Data analysis based internet of things sensor deployment scheme analysis method and system

Abstract

The application provides an Internet of Things sensor deployment scheme analysis method and system based on data analysis, and relates to the technical field of data analysis.In the application, firstly, based on the griding result of a target area, a pending deployment scheme of an Internet of Things sensor is subjected to matrix processing to form a sensor scheme representation matrix;secondly, the sensor scheme representation matrix is converted into a scheme semantic vector;then, the global historical spatial data of the target area are converted into a multi-granularity spatial semantic vector;finally, the scheme semantic vector and the multi-granularity spatial semantic vector are subjected to semantic space conversion and aggregation based on modal weights, and the scheme space aggregation vector obtained through the semantic space conversion and the aggregation based on the modal weights is mapped and output to obtain a target analysis result.Based on the above method, the problem that the reliability of the analysis of the deployment scheme of the Internet of Things sensor is relatively low in the prior art can be improved.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and more specifically, to a method and system for analyzing IoT sensor deployment schemes based on data analysis. Background Technology

[0002] IoT sensors have been widely applied in smart cities, industrial monitoring, environmental governance, smart agriculture, traffic management, and security monitoring. By deploying various types of sensors within a target area, real-time sensing of information such as temperature, humidity, gas concentration, vibration, displacement, light intensity, noise, and human activity can be achieved, providing data support for regional operational status analysis, anomaly warning, and intelligent decision-making. In practical applications, the performance of an IoT system is closely related to the sensor deployment scheme. Different deployment locations, densities, and combinations of sensor types directly affect the area's sensing coverage, data acquisition quality, spatial sensing accuracy, and system construction costs. Therefore, before formal deployment, it is usually necessary to analyze and evaluate the proposed deployment scheme to determine its suitability for the target area.

[0003] However, the inventors' research revealed that existing methods for analyzing deployment schemes using discrete rules or single indicators are insufficient to effectively represent the spatial distribution characteristics in complex areas. Especially in large-scale target areas, where significant spatial correlations and heterogeneities exist between different locations, relying solely on traditional analysis methods cannot accurately reflect the fit between the sensor deployment scheme and the actual spatial characteristics of the area, resulting in relatively low reliability of the analysis. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a data analysis-based method and system for analyzing IoT sensor deployment schemes, so as to improve the problem of relatively low reliability of IoT sensor deployment scheme analysis in the prior art.

[0005] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions: An analysis method for IoT sensor deployment schemes based on data analysis, comprising: Based on the gridding results of the target area, the proposed deployment schemes of IoT sensors are matrixed to form a sensor scheme characterization matrix. Each parameter in the sensor scheme characterization matrix is ​​used to characterize whether the grid at the corresponding location has a sensor and the type of sensor. Convert the sensor scheme representation matrix into a scheme semantic vector; The global historical spatial data of the target area is converted into a multi-granularity spatial semantic vector. The global historical spatial data includes the local historical spatial data corresponding to each sensor in the proposed deployment scheme. In the process of converting the data into the multi-granularity spatial semantic vector, the semantic information of each local historical spatial data is mined at least at two different granularities. The semantic vector of the scheme and the semantic vector of the multi-granularity space are subjected to semantic space transformation and aggregation based on modality weights. The scheme space aggregation vector obtained by semantic space transformation and aggregation based on modality weights is mapped and output to obtain the target analysis result. The target analysis result is used to reflect at least the adaptability of the proposed deployment scheme to the target region.

[0006] In some preferred embodiments, in the above-described data analysis-based IoT sensor deployment scheme analysis method, the step of converting the global historical spatial data of the target area into multi-granularity spatial semantic vectors includes: For each local historical space data in the global historical space data, a coarse-grained historical space vector and multiple fine-grained historical space vectors corresponding to each historical space data segment in the local historical space data are mined, thereby outputting the coarse-grained vector combination and the original fine-grained vector combination corresponding to each local historical space data. For each of the local historical spatial data, under the constraint of the coarse-grained vector combination, the original fine-grained vector combination is forward diffused to output a forward fine-grained vector combination. The original coarse-grained vector cluster is forward diffused to output a forward coarse-grained vector cluster; Under the constraint of the original coarse-grained vector cluster, the original fine-grained vector cluster is forward diffused to output a forward fine-grained vector cluster. The original fine-grained vector cluster includes forward fine-grained vector combinations corresponding to multiple local historical spatial data, and the original coarse-grained vector cluster includes coarse-grained vector combinations corresponding to multiple local historical spatial data. The forward coarse-grained vector cluster and the forward fine-grained vector cluster are aggregated to form a multi-granularity spatial semantic vector corresponding to the global historical spatial data.

[0007] In some preferred embodiments, in the above-described data analysis-based IoT sensor deployment scheme analysis method, the step of forward-diffusion of the original fine-grained vector combination under the constraint of the coarse-grained vector combination for each local historical spatial data, thereby outputting a forward fine-grained vector combination, includes: According to the temporal relationship between the historical spatial data segments in the local historical spatial data, the historical spatial coarse-grained vectors in the coarse-grained vector combination are sorted and associated with each time step of the forward diffusion to form a one-to-one correspondence. At each time step, the corresponding coarse-grained vectors of the historical space are sequentially diffused into the original fine-grained vector combination to achieve semantic constraints and form a forward fine-grained vector combination.

[0008] In some preferred embodiments, in the above-described data analysis-based IoT sensor deployment scheme analysis method, the step of sequentially diffusing the corresponding historical spatial coarse-grained vectors into the original fine-grained vector combination at each time step to achieve semantic constraints and form a forward fine-grained vector combination includes: In each time step, the constraint strength control parameters of the semantic constraints for each time step are determined sequentially based on the corresponding coarse-grained vectors of the historical space. Based on the constraint strength control parameters of semantic constraints at each time step, the corresponding coarse-grained vectors of the historical space are sequentially diffused into the original fine-grained vector combination to achieve semantic constraints and form a forward fine-grained vector combination.

[0009] In some preferred embodiments, in the above-described data analysis-based IoT sensor deployment scheme analysis method, the step of determining the constraint strength control parameters of the semantic constraints at each time step based on the corresponding historical space coarse-grained vectors in sequence includes: In the current time step, the corresponding coarse-grained vector of the historical space is subjected to self-attention processing, and the result of the self-attention processing is subjected to linear mapping and nonlinear activation processing to form the initial control parameters of the semantic constraints of the current time step. The constraint strength control parameters of the semantic constraints at each previous time step are averaged to form historical representative control parameters, where the historical representative control parameter corresponding to the first time step is 0. Based on the original adjustment coefficient corresponding to the current time step and the historical representative control parameter, the target adjustment coefficient corresponding to the current time step is determined. There is a positive correlation between the original adjustment coefficient and the corresponding time step, a positive correlation between the original adjustment coefficient and the target adjustment coefficient, and a negative correlation between the historical representative control parameter and the target adjustment coefficient. Based on the target adjustment coefficient, the initial control parameters are adjusted to form the constraint strength control parameters of the semantic constraints at the current time step.

