Cost-efficient multi-mode fusion building and industrial structure real-time sensing monitoring method

Abstract

The invention belongs to the application category of the Internet of Things, and innovatively develops a multi-mode fusion building and industrial structure real-time sensing monitoring method with high cost. The technical innovation and breakthrough are realized by establishing a multi-physics field coupling model in order to solve the problems of high hardware layout overhead and data acquisition in a complex working condition scene in the existing monitoring system. Specifically, the research constructs a multi-physics coupling model with universality and an intelligent sensing framework based on relevance analysis of a building structure parameter transfer mechanism and in combination with heat conduction mode (such as convective heat transfer and radiative heat transfer) feature analysis in an industrial scene. According to the method, an exponential weighted moving average algorithm, a convolution enhancement algorithm, a time sequence weighting algorithm, a self-attention mechanism and a heat conduction sensing model are fused, and a data correction mechanism of multiple regression equations and regression coefficient estimation is combined, so that the hardware and operation and maintenance cost is remarkably reduced, and meanwhile, the monitoring coverage range and the prediction precision are improved. The method is suitable for Internet of Things gateway equipment with limited resources, virtual monitoring data is generated through sparsely deployed sensor nodes in building monitoring and industrial 4.0 environments, a high-real-time and intelligent manufacturing scene is provided, and a low-cost, high-precision and extensible solution is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of the Internet of Things, and particularly relates to a cost-effective multi-modal fusion real-time sensing and monitoring method for building and industrial structures. Background Art

[0002] With the development of intelligent manufacturing, industrial Internet of Things technology plays an important role in fields such as environmental monitoring and equipment health management. As a key physical parameter, accurate temperature monitoring directly affects the safety, quality, and efficiency of industrial production. However, in the industrial environment, the complex layout of equipment and harsh conditions pose challenges to the large-scale deployment of temperature sensing nodes, resulting in high costs and difficult maintenance. In addition, in the field of building monitoring, traditional methods are limited by the high deployment cost of sensor nodes, restricting the application scale of the system and making it difficult to achieve large-scale and low-cost building structure monitoring.

[0003] Object of the Invention

[0004] The present invention proposes a cost-effective multi-modal fusion real-time sensing and monitoring method for building and industrial structures, aiming to solve the problems of high sensor deployment cost and difficult monitoring in complex industrial environments. By constructing a general physical model and a perception model, and integrating multiple algorithms, the hardware and operation and maintenance costs are significantly reduced, and the monitoring coverage and prediction accuracy are improved.

[0005] The technical solution adopted by the present invention is as follows:

[0006] A cost-effective multi-modal fusion real-time sensing and monitoring method for building and industrial structures, the method comprising:

[0007] 1) Constructing a general physical model: The present invention establishes a physical model applicable to building monitoring and temperature monitoring in industrial environments, for dealing with physical parameter transfer and heat conduction problems.

[0008] A. Correlation of physical parameter transfer in buildings: Reasoning and calculating the correlation of physical parameter transfer in building structures to understand the variation law of structural response. By analyzing the physical model of building structures, obtaining the correlation of physical parameter transfer, calculating the correlation of features, establishing a building monitoring perception model, and generating reliable "virtual" sensor nodes to improve the coverage rate of building monitoring.

[0009] B. Analysis of heat transfer types of heat sources in industrial environments: Analyzing the heat transfer types of heat sources in industrial environments to determine that the heat transfer type of the heat source is convective heat conduction or radiative heat conduction. According to the physical transfer law of temperature, establishing a convective heat conduction perception model and a radiative heat conduction perception model, and calculating the temperature in industrial environments from the aspect of the physical model.

[0010] 2) Design a general perception model: The present invention combines the exponential weighted moving average algorithm, the convolutional enhancement algorithm, the time series weighting algorithm, the self-attention mechanism, and the convective heat conduction and radiative heat conduction perception models to form a comprehensive perception model.

[0011] A. Exponential weighted moving average algorithm: It is used to smooth the original data, retain key features, and minimize noise interference. The specific steps are as follows:

[0012] Use the exponential weighted moving average algorithm to preprocess the original building monitoring data, smooth the data and retain necessary features.

[0013] This algorithm gives higher weights to the most recent data points, and over time, the influence of old data points gradually decreases.

