Multi-sensor fusion temperature detection system and method based on Kalman filtering algorithm

Abstract

The invention discloses a multi-sensor fusion temperature detection system and method based on a Kalman filtering algorithm. The detection system comprises a temperature sensor network, an edge computing node and a cloud server. Wherein the temperature sensor network is used for periodically collecting temperature data of a target area; the edge computing node is used for executing local Kalman filtering preprocessing on the temperature data of the target area, eliminating noise and generating local fusion data; and the cloud server is used for reconstructing a three-dimensional temperature field based on a heat conduction model and a global space-time combined Kalman filtering algorithm according to the local fusion data, and outputting a three-dimensional temperature distribution diagram and an abnormal early warning signal. According to the invention, through the hierarchical edge-cloud collaborative architecture and the space-time joint Kalman filtering algorithm, the problems of data noise interference, limited transmission bandwidth and insufficient anomaly detection precision of existing multiple sensors are solved, and an efficient and reliable temperature monitoring solution is provided for a complex industrial environment.

Description

Technical Field

[0001] The present invention belongs to the technical field of temperature detection and monitoring, and in particular relates to the design of a multi-sensor fusion temperature detection system and method based on a Kalman filter algorithm. Background Art

[0002] Temperature monitoring is crucial for structural health in large concrete dams and industrial equipment. However, existing temperature monitoring systems suffer from several drawbacks: noise interference in multi-sensor data reduces detection accuracy; limited data transmission bandwidth makes real-time monitoring difficult; and weak outlier detection capabilities can easily miss critical temperature changes.

[0003] In recent years, multi-source sensors (such as thermocouples and infrared sensors) have been susceptible to noise due to factors such as environmental electromagnetic interference and device drift, resulting in a decrease in the signal-to-noise ratio (SNR) of the data. Although some studies have attempted to use the Kalman filter algorithm for data fusion, the algorithm is highly complex and requires high-performance hardware, making it difficult to run efficiently on edge computing nodes. Furthermore, variations in sensor accuracy and the uncertainty of process noise further reduce the reliability of the fusion results. Especially under dynamic conditions, fixed-parameter models are prone to deviating from the actual temperature field, leading to cumulative prediction errors.

[0004] Furthermore, traditional centralized cloud-based processing requires the transmission of massive amounts of raw data, placing significant pressure on network bandwidth and making it difficult to meet real-time requirements. While existing edge computing solutions alleviate bandwidth pressure through data compression, complex compression algorithms (such as wavelet transforms and tuple processing) place stringent demands on edge computing power. Furthermore, abnormal data fluctuations can lead to distortion in feature extraction, compromising the integrity of compressed data, particularly in environments with strong interference.

[0005] Existing systems often rely on static thresholds or fixed models to detect anomalies, making them incapable of handling transient temperature fluctuations or changes in environmental conditions. For example, when a local temperature rises suddenly due to structural damage, traditional methods, lacking adaptive parameter adjustment mechanisms, are prone to missed detections or misjudgments. Furthermore, anomaly detection is disconnected from data fusion, preventing real-time optimization of sensor weights through feedback mechanisms, resulting in critical signals being overwhelmed by noise.

[0006] These issues severely limit the effectiveness of temperature monitoring in complex industrial scenarios, potentially delaying early warning of structural safety hazards. Therefore, a fusion system that combines noise suppression, bandwidth optimization, and dynamic anomaly detection is urgently needed to achieve highly reliable, low-latency temperature field reconstruction and risk warning. Summary of the Invention

[0007] The purpose of the present invention is to propose a multi-sensor fusion temperature detection system and method based on the Kalman filter algorithm, which solves the problems of multi-sensor data noise interference, limited transmission bandwidth and insufficient anomaly detection accuracy through a layered edge-cloud collaborative architecture and a spatiotemporal joint Kalman filter algorithm.

[0008] The technical solution of the present invention is as follows: In the first aspect, the present invention provides a multi-sensor fusion temperature detection system based on the Kalman filter algorithm, including a temperature sensor network, an edge computing node and a cloud server.

[0009] The temperature sensor network is distributed in the target area and is used to periodically collect temperature data of the target area.

[0010] The edge computing node is connected to the temperature sensor network to perform local Kalman filter preprocessing on the temperature data of the target area, eliminate noise and generate local fusion data.

[0011] The cloud server communicates with the edge computing node through the MQTT protocol, and is used to reconstruct the three-dimensional temperature field based on local fusion data, the heat conduction model and the global spatiotemporal joint Kalman filter algorithm, and output a three-dimensional temperature distribution map and abnormal warning signals.

[0012] Furthermore, the temperature sensor network includes thermocouples, infrared sensors and optical fiber temperature sensors.

[0013] Furthermore, the edge computing node integrates a local Kalman filter module, a dynamic weight adjustment module, and an encryption communication and data packaging module.

