Region monitoring method and device

Abstract

The embodiment of the invention provides a region monitoring method and device, and the method comprises the steps: carrying out the grid division of a target region according to the geographic information of the target region and a monitoring granularity demand, and obtaining a target region comprising a plurality of grid sub-regions; acquiring monitoring data in partial grid sub-regions; preprocessing the geographic information of the target area and the monitoring data in the partial grid sub-areas to obtain geographic information features and monitoring data features; and inputting the geographic information features and the monitoring data features into a pre-constructed global perception model, and outputting monitoring data in all grid sub-regions by the global perception model. The global monitoring data can be deduced based on the limited monitoring data, the cost and the system complexity are reduced, and the accuracy of the deduced monitoring data is ensured.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of information processing technology, and in particular, to a method and device for regional monitoring. Background Art

[0002] With the rapid development of the Internet of Things and sensor technology, the demand for monitoring specific areas is increasing in many fields such as environmental monitoring, disaster warning, and urban management. The traditional monitoring method generally deploys sensors at multiple monitoring points in a specific area. When the scope of a specific area is large, the number of sensors required to be deployed increases, the deployment and maintenance costs are high, and the complexity of the system structure and data processing also increases. Summary of the invention

[0003] In view of this, an object of the embodiments of the present application is to provide a regional monitoring method and device to solve the problem of regional monitoring.

[0004] Based on the above purpose, the embodiment of the present application provides a regional monitoring method, including:

[0005] According to the geographic information of the target area and the monitoring granularity requirements, the target area is grid-divided to obtain a target area including a plurality of grid sub-areas;

[0006] Obtain monitoring data within some grid sub-areas;

[0007] Preprocessing the geographic information of the target area and the monitoring data in the partial grid sub-areas to obtain geographic information features and monitoring data features;

[0008] The geographic information features and monitoring data features are input into a pre-built global perception model, and the global perception model outputs the monitoring data within all grid sub-areas.

[0009] Optionally, the training method of the global perception model includes:

[0010] Obtain monitoring data samples within some grid sub-areas;

[0011] Preprocessing the geographic information samples and the monitoring data samples of the target area to generate geographic information feature samples and monitoring data feature samples;

[0012] The neural network model is trained based on the geographic information feature samples and the monitoring data feature samples to obtain a trained global perception model.

[0013] Optionally, the global perception model includes a feature extraction module and a feature aggregation module; wherein the feature extraction module is used to extract spatiotemporal correlation features based on the geographic information features of the target area and the monitoring data features of some grid sub-areas; and the feature aggregation module is used to infer the monitoring data features of the target area based on the spatiotemporal correlation features.

[0014] Optionally, the feature aggregation module obtains an output feature H based on a multi-head attention mechanism according to the input spatiotemporal correlation features, and the calculation method is:

[0015] H=V s ·σ((I·W1)W2(W3·I) T +b s ) (5)

[0016] Among them, W1, W2, W3, V s ,b s is the parameter of the model, σ is the activation function, and I represents the spatiotemporal correlation feature; a fully connected layer is used to perform linear transformation to obtain the monitoring data features of the target area, and the method for performing linear transformation of the fully connected layer is:

[0017] Y=linear(H)⊙(1-M)+X⊙M (6)

[0018] Among them, linear(.) is the linear layer, M is the mask matrix, and the element M in the mask matrix gi,j =1 indicates the grid sub-area where monitoring data is collected, M gi,j =0 indicates that the grid sub-area where the monitoring data needs to be inferred, and X is the target indicator item feature in the preprocessed monitoring data feature sample.

[0019] Optionally, after obtaining the trained global perception model, the following is also included:

[0020] Determine a sensor deployment strategy within the target area based on a Markov decision process and a deep reinforcement learning model;

[0021] Deploy sensors in multiple grid sub-areas within the target area according to the sensor deployment strategy;

[0022] The obtaining of monitoring data in some grid sub-areas includes:

[0023] For the grid sub-areas where sensors are deployed, the sensors are used to obtain monitoring data in each grid sub-area.

[0024] Optionally, in a Markov decision process,

[0025] The constructed reward function is expressed as:

[0026] R(s n ,a n )=α(σ-E n (s n ,a n ))-β (7)

[0027] Among them, σ is the error threshold that is expected to be achieved, -β is the negative cost added for each action, α is used to adjust the preference of reducing inference error and deploying fewer sensors, and s n is the sensor deployment state realized in the nth step of the Markov decision process, action a n For state s n Next, the next deployment location of the sensor; E n (s n ,a n ) is the global perception model in state s n The inference error under , is calculated as:

[0028]

[0029] in, is the grid sub-region g in the i-th row and j-th column of the training set i,j The baseline monitoring data, According to the current state s n After the sensors are deployed, the global perception model infers the grid sub-area g based on the monitoring data obtained by the deployed sensors. i,j The monitoring data results are as follows, where I and J are the number of rows and columns of the target area divided into multiple grid sub-areas.

[0030] Optionally, determining the sensor deployment strategy in the target area based on the Markov decision process and the deep reinforcement learning model includes:

[0031] In the process of deploying sensors at each step, the trained global perception model is used as a benchmark, the reward function is used to calculate the reward value, and the deep learning model is used to determine the optimal strategy for sensor deployment based on the reward value and the current state of sensor deployment.

[0032] Optionally, the deep reinforcement learning model determines the optimal strategy for sensor deployment with the goal of maximizing the reward value.