[0010] In some preferred embodiments, in the above-described data analysis-based IoT sensor deployment scheme analysis method, the step of using the constraint strength control parameters based on semantic constraints at each time step to sequentially diffuse the corresponding historical space coarse-grained vectors into the original fine-grained vector combination to achieve semantic constraints and form a forward fine-grained vector combination includes: In the first time step, any one of the historical space fine-grained vectors in the original fine-grained vector combination is used as the corresponding diffusion vector. In other time steps, the diffusion space fine-grained vector output by the previous time step is used as the corresponding diffusion vector. Based on the historical space coarse-grained vector corresponding to the current time step, the corresponding vector to be diffused is subjected to cross-attention processing to achieve semantic diffusion and form the corresponding cross-attention vector. Based on the constraint strength control parameter of the semantic constraint at the current time step and the negative correlation parameter of the constraint strength control parameter, a weighted residual connection is performed on the cross attention vector and the vector to be diffused, and the fine-grained vector of the diffusion space at the current time step is output. After obtaining the diffusion space fine-grained vector at the last time step, this diffusion space fine-grained vector is determined as the corresponding forward fine-grained vector; The forward fine-grained vectors corresponding to each historical space fine-grained vector in the original fine-grained vector combination are combined to form a forward fine-grained vector combination.

[0011] In some preferred embodiments, in the above-described data analysis-based IoT sensor deployment scheme analysis method, the steps of performing semantic space transformation and modality weight-based aggregation on the scheme semantic vector and the multi-granularity spatial semantic vector, and mapping and outputting the scheme space aggregation vector obtained from the semantic space transformation and modality weight-based aggregation to obtain the target analysis result, include: The semantic vector of the scheme and the semantic vector of the multi-granularity space are respectively subjected to semantic space transformation to form a scheme transformation vector and a multi-granularity space transformation vector; Based on the scheme transformation vector and the multi-granularity spatial transformation vector, the scheme mode weights corresponding to the scheme transformation vector and the spatial mode weights corresponding to the multi-granularity spatial transformation vector are determined. The sensor scheme characterization matrix and the global historical spatial data belong to different modes. The scheme mode weights and the spatial mode weights represent the degree of attention to different modes. Based on the scheme modal weights and the spatial modal weights, the scheme semantic vector and the multi-granularity spatial semantic vector, or the scheme transformation vector and the multi-granularity spatial transformation vector, are weighted and aggregated to form a scheme spatial aggregation vector; The spatial aggregation vector of the proposed scheme is mapped and output to obtain the target analysis results.

[0012] In some preferred embodiments, in the above-described data analysis-based IoT sensor deployment scheme analysis method, the step of determining the scheme mode weights corresponding to the scheme transformation vector and the spatial mode weights corresponding to the multi-granularity spatial transformation vector based on the scheme transformation vector and the multi-granularity spatial transformation vector includes: The scheme transformation vector and the multi-granularity space transformation vector are concatenated to form a scheme space concatenated vector. Then, the scheme space concatenated vector is subjected to self-attention processing to extract the associated semantics within the vector and form a scheme space attention vector. Based on the compressed vector of the scheme spatial attention vector, the scheme transformation vector is subjected to cross-attention processing to extract the semantic association between the two vectors and form a scheme association vector. The scheme association vector is linearly mapped and nonlinearly activated to form the scheme mode weights corresponding to the scheme transformation vector; Based on the scheme modal weights, the spatial modal weights corresponding to the multi-granularity spatial transformation vectors are determined, wherein the spatial modal weights are negatively correlated with the scheme modal weights.

[0013] In some preferred embodiments, in the above-described data analysis-based IoT sensor deployment scheme analysis method, the step of converting the sensor scheme representation matrix into a scheme semantic vector includes: Convolve the sensor scheme submatrix representing whether or not a sensor is present in the sensor scheme characterization matrix to form the first scheme semantic vector. Convolve the sensor scheme submatrix representing the sensor type in the sensor scheme characterization matrix to form the second scheme semantic vector; The semantic vectors of the first and second schemes are concatenated to form a scheme connection vector. Based on the semantic vectors of the first and second schemes respectively, the scheme connection vector is gating and adjusted to form a first scheme gating vector and a second scheme gating vector. The first scheme gating vector and the second scheme gating vector are aggregated to form a scheme semantic vector.

[0014] This invention also provides a data analysis-based IoT sensor deployment scheme analysis system, including a processor and a memory. The memory is used to store a computer program, and the processor is used to execute the computer program to implement the above-described data analysis-based IoT sensor deployment scheme analysis method.

[0015] The IoT sensor deployment scheme analysis method and system based on data analysis provided in this invention firstly, based on the gridding results of the target area, performs matrix processing on the proposed deployment schemes of IoT sensors to form a sensor scheme representation matrix; secondly, the sensor scheme representation matrix is ​​converted into scheme semantic vectors; then, the global historical spatial data of the target area is converted into multi-granularity spatial semantic vectors; finally, the scheme semantic vectors and multi-granularity spatial semantic vectors are subjected to semantic space transformation and modality weight-based aggregation, and the scheme spatial aggregation vectors obtained from semantic space transformation and modality weight-based aggregation are mapped and output to obtain the target analysis results. Based on the above method, on the one hand, by performing matrix processing on the proposed deployment schemes, it is easier to fully capture the distributed semantics of sensors in subsequent semantic mining, resulting in a higher semantic representation capability of the formed scheme semantic vectors. On the other hand, since the semantic information of each local historical spatial data at least at two different granularities is mined, the formed multi-granularity spatial semantic vectors take into account both global and local semantic representations. Furthermore, because the semantic vectors of the two modalities undergo semantic space transformation and modality weight-based aggregation before mapping output, the aggregation accuracy is higher, avoiding semantic mismatch issues caused by directly aggregating semantic vectors of different modalities. This ensures the accuracy of subsequent mapping output and leads to reliable analysis results. Based on this, the solution provided by this invention can improve the relatively low reliability of IoT sensor deployment scheme analysis in existing technologies.

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0017] Figure 1 This is a structural block diagram of an IoT sensor deployment scheme analysis system based on data analysis provided in an embodiment of the present invention.

[0018] Figure 2 This is a flowchart illustrating the steps of the data analysis-based IoT sensor deployment scheme analysis method provided in this embodiment of the invention.

[0019] Figure 3 This is a schematic diagram of the meshing result of the target area provided in an embodiment of the present invention.

[0020] Figure 4 This is a schematic diagram of temperature time-series data and temperature frequency domain data provided in an embodiment of the present invention.

[0021] Figure 5 This is a schematic diagram of the sensor scheme characterization matrix provided in an embodiment of the present invention.

[0022] Figure 6 This is a schematic diagram of the semantic transformation of the sensor scheme characterization matrix provided in an embodiment of the present invention.

[0023] Figure 7 This is a schematic diagram illustrating the semantic transformation of global historical spatial data provided in an embodiment of the present invention.

[0024] Figure 8 This is a schematic diagram illustrating the determination of constraint force control parameters provided in an embodiment of the present invention.

[0025] Figure 9 This is a schematic diagram of forward diffusion provided for an embodiment of the present invention.

[0026] Figure 10 This is a schematic diagram illustrating semantic space transformation and modality weight-based aggregation provided in an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0028] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0029] like Figure 1 As shown, this embodiment of the invention provides a data analysis-based IoT sensor deployment scheme analysis system. The data analysis-based IoT sensor deployment scheme analysis system may include a memory and a processor.

[0030] In detail, the memory and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, they can be electrically connected via one or more communication buses or signal lines. The memory may store at least one software functional module (computer program) that exists in the form of software or firmware. The processor can be used to execute the executable computer program stored in the memory, thereby implementing the data analysis-based IoT sensor deployment scheme analysis method provided in this embodiment of the invention.

[0031] Optionally, the memory may be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The processor may be a general-purpose processor, including a Central Processing Unit (CPU), Network Processor (NP), System on Chip (SoC), etc.; it may also be a Digital Signal Processor (DSP), Application-Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0032] and, Figure 1 The structure shown is for illustrative purposes only. The data analysis-based IoT sensor deployment scheme analysis system may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown may include, for example, a communication unit for exchanging information with other devices. In an alternative example, the data analysis-based IoT sensor deployment scheme analysis system may be a server with data processing capabilities.

