results.

[0086] After completing the previous process, the sampling vector h can be obtained (T+1) , the vector corresponds to N grids one by one, and the unsampled grids are initialized to 0 in the sampling vector. Thus, the initial sensing matrix [X (T) ,h (T +1) ]. In the inference process of matrix filling, the matrix decomposition tool is used to focus on extracting two factor matrices that are closely related to the spatiotemporal characteristics of the grid. In order to accurately derive these factor matrices, it is first necessary to obtain the initial spatiotemporal representation of the T+1 time slot grid, and use the spatiotemporal representation to train and output two initial factor matrices. The specific steps are as follows Figure 4 shown.

[0087] First, the augmented grid map and the initial sensing matrix [X (T) ,h (T+1) ] Input the spatiotemporal fusion module and output the spatiotemporal representation Then, through neural network training, from the spatiotemporal representation Derive two initial factor matrices and Then the two initial factor matrices are trained to obtain the final factor matrix. The process is expressed as the following formula:

[0088]

[0089] in,‖‖ F represents the F norm, Z (T+1) is [1,1,…,1,z (T+1) ] is a simple expression, z (T+1) Indicates whether the grid is sampled in the T+1 time slot. If it is sampled, it is marked as 1, otherwise it is marked as 0;

[0090] After getting the final factor matrix U (T+1) and V (T+1) After that, through U (T+1) V (T+1)T The data at the unsampled locations are inferred and combined with the sampled data to form all the collected grid data.

[0091] In practical applications, suppose there is a grid graph G and its historical sensor matrix X collected over T time slots. (T) In order to obtain the data of the T+1 time slot with high accuracy, we first determine the sensor matrix X (T) The rank of the grid graph G is assumed to be r; then the sensor matrix X (T) Input the spatiotemporal fusion module for processing to obtain representation B (T) ; Then use B (T) By performing graph augmentation, the grid graph G can be refined into an augmented grid graph Then, the augmented grid map Perform spectral clustering to obtain r+1 clusters, and select the grid with the largest degree in each cluster as the sampling grid. After sampling, the sensor matrix X (T) and the sampling vector h (T+1) Input the spatiotemporal fusion module to obtain spatiotemporal representation Representing from space and time through neural network training Get two initial factor matrices and Their update process is shown in Algorithm 1:

[0092]

[0093]

[0094] Get the final factor matrix U (T+1) and V (T+1)Then, calculate U (T+1) V (T+1)T , which is similar to the perception matrix X (T +1) , the last column is the inferred perception data of the T+1 time slot, x (T+1) byU (T+1) V (T+1)T The T+1 columns and sampling vector h (T +1) constitute.

[0095] In summary, this embodiment proposes an innovative method for efficient IoT data collection, which transforms the traditional data collection process into an active sampling optimization problem to achieve low-cost and high-precision data collection. This method combines the matrix filling algorithm and optimizes the sampling strategy to ensure that information is obtained from the most critical locations, thereby reducing unnecessary sampling costs.

[0096] Furthermore, this embodiment integrates the grid map and the time series data, and uses the time-space fusion module to mine the deep time-space correlation between grids. This method helps to select more critical sampling locations, thereby improving the accuracy of data reasoning.

[0097] Furthermore, this embodiment performs clustering based on the similarity of grid data patterns and uses the cluster center as the sampling location. This strategy effectively covers global spatiotemporal information, reduces redundant information in data collection, and improves the representativeness and efficiency of data sampling.

[0098] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0099] The above-mentioned embodiments are only preferred embodiments of the present invention and cannot be used to limit the scope of protection of the present invention. Any non-substantial changes and substitutions made by technicians in this field on the basis of the present invention shall fall within the scope of protection required by the present invention.

Claims

1. An efficient method for collecting IoT data based on active sampling, characterized in that: The following steps are involved: The target area is divided into N grids, and the sensor data of each grid in the historical T time slots are obtained to form a sensor matrix X with a scale of N×T (T) ; According to the sensing matrix X (T) The rank r determines that the minimum sampling cost required for the next time slot T+1 is r+1; Augment the grid map and mine the deep relationship and structure between grids through the spatiotemporal fusion module to obtain an augmented grid map. The augmented grid graph is divided into r+1 clusters using the spectral clustering method, and at least one sampling grid is selected from each cluster; Sample the selected sampling grid and obtain the sampling vector h (T+1) and compare it with the sensor matrix X (T) Combined to form the initial T+1 time slot sensing matrix [X (T) ,h (T+1) ]; The augmented grid map and the initial sensor matrix of the T+1 time slot are input into the spatiotemporal fusion module to obtain the spatiotemporal representation, and two initial factor matrices are derived through neural network training; Iteratively update the two initial factor matrices to obtain the final factor matrix U (T+1) and V (T+1) ; Using the final factor matrix U (T+1) and V (T+1) , the data of the unsampled positions are calculated and combined with the sampled data to obtain all the collected grid data.