[0033] Optionally, the deep reinforcement learning model includes a policy network, and the policy network is used to predict the probability distribution of the next sensor deployment position according to the current state of the sensor deployment; wherein the total loss function of the policy network is:

[0034]

[0035] Where N is the total number of deployable sensors, ε0 is the number of sensors deployed in the initial state, and ε n is the number of sensors deployed in the nth step, λ is a constant, P θ represents the probability distribution function of the sensor deployment action at each step of the policy network output, θ is the policy network parameter, Indicates the current state of sensor deployment s n Next, action a n π is the superiority of a sensor deployment policy relative to other possible actions.

[0036] The present application also provides a regional monitoring device, including:

[0037] A grid division module, used to divide the target area into grids according to the geographic information of the target area and the monitoring granularity requirements, to obtain a target area including a plurality of grid sub-areas;

[0038] An acquisition module is used to obtain monitoring data in some grid sub-areas;

[0039] A preprocessing module, used for preprocessing the geographic information of the target area and the monitoring data in the partial grid sub-areas to obtain geographic information features and monitoring data features;

[0040] The global monitoring module is used to input the geographic information features and monitoring data features into a pre-built global perception model, and the global perception model outputs the monitoring data in all grid sub-areas.

[0041] As can be seen from the above, the regional monitoring method and device provided by the embodiment of the present application include: according to the geographic information of the target area and the monitoring granularity requirements, the target area is gridded to obtain a target area including multiple grid sub-areas, the monitoring data in some grid sub-areas are obtained, the geographic information of the target area and the monitoring data in some grid sub-areas are pre-processed to obtain geographic information features and monitoring data features, the geographic information features and monitoring data features are input into a pre-built global perception model, and the global perception model outputs the monitoring data in all grid sub-areas. The present application can infer the global monitoring data in the target area based on limited monitoring data, reduce costs and system complexity, and ensure the accuracy of the inferred monitoring data. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0043] Figure 1 A schematic diagram of a method flow of an embodiment of the present application;

[0044] Figure 2 This is a schematic diagram of the architecture of a global perception model applied to water quality monitoring in an embodiment of the present application;

[0045] Figure 3 A schematic diagram of the distribution of some monitoring points used for water quality monitoring in an embodiment of the present application;

[0046] Figure 4 for Figure 3 Schematic diagram of the change trend of total nitrogen concentration monitored at the monitoring points shown;

[0047] Figure 5 A schematic diagram of the training principle of the global perception model applied to water quality monitoring in an embodiment of the present application;

[0048] Figure 6 A schematic diagram of the training principle of the sensor deployment model according to an embodiment of the present application;

[0049] Figure 7 This is a schematic diagram of the principle of determining the inference error using the global perception model in an embodiment of the present application;

[0050] Figure 8 This is a block diagram of the device structure of an embodiment of the present application;

[0051] Fig. 9 This is a structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to make the objectives, technical solutions and advantages of the present disclosure more clearly understood, the present disclosure is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0053] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present application should be the usual meanings understood by people with ordinary skills in the field to which the present disclosure belongs. The "first", "second" and similar words used in the embodiments of the present application do not represent any order, quantity or importance, but are only used to distinguish different components. "Including" or "comprising" and similar words mean that the elements or objects appearing in front of the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connecting" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to represent relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0054] As described in the background technology section, when monitoring a large area, the more sensors deployed, the higher the cost and the more complex the system.

[0055] In view of this, an embodiment of the present application provides a regional monitoring method, which is based on the monitoring data of limited monitoring points, comprehensively considers multiple factors that affect the monitoring data, and uses intelligent reasoning to restore the monitoring data of all monitoring points in the target area. There is no need to deploy a large number of sensors, which can reduce costs and system complexity and ensure the accuracy of the monitoring data.

[0056] The technical solution of the present application is further described in detail below through specific embodiments.

[0057] like Figure 1 As shown, the embodiment of the present application provides a regional monitoring method, including:

[0058] S101: According to the geographic information of the target area and the monitoring granularity requirement, the target area is divided into grids to obtain a target area including multiple grid sub-areas;

[0059] In this embodiment, the target area to be monitored is gridded and divided into multiple grid sub-areas. In different application scenarios, the target area corresponds to different areas. For example, in a water quality monitoring scenario, the target area is a river basin within a certain range, and the water quality of the river basin needs to be monitored; in an air monitoring scenario, the target area is an environmental area within a certain range, and the air quality of the environmental area needs to be monitored; in an urban management scenario, the target area is a specific urban area, and the traffic flow of the urban area needs to be monitored, etc. The method of this embodiment does not limit the specific application scenario.

[0060] In some implementations, the target area is gridded according to the geographic information of the target area and the monitoring granularity requirements to obtain a target area including multiple grid sub-areas, including:

[0061] According to the geographic information of the target area and the monitoring granularity requirements, the target area is divided into multiple grid sub-areas of the same size;

[0062] According to the geographic information of the target area, coding information for identifying the geographic information of each grid sub-area is determined.

[0063] In this embodiment, a preset grid division method is used to divide the target area into multiple grid sub-areas of the same size, and then each grid sub-area is encoded according to the geographic information of each grid sub-area, so that each grid sub-area has unique coding information. The coding information is used to identify the geographic information of the grid sub-area, accurately locate the geographic location coordinates of the grid sub-area, and identify the geographic features of the grid sub-area.