[0033] Combination Figure 2 This invention also provides a data analysis-based IoT sensor deployment scheme analysis method, which can be applied to the aforementioned data analysis-based IoT sensor deployment scheme analysis system. The method steps defined in the relevant process of the data analysis-based IoT sensor deployment scheme analysis method can be implemented by the data analysis-based IoT sensor deployment scheme analysis system. The following will describe... Figure 2 The specific process shown will be explained in detail.

[0034] Step S110: Based on the gridding results of the target area, the undetermined deployment scheme of the IoT sensor is matrixed to form a sensor scheme characterization matrix.

[0035] In this embodiment of the invention, the data analysis-based IoT sensor deployment scheme analysis system can perform matrix processing on the proposed deployment schemes of IoT sensors based on the gridding results of the target area, forming a sensor scheme characterization matrix. Each parameter in the sensor scheme characterization matrix characterizes whether the grid at the corresponding location has a sensor and the type of sensor. Combined with... Figure 3 The target area can be divided into multiple grid units, such as a 50m×50m or 100m×100m spatial grid, that is, the side length of a grid unit is 50m or 100m.

[0036] Step S120: Convert the sensor scheme representation matrix into a scheme semantic vector.

[0037] In this embodiment of the invention, after obtaining the sensor scheme representation matrix, the IoT sensor deployment scheme analysis system based on data analysis can convert the sensor scheme representation matrix into a scheme semantic vector. That is, the latent semantics in the sensor scheme representation matrix are extracted and represented in vector form to obtain the scheme semantic vector.

[0038] Step S130: Convert the global historical spatial data of the target area into a multi-granularity spatial semantic vector.

[0039] In this embodiment of the invention, the IoT sensor deployment scheme analysis system based on data analysis can also convert the global historical spatial data of the target area into a multi-granularity spatial semantic vector. That is, the potential semantics in the global historical spatial data are extracted and represented in vector form to obtain the multi-granularity spatial semantic vector. The global historical spatial data includes the local historical spatial data corresponding to each type of sensor in the proposed deployment scheme, and during the conversion into the multi-granularity spatial semantic vector, semantic information of each local historical spatial data point is extracted at least at two different granularities. It should be noted that, in the global historical spatial data, for each type of sensor in the proposed deployment scheme, at least one sensor of that type is deployed in the target area to collect data, forming corresponding local historical spatial data. For example, the proposed deployment scheme includes a temperature sensor, a humidity sensor, a wind speed sensor, and a wind direction sensor. Therefore, the global historical spatial data includes at least one local historical spatial data (i.e., temperature time-series data) corresponding to at least one temperature sensor, at least one local historical spatial data (i.e., humidity time-series data) corresponding to at least one humidity sensor, at least one local historical spatial data (i.e., wind speed time-series data) corresponding to at least one wind speed sensor, and at least one local historical spatial data (i.e., wind direction time-series data) corresponding to at least one wind direction sensor. It should also be noted that for time-series data, latent semantics can be extracted either by normalizing the time-series data and then using one-dimensional convolution, or by performing Fourier transform on the time-series data (e.g.,...). Figure 4 This can convert temperature time-series data into temperature frequency domain data, and then perform two-dimensional convolution on the obtained frequency domain data to extract latent semantics.

[0040] Step S140 involves performing semantic space transformation and modality weight-based aggregation on the scheme semantic vector and the multi-granularity spatial semantic vector, and mapping and outputting the scheme space aggregation vector obtained from the semantic space transformation and modality weight-based aggregation to obtain the target analysis result.

[0041] In this embodiment of the invention, after obtaining the scheme semantic vector and the multi-granularity spatial semantic vector, the IoT sensor deployment scheme analysis system based on data analysis can perform semantic space transformation and modality weight-based aggregation on the scheme semantic vector and the multi-granularity spatial semantic vector, and map and output the scheme space aggregation vector obtained by semantic space transformation and modality weight-based aggregation to obtain the target analysis result. The target analysis result at least reflects the fit between the proposed deployment scheme and the target area. A higher fit indicates a higher reliability of the proposed scheme, and sensors can be deployed in the target area based on the proposed deployment scheme. A lower fit indicates a lower reliability of the proposed scheme; a proposed deployment scheme with excessively low fit cannot be used to deploy sensors in the target area, avoiding ineffective monitoring and wasting resources. It should be noted that since the proposed deployment scheme and the global historical space data belong to different modalities, the corresponding semantic vectors are in different semantic spaces. Thus, by performing semantic space transformation, semantic vectors can be represented in similar semantic spaces. Furthermore, during the aggregation process, modal weights can be applied to improve the accuracy of the aggregation.

[0042] Based on the above method, on the one hand, by matrixing the deployment scheme to be determined, it is easier to fully capture the distributed semantics of sensors in subsequent semantic mining, resulting in a higher semantic representation capability of the formed scheme semantic vector. On the other hand, since the semantic information of each local historical spatial data is mined at least at two different granularities, the resulting multi-granularity spatial semantic vector takes into account both global and local semantic representation. Furthermore, before mapping output, the semantic vectors of the two modalities are subjected to semantic space transformation and aggregation based on modality weights, resulting in higher aggregation accuracy and avoiding semantic mismatch problems caused by directly aggregating semantic vectors of different modalities. This ensures the accuracy of subsequent mapping output and thus obtains reliable analysis results. Based on this, the scheme provided by the embodiments of the present invention can improve the problem of relatively low reliability in the analysis of IoT sensor deployment schemes in the prior art.

[0043] Based on the above embodiments, the analysis method for IoT sensor deployment scheme based on data analysis can be further explained, such as being divided into the following four parts.

[0044] In the first part, what needs further explanation for step S110 is that the specific implementation process of matrix processing of the pending deployment scheme of IoT sensors is not restricted, and different selections and configurations can be made according to different needs.

[0045] For example, in one feasible implementation, in order to effectively characterize the desired deployment scheme through the formed sensor scheme characterization matrix, during the matrixing process, it can be determined whether each grid cell has a sensor based on the desired partial scheme, and if a sensor is present, the type of sensor is determined, such as a temperature sensor, humidity sensor, wind speed sensor, wind direction sensor, etc., with different choices possible based on different IoT monitoring needs. Furthermore, it should be noted that in the sensor scheme characterization matrix, combined with... Figure 5 The presence of a sensor can be represented by the value "1", while the absence of a sensor can be represented by the value "0". Furthermore, the type of sensor can be represented by a value greater than 0 and less than 1, such as "0.1" for a temperature sensor, "0.2" for a humidity sensor, "0.3" for a wind speed sensor, and "0.4" for a wind direction sensor. Specifically, two different sensor scheme sub-matrices can be used to represent the presence and type of sensor, respectively. It should also be noted that in the sensor scheme sub-matrices representing sensor types, if a corresponding grid cell has no sensor, it can be represented by the value "0".

[0046] In the second part, what needs further explanation for step S120 is that the specific implementation process of converting the sensor scheme representation matrix into a scheme semantic vector is not limited and can be selected and configured differently according to different needs.

[0047] For example, in one feasible implementation, in order to improve the efficiency of converting the representation matrix into a semantic vector and reduce the consumption of computing resources, the sensor scheme representation matrix can be directly convolved, such as by using a convolutional neural network, to obtain the scheme semantic vector.

[0048] For example, in another possible implementation, in order to improve the accuracy of converting the representation matrix into a semantic vector, so as to fully extract the latent semantics in the sensor scheme representation matrix and improve the semantic representation accuracy of the formed scheme semantic vector, the above step S120 may further include the detailed contents corresponding to the following steps S121, S122, S123 and S124 respectively.

