Temperature and humidity sensor data association analysis method based on deep learning

Abstract

The invention discloses a temperature and humidity sensor data association analysis method based on deep learning, and relates to the technical field of data association analysis. Comprising the following steps: calculating a temperature and humidity gradient difference value of any two adjacent nodes based on a continuous temperature and humidity data sequence acquired by a sensor, and performing partition division in combination with a geographic distance; constructing a correlation coefficient matrix based on a historical time window for all sensor nodes in the same partition, and dividing the nodes into a plurality of correlation groups with high correlation; dividing a master node and a slave node for each association group; and generating a redundancy substitution matrix based on the correlation coefficient matrix, and when the main node is in an abnormal state, screening a standby node as a new main node in combination with the real-time data difference degree, and maintaining the stability of an inference link. The method effectively reduces the node energy consumption and communication redundancy while guaranteeing the prediction accuracy and reliability, is suitable for a large-scale deployment scene, and is especially suitable for a temperature and humidity sensing network analysis task in an edge environment.

Description

Technical Field

[0001] The present invention relates to the technical field of data association analysis, in particular to a method for associating and analyzing temperature and humidity sensor data based on deep learning. Background Art

[0002] In the context of the continuous evolution of intelligent perception and environmental monitoring technologies, temperature and humidity sensors, as basic environmental perception devices, have been widely deployed in multiple application scenarios such as smart agriculture, industrial automation, building management, and weather forecasting. With the deepening development of the Internet of Things architecture, sensor nodes generate a large amount of temperature and humidity data under dense distribution conditions, promoting the development of related monitoring methods from static reading to dynamic analysis and trend prediction. Traditional temperature and humidity monitoring technologies mainly focus on acquisition accuracy and data transmission integrity. In recent years, data-driven intelligent processing methods have gradually been introduced into this field, especially the joint modeling and analysis based on historical data and real-time data, making the prediction and early warning of environmental states more forward-looking and stable. For example, a temperature and humidity sensor and a temperature and humidity monitoring method disclosed in CN117968762A introduce historical temperature and humidity data for threshold calculation and alarm judgment, improving the comprehensiveness of the monitoring system and the accuracy of data processing, and have become one of the typical data fusion methods.

[0003] However, the existing temperature and humidity monitoring methods still have certain limitations. First, the current technologies mostly adopt a per-node static analysis mechanism, without fully considering the spatio-temporal correlation and redundancy between sensor nodes, and it is difficult to exert the synergy effect in a large-scale distributed sensing network. Second, although some methods have introduced historical data modeling (as shown in CN117968762A), they do not perform grouped modeling and hierarchical response on the correlation between sensors, and cannot achieve dynamic optimization allocation of resources at the sensing level. In addition, the temperature and humidity sensor testing method described in the temperature and humidity sensor testing system and method of CN114485757A improves the detection efficiency of individual sensors, but it still stays at the level of hardware testing and static index analysis, lacking a real-time perception and adaptive control mechanism for the collaborative operation state of the sensor group. Therefore, the current technologies have not truly realized the fundamental transformation from passive perception to intelligent collaborative reasoning.

[0004] Especially in a complex environment with a large number of nodes and a data chain that is easily disturbed, how to reduce the network load and enhance the system robustness while ensuring the monitoring accuracy has become a key issue that urgently needs to be addressed in the current temperature and humidity monitoring field. The existing technologies lack a modeling mechanism for the internal redundancy relationship of the sensor network and do not fully introduce deep learning models to infer trends and evaluate uncertainties, resulting in difficulty in achieving a fast and stable alternative response when the main node fails or data is abnormal. In addition, the frequent wake-up and repeated measurements between nodes also increase the energy consumption and communication overhead, restricting its further application in low-power edge intelligent systems. Summary of the Invention

[0005] In view of the problems existing in the existing temperature and humidity monitoring technologies in terms of large-scale distributed deployment requirements, adaptive stability requirements, and resource dynamic scheduling efficiency, the present invention proposes a method for analyzing the correlation of temperature and humidity sensor data based on deep learning.

[0006] Therefore, the problem to be solved by the present invention is how to implement efficient and robust trend inference and node collaborative control by establishing a correlation model and grouping mechanism between sensors, introducing lightweight deep learning to make predictions and judgments by the main node, and constructing a redundant replacement strategy.