[0064] Among them, the size of the grid sub-area can be determined according to the geographic information of the target area and the monitoring granularity requirements. It is understandable that the smaller the monitoring granularity and the higher the monitoring accuracy, the smaller the divided grid sub-area, and the larger the monitoring granularity and the lower the monitoring accuracy, the larger the divided grid sub-area. The geographic information of the target area includes information such as geographic location coordinates (such as longitude and latitude), geographic features, etc. In different application scenarios, the geographic features of the target area are different. It is necessary to combine the actual application scenarios and divide the grid sub-areas according to the geographic features of the target area. For example, when the target area is a river basin of a certain range, it is necessary to consider the geographical features such as the direction of the river in the river basin, whether there are tributaries, and the grade of the tributaries, and divide the river basin into multiple grid sub-areas of appropriate size in combination with the geographical features of the river basin. When the target area is an urban area, it is necessary to consider the geographical features such as the road direction and road grade of the urban area, and divide the urban area into multiple grid sub-areas of appropriate size in combination with the geographical features of the urban area. When the target area is an environmental area, there may be places such as schools and hospitals in the environmental area, and it does not involve complex geographical features such as tributaries and roads. The environmental area can be directly divided into multiple grid sub-areas of the same size and shape according to the monitoring granularity requirements.

[0065] After the target area is divided into a plurality of grid sub-areas, the coding information of each grid sub-area needs to be determined according to the geographic information of the target area.

[0066] In some application scenarios, when the target area is a complex river basin, in order to achieve accurate water quality monitoring, the grid division of the target area needs to fully consider the geographical features such as river flow direction and tributary level. The coded information of the divided grid sub-areas includes geographical information such as longitude and latitude, river flow direction, and tributary level. In some methods, the coded information includes at least three parts, one part is the coded longitude and latitude, the second part is the coded river flow direction, and the third part is the coded tributary level. Among them, for the coding method of river flow direction, a multi-directional coding method based on D8 flow direction can be used to encode the river flow direction of the grid sub-area. The river flow direction in the form of D8 coding can represent one of eight directions. If there are multiple branch flow directions in the river channel in the grid sub-area, the D8 code of each direction can be combined by bitwise OR operation to generate a multi-directional code. For example, the D8 code of the east direction is 1 (binary form is 0001), the D8 code of the south direction is 4 (binary form is 0100), and the D8 code of the southwest direction is 8 (binary form is 1000). The D8 codes of the three directions are bitwise ORed to obtain a multi-directional code 13 (binary form 1101); when decoding, the multi-directional code 13 is bitwise ANDed respectively, and the D8 code forms of 1, 4, and 8 corresponding to the three directions can be decomposed. For the coding method of tributary level, it can be coded as 1-5 according to the size of the tributary level. For example, the first-level tributary is coded as 1, the second-level tributary is coded as 2, and the third-level tributary is coded as 3, etc. The above encoding method of the encoding information of the grid sub-area is only an exemplary description and is not used to limit the protection scope of the present application.

[0067] In other application scenarios, when the target area is an urban area, in order to realize urban traffic flow monitoring, the grid division of the target area needs to fully consider geographical features such as road direction, road level, traffic facilities (traffic lights, traffic signs, etc.), and snow removal priority. The coded information of the divided grid sub-areas includes geographical location information such as longitude and latitude, road direction, road level, and traffic facilities. The coded information at least includes longitude and latitude in coded form, road direction in coded form, road in coded form, traffic facilities in coded form, etc. Through the division of multiple grid sub-areas and the coded information of each grid sub-area, data support can be provided for accurate and detailed inference of traffic flow conditions in urban areas.

[0068] S102: Acquire monitoring data in some grid sub-areas;

[0069] S103: Preprocessing the geographic information of the target area and the monitoring data in some grid sub-areas to obtain geographic information features and monitoring data features;

[0070] In this embodiment, sensors and / or information collection systems are deployed in some grid sub-areas of the target area, and monitoring data in the corresponding grid sub-areas are collected by the deployed sensors and / or information collection systems. The collected monitoring data and geographic information of the target area are preprocessed to obtain monitoring data features and geographic information features suitable for processing by the global perception model, so that the monitoring data features and geographic information features can be input into the global perception model, and the model predicts the monitoring data of all grid sub-areas based on the geographic information of the target area and the monitoring data of some grid sub-areas.

[0071] In some methods, the geographic information and monitoring data of the target area are preprocessed to obtain geographic information features and monitoring data features, including:

[0072] Convert the geographic information of the target area into geographic information features suitable for processing by the global perception model;

[0073] For the collected monitoring data, the erroneous values ​​and abnormal values ​​that are not within the preset standard range are eliminated to obtain the correct monitoring data. Based on the correct monitoring data, the missing values ​​are interpolated to obtain the complete monitoring data. Based on the complete monitoring data, normalization processing is performed to obtain the normalized monitoring data. The normalized monitoring data is converted into monitoring data features suitable for processing by the global perception model.

[0074] Optionally, the missing values ​​can be interpolated using the median interpolation method, and the Z-score method can be used for normalization to convert data of different magnitudes to the same magnitude to ensure comparability between data. The formula of the Z-score method is expressed as:

[0075]

[0076] Among them, x is the indicator item corresponding to the current time slice in the monitoring data, u x is the average value of the indicator item calculated in a certain period of time in the monitoring data, σ x is the variance of the indicator item calculated in a certain period of time in the monitoring data, x std It is a normalized indicator item.

[0077] In some implementations, the monitoring data includes target indicators that need to be monitored in the target area, and associated indicators that affect the target indicators, and the geographic information of the target area includes the coding information of all grid sub-areas. The target indicators and associated indicators can be acquired by sensors or information acquisition systems. In different application scenarios, the target indicators and associated indicators are different, and the geographic information of the target area can be acquired by a positioning system and a geographic information system.

[0078] For example, in the water quality monitoring scenario, the target indicators that need to be monitored include water quality indicators such as pH, dissolved oxygen, water temperature, turbidity, ammonia nitrogen, total nitrogen, total phosphorus, chemical oxygen demand, biochemical oxygen demand, and heavy metal content. The related indicators that affect the above target indicators include season, water pipeline status, plant distribution status, etc. Each target indicator can be collected and obtained using the corresponding sensor, the water pipeline status can be obtained from the water delivery system, and the plant distribution status can be obtained from the plant monitoring system.