[0049] Step S121: Convolve the sensor scheme submatrix representing whether or not a sensor is present in the sensor scheme representation matrix to form the first scheme semantic vector.

[0050] In this embodiment of the invention, combined with Figure 6The sensor scheme submatrix representing whether or not a sensor is present in the sensor scheme representation matrix can be convolved to form a first scheme semantic vector. This convolution can be implemented using a convolutional neural network (CNN). It should be noted that in some implementations, the convolution result can also be processed by pooling (e.g., Max Pooling) and non-linear activation (e.g., ReLU).

[0051] Step S122: Convolve the sensor scheme submatrix representing the sensor type in the sensor scheme characterization matrix to form the second scheme semantic vector.

[0052] In this embodiment of the invention, the sensor scheme submatrix representing the sensor type in the sensor scheme characterization matrix can also be convolved to form a second scheme semantic vector.

[0053] Step S123: Connect the first scheme semantic vector and the second scheme semantic vector to form a scheme connection vector; and perform gating adjustment on the scheme connection vector based on the first scheme semantic vector and the second scheme semantic vector respectively to form a first scheme gating vector and a second scheme gating vector.

[0054] In this embodiment of the invention, after obtaining the first scheme semantic vector and the second scheme semantic vector, since the two semantic vectors represent the presence / absence and type of the sensor, respectively, belonging to different dimensions of semantic information, the first scheme semantic vector and the second scheme semantic vector can be connected (to achieve preliminary fusion of semantic information of different dimensions, such as by performing calculations such as averaging or summing to achieve connection), forming a scheme connection vector. Furthermore, based on the first scheme semantic vector and the second scheme semantic vector, the scheme connection vector is gated to form a first scheme gate vector and a second scheme gate vector. The gated adjustment method can refer to relevant existing technologies. For example, the first scheme semantic vector can be gated to obtain a corresponding gate weight. Then, the gate weight is multiplied by the scheme connection vector to form the first scheme gate vector. This allows the formed first scheme gate vector to represent both dimensions of semantic information while emphasizing or tending to represent the semantic information of the presence / absence of the sensor. Similarly, the formed second scheme gate vector represents both dimensions of semantic information while emphasizing or tending to represent the semantic information of the sensor type.

[0055] Step S124: Aggregate the first scheme gate vector and the second scheme gate vector to form a scheme semantic vector.

[0056] In this embodiment of the invention, after obtaining the first scheme gating vector and the second scheme gating vector, they can be aggregated (e.g., simple aggregation such as averaging or summing, or convolution and pooling after concatenation to further capture abstract and complex semantic information) to form a scheme semantic vector. Based on this, since both the first and second scheme gating vectors can represent semantic information in two dimensions, the richness of semantic information representation can be considered. Furthermore, since the first and second scheme gating vectors focus on representing semantic information in different dimensions, the precision or accuracy of semantic information representation can also be considered. Thus, the semantic representation capability of the formed scheme semantic vector can be fully guaranteed.

[0057] In the third part, what needs further explanation for step S130 is that the specific implementation process of converting the global historical spatial data of the target area into multi-granularity spatial semantic vectors is not limited and can be selected and configured differently according to different needs.

[0058] For example, in one feasible implementation, in order to improve the efficiency of converting spatial data into semantic vectors and reduce the consumption of computing resources, each local historical spatial data in the global historical spatial data can be convolved (either a one-dimensional convolution in the time domain or a two-dimensional convolution in the frequency domain) to obtain the corresponding local convolution vector. Then, the local convolution vectors can be aggregated by averaging, summing, or concatenating to obtain a coarse-grained vector. Then, the coarse-grained vector can be downsampled and self-attention processed to obtain a fine-grained vector. Finally, the coarse-grained vector and the fine-grained vector can be aggregated by averaging, summing, or concatenating to obtain a multi-granularity spatial semantic vector.

[0059] For example, in another possible implementation, in order to improve the reliability of the conversion into semantic vectors, that is, to ensure the accuracy of semantic information representation while achieving semantic representation richness through semantic information of multiple granularities, the above step S130 may further include the detailed contents corresponding to the following steps S131, S132, S133, S134 and S135 respectively.

[0060] Step S131: For each local historical spatial data in the global historical spatial data, mine a historical spatial coarse-grained vector and multiple historical spatial fine-grained vectors corresponding to each historical spatial data segment in the local historical spatial data, and output the coarse-grained vector combination and the original fine-grained vector combination corresponding to each local historical spatial data.

[0061] In this embodiment of the invention, combined with Figure 7 For each local historical spatial data in the global historical spatial data, a coarse-grained historical spatial vector and multiple fine-grained historical spatial vectors corresponding to each historical spatial data segment in the local historical spatial data are mined, thereby outputting the coarse-grained vector combination and the original fine-grained vector combination corresponding to each local historical spatial data. For example, local historical spatial data A includes historical spatial data segment 1 (such as temperature data on the first day) and historical spatial data segment 2 (such as temperature data on the second day). By performing convolution processing on historical spatial data segment 1 as a whole (either by directly performing one-dimensional convolution on the time-domain data or by converting the time-domain data to the frequency domain and then performing two-dimensional convolution), a coarse-grained historical spatial vector (focusing on the representation of global semantics) corresponding to historical spatial data segment 1 can be obtained. In addition, historical spatial data segment 1 can be segmented into multiple window data, and then convolution processing can be performed on each window data separately to form multiple fine-grained historical spatial vectors corresponding to historical spatial data segment 1 (focusing on the representation of local semantics). Similarly, by performing convolution processing on historical spatial data segment 2 as a whole (either by directly performing one-dimensional convolution on the time-domain data or by converting the time-domain data to the frequency domain and then performing two-dimensional convolution), a coarse-grained historical spatial vector (focusing on the representation of local semantics) can be obtained. After converting the data from the domain to the frequency domain, two-dimensional convolution is performed to obtain the historical spatial coarse-grained vector corresponding to historical spatial data segment 2 (focusing on the representation of global semantics). In addition, historical spatial data segment 2 can be segmented by sliding window to form multiple window data. Then, convolution processing can be performed on each window data to form multiple historical spatial fine-grained vectors corresponding to historical spatial data segment 2 (focusing on the representation of local semantics). Then, the historical spatial coarse-grained vectors corresponding to historical spatial data segment 1 and historical spatial data segment 2 can be combined to form the coarse-grained vector combination corresponding to local historical spatial data A. And, the historical spatial fine-grained vectors corresponding to historical spatial data segment 1 and historical spatial data segment 2 can be combined to form the fine-grained vector combination corresponding to local historical spatial data A.

[0062] Step S132: For each of the local historical spatial data, under the constraint of the coarse-grained vector combination, the original fine-grained vector combination is forward diffused to output a forward fine-grained vector combination.

[0063] In this embodiment of the invention, after obtaining the coarse-grained vector combination and the original fine-grained vector combination, for each local historical spatial data, the original fine-grained vector combination can be forward-diverged under the constraint of the coarse-grained vector combination to output a forward-diverged fine-grained vector combination. Forward-divergence can refer to deep semantic extraction, obtaining deep semantics, complex semantics, high-level semantics, or abstract semantics. It should be noted that semantic distortion may occur during deep extraction; therefore, the coarse-grained vector combination can be used as a constraint to ensure the reliability of deep extraction.

[0064] Step S133: Forward diffusion is performed on the original coarse-grained vector cluster to output a forward coarse-grained vector cluster.