[0007] To solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides a method for analyzing the correlation of temperature and humidity sensor data based on deep learning, which includes calculating the temperature and humidity gradient difference between any two adjacent nodes based on the continuous temperature and humidity data sequence collected by the sensors, and partitioning the area in combination with the geographical distance; for all sensor nodes within the same partition, constructing a correlation coefficient matrix based on the historical time window, and dividing the nodes into several associated groups with a high degree of correlation; for each associated group, dividing the main node and the slave nodes; wherein, the main node accesses a lightweight deep learning model to infer the temperature and humidity trend of the partition; the slave nodes default to sleep and are only awakened to participate in the multi-node verification and judgment of the current trend when the confidence level of the prediction result output by the main node is lower than the threshold; based on the correlation coefficient matrix, generating a redundant replacement matrix, and when the main node is in an abnormal state, screening standby nodes as the new main node in combination with the real-time data difference degree to maintain the stability of the inference link.

[0009] As a preferred embodiment of the method for analyzing the correlation of temperature and humidity sensor data based on deep learning according to the present invention, the partitioning includes: fusing the normalized ratio of the temperature and humidity gradient difference and the geographical distance between the corresponding nodes to construct a weighted environmental difference index; constructing a similarity graph structure based on the weighted environmental difference index value, and performing spatial partitioning on the node set in the graph structure to form multiple dynamically variable partition clusters, and generating a unique environmental coding identifier for each dynamically formed partition cluster.

[0010] As a preferred solution of the method for analyzing the correlation of temperature and humidity sensor data based on deep learning according to the present invention, wherein: the construction of the correlation coefficient matrix includes: for each set of temperature and humidity sensor nodes bound to the environmental coding identifier, extracting the temperature and humidity data sequences within the most recent sliding time window, calculating the Pearson correlation coefficients for the temperature component and the humidity component respectively to form a two-dimensional correlation matrix group, and forming a comprehensive trend coupling matrix through asymmetric normalization synthesis.

[0011] As a preferred solution of the method for analyzing the correlation of temperature and humidity sensor data based on deep learning according to the present invention, wherein: the division of the association groups includes: converting the comprehensive trend coupling matrix into an undirected weighted graph structure, adopting a local connection strategy controlled by a density threshold, aggregating nodes according to the edge weight strength, and constructing a set of multiple highly coupled node subgraphs, with each subgraph regarded as an independent association group.

[0012] As a preferred solution of the method for analyzing the correlation of temperature and humidity sensor data based on deep learning according to the present invention, wherein: the division of the master node and the slave nodes includes: for each association group, calculating the average trend coupling degree value of each node relative to all other nodes within the group, and selecting the node corresponding to the maximum value as the master node, with the remaining nodes marked as the slave node set.

[0013] As a preferred solution of the method for analyzing the correlation of temperature and humidity sensor data based on deep learning according to the present invention, wherein: the multi-node verification and judgment includes: activating the top K slave nodes in sequence according to the correlation ranking in the correlation coefficient matrix to form a temporary verification group; each slave node within the verification group outputs a prediction result respectively, and constructs a local average prediction output vector; calculating the residual between the master node prediction value and the local average prediction output vector, and when the residual value is greater than the residual tolerance threshold, marking the master node as an abnormal state.

[0014] As a preferred solution of the method for analyzing the correlation of temperature and humidity sensor data based on deep learning according to the present invention, wherein: the generation of the redundant substitution matrix is based on the correlation coefficient matrix, extracting the maximum correlation degree paths between each node and the remaining nodes, and constructing the redundant substitution matrix.

[0015] As a preferred solution of the method for analyzing the correlation of temperature and humidity sensor data based on deep learning according to the present invention, wherein: the step of screening standby nodes as new master nodes in combination with the real-time data difference degree includes: after the master node is marked as an abnormal state, extracting its standby node set from the redundant substitution matrix, calculating the dynamic time warping distance between each standby node and the master node's most recent period data, and selecting the node corresponding to the minimum distance as the new master node; the node selected as the new master node will be connected to the lightweight deep learning inference link, and the original master node will be transferred to the state to be tested and enter the short-term recovery observation stage.

[0016] In a second aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program instructions are executed by the processor, the steps of the method for associative analysis of temperature and humidity sensor data based on deep learning as described in the first aspect of the present invention are implemented.

[0017] In a third aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program instructions are executed by the processor, the steps of the method for associative analysis of temperature and humidity sensor data based on deep learning as described in the first aspect of the present invention are implemented.