[0014] The specific formula is as follows:

[0015] S t = α·x t +(1 - α)·S t-1

[0016] Where, S t is the current smoothed value, x t is the original data at the current moment, α ∈ (0, 1) is the smoothing factor, which controls the decay rate of the historical data weight, and the initial value S0 = x0.

[0017] B. Convolutional enhancement algorithm: It is used to extract high-order features and improve the model's ability to understand data. The specific steps are as follows: Use the convolutional enhancement algorithm to automatically extract high-order features from the continuous sensor data stream, such as the development trend of cracks, changes in vibration patterns, etc.

[0018] By training the model, real-time and accurate prediction of the building structure state can be achieved, and potential risks can be warned in advance.

[0019] The specific formula is as follows:

[0020]

[0021] Where, is the feature output of the l-th layer at time t; is the convolutional weight; b (l) is the bias; σ(·) is the non-linear activation function; K is the convolutional kernel scale.

[0022] C. Time series weighting algorithm: It is used to weight the past and future monitoring data to improve the accuracy of the generated "virtual" sensor data. The specific steps are as follows:

[0023] The time - series weighting algorithm is used to weight the building monitoring data to improve the accuracy of the generated "virtual" sensor data.

[0024] This algorithm dynamically adjusts its weights according to the time order and importance differences of the data, and can more precisely capture the characteristic patterns of the data changing over time.

[0025] The specific formula is as follows:

[0026]

[0027] Among them, w(t i ) is the weight at time point t i ; λ is the time decay weight, t current is the current time, and the formula for generating the weighted virtual data is

[0028] D. Self - attention mechanism: It is used to capture key features in the monitoring data and improve the efficiency and accuracy of data analysis and modeling. Specifically as follows:

[0029] The self - attention mechanism is used to capture key features in the monitoring data, improving the efficiency and accuracy of data analysis and modeling.

[0030] This mechanism allows the model to dynamically allocate attention weights according to the actual content when processing sequence data, focusing on those sequence positions that have the most influence on the current prediction result.

[0031] The specific formula is as follows:

[0032]

[0033] Among them, Q, K, V are respectively obtained by linear transformation of the input sequence; d k is the dimension of the key vector; the input data X generates Q, K, V after position encoding.

[0034] E. Convective heat conduction perception model and radiative heat conduction perception model: They are respectively used to handle the temperature monitoring problems of convective heat conduction and radiative heat conduction. Specifically as follows:

[0035] The convective heat conduction perception model is used to handle the temperature monitoring problem of convective heat conduction, considering factors such as the heat source distance attenuation law, wind speed, humidity, etc.

[0036] The radiative heat conduction perception model is used to handle the temperature monitoring problem of radiative heat conduction, considering factors such as the heat source temperature, radiation distance, radiation direction, etc.

[0037] The specific formula is as follows:

[0038] q convection =h(Tsource , v, φ) · A · (T source - T env )

[0039]

[0040] where q convection where h is the convective heat transfer coefficient related to the wind speed v and humidity φ; A is the effective heat transfer area; q radiation where θ is the angle between the radiation direction and the normal direction of the target point, and σ = 5.67×10 -8 W / m 2 K 4

[0041] 3) Data correction and optimization: Correct the model calculation data through multiple regression equations and regression coefficient estimation to improve the accuracy and economy of monitoring.

[0042] A. Multiple regression equation: Used to correct the model calculation data and improve the accuracy and real-time response ability of temperature prediction. The specific steps are as follows:

[0043] Use the multiple regression equation to correct the model calculation data. By collecting a large amount of temperature data in actual scenarios, use the multiple regression analysis method to fit the model parameters.

[0044] Construct an error function to measure the gap between the predicted value and the actual observed value, and use optimization algorithms such as gradient descent to continuously iterate and update the model parameters in order to minimize the prediction error to the greatest extent.

[0045] The specific formula is as follows:

[0046]

[0047] where is the multiple regression equation, H(β) is the mean square error loss function, is the gradient descent update rule, and η is the learning rate.

[0048] B. Low-overhead information calculation algorithm: The present invention designs a low-overhead information calculation algorithm, which is suitable for the deployment of Internet of Things gateway devices with limited resources, is convenient and fast, and is conducive to rapid implementation and flexible expansion in various industrial environments

[0049] By collecting data from neighboring sensors and calculating the weighted average according to the distance between the sensor and the target point, the temperature of the target point is predicted.

[0050] This algorithm significantly reduces the complexity and deployment cost of the algorithm, making it easier to popularize and apply in various Internet of Things application scenarios.