[0079] In the urban management scenario, the target indicators that need to be monitored include traffic flow indicators such as vehicle volume, and the related indicators that affect the above target indicators include time period, accident situation, vehicle speed, congestion status, etc. Each target indicator and related indicator can be obtained from the relevant sensors of the vehicle system and the urban road system.

[0080] In the air monitoring scenario, the target indicators that need to be monitored include sulfur dioxide, nitrogen monoxide, hydrocarbons, floating dust and other air indicators. The related indicators that affect the above target indicators include precipitation, temperature, humidity, wind speed, etc. Each target indicator can be collected and obtained using the corresponding sensors, and the related indicators can be obtained from the meteorological system.

[0081] In some embodiments, the target region G is divided into grid sub-regions with I rows and J columns, represented as G = {g i,j |1≤i≤I,1≤j≤J},g i,j is the grid sub-region of the i-th row and j-th column, I is the number of rows, J is the number of columns, and the target region G can also be regarded as a set of grid sub-regions. Since the monitoring data is dynamically changing, the monitoring data can be regarded as a dynamic feature. The monitoring data is periodically collected in each time segment, and the monitoring data features are obtained after preprocessing the monitoring data.

[0082] Among them, the target index item features after preprocessing can be expressed in the form of vector: It represents the target indicator item features of a specific type in the type set K collected by the grid sub-region in the grid sub-region set G in the time segment in the time set T; according to the feature dimension, the target indicator item features can be expressed as Indicates that the type of time segment collection of the grid sub-region in the grid sub-region set G in the time set T is k d The target indicator item characteristics; according to the spatial dimension, the target indicator item characteristics can be expressed as Represents the grid sub-region g i,j The target indicator item features of a specific type in the type set K collected in the time segment of the time set T; divided according to the time dimension, the target indicator item features can be expressed as Represents the grid sub-region in the grid sub-region set G in the time segment T t The target indicator item features of a specific type in the collected type set K.

[0083] The characteristics of the associated index items after preprocessing can be expressed as It represents the associated index item features of a specific type in the type set L collected by the time segment in the time set T of the grid sub-region in the grid sub-region set G; according to the feature dimension, the associated index item features can be expressed as Indicates that the type of time segment collection of the grid sub-region in the grid sub-region set G in the time set T is l e According to the spatial dimension, the characteristics of the associated index items can be expressed as Represents the grid sub-region g i,j The characteristics of the associated index items of a specific type in the type set L collected in the time segment of the time set T; divided according to the time dimension, the characteristics of the associated index items can be expressed as Represents the grid sub-region in the grid sub-region set G in the time segment T t The associated index item features of a specific type in the collected type set L.

[0084] Since the geographical coordinates and geographical features of the target area generally do not change, the geographical information of the target area can be regarded as static features. The preprocessed geographical information features can be expressed as ξ={ξ1,ξ2,…,ξ m},ξ m Represents the mth geographic information feature.

[0085] For example, in the water quality monitoring scenario, the input of the global perception model is (X, V, ξ), where X represents the characteristics of water quality indicators in some grid sub-regions, the covariate V represents the characteristics of water quality related indicators in some grid sub-regions, and the covariate ξ represents the geographic information characteristics of the target area. The model output is the inference result of water quality indicators in all grid sub-regions in the entire target area. In this way, the global perception model can be used to accurately infer the water quality indicators of the entire complex basin based on the water quality indicators of some monitoring points and the external related factors that affect the water quality indicators.

[0086] It should be noted that if Figure 2 , 5As shown in the figure, the monitoring data of all grid sub-areas that can be inferred by the global perception model refer to all grid sub-areas in the target area that can collect and obtain monitoring data. For grid sub-areas that do not need to obtain monitoring data, there is no need to infer monitoring data. For example, when the target area is a river basin, among all grid sub-areas in the target area, the grid sub-areas occupied by the land part do not need to infer water quality indicators; when the target area is an urban area, the grid sub-areas occupied by buildings do not need to infer traffic flow indicators, etc.

[0087] In the urban management scenario, the input of the global perception model is (X, V, ξ), where X represents the traffic flow index item characteristics of some grid sub-areas, and the traffic flow related index item characteristics v can be expressed as represents the grid sub-region g at time segment t i,j The weighted sum of the number and severity of accidents within represents the grid sub-region g at time segment t i,j The average speed of vehicles in represents the grid sub-region g at time segment t i,j Whether it is in the morning rush hour or evening rush hour, ξ represents the geographical information characteristics of the urban area, including road types, distribution of transportation facilities, etc. The output of the model is the traffic flow inference result of all grid sub-areas in the urban area.

[0088] In the air monitoring scenario, the input of the global perception model is (X, V, ξ), where ξ represents the geographic information characteristics of the environmental area, including the type, name, and attributes of the place, X represents the air index item characteristics of some grid sub-areas, and the air-related index item characteristics V can be expressed as represents the grid sub-region g at time segment t i,j Meteorological characteristics such as temperature, humidity, pressure, wind speed, precipitation, etc. represents the grid sub-region g at time segment t i,j The output of the model is the inference result of air index items of all grid sub-areas in the environmental area.

[0089] S104: Input geographic information features and monitoring data features into a pre-built global perception model, and the global perception model outputs monitoring data within all grid sub-areas.