[0014] The local Kalman filter module is used to perform a spatiotemporal joint Kalman filter algorithm on the temperature data of the target area to eliminate noise. The state prediction formula of the spatiotemporal joint Kalman filter algorithm is: in represents the state vector at the current moment, Represents the state transfer matrix from the previous moment to the current moment, represents the state vector at the previous moment, represents the control matrix at the current moment, represents the control input vector at the current moment, Represents the process noise vector at the current moment.

[0015] The state update formula of the spatiotemporal joint Kalman filter algorithm is: in Represents the state estimation value at the current moment, that is, the generated local fusion data, represents the Kalman gain at the current moment, represents the observation value at the current moment, Represents the observation matrix at the current moment.

[0016] The dynamic weight adjustment module is used to dynamically assign weights to sensors in the temperature sensor network: in Represents the first i The weight of each sensor, Represents the first i The historical error variance of each sensor, Represents the first j The historical error variance of each sensor, n Indicates the number of sensors in the temperature sensor network.

[0017] The encrypted communication and data packaging module is used to generate lightweight data packets based on local fusion data, and upload them to the cloud server after encryption using the AES-256 algorithm.

[0018] Furthermore, the cloud server integrates a global Kalman filter module, a three-dimensional temperature field reconstruction module, an anomaly detection and self-correction module, and a temperature distribution warning module.

[0019] The global Kalman filter module is used to fuse local fusion data from multiple edge computing nodes. Through spatiotemporal joint Kalman filter optimization, it achieves the unification of cross-regional temperature information and eliminates the discontinuity of temperature fields between different target areas.

[0020] The 3D temperature field reconstruction module is used to construct a 3D temperature field based on the heat conduction model. The heat conduction model is: in T represents the temperature distribution function, t Indicates time, represents the thermal diffusivity, ∇ 2 T The Laplacian operator representing temperature.

[0021] The anomaly detection and self-correction module is used to detect sensor anomalies based on local fusion data. In response to the detection of sensor anomalies, the process noise covariance matrix, the observation noise covariance matrix, and the sensor weights are adaptively adjusted: in represents the process noise covariance matrix at the current moment, represents the smoothing coefficient, represents the process noise covariance matrix at the previous moment, represents the state prediction error at the current moment, represents the observation noise covariance matrix at the current moment, represents the attenuation factor, represents the observation noise covariance matrix at the previous moment, represents the observation value at the current moment, represents the observation matrix, represents the estimated value of the state at the current moment, Indicates the current temperature sensor network i The weight of each sensor, Represents the first temperature sensor in the adjusted temperature sensor network i The weight of each sensor, Indicates temperature deviation, Indicates the preset temperature threshold.

[0022] The temperature distribution warning module is used to output an abnormal warning signal when the temperature deviation exceeds the preset temperature threshold.

[0023] In a second aspect, the present invention provides a multi-sensor fusion temperature detection method based on a Kalman filter algorithm, comprising the following steps: S1. Periodically collect temperature data of the target area through the temperature sensor network.

[0024] S2. Perform local Kalman filter preprocessing on the temperature data of the target area at the edge computing node to eliminate noise and generate local fusion data.

[0025] S3, upload the local fusion data to the cloud server through the MQTT protocol.

[0026] S4. In the cloud server, based on the local fusion data, the three-dimensional temperature field is reconstructed based on the heat conduction model and the global spatiotemporal joint Kalman filtering algorithm.

[0027] S5. In the cloud server, sensor anomaly detection is performed based on the local fusion data, and Kalman filter parameters and sensor weights are adaptively adjusted.

[0028] S6. Output three-dimensional temperature distribution map and abnormal warning signal.

[0029] Furthermore, in step S2, local Kalman filter preprocessing is implemented by the spatiotemporal joint Kalman filter algorithm. The state prediction formula of the spatiotemporal joint Kalman filter algorithm is: in represents the state vector at the current moment, Represents the state transfer matrix from the previous moment to the current moment, represents the state vector at the previous moment, represents the control matrix at the current moment, represents the control input vector at the current moment, Represents the process noise vector at the current moment.

[0030] The state update formula of the spatiotemporal joint Kalman filter algorithm is: in Represents the state estimation value at the current moment, that is, the generated local fusion data, represents the Kalman gain at the current moment, represents the observation value at the current moment, Represents the observation matrix at the current moment.

[0031] Furthermore, step S4 includes the following sub-steps: S41. Establish a heat conduction model: in T represents the temperature distribution function, t Indicates time, represents the thermal diffusivity, ∇ 2 T The Laplacian operator representing temperature.

[0032] S42. The finite difference method is used to discretize the heat conduction model, divide the target monitoring area into equal-volume voxel units, set the time step, and construct a time-space coupled temperature evolution model.

[0033] S43. Input the local fusion data as the initial boundary condition into the temperature evolution model, and combine the global spatiotemporal joint Kalman filtering algorithm to fuse the local fusion data of multiple edge computing nodes, eliminate the discontinuity and data holes in the temperature field between different target areas, and improve the global consistency and accuracy of temperature estimation through iterative optimization to generate a preliminary three-dimensional temperature field.