[0018] The beneficial effects of the present invention are as follows: By introducing the fusion of temperature and humidity gradient difference and geographical distance to construct a weighted environmental difference index, the present invention realizes the spatial partitioning and dynamic clustering of sensor nodes, improves the similarity and correlation of data within the region; and combines a sliding time window to construct a trend coupling matrix to form multiple highly coupled sensor association groups. Compared with the prior art, the present invention introduces a lightweight deep learning model to predict the trend of the main nodes of the highly correlated group, and at the same time intelligently activates the slave nodes through a confidence mechanism, realizing the dynamic balance between prediction accuracy and energy consumption; in the abnormal state of the main node, based on the redundant replacement matrix and the dynamic time warping distance, a standby node is intelligently selected for replacement, significantly enhancing the stability and robustness of the prediction link. Through the above mechanism, while ensuring the prediction accuracy and reliability, the present invention effectively reduces the node energy consumption and communication redundancy, is applicable to large-scale deployment scenarios, and is particularly suitable for temperature and humidity perception network analysis tasks in edge environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0020] Figure 1 It is a schematic flowchart of the method for associative analysis of temperature and humidity sensor data based on deep learning;

[0021] Figure 2 It is a flowchart of multi-node verification and judgment of the method for associative analysis of temperature and humidity sensor data based on deep learning;

[0022] Figure 3 It is a flowchart of selecting a standby node as a new main node by combining the real-time data difference degree of the method for associative analysis of temperature and humidity sensor data based on deep learning. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following provides a detailed description of the specific embodiments of the present invention in conjunction with the accompanying drawings of the specification.

[0024] In the following description, many specific details are set forth to facilitate a thorough understanding of the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0025] Secondly, the so-called "one embodiment" or "embodiment" herein refers to specific features, structures, or characteristics that can be included in at least one implementation manner of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that mutually excludes other embodiments.

[0026] Embodiment 1

[0027] Referring to Figures 1 to 3 , which is the first embodiment of the present invention, this embodiment provides a method for associative analysis of temperature and humidity sensor data based on deep learning. As Figure 1 shown, it includes

[0028] S1: Based on the continuous temperature and humidity data sequence collected by the sensor, calculate the temperature and humidity gradient difference between any two adjacent nodes, and combine the geographical distance to divide the area.

[0029] Preferably, the division of the area includes: normalizing and fusing the temperature and humidity gradient difference with the geographical distance between the corresponding nodes to construct a weighted environmental difference index; constructing a similarity graph structure based on the weighted environmental difference index value, and performing spatial partitioning on the node set in the graph structure to form multiple dynamically variable area clusters, and generating a unique environmental coding identifier for each dynamically formed area cluster.

[0030] Specifically, calculate the temperature and humidity gradient difference vector between all adjacent node pairs based on the temperature and humidity data sequence continuously collected by the sensor nodes. Among them, the calculation of the temperature and humidity gradient difference vector is the square root of the sum of the square of the temperature difference and the square of the humidity difference between the adjacent node pairs at this moment. This gradient difference vector not only describes the degree of temperature and humidity distribution difference between adjacent nodes at a specific moment but also reflects the severity of local environmental microclimate characteristic changes.

[0031] Furthermore, the present invention introduces the concept of weighted environmental difference index on the basis of traditional gradient analysis. The index comprehensively considers the temperature and humidity gradient difference and the geographical space distance between nodes, and constructs a more physically meaningful and engineering-valuable environmental heterogeneity measurement method. Considering that even if two nodes have significant differences in temperature and humidity, if they are far apart in space, they can be classified as different regions in partitioning. Therefore, in order to add spatial geographical restriction factors on the basis of gradient difference, a weighted environmental difference index is proposed. Among them, the weighted environmental difference index represents a comprehensive index between the temperature and humidity gradient difference between adjacent node pairs and their geographical distance at the current moment, which can be defined as the temperature and humidity gradient difference vector divided by the geographical distance of the adjacent node pairs plus 1. By introducing the distance factor in the denominator, the node pairs with long distance but large environmental differences can be downgraded, so that in the subsequent partitioning process, more attention is paid to the environmental consistency of adjacent spatial regions, effectively avoiding the misdivision caused by a single gradient mutation.

[0032] Furthermore, after obtaining the weighted environmental difference index between all pairs of nodes, the present invention completes the spatial partitioning by constructing an environmental similarity graph structure. Specifically, all sensor nodes are taken as the node set in the graph, and the value between each pair of nodes is used as the weight of the edge to construct a weighted undirected graph. In this graph, the smaller the weight of the edge, the more similar the environment between the nodes; then, the similarity graph is processed by a graph partitioning algorithm based on spectral clustering, and several partition clusters are automatically divided. Compared with conventional partitioning strategies based on geographic grids or regular radius methods, this method has stronger environmental sensitivity and adaptability, and can form smaller-grained partitions in areas with drastic changes in temperature and humidity, and form larger-scale partitions in areas with stable changes, thereby dynamically adapting to the actual characteristics of environmental distribution.