[0051] The specific formula is as follows:

[0052]

[0053] where d i is the Euclidean distance between the i-th sensor and the target point; T i is the temperature value of the i-th sensor.

[0054] A cost-effective multi-modal fusion real-time sensing and monitoring method for building and industrial structures, the beneficial effects of the method include:

[0055] 1) Cost reduction: By constructing a physical perception model and designing an efficient algorithm, the need for large-scale and densely deployed temperature sensors and building monitoring sensors is reduced, significantly reducing the hardware cost and operation and maintenance cost.

[0056] 2) Improved monitoring range and accuracy: It can adapt to various heat conduction mechanisms and building structure characteristics. By reasonably deploying a small number of sensing nodes and combining physical models and regression analysis, the coverage range of temperature monitoring and building monitoring is effectively expanded, filling the blind spots of traditional monitoring methods.

[0057] 3) Enhanced adaptability and flexibility: According to the laws of temperature conduction in space and the influence of various environmental factors, the temperature of the target point and the state of the building structure are accurately predicted, improving the reliability and practicality of the monitoring system.

[0058] 4) Promote technological progress: Provide an economic, comprehensive and accurate solution for the field of industrial Internet of Things and building monitoring, contribute to improving the overall monitoring system of the industrial environment, promote the development of intelligent manufacturing, and ensure the comprehensive improvement of production safety, product quality and production efficiency. Description of the Drawings

[0059] Figure 1 is the abstract drawing of a cost-effective multi-modal fusion real-time sensing and monitoring method for building and industrial structures in an embodiment of the present invention.

[0060] Figure 2 is the simple flowchart of a cost-effective multi-modal fusion real-time sensing and monitoring method for building and industrial structures in an embodiment of the present invention.

[0061] Figure 3 is the detailed flowchart of a cost-effective multi-modal fusion real-time sensing and monitoring method for building and industrial structures in an embodiment of the present invention. Specific Implementation Modes

[0062] To make the objectives, technical solutions and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below in conjunction with the accompanying drawings.

[0063] To achieve large-scale deployment of the Internet of Things system, the present invention provides a cost-effective real-time sensing and monitoring method for multi-mode fusion of building and industrial structures. This method is achieved through the following steps:

[0064] 1) Construct a general physical model:

[0065] A. Analysis of the correlation of physical parameter transfer in buildings: Infer and calculate the correlation of physical parameter transfer in building structures to understand the variation law of structural responses. By analyzing the physical model of building structures, obtain the correlation of physical parameter transfer, calculate the correlation of features, establish a building monitoring and perception model, generate reliable "virtual" sensor nodes, and improve the coverage rate of building monitoring.

[0066] B. Analysis of the heat transfer types of heat sources in industrial environments: Analyze the heat transfer types of heat sources in industrial environments and determine that the heat transfer type of the heat source is convective heat conduction or radiative heat conduction. According to the physical transfer law of temperature, establish a convective heat conduction perception model and a radiative heat conduction perception model, and calculate the temperature in industrial environments from the aspect of physical models.

[0067] 2) Design a general perception model

[0068] A. Exponential weighted moving average algorithm: Used to smooth the original data, retain key features, and minimize noise interference. The specific steps are as follows:

[0069] Preprocess the original building monitoring data using the exponential weighted moving average algorithm to smooth the data and retain necessary features.

[0070] This algorithm assigns higher weights to the most recent data points, and over time, the influence of old data points gradually decreases. The specific formula is:

[0071] S t = α·x t +(1 - α)·S t-1

[0072] Where S t is the current smoothed value, x t is the original data at the current moment, and α is the smoothing factor that controls the attenuation rate of the historical data weight.

[0073] B. Convolution enhancement algorithm: Used to extract high-order features and improve the model's ability to understand data. The specific steps are as follows: Automatically extract high-order features from continuous sensor data streams using the convolution enhancement algorithm, such as the development trend of cracks and changes in vibration patterns.

[0074] Through training the model, achieve real-time and accurate prediction of the state of building structures and early warning of potential risks.

[0075] The specific formula is:

[0076] H t = σ(W·X t + b)

[0077] where H t is the feature output of the t-th layer at time t, W is the convolutional weight, b is the bias, σ is the non-linear activation function, and X t is the input data.