[0034] S44. For blind areas where sensors have not yet been deployed, an adaptive Kriging interpolation algorithm is introduced on the basis of the preliminary three-dimensional temperature field. Interpolation weights are dynamically allocated based on the temperature gradient and spatial correlation of the adjacent areas to fill the data in the blind areas, improve the continuity and physical consistency of the preliminary three-dimensional temperature field, and obtain a three-dimensional temperature field with global consistency and high spatial resolution.

[0035] Furthermore, the method for performing sensor anomaly detection based on the local fusion data in step S5 is specifically as follows: In response to a temperature deviation of Δ in any target area for three consecutive cycles T >2℃, calculate the temperature data variance of the target area : in Indicates the number of temperature sensors in the target area, Indicates the i The temperature measurement value of each temperature sensor in the current detection cycle, Indicates the arithmetic mean of all temperature measurements in the current detection cycle.

[0036] In response to , it is determined that the sensor is abnormal.

[0037] Furthermore, the Kalman filter parameters in step S5 include the process noise covariance matrix and the observation noise covariance matrix. The adaptive adjustment formulas of the process noise covariance matrix and the observation noise covariance matrix are: in represents the process noise covariance matrix at the current moment, represents the smoothing coefficient, represents the process noise covariance matrix at the previous moment, represents the state prediction error at the current moment, represents the observation noise covariance matrix at the current moment, represents the attenuation factor, represents the observation noise covariance matrix at the previous moment, represents the observation value at the current moment, represents the observation matrix, Indicates the estimated value of the state at the current moment.

[0038] Furthermore, the adaptive adjustment formula of the sensor weight in step S5 is: in Indicates the current temperature sensor network i The weight of each sensor, Represents the first temperature sensor in the adjusted temperature sensor network i The weight of each sensor, Indicates temperature deviation, Indicates the preset temperature threshold.

[0039] The beneficial effects of the present invention are: (1) The present invention constructs a layered processing architecture based on edge computing nodes and cloud servers, which can effectively optimize the utilization of system resources. Through localized data processing of edge computing nodes, the spatiotemporal alignment and noise filtering of multi-sensor information are completed near the data source. This architecture significantly reduces the data throughput pressure of the cloud server and ensures the reliability of data transmission through a lightweight communication protocol. At the same time, the layered processing mechanism enables the system to adapt to network fluctuations in complex industrial scenarios and improve the overall operational stability.

[0040] (2) The system in the present invention intelligently adjusts the fusion weight of multi-source information based on sensor data, effectively suppressing the impact of abnormal data on detection accuracy. At the same time, the adaptive parameter correction mechanism enables the Kalman filter algorithm to dynamically adapt to environmental changes and maintain stable temperature tracking capabilities under complex working conditions such as electromagnetic interference and mechanical vibration, significantly improving the accuracy of anomaly detection and the robustness of the system.

[0041] (3) The present invention combines the physical laws of heat conduction with data-driven algorithms, breaking through the dependence of traditional interpolation methods on sensor deployment density, and generating high-reliability temperature data in blind areas where sensors are not deployed. At the same time, the global spatiotemporal joint Kalman filter optimization algorithm effectively eliminates data discontinuities between regions and outputs a three-dimensional temperature distribution with physical consistency, providing refined temperature field visualization support for structural health assessment and achieving high-fidelity temperature field reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 The figure shows a structural diagram of a multi-sensor fusion temperature detection system based on a Kalman filter algorithm provided in the first embodiment of the present invention.

[0043] Figure 2 Shown is a schematic diagram of temperature field reconstruction provided by the first and second embodiments of the present invention.

[0044] Figure 3 The figure shows a flow chart of a multi-sensor fusion temperature detection method based on a Kalman filter algorithm provided in the second embodiment of the present invention. DETAILED DESCRIPTION

[0045] The exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be understood that the embodiments shown and described in the accompanying drawings are merely exemplary and are intended to illustrate the principles and spirit of the present invention, rather than to limit the scope of the present invention.

[0046] Example 1: The embodiment of the present invention provides a multi-sensor fusion temperature detection system based on the Kalman filter algorithm. Figure 1 As shown, it includes a temperature sensor network, edge computing nodes, and cloud servers.

[0047] The temperature sensor network is distributed in the target area and is used to periodically collect temperature data of the target area.

[0048] The edge computing node is connected to the temperature sensor network to perform local Kalman filter preprocessing on the temperature data of the target area, eliminate noise and generate local fusion data.

[0049] The cloud server communicates with the edge computing node through the lightweight Message Queuing Telemetry Transport (MQTT) protocol. It is used to reconstruct the three-dimensional temperature field based on local fusion data, a heat conduction model and a global spatiotemporal joint Kalman filter algorithm, and output a three-dimensional temperature distribution map and abnormal warning signals.