[0033] It is particularly worth emphasizing that the graph structure division method has dynamic evolution characteristics. Since the data collected by the sensor is time series data, the temperature and humidity gradient difference and its weighted environmental difference index will also change dynamically over time. To this end, the present invention designs a partition evolution control mechanism based on the division: when the mean fluctuation of the weighted environmental difference index of the node pair in a partition exceeds a preset threshold (such as 10%) within multiple consecutive time steps, the partition redrawing mechanism is triggered, and it is split into two or more new partitions; conversely, if the mean difference between multiple adjacent partitions is lower than the threshold, they are merged into a unified partition. This mechanism can effectively respond to scenarios such as emergencies and ensure the timeliness and sensitivity of the partition structure to environmental changes.

[0034] After the graph partitioning is completed, the present invention also generates a unique environmental encoding identifier for each dynamically generated partition cluster, which is used for binding references in multiple stages such as subsequent node clustering and data association analysis. This identifier can be encoded based on various parameters such as the timestamp of partition generation, node number hash, geographical center coordinates, etc., to ensure uniqueness and traceability in system-level management.

[0035] In summary, in this step, by fusing the temperature and humidity gradient differences and geographical distance information, introducing a weighted environmental difference index, and constructing an environmental heterogeneity measurement method with physical significance, the scientificity and accuracy of spatial partitioning are improved. Among them, by constructing a weighted similarity graph and using the spectral clustering algorithm to achieve dynamic partitioning, the partition structure can more truly reflect the characteristics of local microclimate changes. Compared with the existing methods based on fixed grids or radius thresholds, this step has stronger adaptability, can form fine partitions in areas with drastic environmental changes, and form large-scale aggregations in stable areas, effectively improving the local consistency and operation efficiency of subsequent analysis. In addition, combined with the design of the partition evolution mechanism and the unique environmental encoding, continuous response and tracking of the time-varying environment are realized, enhancing the sensitivity and manageability to abnormal changes.

[0036] S2: For all sensor nodes within the same partition, construct a correlation coefficient matrix based on the historical time window, and divide the nodes into several highly correlated association groups.

[0037] Specifically, the construction of the correlation coefficient matrix includes: for the set of temperature and humidity sensor nodes bound to each environmental encoding identifier, extract the temperature and humidity data sequences within the most recent sliding time window, and calculate the Pearson correlation coefficients for the temperature component and the humidity component respectively to form a two-dimensional correlation matrix group, and synthesize them through asymmetric normalization to form a comprehensive trend coupling matrix.

[0038] First, it is necessary to extract all sensor nodes under each dynamic partition (i.e., the set of nodes with a unified environmental encoding identifier), denoted as set N k , and record the temperature and humidity sequences collected by each node in the set within the most recent sliding time window [t - τ, t] and where t is the current time, τ is the window length, T is the temperature data, and H is the humidity data. The sliding time window adopted in the present invention can not only capture the short-term change trends of nodes, but also effectively smooth out instantaneous perturbations, improving the stability of correlation calculation. Among them, the value of the time window size can be adjusted according to the actual scenario.

[0039] Subsequently, calculate the Pearson correlation coefficient matrices for the extracted temperature sequence and humidity sequence respectively. Specifically, for any two nodes i and j, the temperature trend correlation coefficient can be obtained through the following formula:

[0040]

[0041] Among them, cov(.) represents the covariance function, σ(.) represents the standard deviation, and both i and j are node numbers. This formula reflects the linear dependence degree of two nodes in the temperature change trend, and the value range is [-1, 1], where approaching 1 indicates a complete positive correlation, approaching -1 indicates a complete negative correlation, and approaching 0 indicates no obvious linear correlation. Humidity trend correlation coefficient The calculation is carried out in the same way. Finally, two symmetric correlation matrices are obtained respectively, which are used to describe the temporal relationship network of temperature and humidity respectively.

[0042] After obtaining the two-dimensional trend correlation matrix, the present invention further introduces a construction method of a comprehensive trend coupling matrix to comprehensively depict the dynamic association structure of temperature and humidity in the same partition.