2. The method according to claim 1, characterized in that: When determining the minimum sampling cost, the historical sensing matrix X is known. (T) , assuming its rank is r, X (T+1) (X (T+1) =[X (T) , x (T +1) ]) is at most r+1, so the minimum sampling cost is r+1 according to the basic properties of the matrix; where X (T+1) The space of is composed of r+1 linearly independent vectors.

3. The method according to claim 1, characterized in that: When performing augmentation processing, the deep relationship and structure between grids are mined according to the historical data of the grids, that is, the grid graph G = (V, ε) is augmented to obtain the augmented grid graph, where V = {v1, v2, …, v N } is the set of grids, ε is the set of edges; the adjacency matrix of the grid graph G is denoted by A, a ij ∈A, if v i , v j ∈V,ε ij ∈ε, then a ij =1, otherwise 0.

4. The method according to claim 3, characterized in that: For the grid v i (1≤i≤N), whose temporal representation s spans multiple time steps i It can be expressed as s i =TC (m) (b i ), where TC represents temporal convolution, m represents multiple temporal convolutions to aggregate information from all time steps, It is the intermediate product of the space-time fusion module and is expressed as the following formula: CT(X (T) )→B (T) TC(SC(B (T) ,A))→Y (T) Among them, TC represents temporal convolution, SC represents spatial convolution, and They represent different stages of representation, D is the representation dimension, T′ and T″ are the time dimensions after convolution, and after stacking multiple blocks, the final representation of the grid can be obtained.

5. The method according to claim 4, characterized in that: The cosine distance of the temporal representation is used to measure the similarity between two meshes, mesh v i and grid v j The similarity between (1≤j≤N) can be expressed as the following formula: According to the formula, we can get a similarity matrix All diagonal elements are equal to 1, w ij The larger it is, the more similar the data patterns between the grids are, and vice versa.

6. The method according to claim 5, characterized in that: According to the similarity matrix W, the adjacency relationship between grids is re-evaluated, and the grid v i and grid v j The edge between ij The probability of adding or removing follows the Bernoulli distribution with probability p ij , expressed as the following formula: Among them, max(W·(1-I)) represents the maximum value of the matrix W except the diagonal elements, represents their average value.

7. The method according to claim 6, characterized in that: If the grid v i and grid v j Physical locations are adjacent, with probability ρ ij Delete edge ε ij , ρ ij With the similarity w ij decreases with the increase of; if the grid v i and grid v j If the physical locations are not adjacent, then 1-ρ ij The probability of adding an edge between the two; the probability of adding an edge is related to the similarity w ij The trend remains consistent; Get the augmented grid map 8. The method according to any one of claims 1 to 7, characterized in that: When the spectral clustering method is used to divide the augmented grid graph, the division basis is based on the Laplace matrix L, which is derived from the similarity matrix W, that is, L = DW, and D is obtained by the following formula: After obtaining the Laplace matrix L, first decompose its eigenvalues ​​to obtain its eigenvectors The new matrix is ​​composed of ; then the K-Means algorithm is used to divide the grid into r+1 clusters according to the new matrix.

9. The method according to any one of claims 1 to 7, characterized in that: For each cluster, perform the following steps to select cluster centers as the sampling grid: Calculate node degree: For each node in the cluster, calculate its degree, that is, count the number of edges directly connected to the node, which is used as a measure of node degree centrality; Sorting nodes: All nodes in the cluster are sorted in descending order according to the calculated node degree. Nodes with higher degrees are ranked higher, indicating that they have stronger connectivity in the cluster and higher importance. Select cluster centers: From the sorted node list, select one or more nodes with the highest degree as the cluster centers of the cluster and as representative points of the sampling grid; Forming a sampling grid: The selected cluster centers in each cluster are combined to form a sampling grid for the entire data set. The sampling grid can reflect the structural characteristics of the data set and is used for subsequent data analysis, sampling, or visualization of clustering results.

10. The method according to claim 7, characterized in that: Augment the grid and the initial sensing matrix [X (T) ,h (T+1) ] Input the spatiotemporal fusion module and output the spatiotemporal representation Through neural network training, from spatiotemporal representation Derive two initial factor matrices and The two initial factor matrices are trained to obtain the final factor matrix. The process is expressed as the following formula: in,‖ ‖ F represents the F norm, Z (T+1) is [1,1,…,1,z (T+1) ] is a simple expression, z (T+1) Indicates whether the grid is sampled in the T+1 time slot. If it is sampled, it is marked as 1, otherwise it is marked as 0; After getting the final factor matrix U (T+1) and V (T+1) After that, through U (T+1) V (T+1)T The data at the unsampled locations are inferred and combined with the sampled data to form all the collected grid data.

Images