[0090] In this embodiment, after the geographic information features and monitoring data features are obtained through preprocessing, the geographic information features and monitoring data features are input into a pre-built global perception model, and the global perception model is used to predict the monitoring data of the entire target area based on the geographic information of the target area and the monitoring data of the acquired partial area, that is, based on the monitoring data of the limited grid sub-areas, the monitoring data of all grid sub-areas in the target area are restored through intelligent reasoning of the global perception model. In this way, only a limited number of sensors or information collection systems need to be deployed to intelligently infer the monitoring data of the entire target area, thereby reducing costs and system complexity.

[0091] In some embodiments, the training method of the global perception model includes:

[0092] Obtain monitoring data samples within some grid sub-areas;

[0093] Preprocessing the geographic information samples and monitoring data samples of the target area to generate geographic information feature samples and monitoring data feature samples;

[0094] The neural network model is trained based on geographic information feature samples and monitoring data feature samples to obtain a trained global perception model.

[0095] In this embodiment, the global perception model is used to infer the monitoring data of all grid sub-areas based on the geographic information of the target area and the monitoring data of some grid sub-areas. How to mine the temporal and spatial correlation of the monitoring data in different grid sub-areas based on the monitoring data of some grid sub-areas and realize the accurate inference of global monitoring data is the key technology to be solved. The training method of the global perception model includes obtaining monitoring data samples in some grid sub-areas in the target area, preprocessing the geographic information samples and monitoring data samples of the target area, generating geographic information feature samples and monitoring data feature samples, inputting the geographic information feature samples and monitoring data feature samples into the neural network model for training, and obtaining the trained global perception model.

[0096] In some methods, the deployed sensors or information collection systems can be used to obtain monitoring data samples of all grid sub-areas. In the training stage, in order to improve the model effect, as many monitoring data samples of all grid sub-areas as possible can be obtained. The geographic information samples and monitoring data samples of the target area are preprocessed to generate geographic information feature samples and monitoring data feature samples suitable for model processing. The data samples are divided into training sets and validation sets according to a certain ratio for training the neural network model. After training, a global perception model is obtained.

[0097] The training set includes monitoring data samples of all grid sub-areas. The model is trained using a labeled training set. During the training process, the model calculates the loss value based on the output monitoring data of the grid sub-area and the real data corresponding to the grid sub-area in the training set, and adjusts the model parameters based on the loss value. The trained global perception model can output the inferred monitoring data of all grid sub-areas as long as the monitoring data of a limited number of monitoring points is input.

[0098] In some embodiments, to achieve globally accurate inference, the global perception model includes a feature extraction module based on a multidimensional convolution structure and a feature aggregation module based on an attention mechanism. The feature extraction module is used to extract spatiotemporal correlation features based on the geographic information features of the target area and the monitoring data features of some grid sub-areas, and the feature aggregation module is used to infer the monitoring data features of the target area based on the spatiotemporal correlation features.

[0099] In some ways, such as Figure 2 , 5 As shown in the figure, the feature extraction module uses a multi-dimensional convolution kernel to compress the information in the geographic information features and the monitoring data features of some grid sub-areas into a latent vector, extracts the time features and spatial features from the latent vector respectively, and concatenates the time features and spatial features to obtain the spatiotemporal correlation features. Among them, the extracted multi-channel time features are expressed as:

[0100]

[0101] In the formula, f TCNN () represents time convolution, I t Represents the extracted temporal features.

[0102] The extracted multi-channel spatial features are expressed as:

[0103]

[0104] In the formula, f SCNN () represents spatial convolution, I s Represents the extracted spatial features.

[0105] Cascading the temporal and spatial features, we get the spatiotemporal correlation features, which are expressed as:

[0106]

[0107] Among them, I represents the spatiotemporal correlation feature after cascading, || represents the cascade operation of vectors, represents the D-th spatial feature, represents the Dth time feature.

[0108] After the spatiotemporal correlation features are extracted by the feature extraction module, the spatiotemporal correlation features are input into the feature aggregation module based on the attention mechanism. The feature aggregation module automatically learns the importance of each hidden state through the added attention layer, and obtains the output feature H based on the multi-head attention mechanism according to the spatiotemporal correlation features of the current input. The calculation method is:

[0109] H=V s ·σ((I·W1)W2(W3·I) T +b s ) (5)

[0110] Among them, W1, W2, W3, V s ,b s is the parameter of the model, and σ is the activation function.

[0111] After the feature aggregation module calculates the output feature H of feature fusion, it is linearly transformed by the fully connected layer to obtain the final inference result, that is, the global monitoring data of the target area output by the global perception model. Among them, the method of linear transformation of the fully connected layer is:

[0112] Y=linear(H)⊙(1-M)+X⊙M (6)

[0113] Among them, linear(.) is a linear layer, M is a mask matrix, and the elements in the mask matrix are represents the grid sub-area where monitoring data is collected, Represents the grid sub-region where monitoring data needs to be inferred.

[0114] In some embodiments, after obtaining the trained global perception model, the method further includes:

[0115] Determine the sensor deployment strategy in the target area based on Markov decision process and deep reinforcement learning model;

[0116] According to the sensor deployment strategy, sensors are deployed in multiple grid sub-areas within the target area;

[0117] Obtain monitoring data for some grid sub-areas, including:

[0118] For the grid sub-areas where sensors are deployed, the sensors are used to obtain monitoring data in each grid sub-area.

[0119] In this embodiment, after the global perception model is obtained through training, it is necessary to further use the global perception model to determine a reasonable sensor deployment strategy in the target area. It can be understood that although the global perception model can obtain global monitoring data in the target area based on the monitoring data of limited monitoring points through intelligent reasoning, however, considering multiple factors such as economy, system complexity, correlation of monitoring points, and the impact of monitoring points on the inference results of the global perception model, how to reasonably deploy a limited number of sensors in the target area and obtain accurate global monitoring data based on the limited monitoring data obtained by the reasonably deployed sensors through reasoning with the global perception model is a key problem to be solved.