[0065] In this embodiment of the invention, after obtaining the forward fine-grained vector combination, the original coarse-grained vector cluster can be forward-diverged to output a forward coarse-grained vector cluster. The original coarse-grained vector cluster includes multiple coarse-grained vector combinations corresponding to local historical spatial data. It should be noted that during the forward diffraction process, the forward diffraction of the original coarse-grained vector cluster can be constrained based on the original coarse-grained vector cluster itself, thereby obtaining the forward coarse-grained vector cluster, i.e., capturing the deep semantics, complex semantics, high-level semantics, or abstract semantics in the coarse-grained vectors. Furthermore, the method for forward diffraction of the original coarse-grained vector cluster can be the same as the method for forward diffraction of the original fine-grained vector combination.

[0066] Step S134: Under the constraint of the original coarse-grained vector cluster, the original fine-grained vector cluster is forward diffused to output a forward fine-grained vector cluster.

[0067] In this embodiment of the invention, the original fine-grained vector cluster can be forward-diverged under the constraints of the original coarse-grained vector cluster to output a forward fine-grained vector cluster. The original fine-grained vector cluster includes multiple forward fine-grained vector combinations corresponding to local historical spatial data. It should be noted that since the forward fine-grained vector combination is formed by diffusion under the constraints of the coarse-grained vector combination, the semantics of the constraint is still within the global dimension of the local historical spatial data, meaning its globality is not high. In reality, there may still be potential semantic relationships between different local historical spatial data (i.e., between different sensors). Therefore, constraints can also be applied from the higher global dimension of the original coarse-grained vector cluster (i.e., the global historical spatial data) to obtain a forward fine-grained vector cluster with higher constraint accuracy. Furthermore, the method of forward-diverging the original fine-grained vector cluster can be the same as the method of forward-diverging the original fine-grained vector combination.

[0068] Step S135: Aggregate the forward coarse-grained vector cluster and the forward fine-grained vector cluster to form a multi-granular spatial semantic vector corresponding to the global historical spatial data.

[0069] In this embodiment of the invention, after obtaining the forward coarse-grained vector cluster and the forward fine-grained vector cluster, they can be aggregated to form a multi-granularity spatial semantic vector corresponding to the global historical spatial data. It should be noted that since both the forward coarse-grained vector cluster and the forward fine-grained vector cluster are actually semantic vectors formed under global constraints within the dimension of the global historical spatial data, they can be directly aggregated using methods such as averaging and summing to form a multi-granularity spatial semantic vector. Thus, the forward coarse-grained vectors in the forward coarse-grained vector cluster and the forward fine-grained vectors in the forward fine-grained vector cluster can be processed using methods such as averaging and summing.

[0070] Furthermore, in steps S132, S133, and S134 described above, the specific process of forward diffusion is not limited and can be selected and configured differently according to different needs. Since the forward diffusion method can be the same in each step, in this embodiment of the invention, step S132 is used as an example to explain the specific process of forward diffusion.

[0071] For example, in one feasible implementation, considering that there is a temporal relationship between different historical spatial data fragments, and that this relationship is important for capturing semantic information, for example, the semantic information is more important the later the time, the more important the semantic information is. Therefore, the above step S132 can further include the detailed contents corresponding to the following steps S132a and S132b respectively.

[0072] Step S132a: According to the temporal relationship between the historical spatial data segments in the local historical spatial data, sort the historical spatial coarse-grained vectors in the coarse-grained vector combination, and associate them with the forward diffusion time steps to form a one-to-one correspondence.

[0073] In this embodiment of the invention, the historical spatial coarse-grained vectors in the coarse-grained vector combination can be sorted according to the temporal order of the historical spatial data segments in the local historical spatial data, and associated with each time step of the forward diffusion to form a one-to-one correspondence. For example, the historical spatial coarse-grained vector corresponding to the earliest historical spatial data segment corresponds to the first time step, and the historical spatial coarse-grained vector corresponding to the latest historical spatial data segment corresponds to the last time step.

[0074] In step S132b, at each time step, the corresponding historical space coarse-grained vectors are sequentially diffused into the original fine-grained vector combination to achieve semantic constraints and form a forward fine-grained vector combination.

[0075] In this embodiment of the invention, after associating time steps with historical space coarse-grained vectors, the corresponding historical space coarse-grained vectors can be sequentially diffused into the original fine-grained vector combination at each time step to achieve semantic constraints and form a forward fine-grained vector combination. For example, in the first time step, the historical space coarse-grained vector corresponding to the earliest historical space data segment can be diffused into the original fine-grained vector combination; in the last time step, the historical space coarse-grained vector corresponding to the latest historical space data segment can be diffused into the original fine-grained vector combination.

[0076] Furthermore, in step S132b above, the specific implementation process of sequentially diffusing the corresponding historical space coarse-grained vectors into the original fine-grained vector combination is not limited. For example, in one feasible implementation, in order to improve the accuracy of semantic diffusion, step S132b above may further include the contents corresponding to steps b1 and b2 respectively.

[0077] Step b1: In each time step, the constraint strength control parameters of the semantic constraints for each time step are determined sequentially based on the corresponding coarse-grained vectors of the historical space.

[0078] In this embodiment of the invention, at each time step, the constraint strength control parameter for the semantic constraints at each time step is determined sequentially based on the corresponding coarse-grained vectors of the historical space. The constraint strength control parameter is used to constrain the strength of semantic diffusion, achieving different levels of constraint strength at different time steps. This ensures that the constraints are more easily adapted to the actual semantic diffusion situation, thereby guaranteeing the accuracy of semantic diffusion.

[0079] Step b2: Based on the constraint strength control parameters of the semantic constraints at each time step, the corresponding coarse-grained vectors of the historical space are sequentially diffused into the original fine-grained vector combination to achieve semantic constraints and form a forward fine-grained vector combination.

[0080] In this embodiment of the invention, after obtaining the constraint strength control parameters, the corresponding historical space coarse-grained vectors can be sequentially diffused into the original fine-grained vector combination based on the constraint strength control parameters of the semantic constraints at each time step, so as to realize semantic constraints and form a forward fine-grained vector combination.

[0081] Furthermore, in step b1 above, the specific implementation process of determining the constraint strength control parameters of semantic constraints at each time step is not limited. For example, in one feasible implementation, in order to improve the accuracy of semantic diffusion, step b1 above may further include the detailed contents corresponding to the following steps b11, b12, b13 and b14 respectively.

[0082] Step b11: In the current time step, the corresponding historical space coarse-grained vector is subjected to self-attention processing, and the result of the self-attention processing is subjected to linear mapping and nonlinear activation processing to form the initial control parameters of the semantic constraints of the current time step.

[0083] In this embodiment of the invention, combined with Figure 8 In the current time step, the corresponding coarse-grained vector of the historical space undergoes self-attention processing. The result of this self-attention processing is then subjected to linear mapping and nonlinear activation processing to form the initial control parameters for the semantic constraints of the current time step. For example, in the first time step, constraints need to be applied based on the first coarse-grained vector of the historical space. Therefore, the first coarse-grained vector of the historical space can be processed to form the initial control parameters for the semantic constraints of the first time step. Through self-attention processing, important semantic information in the coarse-grained vector of the historical space can be captured. Then, the captured important semantic information is linearly mapped and nonlinearly activated to form the corresponding initial control parameters. These initial control parameters are related to the important semantic information in the coarse-grained vector of the historical space. Thus, when constraints are applied based on these initial control parameters, they are actually based on the important semantic information in the coarse-grained vector of the historical space, reducing the interference caused by unimportant semantic information in the coarse-grained vector of the historical space on the constraints. Furthermore, linear mapping can be implemented using a fully connected network, and nonlinear activation processing can be implemented using functions such as sigmoid. The vector output by the fully connected network has a size of 1. 1. Then, it can be mapped to a parameter belonging to 0-1 through functions such as sigmoid, that is, the initial control parameter.