[0050] In the embodiment of the present invention, Figure 1 As shown, the temperature sensor network includes thermocouples, infrared sensors, and fiber optic temperature sensors. The thermocouple sensors have a measurement accuracy of ±0.5°C and are housed in 304 stainless steel. An infrared sensor array with a resolution of 0.1°C is deployed every 2 meters along the dam's axis. Distributed fiber optic temperature sensors are deployed at 0.5-meter intervals along the boundaries of the target area. Electromagnetic interference-resistant fiber optic temperature sensors are wired in a serpentine pattern along key structural sections, forming a star topology. The master node connects to the child nodes via a CAN bus, ensuring automatic switching to a backup link in the event of a single point of failure. All sensors are synchronized (error ≤5ms) using the Network Time Protocol (NTP), with synchronization errors controlled to within 3ms.

[0051] In the embodiment of the present invention, Figure 1 As shown in the figure, the edge computing node integrates a local Kalman filter module, a dynamic weight adjustment module, and an encryption communication and data packaging module.

[0052] Among them, the local Kalman filter module is used to perform a spatiotemporal joint Kalman filter algorithm on the temperature data of the target area to eliminate noise. The state prediction formula of the spatiotemporal joint Kalman filter algorithm is: in Indicates the current moment (i.e. k The state vector at time , Indicates the previous moment (i.e. k −1 moment) to the current moment, reflecting the state transition matrix of the system from k −1 time has passed k The state change law of the moment, represents the state vector at the previous moment, Represents the control matrix at the current moment, describing the impact of the control input on the system state, represents the control input vector at the current moment, Represents the process noise vector at the current moment.

[0053] In the embodiment of the present invention, the state transition matrix According to the heat conduction characteristics, it is set as: in Represents the discretized time step.

[0054] The state update formula of the spatiotemporal joint Kalman filter algorithm is: in Represents the state estimation value at the current moment, that is, the generated local fusion data, represents the Kalman gain at the current moment, represents the observation value at the current moment, Represents the observation matrix at the current moment.

[0055] The dynamic weight adjustment module is used to dynamically assign weights to sensors in the temperature sensor network: in Represents the first i The weight of each sensor, Represents the first i The historical error variance of each sensor, Represents the first j The historical error variance of each sensor, n Represents the number of sensors in the temperature sensor network. The above dynamic weight allocation formula ensures that the smaller the variance (ie, the more reliable) the sensor, the greater the weight. In this embodiment of the present invention, when the historical error variance of a sensor is When the traffic suddenly increases by 200%, its weight is automatically reduced to below 0.2.

[0056] The encrypted communication and data packaging module is used to generate a lightweight data packet based on the locally fused data, which is encrypted using the AES-256 algorithm and uploaded to the cloud server. In this embodiment of the present invention, the lightweight data packet includes the locally fused data, filtered temperature values, weight coefficients, and coordinate information. Low-bandwidth transmission is achieved between the edge computing node and the cloud server via the MQTT protocol, while bidirectional certificate authentication is established, keeping communication latency to less than 300ms. This solution reduces the cloud network load while preventing man-in-the-middle attacks and data tampering.

[0057] In the embodiment of the present invention, Figure 1 As shown in the figure, the cloud server integrates a global Kalman filter module, a three-dimensional temperature field reconstruction module, an anomaly detection and self-correction module, and a temperature distribution warning module.

[0058] The global Kalman filter module is used to fuse local fusion data from multiple edge computing nodes. Through spatiotemporal joint Kalman filter optimization, it achieves the unification of cross-regional temperature information and eliminates the discontinuity of temperature fields between different target areas.

[0059] The 3D temperature field reconstruction module is used to construct a 3D temperature field based on the heat conduction model. The heat conduction model is: in T represents the temperature distribution function, t Indicates time, represents the thermal diffusivity, ∇ 2 T The Laplace operator representing temperature describes the trend of temperature change in space (i.e., heat conduction).

[0060] The specific steps for constructing a three-dimensional temperature field based on the heat conduction model are: First, the above heat conduction model is established; secondly, the heat conduction model is discretized using the finite difference method, and the target monitoring area is divided into equal volume voxel units (0.1m in the embodiment of the present invention). 3 ), and set the time step (10 seconds in the embodiment of the present invention) to construct a spatiotemporal coupled temperature evolution model; then, the local fused data is input into the temperature evolution model as the initial boundary condition, and the local fused data of multiple edge computing nodes are fused in combination with the global spatiotemporal joint Kalman filtering algorithm to eliminate the discontinuity and data holes in the temperature fields between different target areas. The global consistency and accuracy of temperature estimation are improved through iterative optimization to generate a preliminary three-dimensional temperature field; finally, for blind areas where sensors have not yet been deployed, an adaptive Kriging interpolation algorithm is introduced on the basis of the preliminary three-dimensional temperature field. Interpolation weights are dynamically allocated based on the temperature gradients and spatial correlations of neighboring areas to fill the data in the blind areas, improve the continuity and physical consistency of the preliminary three-dimensional temperature field, and obtain a three-dimensional temperature field with global consistency and high spatial resolution.