[0120] Combination Figure 3 , 4 As shown in the figure, the changing trend of total nitrogen concentration monitored at several monitoring points in a river basin is presented. The total nitrogen concentration shows obvious dynamic characteristics. The monitoring results of different monitoring points at different time points are different. The changing trend of total nitrogen concentration at each monitoring point is unique, reflecting the complexity of nitrogen cycle in aquatic environment.

[0121] Table 1 Correlation of monitoring data at different monitoring points

[0122] Monitoring point 1 Monitoring point 2 Monitoring point 3 Monitoring point 4 Monitoring point 1 1.000 0.815 0.281 0.594 Monitoring point 2 0.815 1.000 0.179 0.525 Monitoring point 3 0.281 0.179 1.000 0.186 Monitoring point 4 0.594 0.525 0.186 1.000

[0123] As shown in Table 1, the total nitrogen concentrations monitored at different monitoring points have different degrees of correlation. For example, the correlation between the total nitrogen concentrations monitored at monitoring point 1 and monitoring point 2 is 0.815, which is a high correlation, and the correlation between the total nitrogen concentrations monitored at monitoring point 3 and other monitoring points is low. This shows that deploying sensors at different monitoring points has different values, that is, there are differences in the value of the deployment locations of sensors, and the differences in the value of the deployment locations of sensors have a direct relationship to the accuracy and effectiveness of sensor deployment. In addition, the geographic information of different monitoring points directly affects the inference results of the global perception model, so the deployment location of the sensor directly affects the inference results of the global perception model.

[0124] In order to solve the problem of reasonable deployment of sensors in the target area, this embodiment determines the sensor deployment strategy in the target area based on the Markov decision process and the deep reinforcement learning model, and deploys sensors in the corresponding grid sub-areas in the target area according to the determined sensor deployment strategy; then, the monitoring data in some grid sub-areas in step S102 is to use the deployed sensors to obtain the monitoring data of the grid sub-areas where the sensors are deployed, and then based on the obtained limited monitoring data, use the global perception model for intelligent reasoning to obtain the global monitoring data of the target area. In this way, limited monitoring data is obtained based on the limited number of reasonably deployed sensors, and accurate global monitoring data can be inferred using the global perception model.

[0125] Among them, the Markov decision process is a sequential decision problem. Given the current state, the future state transition depends entirely on the current state and will not be affected by the past state. For the sensor deployment problem, before obtaining the final deployment strategy, the future state of the sensor deployment depends only on the current deployment location state.

[0126] The Markov decision process includes constructing the state space, the action space, and the reward function. The state space construction process is to regard the sensor deployment process as a multi-step process, each step realizes the deployment of a specific number of sensors, and the sensor deployment state realized in the nth step is represented as s n , state s n ∈S, state matrix Each element of can be represented as 0 or 1. In the nth step, if the grid sub-area g i,j If sensors are deployed, otherwise

[0127] The process of constructing the action space is to use action a n Indicates state s n Next, the next deployment location of the sensor, is the position of the Qth sensor deployed, for state s n , the corresponding legal action a n ∈A, a legal action in A means that the location selected by the current action cannot have sensors deployed.

[0128] The constructed reward function is expressed as:

[0129] R(s n ,a n )=α(σ-E n (s n ,a n ))-β (7)

[0130] Among them, σ represents the error threshold expected to be achieved, -β is the negative cost added for each action, and α is used to adjust the preference for reducing inference error and deploying fewer sensors. Larger α and smaller β indicate fewer inference errors and larger rewards, and smaller α and larger β indicate more emphasis on the number of deployed sensors.

[0131] E n (s n ,a n ) represents the global perception model in the current state s n The inference error under is calculated as:

[0132]

[0133] Among them, ygi,j represents the grid sub-region g in the training set i,j The baseline monitoring data, Indicates that according to the current state s n After the sensors are deployed, the global perception model infers the grid sub-area g based on the monitoring data obtained by the deployed sensors. i,j The monitoring data results.

[0134] Specific, combined Figure 7 As shown in FIG. 1 , in the Markov decision process, the method of combining the global perception model to determine the reward value of each step of the sensor deployment strategy is to use the monitoring data of all grid sub-areas collected in the training set as the benchmark. In the Markov decision process, after each step of deploying sensors in a specific grid sub-area, the deployed sensors are used to obtain the monitoring data of some grid sub-areas. The global perception model is used to obtain the monitoring data of all grid sub-areas based on the acquired monitoring data and the geographic information of the target area through intelligent reasoning. Then, according to formula (8), the benchmark monitoring data in each grid sub-area and the current state s of the sensor deployment are calculated respectively. n The inference error between the monitoring data of the corresponding grid sub-areas obtained by inference is calculated. After the inference error is calculated, the reward value R (s) that can be obtained in the current state of sensor deployment is calculated according to formula (7) n ,a n ).

[0135] In some methods, deep reinforcement learning models are used to enable the intelligent agent to continuously interact with the environment to learn the optimal strategy to obtain the maximum cumulative reward. That is, in each step of sensor deployment, the reward function is used to calculate the reward value based on the trained global perception model, and the deep learning model is used to determine the optimal strategy for sensor deployment based on the reward value and the current state of sensor deployment.