[0084] Step b12: Average the constraint strength control parameters of the semantic constraints at each previous time step to form historical representative control parameters.

[0085] In this embodiment of the invention, the constraint strength control parameters of the semantic constraints at previous time steps can be averaged to form historical representative control parameters. Specifically, the historical representative control parameter for the first time step is 0, the historical representative control parameter for the second time step is the constraint strength control parameter of the semantic constraints at the first time step, and the historical representative control parameter for the third time step is the average of the constraint strength control parameters of the semantic constraints at the first and second time steps. In other words, the constraint strength control parameters of the previous time steps are fused to represent the historical constraint conditions.

[0086] Step b13: Based on the original adjustment coefficient corresponding to the current time step and the historical representative control parameters, determine the target adjustment coefficient corresponding to the current time step.

[0087] In this embodiment of the invention, after obtaining the historical representative control parameters, the target adjustment coefficient corresponding to the current time step can be determined based on the original adjustment coefficient corresponding to the current time step and the historical representative control parameters. There is a positive correlation between the original adjustment coefficient and the corresponding time step; that is, as the time step increases, the original adjustment coefficient also increases. This is related to the time corresponding to the historical space data segment of the historical space coarse-grained vector used as the constraint basis. That is, the later the time, the more relevant the corresponding historical space coarse-grained vector is to the current situation, the more important the semantic information, and the more important the constraint. There is a positive correlation between the original adjustment coefficient and the target adjustment coefficient, and a negative correlation between the historical representative control parameters and the target adjustment coefficient. For example, the target adjustment coefficient = original adjustment coefficient - historical representative control parameters. It should be noted that the historical representative control parameters represent the overall strength of constraints applied historically. A larger overall strength indicates a deeper constraint, causing the forward diffusion results to focus on the representation of coarse-grained semantics. However, what is actually needed is fine-grained semantics; it's just that fine-grained semantics require constraints from coarse-grained semantics. Therefore, by setting the historical representative control parameters and the target adjustment coefficient to a negative correlation, a balance of constraint intensity across all time steps can be achieved. That is, if the constraint intensity is greater in the early stages, the constraint intensity in the later stages can be reduced accordingly, and if the constraint intensity is smaller in the early stages, the constraint intensity in the later stages can be increased accordingly.

[0088] Step b14: Based on the target adjustment coefficient, adjust the initial control parameters to form the constraint strength control parameters of the semantic constraint at the current time step.

[0089] In this embodiment of the invention, after obtaining the target adjustment coefficient, the initial control parameters can be adjusted based on the target adjustment coefficient to form the constraint strength control parameters of the semantic constraints at the current time step. For example, the target adjustment coefficient and the initial control parameters can be multiplied to obtain the corresponding constraint strength control parameters. Based on this, it can be determined that the historical strength control parameters of the semantic constraints at the current time step are related to important semantic information in the corresponding historical coarse-grained vector (positive correlation), related to the time step itself (positive correlation), and related to the historical constraint strength control parameters (negative correlation).

[0090] Furthermore, in step b2 above, the specific implementation process of sequentially diffusing the corresponding historical space coarse-grained vectors into the original fine-grained vector combination is not limited. For example, in one feasible implementation, in order to improve the accuracy of semantic diffusion, step b2 above may further include the detailed contents corresponding to the following steps b21, b22, b23, b24 and b25 respectively.

[0091] Step b21: In the first time step, any historical space fine-grained vector in the original fine-grained vector combination is used as the corresponding diffusion vector. In other time steps, the diffusion space fine-grained vector output by the previous time step is used as the corresponding diffusion vector.

[0092] In this embodiment of the invention, combined with Figure 9 In the first time step, any historical space fine-grained vector from the original fine-grained vector combination can be used as the corresponding vector to be diffused. In other time steps, the diffused space fine-grained vector output from the previous time step is used as the corresponding vector to be diffused. It should be noted that, in fact, each historical space fine-grained vector in the original fine-grained vector combination undergoes semantic forward diffusion separately. Therefore, each historical space fine-grained vector has a corresponding first time step and other time steps.

[0093] Step b22: Based on the historical space coarse-grained vector corresponding to the current time step, perform cross-attention processing on the corresponding vector to be diffused to achieve semantic diffusion and form the corresponding cross-attention vector.

[0094] In this embodiment of the invention, after determining the vector to be diffused at the current time step, cross-attention processing can be performed on the corresponding vector to be diffused based on the historical space coarse-grained vector corresponding to the current time step to achieve semantic diffusion and form a corresponding cross-attention vector. That is, semantic information with a correlation to the historical space coarse-grained vector can be extracted from the vector to be diffused, thereby achieving the constraint of the vector to be diffused based on the historical space coarse-grained vector to obtain the corresponding cross-attention vector. Specifically, the historical space coarse-grained vector can be mapped to a query vector, and the vector to be diffused can be mapped to a key vector and a value vector; subsequent calculation processes can refer to relevant prior art.

[0095] Step b23: Based on the constraint strength control parameter of the semantic constraint at the current time step and the negative correlation parameter of the constraint strength control parameter, perform a weighted residual connection on the cross attention vector and the vector to be diffused, and output the fine-grained vector of the diffusion space at the current time step.

[0096] In this embodiment of the invention, after obtaining the cross-attention vector corresponding to the current time step, a weighted residual connection can be performed on the cross-attention vector and the vector to be diffused based on the constraint strength control parameter of the semantic constraint of the current time step and the negative correlation parameter of the constraint strength control parameter, to output the fine-grained vector of the diffusion space of the current time step. For example, the fine-grained vector of the diffusion space of the current time step = the cross-attention vector corresponding to the current time step. The constraint strength control parameter of the semantic constraint at the current time step + the vector to be diffused at the current time step The negative correlation parameter of the constraint strength control parameter of the semantic constraint at the current time step is calculated as follows: the sum of the constraint strength control parameter of the semantic constraint at the current time step and the negative correlation parameter of the constraint strength control parameter of the semantic constraint at the current time step equals 1. In other words, the larger the constraint strength control parameter of the semantic constraint at the current time step, the more important it is to the fine-grained vector of the diffusion space formed at the current time step; conversely, the smaller the constraint strength control parameter of the semantic constraint at the current time step, the less important it is to the fine-grained vector of the diffusion space formed at the current time step, and the more important the vector to be diffused at the current time step.

[0097] Step b24: After obtaining the diffusion space fine-grained vector of the last time step, determine the diffusion space fine-grained vector as the corresponding forward fine-grained vector.

[0098] In this embodiment of the invention, after obtaining the diffusion space fine-grained vector of the last time step, the diffusion space fine-grained vector can be determined as the corresponding forward fine-grained vector.

[0099] Step b25: Combine the forward fine-grained vectors corresponding to each historical space fine-grained vector in the original fine-grained vector combination to form a forward fine-grained vector combination.

[0100] In this embodiment of the invention, after obtaining the forward fine-grained vector, the forward fine-grained vectors corresponding to each historical space fine-grained vector in the original fine-grained vector combination can be combined to form a forward fine-grained vector combination. As mentioned above, each historical space fine-grained vector in the original fine-grained vector combination undergoes forward diffusion under constraints based on the coarse-grained vector combination. Therefore, the forward fine-grained vectors formed by forward diffusion of each historical space fine-grained vector can be combined to form a forward fine-grained vector combination.

[0101] In the fourth part, what needs further explanation for step S140 is that the specific implementation process of performing semantic space transformation and aggregation and mapping output based on modality weights to obtain the target analysis results is not limited and can be selected and configured differently according to different needs.