[0061] In the reconstructed three-dimensional temperature field, the system supports the visualization of the positions and monitoring contributions of various sensors to enhance the understanding and explanation of the sensor layout and temperature measurement mechanism. For example: abnormal points are marked as locations with local temperature mutations or significant temperature differences with the surrounding areas in the thermal map, which intuitively reflect potential structural risk areas; thermocouples are used as contact-type high-precision measurement devices and are deployed at locations where temperature changes drastically or at key structural positions (such as planes at different height levels of pyramid-shaped structural models, edge nodes, etc.) to capture temperature changes at key positions; infrared sensors are deployed in a non-contact manner at an appropriate distance outside the model, covering a wider area, and completing the overall scan of the temperature field by scanning surface thermal radiation; fiber optic temperature sensors can be deployed along the structural lines of the model and are suitable for continuous temperature change monitoring, such as being set near the edges of the structure, internal channels or key lines to achieve high-resolution temperature sensing along the line. Finally, the cloud server can generate a three-dimensional thermal map that integrates the above measurement results, achieving high-resolution, full-view display of spatial temperature distribution, and serving as an important basis for abnormal warning and structural safety assessment, such as Figure 2 shown.

[0062] The anomaly detection and self-correction module is used to detect sensor anomalies based on local fusion data. In response to the detection of sensor anomalies, the process noise covariance matrix, the observation noise covariance matrix, and the sensor weights are adaptively adjusted: in Represents the process noise covariance matrix at the current moment. In the embodiment of the present invention, The initial value is set to 0.01I, where I is the identity matrix, represents the smoothing coefficient, represents the process noise covariance matrix at the previous moment, represents the state prediction error at the current moment, represents the observation noise covariance matrix at the current moment, represents the attenuation factor, represents the observation noise covariance matrix at the previous moment, represents the observation value at the current moment, represents the observation matrix, represents the estimated value of the state at the current moment, Indicates the current temperature sensor network i The weight of each sensor, Represents the first temperature sensor in the adjusted temperature sensor network i The weight of each sensor, Indicates temperature deviation, Indicates the preset temperature threshold, which is set to 3°C.

[0063] In the embodiment of the present invention, the method for detecting sensor abnormality is specifically as follows: In response to a temperature deviation of Δ in any target area for three consecutive cycles T >2℃, calculate the temperature data variance of the target area : in Indicates the number of temperature sensors in the target area, Indicates the i The temperature measurement value of each temperature sensor in the current detection cycle, Indicates the arithmetic mean of all temperature measurements in the current detection cycle.

[0064] In response to , it is determined that the sensor is abnormal.

[0065] The temperature distribution warning module is used to output an abnormal warning signal when the temperature deviation exceeds the preset temperature threshold.

[0066] In this embodiment of the present invention, a visualization rendering module is also deployed in the cloud server to graphically process the reconstructed three-dimensional temperature field. Temperature distribution information is overlaid via a heat map, providing an intuitive presentation of temperature trends. The system supports multi-angle rotation and multi-level zooming, facilitating in-depth analysis of the temperature status of specific areas. Temperature anomalies identified by the system are highlighted to improve risk identification efficiency and assist operations and maintenance personnel in quickly locating potential hazards. The temperature distribution warning module utilizes a multi-level response mechanism, automatically triggering appropriate system behavior adjustments based on the degree of temperature deviation. For example, when temperature fluctuations approach a preset threshold, the system can adaptively increase the data sampling frequency of edge computing nodes to enhance local monitoring sensitivity. When an abnormal change is detected exceeding the preset temperature threshold, the system triggers a coordinated control process, sending safety instructions to relevant control units to implement protective control of the operating status of critical equipment or structures, mitigating accident risks. Furthermore, warning information can be synchronized in real time via the cloud server to PCs or mobile devices, and alerts are provided through various channels (including pop-up windows, SMS notifications, and audio and visual signals), ensuring that relevant management personnel receive critical information immediately. At the same time, the system supports complete recording of all warning events and response processes, and combines historical data for compressed storage and archiving management, providing data support and decision-making basis for subsequent safety assessments and fault tracing.

[0067] Example 2: The embodiment of the present invention provides a multi-sensor fusion temperature detection method based on the Kalman filter algorithm. Figure 3 As shown, the following steps S1 to S6 are included: S1. Periodically collect temperature data of the target area through the temperature sensor network.

[0068] In an embodiment of the present invention, the temperature sensor network includes thermocouples, infrared sensors, and fiber optic temperature sensors. The thermocouple sensor has a measurement accuracy of ±0.5°C and uses a 304 stainless steel housing. An array of infrared sensors is arranged every 2 meters along the axis of the dam, with a resolution of 0.1°C. Distributed fiber optic temperature sensors are deployed at the boundary of the target area at intervals of 0.5 meters. Anti-electromagnetic interference fiber optic temperature sensors are used, and the network is serpentinely wired along the key sections of the structure. A star topology is adopted, and the master node is connected to the child node via the CAN bus to ensure automatic switching to the backup link in the event of a single point failure. All sensors achieve time synchronization (error ≤5ms) through the Network Time Protocol (NTP), and the synchronization error is controlled within 3ms.