[0136] Among them, Figure 6 As shown in the figure, the sensor deployment model is trained based on the deep reinforcement learning model. The deep reinforcement learning model includes a policy network and a value network. The policy network is used to predict the probability distribution of the next sensor deployment location based on the current state of the sensor deployment, and the value network is used to evaluate the overall value of the current state of the sensor deployment by estimating the possible cumulative rewards in the future. By training the deep reinforcement learning model of the sensor deployment strategy, the optimal deployment strategy under different sensor numbers and inference accuracy constraints can be achieved, thereby maximizing resource utilization.

[0137] Among them, the deep reinforcement learning model aims to maximize the expectation of long-term returns, that is, to determine the optimal deployment strategy with the goal of maximizing the reward value, and this goal is defined as:

[0138]

[0139] Among them, π is the sensor deployment strategy, R(s n ,a n ) is the reward function, which is related to the inference error and inference stability. That is, the fewer sensors deployed, the lower the inference error, and the greater the reward. Considering action A as a random variable, A~P π Represents the distribution P v (.) in which we sample action A, and the strategy π determines the n The probability distribution of the action selected is, therefore, P v (.) is the distribution of state-action pairs determined by the policy π, represents the probability density function P of action A under strategy π π (A) is the expectation. The above formula sets a goal in policy optimization, which is to find the optimal policy π * , so that under a given strategy π * and the associated state-action distribution P π Maximize the expected reward under .

[0140] In some methods, during the training of the policy network, The network parameters are optimized using the A2C algorithm gradient estimator with an advantage function, expressed as:

[0141]

[0142] Among them, P θ represents the probability distribution function of the sensor deployment action at each step of the policy network output, θ is the policy network parameter, Indicates the current state of sensor deployment s n Next, action a n The superiority of an action relative to other possible actions can be calculated using the advantage function. Indicates that in the current state s n Next, we select strategy π according to the policy network parameters θ θ The logarithmic probability of .

[0143] In order to ensure full exploration of the sensor deployment space, an entropy loss function is introduced into the policy network based on the objective function. The entropy loss function is used to measure the randomness of the policy's action selection in a given state, and encourages the policy network to conduct more exploration in the early stages of training to discover the potential best deployment strategy. Through the expected calculation, entropy is applied to all possible states, thereby ensuring that the policy has exploration capabilities in the entire state space. Among them, the total loss function of the policy network can be expressed as:

[0144]

[0145] In some implementations, the advantage function is estimated in the form of TD-Error, which is the reward value obtained from the current state of the sensor deployment and the estimated value of the state of the sensor deployment after the next deployment action is executed minus the estimated value of the current state of the sensor deployment. The advantage function calculation method for step n is:

[0146]

[0147] Among them, the value function V Φ (s n ) is used to estimate starting from step n, by being in state s n The expected total reward that can be obtained is calculated as follows:

[0148]

[0149] The embodiment of the present application provides a regional monitoring method, taking into account that for a large area, if a large number of sensors are deployed, the cost and system complexity will increase. Therefore, based on the Markov decision process and the deep learning model, the optimal deployment strategy of a limited number of sensors in the target area is determined, and the sensors are deployed in the specific grid sub-area of ​​the target area according to the optimal deployment strategy; the monitoring data of a part of the target area is obtained by using the deployed limited number of sensors, and the geographic information of the target area and the acquired monitoring data of the part of the area are input into the global perception model, and the monitoring data of the target area is inferred by the global perception model, so that the global monitoring data in the target area can be obtained without deploying a large number of sensors, and the data accuracy is high, which can reduce the cost and system complexity.

[0150] It should be noted that the method of the embodiment of the present application can be performed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the embodiment of the present application, and the multiple devices will interact with each other to complete the described method.

[0151] It should be noted that the above is a description of a specific embodiment of the present specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0152] like Figure 8As shown, the embodiment of the present application also provides a regional monitoring device, including:

[0153] A grid division module is used to divide the target area into grids according to the geographic information of the target area and the monitoring granularity requirements, so as to obtain a target area including multiple grid sub-areas;

[0154] An acquisition module is used to obtain monitoring data in some grid sub-areas;

[0155] A preprocessing module is used to preprocess the geographic information of the target area and the monitoring data in some grid sub-areas to obtain geographic information features and monitoring data features;

[0156] The global monitoring module is used to input geographic information features and monitoring data features into a pre-built global perception model, and the global perception model outputs the monitoring data in all grid sub-areas.

[0157] For the convenience of description, the above devices are described in terms of functions divided into various modules. Of course, when implementing the embodiments of the present application, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0158] The apparatus of the above-mentioned embodiment is used to implement the corresponding method in the above-mentioned embodiment, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0159] Fig. 9 A more specific schematic diagram of the hardware structure of an electronic device provided in this embodiment is shown, and the device may include: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are connected to each other through the bus 1050 in the device.

[0160] The processor 1010 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0161] The memory 1020 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 may store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program codes are stored in the memory 1020 and are called and executed by the processor 1010.

[0162] The input / output interface 1030 is used to connect the input / output module to realize information input and output. The input / output module can be configured in the device as a component (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.

[0163] The communication interface 1040 is used to connect a communication module (not shown) to realize communication interaction between the device and other devices. The communication module can realize communication through a wired mode (such as USB, network cable, etc.) or a wireless mode (such as mobile network, WIFI, Bluetooth, etc.).

[0164] The bus 1050 includes a path that transmits information between the various components of the device (eg, the processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040).

[0165] It should be noted that, although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040 and the bus 1050, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include the components necessary for implementing the embodiments of the present specification, and does not necessarily include all the components shown in the figure.