[0102] For example, in one feasible implementation, in order to ensure that the formed scheme space aggregation vector has a high semantic representation capability, thereby improving the reliability of the target analysis results of the mapping output, the above step S140 may further include the detailed contents corresponding to the following steps S141, S142, S143 and S144 respectively.

[0103] Step S141: Perform semantic space transformation on the scheme semantic vector and the multi-granularity space semantic vector respectively to form a scheme transformation vector and a multi-granularity space transformation vector.

[0104] In this embodiment of the invention, combined with Figure 10 The semantic space transformation can be performed on the scheme semantic vector and the multi-granularity space semantic vector respectively to form a scheme transformation vector and a multi-granularity space transformation vector. That is, the scheme semantic vector can be semantically transformed to form a scheme transformation vector, and the multi-granularity space semantic vector can be semantically transformed to form a multi-granularity space transformation vector. This allows the scheme transformation vector and the multi-granularity space transformation vector to be represented in similar semantic spaces, thus improving the semantic mismatch problem that easily occurs when semantically aggregating data from different modalities. Furthermore, the semantic space transformation can be achieved through a linear mapping function, such as the scheme semantic vector being represented by "F(X1) = A1". X1+B1” can be implemented, and the multi-granularity spatial transformation vector can be achieved through “F(X2)=A2”. The process of "X2+B2" is implemented, where A1, B1, A2, and B2 can be formed during the learning and training process of the corresponding neural network model. In other words, each step in the embodiments of this invention can be implemented through the neural network model.

[0105] Step S142: Based on the scheme transformation vector and the multi-granularity spatial transformation vector, determine the scheme mode weights corresponding to the scheme transformation vector and the spatial mode weights corresponding to the multi-granularity spatial transformation vector.

[0106] In this embodiment of the invention, after obtaining the scheme transformation vector and the multi-granularity spatial transformation vector, the scheme mode weights corresponding to the scheme transformation vector and the spatial mode weights corresponding to the multi-granularity spatial transformation vector can be determined based on the scheme transformation vector and the multi-granularity spatial transformation vector. The sensor scheme representation matrix and the global historical spatial data belong to different modes, and the scheme mode weights and spatial mode weights represent the degree of attention paid to different modes. It should be noted that, in this embodiment of the invention, it is essentially still a scheme evaluation; therefore, the degree of attention to the semantic information corresponding to the sensor scheme representation matrix should be higher. Thus, constraints can be placed on the scheme mode weights and the spatial mode weights, such as limiting the scheme mode weights to be greater than the spatial mode weights.

[0107] Step S143: Based on the scheme modal weights and the spatial modal weights, the scheme semantic vector and the multi-granularity spatial semantic vector, or the scheme transformation vector and the multi-granularity spatial transformation vector, are weighted and aggregated to form a scheme spatial aggregation vector.

[0108] In this embodiment of the invention, after obtaining the scheme modal weights and the spatial modal weights, the scheme semantic vector and the multi-granularity spatial semantic vector, or the scheme transformation vector and the multi-granularity spatial transformation vector, can be weighted and aggregated based on the scheme modal weights and the spatial modal weights to form a scheme spatial aggregate vector. For example, the scheme spatial aggregate vector = scheme modal weights. Scheme semantic vector + spatial modality weights Multi-granularity spatial semantic vector, or, solution space aggregation vector = solution modal weights Scheme Transformation Vector + Spatial Modal Weights Multi-granularity spatial transformation vector.

[0109] Step S144: Map the aggregated vector of the scheme space and output the target analysis result.

[0110] In this embodiment of the invention, after obtaining the scheme space aggregation vector, the scheme space aggregation vector can be mapped and output to obtain the target analysis result. The mapping and output process can refer to relevant existing technologies; for example, the scheme space aggregation vector can be fully connected to obtain a 1... A vector of 1 is obtained, and then the vector is processed by functions such as sigmoid to obtain the target analysis result. Alternatively, one parameter of the vector can be directly determined as the target analysis result. The larger the parameter, the higher the fit.

[0111] Furthermore, in step S142 above, the specific implementation process of determining the scheme mode weights corresponding to the scheme transformation vector and the spatial mode weights corresponding to the multi-granularity spatial transformation vector is not limited. For example, in one feasible implementation, in order to improve the reliability of the mode weight determination, step S142 above may further include the detailed contents corresponding to the following steps S142a, S142b, S142c and S142d respectively.

[0112] Step S142a: The scheme transformation vector and the multi-granularity space transformation vector are concatenated to form a scheme space concatenated vector. Then, the scheme space concatenated vector is subjected to self-attention processing to extract the associated semantics within the vector and form a scheme space attention vector.

[0113] In this embodiment of the invention, the scheme transformation vector and the multi-granularity space transformation vector can be concatenated to form a scheme space concatenation vector, such as (scheme transformation vector; multi-granularity space transformation vector). Furthermore, the scheme space concatenation vector can be subjected to self-attention processing to extract the associated semantics within the vector, thereby forming a scheme space attention vector.

[0114] Step S142b: Based on the compressed vector of the scheme spatial attention vector, perform cross-attention processing on the scheme transformation vector to extract the semantic association between the two vectors and form a scheme association vector.

[0115] In this embodiment of the invention, after obtaining the scheme spatial attention vector, cross-attention processing can be performed on the scheme transformation vector based on the compressed vector of the scheme spatial attention vector to extract the semantic association between the two vectors and form a scheme association vector. That is, the scheme spatial attention vector can be compressed (e.g., by pooling) so that the size of the resulting compressed vector is the same as the scheme transformation vector. Thus, cross-attention processing can be performed on the scheme transformation vector based on this compressed vector to extract the semantic association between the two vectors and form a scheme association vector. Since the compressed vector carries the semantic information of both the scheme transformation vector and the multi-granularity spatial transformation vector, cross-attention processing can capture both the semantic association within the scheme transformation vector and the semantic association between the multi-granularity spatial transformation vector and the scheme transformation vector, thereby further improving the accuracy of the formed scheme association vector in representing the semantic information of the scheme dimension.

[0116] Step S142c: Perform linear mapping and nonlinear activation on the scheme association vector to form the scheme mode weights corresponding to the scheme transformation vector.

[0117] In this embodiment of the invention, after obtaining the scheme association vector, the scheme association vector is linearly mapped (e.g., through a fully connected network) and nonlinearly activated (e.g., through a sigmoid function) to form the scheme mode weights corresponding to the scheme transformation vector.

[0118] Step S142d: Based on the modal weights of the scheme, determine the spatial modal weights corresponding to the multi-granularity spatial transformation vector.

[0119] In this embodiment of the invention, after obtaining the scheme modal weights, the spatial modal weights corresponding to the multi-granularity spatial transformation vector can be determined based on the scheme modal weights. The spatial modal weights and the scheme modal weights are negatively correlated; for example, spatial modal weights + scheme modal weights = 1.