[0069] S2. Perform local Kalman filter preprocessing on the temperature data of the target area at the edge computing node to eliminate noise and generate local fusion data.

[0070] In the embodiment of the present invention, local Kalman filter preprocessing is implemented by a spatiotemporal joint Kalman filter algorithm. The state prediction formula of the spatiotemporal joint Kalman filter algorithm is: in Indicates the current moment (i.e. k The state vector at time , Indicates the previous moment (i.e. k −1 moment) to the current moment, reflecting the state transition matrix of the system from k −1 time has passed k The state change law of the moment, represents the state vector at the previous moment, Represents the control matrix at the current moment, describing the impact of the control input on the system state, represents the control input vector at the current moment, Represents the process noise vector at the current moment.

[0071] In the embodiment of the present invention, the state transition matrix According to the heat conduction characteristics, it is set as: The state update formula of the spatiotemporal joint Kalman filter algorithm is: in Represents the state estimation value at the current moment, that is, the generated local fusion data, represents the Kalman gain at the current moment, represents the observation value at the current moment, Represents the observation matrix at the current moment.

[0072] S3, upload the local fusion data to the cloud server through the MQTT protocol.

[0073] In this embodiment of the present invention, low-bandwidth transmission between edge computing nodes and cloud servers is achieved through the MQTT protocol, while establishing two-way certificate authentication and controlling communication latency to less than 300ms. This solution reduces cloud network load while preventing man-in-the-middle attacks and data tampering.

[0074] S4. In the cloud server, based on the local fusion data, the three-dimensional temperature field is reconstructed based on the heat conduction model and the global spatiotemporal joint Kalman filtering algorithm.

[0075] Step S4 includes the following sub-steps S41 to S44: S41. Establish a heat conduction model: in T represents the temperature distribution function, t Indicates time, represents the thermal diffusivity, ∇ 2 T The Laplacian operator representing temperature.

[0076] S42, using the finite difference method to discretize the heat conduction model, and dividing the target monitoring area into equal volume voxel units (0.1m in the embodiment of the present invention). 3 ), and set the time step (10 seconds in the embodiment of the present invention) to build a spatiotemporal coupled temperature evolution model.

[0077] S43. Input the local fusion data as the initial boundary condition into the temperature evolution model, and combine the global spatiotemporal joint Kalman filtering algorithm to fuse the local fusion data of multiple edge computing nodes, eliminate the discontinuity and data holes in the temperature field between different target areas, and improve the global consistency and accuracy of temperature estimation through iterative optimization to generate a preliminary three-dimensional temperature field.

[0078] S44. For blind areas where sensors have not yet been deployed, an adaptive Kriging interpolation algorithm is introduced on the basis of the preliminary three-dimensional temperature field. Interpolation weights are dynamically allocated based on the temperature gradient and spatial correlation of the adjacent areas to fill the data in the blind areas, improve the continuity and physical consistency of the preliminary three-dimensional temperature field, and obtain a three-dimensional temperature field with global consistency and high spatial resolution.

[0079] In the reconstructed three-dimensional temperature field, the system supports the visualization of the positions and monitoring contributions of various sensors to enhance the understanding and explanation of the sensor layout and temperature measurement mechanism. For example: abnormal points are marked as locations with local temperature mutations or significant temperature differences with the surrounding areas in the thermal map, which intuitively reflect potential structural risk areas; thermocouples are used as contact-type high-precision measurement devices and are deployed at locations where temperature changes drastically or at key structural positions (such as planes at different height levels of pyramid-shaped structural models, edge nodes, etc.) to capture temperature changes at key positions; infrared sensors are deployed in a non-contact manner at an appropriate distance outside the model, covering a wider area, and completing the overall scan of the temperature field by scanning surface thermal radiation; fiber optic temperature sensors can be deployed along the structural lines of the model and are suitable for continuous temperature change monitoring, such as being set near the edges of the structure, internal channels or key lines to achieve high-resolution temperature sensing along the line. Finally, the cloud server can generate a three-dimensional thermal map that integrates the above measurement results, achieving high-resolution, full-view display of spatial temperature distribution, and serving as an important basis for abnormal warning and structural safety assessment, such as Figure 2 shown.

[0080] S5. In the cloud server, sensor anomaly detection is performed based on the local fusion data, and Kalman filter parameters and sensor weights are adaptively adjusted.

[0081] In an embodiment of the present invention, the method for detecting sensor anomalies based on local fusion data is specifically as follows: In response to a temperature deviation of Δ in any target area for three consecutive cycles T >2℃, calculate : In response to , it is determined that the sensor is abnormal.