[0166] The electronic device of the above embodiment is used to implement the corresponding method in the above embodiment, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0167] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0168] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present disclosure (including the claims) is limited to these examples. Based on the concept of the present disclosure, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0169] In addition, to simplify the description and discussion, and in order not to make the embodiments of the present application difficult to understand, the known power supply / ground connection with the integrated circuit (IC) chip and other components may or may not be shown in the provided drawings. In addition, the device can be shown in the form of a block diagram to avoid making the embodiments of the present application difficult to understand, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform to be implemented in the embodiments of the present application (that is, these details should be fully within the scope of understanding of those skilled in the art). In the case of elaborating specific details (e.g., circuits) to describe exemplary embodiments of the present disclosure, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details or when these specific details are changed. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0170] Although the present disclosure has been described in conjunction with specific embodiments of the present disclosure, many replacements, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.

[0171] The embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the scope of protection of the present disclosure.

Claims

1. A regional monitoring method, characterized in that: include: According to the geographic information of the target area and the monitoring granularity requirements, the target area is grid-divided to obtain a target area including a plurality of grid sub-areas; Obtain monitoring data within some grid sub-areas; Preprocessing the geographic information of the target area and the monitoring data in the partial grid sub-areas to obtain geographic information features and monitoring data features; The geographic information features and monitoring data features are input into a pre-built global perception model, and the global perception model outputs the monitoring data within all grid sub-areas.

2. The method according to claim 1, characterized in that The training method of the global perception model includes: Obtain monitoring data samples within some grid sub-areas; Preprocessing the geographic information samples and the monitoring data samples of the target area to generate geographic information feature samples and monitoring data feature samples; The neural network model is trained based on the geographic information feature samples and the monitoring data feature samples to obtain a trained global perception model.

3. The method according to claim 2, characterized in that The global perception model includes a feature extraction module and a feature aggregation module; wherein the feature extraction module is used to extract spatiotemporal correlation features based on the geographic information features of the target area and the monitoring data features of some grid sub-areas; the feature aggregation module is used to infer the monitoring data features of the target area based on the spatiotemporal correlation features.

4. According to claim 3, it is characterized in that: The feature aggregation module obtains the output feature H based on the multi-head attention mechanism according to the input spatiotemporal correlation features. The calculation method is: H=V s ·σ((I·W1)W2(W3·I) T +b s ) (5) Among them, W1, W2, W3, V s ,b s is the parameter of the model, σ is the activation function, and I represents the spatiotemporal correlation feature; a linear transformation is performed using a fully connected layer to obtain the monitoring data features of the target area. The method for performing linear transformation on the fully connected layer is: Y=linear(H)⊙(1-M)+X⊙M (6) Among them, linear(.) is a linear layer, M is a mask matrix, and the elements in the mask matrix are represents the grid sub-area where monitoring data is collected, It represents the grid sub-area where the monitoring data needs to be inferred, and X is the target indicator item feature in the preprocessed monitoring data feature sample.

5. The method according to claim 2, characterized in that: After obtaining the trained global perception model, it also includes: Determine a sensor deployment strategy within the target area based on a Markov decision process and a deep reinforcement learning model; Deploy sensors in multiple grid sub-areas within the target area according to the sensor deployment strategy; The obtaining of monitoring data in some grid sub-areas includes: For the grid sub-areas where sensors are deployed, the sensors are used to obtain monitoring data in each grid sub-area.

6. The method according to claim 5, characterized in that In the Markov decision process, the constructed reward function is expressed as: R(s n ,a n )=α(σ-E b (s n ,a n ))-b (7) Among them, σ is the error threshold that is expected to be achieved, -β is the negative cost added for each action, α is used to adjust the preference of reducing inference error and deploying fewer sensors, and s n is the sensor deployment state realized in the nth step of the Markov decision process, action a n For state s n Next, the next deployment location of the sensor; E n (s n ,a n ) is the global perception model in state s n The inference error under , is calculated as: in, is the grid sub-region g in the i-th row and j-th column of the training set i,j The baseline monitoring data, According to the current state s n After the sensors are deployed, the global perception model infers the grid sub-area g based on the monitoring data obtained by the deployed sensors. i,j The monitoring data results are as follows, where I and J are the number of rows and columns of the target area divided into multiple grid sub-areas.

7. The method according to claim 6, characterized in that The step of determining the sensor deployment strategy in the target area based on the Markov decision process and the deep reinforcement learning model includes: In the process of deploying sensors at each step, the trained global perception model is used as a benchmark, the reward function is used to calculate the reward value, and the deep learning model is used to determine the optimal strategy for sensor deployment based on the reward value and the current state of sensor deployment.

8. The method according to claim 7, characterized in that The deep reinforcement learning model determines the optimal strategy for sensor deployment with the goal of maximizing the reward value.

9. The method according to claim 5, characterized in that The deep reinforcement learning model includes a policy network, which is used to predict the probability distribution of the next sensor deployment position according to the current state of the sensor deployment; wherein the total loss function of the policy network is: Where N is the total number of deployable sensors, ε0 is the number of sensors deployed in the initial state, and ε n is the number of sensors deployed in the nth step, λ is a constant, P θ represents the probability distribution function of the sensor deployment action at each step of the policy network output, θ is the policy network parameter, Indicates the current state of sensor deployment s n Next, action a n π is the superiority of a sensor deployment policy relative to other possible actions.

10. A regional monitoring device, characterized in that: include: A grid division module, used to divide the target area into grids according to the geographic information of the target area and the monitoring granularity requirements, to obtain a target area including a plurality of grid sub-areas; An acquisition module is used to obtain monitoring data in some grid sub-areas; A preprocessing module, used for preprocessing the geographic information of the target area and the monitoring data in the partial grid sub-areas to obtain geographic information features and monitoring data features; The global monitoring module is used to input the geographic information features and monitoring data features into a pre-built global perception model, and the global perception model outputs the monitoring data in all grid sub-areas.

Images