[0120] In summary, the IoT sensor deployment scheme analysis method and system based on data analysis provided by this invention firstly, based on the gridding results of the target area, performs matrix processing on the proposed deployment schemes of IoT sensors to form a sensor scheme representation matrix; secondly, the sensor scheme representation matrix is ​​converted into scheme semantic vectors; then, the global historical spatial data of the target area is converted into multi-granularity spatial semantic vectors; finally, the scheme semantic vectors and multi-granularity spatial semantic vectors are subjected to semantic space transformation and modality weight-based aggregation, and the scheme spatial aggregation vectors obtained from semantic space transformation and modality weight-based aggregation are mapped and output to obtain the target analysis results. Based on the above method, on the one hand, by performing matrix processing on the proposed deployment schemes, it is easier to fully capture the distributed semantics of sensors in subsequent semantic mining, resulting in a higher semantic representation capability of the formed scheme semantic vectors. On the other hand, since the semantic information of each local historical spatial data at least at two different granularities is mined, the formed multi-granularity spatial semantic vectors take into account both global and local semantic representations. Furthermore, because the semantic vectors of the two modalities undergo semantic space transformation and modality weight-based aggregation before mapping output, the aggregation accuracy is higher, avoiding semantic mismatch issues caused by directly aggregating semantic vectors of different modalities. This ensures the accuracy of subsequent mapping output and leads to reliable analysis results. Based on this, the solution provided by this invention can improve the relatively low reliability of IoT sensor deployment scheme analysis in existing technologies.

[0121] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A data analysis-based method for analyzing IoT sensor deployment schemes, characterized in that, include: Based on the gridding results of the target area, the proposed deployment schemes of IoT sensors are matrixed to form a sensor scheme characterization matrix. Each parameter in the sensor scheme characterization matrix is ​​used to characterize whether the grid at the corresponding location has a sensor and the type of sensor. Convert the sensor scheme representation matrix into a scheme semantic vector; For each local historical spatial data point in the global historical spatial data of the target region, a coarse-grained historical spatial vector and multiple fine-grained historical spatial vectors corresponding to each historical spatial data segment are mined, thereby outputting a coarse-grained vector combination and an original fine-grained vector combination corresponding to each local historical spatial data point. According to the temporal relationship between the historical spatial data segments in the local historical spatial data, the coarse-grained historical spatial vectors in the coarse-grained vector combination are sorted and associated with each time step of the forward diffusion, forming a one-to-one correspondence. In each time step, based on the corresponding coarse-grained historical spatial vectors, the constraint strength control parameters of the semantic constraints for each time step are determined sequentially. Based on the constraint strength control parameters of the semantic constraints for each time step, the corresponding historical spatial data segments are sequentially... A coarse-grained vector is diffused into the original fine-grained vector combination to achieve semantic constraints, forming a forward fine-grained vector combination. The original coarse-grained vector cluster is forward diffused to output a forward coarse-grained vector cluster. Under the constraints of the original coarse-grained vector cluster, the original fine-grained vector cluster is forward diffused to output a forward fine-grained vector cluster. The original fine-grained vector cluster includes forward fine-grained vector combinations corresponding to multiple local historical spatial data, and the original coarse-grained vector cluster includes coarse-grained vector combinations corresponding to multiple local historical spatial data. The forward coarse-grained vector cluster and the forward fine-grained vector cluster are aggregated to form a multi-granular spatial semantic vector corresponding to the global historical spatial data. The global historical spatial data includes local historical spatial data corresponding to each sensor in the proposed deployment scheme. The semantic vector of the scheme and the semantic vector of the multi-granularity space are respectively subjected to semantic space transformation to form a scheme transformation vector and a multi-granularity space transformation vector; the scheme transformation vector and the multi-granularity space transformation vector are concatenated to form a scheme space concatenated vector; and self-attention processing is applied to the scheme space concatenated vector to extract the associated semantics within the vector, forming a scheme space attention vector; based on the compressed vector of the scheme space attention vector, cross-attention processing is applied to the scheme transformation vector to extract the associated semantics between the two vectors, forming a scheme association vector; linear mapping and nonlinear activation are applied to the scheme association vector to form the scheme modality weights corresponding to the scheme transformation vector; based on the scheme modality weights, the scheme transformation vector is determined... The spatial modality weights corresponding to the multi-granularity spatial transformation vectors are determined, wherein the spatial modality weights are negatively correlated with the scheme modality weights, the sensor scheme characterization matrix and the global historical spatial data belong to different modalities, and the scheme modality weights and the spatial modality weights represent the degree of attention to different modalities; based on the scheme modality weights and the spatial modality weights, the scheme semantic vector and the multi-granularity spatial semantic vector or the scheme transformation vector and the multi-granularity spatial transformation vector are weighted and aggregated to form a scheme spatial aggregation vector; the scheme spatial aggregation vector is mapped and output to obtain the target analysis result, wherein the target analysis result is at least used to reflect the adaptability of the proposed deployment scheme to the target region.

2. The method for analyzing IoT sensor deployment schemes based on data analysis as described in claim 1, characterized in that, The step of determining the constraint strength control parameters of the semantic constraints at each time step based on the corresponding coarse-grained vectors of the historical space in sequence includes: In the current time step, the corresponding coarse-grained vector of the historical space is subjected to self-attention processing, and the result of the self-attention processing is subjected to linear mapping and nonlinear activation processing to form the initial control parameters of the semantic constraints of the current time step. The constraint strength control parameters of the semantic constraints at each previous time step are averaged to form historical representative control parameters, where the historical representative control parameter corresponding to the first time step is 0. Based on the original adjustment coefficient corresponding to the current time step and the historical representative control parameter, the target adjustment coefficient corresponding to the current time step is determined. There is a positive correlation between the original adjustment coefficient and the corresponding time step, a positive correlation between the original adjustment coefficient and the target adjustment coefficient, and a negative correlation between the historical representative control parameter and the target adjustment coefficient. Based on the target adjustment coefficient, the initial control parameters are adjusted to form the constraint strength control parameters of the semantic constraints at the current time step.

3. The method for analyzing IoT sensor deployment schemes based on data analysis as described in claim 1, characterized in that, The step of controlling the constraint strength based on the semantic constraints at each time step, sequentially spreading the corresponding coarse-grained vectors of the historical space to the original fine-grained vector combination to achieve semantic constraints and form a forward fine-grained vector combination, includes: In the first time step, any one of the historical space fine-grained vectors in the original fine-grained vector combination is used as the corresponding diffusion vector. In other time steps, the diffusion space fine-grained vector output by the previous time step is used as the corresponding diffusion vector. Based on the historical space coarse-grained vector corresponding to the current time step, the corresponding vector to be diffused is subjected to cross-attention processing to achieve semantic diffusion and form the corresponding cross-attention vector. Based on the constraint strength control parameter of the semantic constraint at the current time step and the negative correlation parameter of the constraint strength control parameter, a weighted residual connection is performed on the cross attention vector and the vector to be diffused, and the fine-grained vector of the diffusion space at the current time step is output. After obtaining the diffusion space fine-grained vector at the last time step, this diffusion space fine-grained vector is determined as the corresponding forward fine-grained vector; The forward fine-grained vectors corresponding to each historical space fine-grained vector in the original fine-grained vector combination are combined to form a forward fine-grained vector combination.

4. The method for analyzing IoT sensor deployment schemes based on data analysis as described in any one of claims 1-3, characterized in that, The step of converting the sensor scheme representation matrix into a scheme semantic vector includes: Convolve the sensor scheme submatrix representing whether or not a sensor is present in the sensor scheme characterization matrix to form the first scheme semantic vector. Convolve the sensor scheme submatrix representing the sensor type in the sensor scheme characterization matrix to form the second scheme semantic vector; The semantic vectors of the first and second schemes are connected to form a scheme connection vector. Based on the semantic vectors of the first and second schemes respectively, the scheme connection vector is gating and adjusted to form a first scheme gating vector and a second scheme gating vector. The first scheme gating vector and the second scheme gating vector are aggregated to form a scheme semantic vector.

5. A data analysis-based IoT sensor deployment scheme analysis system, characterized in that, It includes a processor and a memory, the memory being used to store a computer program, and the processor being used to execute the computer program to implement the data analysis-based IoT sensor deployment scheme analysis method according to any one of claims 1-4.

Images