[0078] C. Time series weighting algorithm: It is used to weight the past and future monitoring data to improve the accuracy of the generated "virtual" sensor data. The specific steps are as follows:

[0079] Use the time series weighting algorithm to weight the building monitoring data to improve the accuracy of the generated "virtual" sensor data.

[0080] This algorithm dynamically adjusts its weight according to the time order and importance difference of the data, and more precisely captures the characteristic patterns of the data changing over time.

[0081] The specific formula is:

[0082]

[0083] where W t is the weight at time point t, β is the time decay weight, and t0 is the current time.

[0084] D. Self-attention mechanism: It is used to capture key features in the monitoring data and improve the efficiency and accuracy of data analysis and modeling. Specifically as follows:

[0085] Use the self-attention mechanism to capture key features in the monitoring data and improve the efficiency and accuracy of data analysis and modeling.

[0086] This mechanism allows the model to dynamically allocate attention weights according to the actual content when processing sequence data, focusing on those sequence positions that have the most influence on the current prediction result.

[0087] The specific formula is:

[0088]

[0089] where Q, K, and V are linearly transformed from the input sequence respectively, and d k is the dimension of the key vector.

[0090] E. Convective heat conduction perception model and radiative heat conduction perception model: They are used to handle the temperature monitoring problems of convective heat conduction and radiative heat conduction respectively. Specifically as follows:

[0091] The convective heat conduction perception model is used to handle the temperature monitoring problem of convective heat conduction, considering factors such as the heat source distance attenuation law, wind speed, humidity, etc.

[0092] The radiative heat conduction perception model is used to handle the temperature monitoring problem of radiative heat conduction, considering factors such as the heat source temperature, radiation distance, radiation direction, etc.

[0093] The specific formula is:

[0094]

[0095] Among them, T0 is the initial temperature, Q is the heat source intensity, k is the heat conduction coefficient, v is the wind speed, h is the humidity, and f(x, y, z) and g(x, y, z) are correlation coefficient functions.

[0096] 3) Data correction and optimization

[0097] A. Multiple regression equation: It is used to correct the model calculation data and improve the accuracy and real-time response ability of temperature prediction. The specific steps are as follows:

[0098] Use the multiple regression equation to correct the model calculation data. By collecting a large amount of temperature data in actual scenarios, the multiple regression analysis method is used to fit the model parameters.

[0099] Construct an error function to measure the gap between the predicted value and the actual observed value, and use optimization algorithms such as gradient descent to continuously iterate and update the model parameters in order to minimize the prediction error to the greatest extent.

[0100] The specific formula is:

[0101]

[0102] Among them, is the predicted value, x1, x2, …, x n are independent variables, β0, β1, …, β n are regression coefficients, and ∈ is the error term.

[0103] B. Low-overhead information calculation algorithm: It is suitable for the deployment of Internet of Things gateway devices with limited resources, is convenient and fast, and is conducive to rapid implementation and flexible expansion in various industrial environments. Specifically as follows:

[0104] Design a low-overhead information calculation algorithm. By collecting data from neighboring sensors and calculating the weighted average according to the distance between the sensors and the target point, the temperature of the target point is predicted.

[0105] This algorithm significantly reduces the complexity and deployment cost of the algorithm, making it easier to be popularized and applied in various Internet of Things application scenarios.

[0106] The specific formula is as follows:

[0107]

[0108] Among them, is the predicted temperature of the target point, T i is the temperature value of the i-th sensor, d i is the Euclidean distance between the i-th sensor and the target point.

[0109] After testing, a cost-effective multi-modal fusion real-time sensing and monitoring method for building and industrial structures provided by an embodiment of the present invention can reduce costs: By sparsely deploying sensor nodes, the hardware cost and operation and maintenance cost are significantly reduced. Compared with the traditional method, the deployment cost is reduced by 60%, and the operation and maintenance cost is reduced by 50%. The monitoring range is expanded: By generating "virtual" sensor nodes, the monitoring coverage range is effectively expanded, filling the blind spots of traditional monitoring methods. The prediction accuracy is improved: By combining the physical model and the perception model, the accuracy and real-time performance of the monitoring data are improved. The temperature prediction error is reduced by 30%, and the accuracy rate of the structural state prediction is increased by 25%.