[0043] Specifically, the calculation of the comprehensive trend coupling matrix includes the temperature trend correlation coefficient matrix and the humidity trend correlation coefficient matrix being fused in a weighted manner, adopting an asymmetric weighted fusion mechanism, and the fusion weight is controlled by an adaptive factor. Among them, the adaptive fusion factor is dynamically adjusted according to the variance of the temperature gradient of each node in the partition to enhance the expression ability of the temperature-dominated trend.

[0044] Preferably, the division of the associated groups includes: converting the comprehensive trend coupling matrix into an undirected weighted graph structure, adopting a local connection strategy controlled by a density threshold, aggregating nodes according to the edge weight intensity, and constructing a set of multiple highly coupled node subgraphs, and each subgraph is regarded as an independent associated group.

[0045] Specifically, after obtaining the comprehensive trend coupling matrix, it is further converted into an undirected weighted graph structure, and its expression form is: g k =(N k , ε k ), where the node set is N k , and each edge (i, j) in the edge set ε k corresponds to a weight of This graph represents the distribution of the coupling strength between nodes and is the basis for subsequent node aggregation analysis.

[0046] To realize the construction of highly correlated associated groups, the present invention adopts a local connection strategy based on density threshold control. The core idea of this strategy is: in the graph structure, first select the coupling degree greater than the set threshold θ cFor the edges, the corresponding node pairs are regarded as strongly correlated connections. Subsequently, starting from any highly coupled node, all its strongly correlated adjacent nodes are recursively expanded through methods such as depth-first search until no further expansion is possible, forming a high-density coupled subgraph. Each coupled subgraph is regarded as an independent node association group.

[0047] In particular, to avoid low-coupled nodes from forming an isolated state, the present invention also sets a boundary acceptance mechanism: for a node that has not been classified, if its coupling degree with the nodes in any existing association group exceeds a preset threshold θ b (θ b <θ c ), then it is accepted into this association group. Through the master-slave threshold control strategy, both the strong correlation within the aggregation group can be ensured, and the problem of node isolation can be minimized as much as possible, thereby improving the data utilization efficiency and modeling integrity.

[0048] It can be seen that this step is based on the node set bound by each environmental coding identifier, constructs the temperature and humidity correlation matrix under the historical time window, extracts the trend coupling relationship, and constructs a set of node subgraphs with high coupling degree based on this. To sum up, the present invention has successfully achieved a highly reliable structure division of nodes in the same region by constructing the trend correlation matrix under the historical time window, and combining the adaptive coupling mechanism and the density-aware clustering strategy, providing a more refined and clearly structured node organization form for the subsequent algorithm. The entire process strengthens the temporal trend expression while maintaining spatial consistency.

[0049] S3: For each association group, divide the master node and the slave node; among them, the master node accesses the lightweight deep learning model to infer the temperature and humidity trend of the partition; the slave node defaults to sleep and is only awakened to participate in the multi-node verification and judgment of the current trend when the confidence level of the prediction result output by the master node is lower than the threshold.

[0050] After completing the construction of the node association group, the present invention enters the stage of dividing the functional roles of each association group, that is, distinguishing the nodes in the group into master nodes and slave nodes, and further constructing a dynamic wake-up mechanism based on the prediction confidence level to achieve the dual goals of energy consumption optimization and enhanced inference credibility.

[0051] Specifically, the division of the master node and slave nodes includes: for each associated group, calculate the average trend coupling degree value of each node in it relative to all other nodes in the group, and select the node corresponding to the maximum value as the master node, and mark the remaining nodes as the slave node set. Among them, the calculation of the average trend coupling degree value is as follows: Suppose the i-th node is in the associated group, and its trend coupling degree relative to all other nodes in the group is defined as the mean difference between its row vector corresponding to other nodes in the group in the comprehensive trend coupling matrix. This method ensures that the selected master node has the maximum aggregation consistency in the group and can better represent the overall trend of the group, and can be used to evaluate the centrality or representativeness strength of the node in the group.

[0052] After the master node is selected, a set of lightweight deep learning prediction models will be configured for it, such as temporal CNN or gated recurrent unit (GRU), to predict the temperature and humidity trends at the next moment or within several future time points.

[0053] Exemplarily, the input of the lightweight deep learning prediction model is the temperature and humidity sequence of the master node in the recent M moments; the output is the temperature and humidity prediction vector for a future period of time (such as the next 5 minutes); at the same time, the lightweight deep learning prediction model also outputs a confidence index, which reflects the credibility of the current prediction of the model, usually obtained through methods such as attention coefficient distribution, entropy index or Bayesian uncertainty estimation. In this embodiment, the acquisition method is not uniquely limited; if the confidence index is greater than or equal to the confidence threshold, the prediction result of the master node will be directly adopted; otherwise, the slave node verification mechanism will be triggered.