[0082] In the embodiment of the present invention, the Kalman filter parameters include the process noise covariance matrix and the observation noise covariance matrix. The adaptive adjustment formulas of the process noise covariance matrix and the observation noise covariance matrix are as follows: in Represents the process noise covariance matrix at the current moment. In the embodiment of the present invention, The initial value is set to 0.01I, where I is the identity matrix, represents the smoothing coefficient, represents the process noise covariance matrix at the previous moment, represents the state prediction error at the current moment, represents the observation noise covariance matrix at the current moment, represents the attenuation factor, represents the observation noise covariance matrix at the previous moment, represents the observation value at the current moment, represents the observation matrix, Indicates the estimated value of the state at the current moment.

[0083] In the embodiment of the present invention, the adaptive adjustment formula of the sensor weight is: in Indicates the current temperature sensor network i The weight of each sensor, Represents the first temperature sensor in the adjusted temperature sensor network i The weight of each sensor, Indicates temperature deviation, Indicates the preset temperature threshold, which is set to 3°C.

[0084] S6. Output three-dimensional temperature distribution map and abnormal warning signal.

[0085] In an embodiment of the present invention, a visualization rendering module is also deployed in the cloud server to graphically process the reconstructed three-dimensional temperature field. Temperature distribution information is superimposed via a heat map to provide an intuitive representation of temperature trends. The system supports multi-angle rotation and multi-level zooming, allowing users to conduct in-depth analysis of the temperature status of specific areas. Temperature anomalies identified by the system are highlighted to improve risk identification efficiency and assist operations and maintenance personnel in quickly locating potential hazards.

[0086] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in the present invention without departing from the essence of the present invention, and such variations and combinations are still within the scope of protection of the present invention.

Claims

1. A multi-sensor fusion temperature detection system based on Kalman filter algorithm, characterized in that: Includes temperature sensor network, edge computing nodes and cloud servers; The temperature sensor network is distributed in the target area and is used to periodically collect temperature data of the target area; The edge computing node is communicatively connected to the temperature sensor network, and is used to perform local Kalman filter preprocessing on the temperature data of the target area to eliminate noise and generate local fusion data; The cloud server is connected to the edge computing node through the MQTT protocol, and is used to reconstruct the three-dimensional temperature field based on the local fusion data, the heat conduction model and the global space-time joint Kalman filtering algorithm, and output the three-dimensional temperature distribution map and abnormal warning signal.

2. The multi-sensor fusion temperature detection system based on the Kalman filter algorithm according to claim 1 is characterized in that: The temperature sensor network includes thermocouples, infrared sensors and optical fiber temperature sensors.

3. The multi-sensor fusion temperature detection system based on the Kalman filter algorithm according to claim 1 is characterized in that: The edge computing node is integrated with a local Kalman filter module, a dynamic weight adjustment module, and an encryption communication and data packaging module; The local Kalman filter module is used to perform a spatiotemporal joint Kalman filter algorithm on the temperature data of the target area to eliminate noise. The state prediction formula of the spatiotemporal joint Kalman filter algorithm is: in represents the state vector at the current moment, Represents the state transfer matrix from the previous moment to the current moment, represents the state vector at the previous moment, represents the control matrix at the current moment, represents the control input vector at the current moment, represents the process noise vector at the current moment; The state update formula of the spatiotemporal joint Kalman filter algorithm is: in Represents the state estimation value at the current moment, that is, the generated local fusion data, represents the Kalman gain at the current moment, represents the observation value at the current moment, Represents the observation matrix at the current moment; The dynamic weight adjustment module is used to dynamically assign weights to sensors in the temperature sensor network: in Represents the first i The weight of each sensor, Represents the first i The historical error variance of each sensor, Represents the first j The historical error variance of each sensor, n Indicates the number of sensors in the temperature sensor network; The encryption communication and data packaging module is used to generate a lightweight data packet based on the local fusion data, and upload it to the cloud server after encryption using the AES-256 algorithm.

4. The multi-sensor fusion temperature detection system based on the Kalman filter algorithm according to claim 1 is characterized in that: The cloud server is integrated with a global Kalman filter module, a three-dimensional temperature field reconstruction module, an anomaly detection and self-correction module, and a temperature distribution early warning module; The global Kalman filter module is used to fuse local fusion data from multiple edge computing nodes, and achieve the unification of cross-regional temperature information through spatiotemporal joint Kalman filter optimization, eliminating the discontinuity of temperature fields between different target areas; The three-dimensional temperature field reconstruction module is used to construct a three-dimensional temperature field according to a heat conduction model. The heat conduction model is: in T represents the temperature distribution function, t Indicates time, represents the thermal diffusivity, ∇ 2 T Laplace operator representing temperature; The anomaly detection and self-correction module is used to detect sensor anomalies based on local fusion data, and in response to detecting sensor anomalies, adaptively adjust the process noise covariance matrix, the observation noise covariance matrix, and the sensor weights: in represents the process noise covariance matrix at the current moment, represents the smoothing coefficient, represents the process noise covariance matrix at the previous moment, represents the state prediction error at the current moment, represents the observation noise covariance matrix at the current moment, represents the attenuation factor, represents the observation noise covariance matrix at the previous moment, represents the observation value at the current moment, represents the observation matrix, represents the estimated value of the state at the current moment, Indicates the current temperature sensor network i The weight of each sensor, Represents the first temperature sensor in the adjusted temperature sensor network i The weight of each sensor, Indicates temperature deviation, Indicates the preset temperature threshold; The temperature distribution warning module is used to output an abnormal warning signal when the temperature deviation exceeds a preset temperature threshold.