[0054] Preferably, as Figure 2 shown, the multi-node verification and judgment includes: according to the relevance ranking in the correlation coefficient matrix, activate the top K slave nodes in sequence to form a temporary verification group; each slave node in the verification group outputs a prediction result respectively, and constructs a local average prediction output vector; calculate the residual between the master node prediction value and the local average prediction output vector. When the residual value is greater than the residual tolerance threshold, mark the master node as an abnormal state.

[0055] Exemplarily, use the other element values in the row where the master node is located in the comprehensive trend coupling matrix to sort the slave nodes in descending order of coupling strength; activate the top K slave nodes (such as K = 2 - 4) to form a temporary verification group; each verification slave node uses its own historical observation sequence as the input, and outputs a local prediction result through a lightweight model with the same structure as the master node; perform weighted averaging on these prediction results to form a local average prediction vector; further, compare the prediction output of the master node with the local average result, calculate the residual index, and if any of the two residuals exceeds the residual tolerance threshold, it is considered that the master node prediction deviates from the group trend, and mark its current state as abnormal.

[0056] It can be seen that in this step, by introducing the master-slave node division and dynamic wake-up mechanism, the goal of significantly reducing energy consumption while maintaining prediction accuracy is achieved. The most representative master node is selected by calculating the average trend coupling degree to ensure that the master node can accurately reflect the overall environmental trend of the affiliated partition and improve the prediction representativeness. The master node accesses the lightweight deep learning model for trend prediction and combines the confidence index to judge the reliability of the result, ensuring the adaptability and intelligence of the prediction process; when the prediction confidence is insufficient, some slave nodes are awakened to participate in the verification, and a local consensus is constructed based on the multi-node collaborative verification, effectively preventing decision-making biases caused by misjudgment of the master node. In addition, this mechanism has good characteristics of saving computing resources. The slave nodes are in the sleep state when not necessary, significantly reducing energy consumption and providing an efficient and reliable inference solution for the sensor network deployed in resource-constrained environments.

[0057] S4: Based on the correlation coefficient matrix, generate a redundant replacement matrix. When the master node is in an abnormal state, screen the standby nodes as the new master node in combination with the real-time data difference degree to maintain the stability of the inference link.

[0058] To ensure stable operation in the event of anomalies such as failure, abnormal fluctuations, or communication interruptions of the master node, a master node redundant replacement mechanism based on correlation path optimization and real-time difference degree analysis is proposed. This mechanism can not only enable a more reliable replacement node as the master node in a timely manner to maintain the continuity of the inference link within the partition, but also construct a feedback-type trend label based on the reasoning collaboration of the master-slave nodes to realize the online reinforcement learning-style update of the model input data and improve the dynamic adaptation ability of the prediction.

[0059] The present invention proposes to expand the existing comprehensive trend coupling matrix into a redundant replacement matrix for establishing a high-correlation path graph between each master node and its potential replacement nodes, so that when the master node is abnormal, the optimal alternative node can be quickly found.

[0060] Preferably, the generation of the redundant replacement matrix is based on the correlation coefficient matrix, extracting the maximum correlation degree path between each node and the remaining nodes to construct the redundant replacement matrix.

[0061] Specifically, for each master node, extract its corresponding high-coupling path within its associated group; the so-called high-coupling path refers to the longest single path (or sub-optimal path) composed of adjacent nodes starting from the master node with an edge weight greater than the preset threshold θ c ; take the set of all nodes except the master node in the path as its standby node candidate set and record this set in the redundant replacement matrix. This matrix pre-establishes the candidate replacement paths for all master nodes without real-time full-scale calculation; the replacement paths are constructed based on the long-term trend stability and have structural robustness; and in the subsequent selection process, more optimal replacement nodes are further screened through local dynamic comparison.

[0062] To improve the consistency between the representativeness of the alternative node and the current trend, after the main node is marked as abnormal, instead of directly adopting the candidates in the redundant alternative matrix, a dynamic time warping distance (DTW) mechanism is further introduced to measure the similarity between the latest observation sequences of the main node and the candidate nodes.

[0063] Preferably, as Figure 3 shown, screening standby nodes as the new main node in combination with the real-time data difference degree includes: after the main node is marked as an abnormal state, extracting its set of standby nodes from the redundant alternative matrix, and calculating the dynamic time warping distance for the data of the nearest period between each standby node and the main node, and selecting the one with the smallest distance as the new main node; the node selected as the new main node will be connected to the lightweight deep learning inference link, the original main node will be transferred to the state to be tested, and enter the short-term recovery observation stage.

[0064] Exemplarily, extract the set of standby nodes of the current main node from the redundant alternative matrix; for each candidate node, extract its temperature and humidity observation sequences for the nearest τ d moments; calculate the dynamic time warping distance between the candidate node sequence and the historical sequence of the current main node, that is, for the temperature data sequences of the current main node and the candidate standby node in the nearest period of time, use the dynamic time warping algorithm to calculate the time series difference degree between the two to obtain the distance D T(i) in the temperature dimension; process the humidity data in the same way to obtain the difference degree D H(i) between the humidity sequences of the main node and the candidate node; balance the influence of temperature and humidity in the decision-making through the temperature and humidity fusion weight β, and this weight can be dynamically set according to the variance of the historical temperature and humidity gradients. Finally, calculate the comprehensive difference degree of each candidate standby node: The smaller the value, the closer the behavior pattern of the standby node is to the current main node, and it is a more ideal replacement.

[0065] Among all candidate nodes, select the node corresponding to the smallest distance as the new main node; the new main node is connected to the original inference link and loads the lightweight deep learning model used by the main node; the original main node is transferred to the state to be tested, continues to receive data in the background, but does not participate in the inference, and is only used to monitor whether it returns to normal; if the prediction residuals and confidence levels of the original main node return to the normal level in multiple consecutive cycles, it can re-enter the master-slave candidate pool.

[0066] Optionally, to enhance the adaptive ability of the model, the main node prediction output can be combined with the secondary inference results of the slave nodes to form a temperature and humidity change trend label with confidence feedback, and continuously update the input data set of the lightweight deep learning model. For example, in each cycle, the following three outputs can be obtained, including the main node prediction output vector, the average prediction vector of the verification group, and the model confidence and residual values. A combined trend label can be constructed, and the combined trend label can be continuously accumulated into the model training sample cache queue for periodic or sliding window model fine-tuning, forming a training closed-loop of master-slave collaboration.

[0067] It should be noted that this step significantly enhances the robustness and adaptive ability in the case of main node anomalies by introducing a redundant substitution matrix and a real-time difference degree screening mechanism. To improve the current representativeness of the standby node, the dynamic time warping distance (DTW) is introduced to compare the recent observation data of the main node and the candidate nodes, and the real-time behavior consistency index of the temperature and humidity dimension is fused. The most suitable replacement node for the current trend is dynamically selected to access the inference link, ensuring that the replacement node not only has a high structural correlation but also highly matches the real-time trend, greatly improving the reliability of the replacement effect. In addition, this mechanism also has self-recovery ability. The original main node is continuously monitored in the background. When its state returns to the normal range, it can re-participate in the master-slave role rotation, further improving the dynamic balance. Furthermore, by constructing a feedback-type trend label from indicators such as the master-slave node prediction results and confidence residuals and using it to continuously optimize the model input data set, a closed-loop training mechanism of master-slave collaboration is realized, enabling online learning and adaptation ability, so as to achieve the comprehensive beneficial effects of stable long-term inference performance, error convergence, and enhanced environmental adaptation.

[0068] This embodiment also provides a computer device applicable to the case of the temperature and humidity sensor data correlation analysis method based on deep learning, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the temperature and humidity sensor data correlation analysis method based on deep learning as proposed in the above embodiment.

[0069] The computer device can be a terminal, which includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, a trackball, or a touchpad set on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0070] This embodiment also provides a storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the method for realizing the correlation analysis of temperature and humidity sensor data based on deep learning proposed in the above embodiment.

[0071] In summary, the present invention constructs a weighted environmental difference index by introducing the fusion of temperature and humidity gradient difference and geographical distance, realizes the spatial partitioning and dynamic clustering of sensor nodes, and improves the similarity and correlation of data within the region; and combines a sliding time window to construct a trend coupling matrix to form multiple highly coupled sensor association groups. Compared with the prior art, the present invention introduces a lightweight deep learning model to predict the trend of the main node of the highly correlated group, and at the same time intelligently activates the slave nodes through a confidence mechanism, realizing the dynamic balance between prediction accuracy and energy consumption; in the abnormal state of the main node, based on the redundant replacement matrix and the dynamic time warping distance, a standby node is intelligently selected for replacement, significantly enhancing the stability and robustness of the prediction link. Through the above mechanism, the present invention effectively reduces the node energy consumption and communication redundancy while ensuring the prediction accuracy and reliability, is applicable to large-scale deployment scenarios, and is particularly suitable for temperature and humidity perception network analysis tasks in edge environments.