5. A multi-sensor fusion temperature detection method based on Kalman filter algorithm, characterized in that: The following steps are involved: S1, periodically collect temperature data of the target area through the temperature sensor network; S2. Perform local Kalman filter preprocessing on the temperature data of the target area at the edge computing node to eliminate noise and generate local fusion data; S3, upload the local fusion data to the cloud server via the MQTT protocol; S4. In the cloud server, based on the local fusion data, the three-dimensional temperature field is reconstructed using the heat conduction model and the global spatiotemporal joint Kalman filter algorithm; S5. In the cloud server, sensor anomaly detection is performed based on the local fusion data, and Kalman filter parameters and sensor weights are adaptively adjusted; S6. Output three-dimensional temperature distribution map and abnormal warning signal.

6. The multi-sensor fusion temperature detection method based on the Kalman filter algorithm according to claim 5 is characterized in that: In step S2, local Kalman filter preprocessing is implemented by using a spatiotemporal joint Kalman filter algorithm. The state prediction formula of the spatiotemporal joint Kalman filter algorithm is: in represents the state vector at the current moment, Represents the state transfer matrix from the previous moment to the current moment, represents the state vector at the previous moment, represents the control matrix at the current moment, represents the control input vector at the current moment, represents the process noise vector at the current moment; The state update formula of the spatiotemporal joint Kalman filter algorithm is: in Represents the state estimation value at the current moment, that is, the generated local fusion data, represents the Kalman gain at the current moment, represents the observation value at the current moment, Represents the observation matrix at the current moment.

7. The multi-sensor fusion temperature detection method based on the Kalman filter algorithm according to claim 5 is characterized in that: The step S4 comprises the following sub-steps: S41. Establish a heat conduction model: in T represents the temperature distribution function, t Indicates time, represents the thermal diffusivity, ∇ 2 T Laplace operator representing temperature; S42. Use the finite difference method to discretize the heat conduction model, divide the target monitoring area into equal-volume voxel units, set the time step, and construct a spatiotemporal coupled temperature evolution model; S43. Input the local fused data as the initial boundary conditions into the temperature evolution model. Combined with the global spatiotemporal joint Kalman filter algorithm, the local fused data of multiple edge computing nodes are fused to eliminate the discontinuity and data holes in the temperature field between different target areas. The global consistency and accuracy of the temperature estimation are improved through iterative optimization to generate a preliminary three-dimensional temperature field. S44. For blind areas where sensors have not yet been deployed, an adaptive Kriging interpolation algorithm is introduced on the basis of the preliminary three-dimensional temperature field. Interpolation weights are dynamically allocated based on the temperature gradient and spatial correlation of the adjacent areas to fill the data in the blind areas, improve the continuity and physical consistency of the preliminary three-dimensional temperature field, and obtain a three-dimensional temperature field with global consistency and high spatial resolution.

8. The multi-sensor fusion temperature detection method based on the Kalman filter algorithm according to claim 5 is characterized in that: The method for performing sensor anomaly detection based on local fusion data in step S5 is specifically as follows: In response to a temperature deviation of Δ in any target area for three consecutive cycles T >2℃, calculate the temperature data variance of the target area : in Indicates the number of temperature sensors in the target area, Indicates the i The temperature measurement value of each temperature sensor in the current detection cycle, Indicates the arithmetic mean of all temperature measurements within the current detection cycle; In response to , it is determined that the sensor is abnormal.

9. The multi-sensor fusion temperature detection method based on the Kalman filter algorithm according to claim 5 is characterized in that: The Kalman filter parameters in step S5 include the process noise covariance matrix and the observation noise covariance matrix. The adaptive adjustment formula of the process noise covariance matrix and the observation noise covariance matrix is: in represents the process noise covariance matrix at the current moment, represents the smoothing coefficient, represents the process noise covariance matrix at the previous moment, represents the state prediction error at the current moment, represents the observation noise covariance matrix at the current moment, represents the attenuation factor, represents the observation noise covariance matrix at the previous moment, represents the observation value at the current moment, represents the observation matrix, Indicates the estimated value of the state at the current moment.

10. The multi-sensor fusion temperature detection method based on the Kalman filter algorithm according to claim 5, characterized in that: The adaptive adjustment formula of the sensor weight in step S5 is: in Indicates the current temperature sensor network i The weight of each sensor, Represents the first temperature sensor in the adjusted temperature sensor network i The weight of each sensor, Indicates temperature deviation, Indicates the preset temperature threshold.

Images