[0072] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for correlation analysis of temperature and humidity sensor data based on deep learning, characterized in that: Including: Based on the continuous temperature and humidity data sequence collected by sensors, calculate the temperature and humidity gradient difference between any two adjacent nodes, and combine the geographical distance to divide the area; For all sensor nodes within the same area, construct a correlation coefficient matrix based on the historical time window, and divide the nodes into several highly correlated association groups; For each association group, divide the master node and the slave nodes; among them, the master node accesses a lightweight deep learning model to infer the temperature and humidity trend of the area; the slave nodes are default in the sleep state and are only awakened to participate in the multi-node verification and judgment of the current trend when the confidence level of the prediction result output by the master node is lower than the threshold; Based on the correlation coefficient matrix, generate a redundant replacement matrix. When the master node is in an abnormal state, combine the real-time data difference degree to screen the standby node as the new master node to maintain the stability of the inference link.

2. The method for correlative analysis of temperature and humidity sensor data based on deep learning according to claim 1, wherein: The division of the area includes: Normalize the ratio of the temperature and humidity gradient difference to the geographical distance between the corresponding nodes, and construct a weighted environment difference index; Based on the weighted environment difference index value, construct a similarity graph structure, and perform spatial partitioning on the node set in the graph structure to form multiple dynamically variable area clusters, and generate a unique environment coding identifier for each dynamically formed area cluster.

3. The method for associative analysis of temperature and humidity sensor data based on deep learning according to claim 1, wherein: The construction of the correlation coefficient matrix includes: For the set of temperature and humidity sensor nodes bound to each environment coding identifier, extract the temperature and humidity data sequence within the recent sliding time window, and calculate the Pearson correlation coefficient for the temperature component and the humidity component respectively to form a two-dimensional correlation matrix group, and form a comprehensive trend coupling matrix through asymmetric normalization synthesis.

4. The method for associative analysis of temperature and humidity sensor data based on deep learning according to claim 3, characterized in that: The division of the association group includes: Convert the comprehensive trend coupling matrix into an undirected weighted graph structure, adopt a local connection strategy controlled by a density threshold, aggregate nodes according to the edge weight strength, and construct a set of multiple highly coupled node subgraphs, and each subgraph is regarded as an independent association group.

5. The method for correlative analysis of temperature and humidity sensor data based on deep learning according to claim 1, characterized in that: The division of the master node and the slave nodes includes: for each association group, calculate the average trend coupling degree value of each node relative to all other nodes in the group, and select the node corresponding to the maximum value as the master node, and the remaining nodes are marked as the slave node set.

6. The method for associative analysis of temperature and humidity sensor data based on deep learning according to claim 5, wherein: The multi-node verification and judgment includes: According to the correlation degree ranking in the correlation coefficient matrix, activate the top K slave nodes in turn to form a temporary verification group; Each slave node in the verification group outputs a prediction result respectively, and constructs a local average prediction output vector; Calculate the residual between the master node prediction value and the local average prediction output vector. When the residual value is greater than the residual tolerance threshold, mark the master node as an abnormal state.

7. The method for associative analysis of temperature and humidity sensor data based on deep learning according to claim 1, characterized in that: The generation of the redundant replacement matrix is based on the correlation coefficient matrix, extracts the maximum correlation degree path between each node and the remaining nodes, and constructs a redundant replacement matrix.

8. The method for analyzing the association of temperature and humidity sensor data based on deep learning according to claim 7, characterized in that: The combination of the real-time data difference degree to screen the standby node as the new master node includes: After the master node is marked as an abnormal state, extract its standby node set from the redundant replacement matrix, and calculate the dynamic time warping distance between each standby node and the master node's recent period data, and select the node corresponding to the minimum distance as the new master node; The node selected as the new master node will be connected to the lightweight deep learning inference link, while the original master node will be transferred to the state to be tested and enter the short-term recovery observation stage.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, the steps of the method for associative analysis of temperature and humidity sensor data based on deep learning according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, the steps of the method for associative analysis of temperature and humidity sensor data based on deep learning according to any one of claims 1 to 8 are implemented.

Images