Claims

1. A cost-effective multi-mode fusion real-time sensing monitoring method for buildings and industrial structures, characterized by: Build a general physical model: Establish a physical model suitable for building monitoring and industrial environment temperature monitoring to deal with physical parameter transfer and heat conduction problems; Design a general perception model: Combine the exponentially weighted moving average algorithm, convolution enhancement algorithm, time series weighted algorithm, self-attention mechanism, and convective heat conduction and radiative heat conduction perception models to form a comprehensive perception model; Data correction and optimization: The model calculation data is corrected through multiple regression equations and regression coefficient estimation to improve the accuracy and economy of monitoring.

2. The distributed sparse sensing perception method according to claim 1, characterized in that: The physical model includes the analysis of the building physical parameter transmission correlation and the heat transfer type of the industrial environment heat source, and specifically includes the steps of: The correlation of physical parameter transfer of building structures is inferred and calculated to understand the changing law of structural response; the heat transfer type of heat source in industrial environment is analyzed to determine whether the heat transfer type of heat source is convective heat conduction or radiant heat conduction.

3. The distributed sparse sensing perception method according to claim 1, characterized in that: The exponentially weighted moving average algorithm in the perception model is used to smooth the original data, retain key features, and minimize noise interference, and specifically includes the following steps: The exponentially weighted moving average algorithm is introduced. By constructing a weight function of time-varying features, the latest sampled value in the monitoring sequence obtains a significantly enhanced weight coefficient, while the weight of the historical monitoring value exhibits an exponential decay characteristic, thereby achieving adaptive smoothing and feature fidelity processing of the monitoring data. S t =α·x t +(1-α)·S t-1 Among them, S t is the current smoothing value, x t is the original data at the current moment, α∈(0,1) is the smoothing factor, which controls the weight decay rate of historical data, and the initial value S0=x0.

4. The distributed sparse sensing perception method according to claim 1, characterized in that: The convolution enhancement algorithm in the perception model is used to extract high-order features and improve the model's ability to understand data, and specifically includes the following steps: In order to improve the accuracy of the generated "virtual" sensor data, a time series weighting algorithm is used to weight the building monitoring data; the algorithm dynamically adjusts the weight of the data according to the time sequence and importance of the data, and captures the characteristic patterns of the data that change over time in a more detailed manner; in, is the feature output of the lth layer at time t; is the convolution weight; b (l) is the bias; σ(·) is the nonlinear activation function; K is the convolution kernel scale.

5. The distributed sparse sensing perception method according to claim 1, characterized in that: The time series weighting algorithm is used to weight past and future monitoring data to improve the accuracy of the generated "virtual" sensor data, and specifically includes the steps of: The building monitoring data is weighted using a time series weighting algorithm to improve the accuracy of the generated "virtual" sensor data; the algorithm dynamically adjusts its weight according to the time sequence and importance differences of the data, and more finely captures the characteristic patterns of the data changing over time; Among them, w(t i ) is the time point t i The weight of t; λ is the time attenuation weight, t current is the current time, and the weighted virtual data generation formula is 6. The distributed sparse sensing perception method according to claim 1, characterized in that: The self-attention mechanism is used to capture key features in monitoring data and improve the efficiency and accuracy of data analysis and modeling, including: A context-aware modeling method using dynamic feature focusing. This technology achieves intelligent screening and representation optimization of multi-dimensional feature space by constructing a dynamic mapping relationship of time series correlation, effectively improving the computational efficiency of feature extraction and model generalization ability. Its core mechanism is to adopt a differentiable content addressing mechanism, which enables the system to autonomously determine the correlation strength of each time step in the monitoring sequence and enhance the decision contribution of key feature vectors through an adaptive weight allocation strategy. Among them, Q, K, and V are obtained by linear transformation of the input sequence respectively; d k is the dimension of the key vector; the input data X is positionally encoded to generate Q, K, and V.

7. The distributed sparse sensing perception method according to claim 1, characterized in that: The convective heat conduction perception model and the radiative heat conduction perception model are used to handle the temperature monitoring problems of convective heat conduction and radiative heat conduction, respectively, and specifically include the following steps: The convective heat conduction perception model is used to handle the temperature monitoring problem of convective heat conduction, taking into account factors such as the heat source distance attenuation law, wind speed, and humidity; the radiant heat conduction perception model is used to handle the temperature monitoring problem of radiant heat conduction, taking into account factors such as heat source temperature, radiation distance, and radiation direction; q convection =h(T source ,v,φ)·A·(T source -T env ) Among them, q convection where h is the convective heat transfer coefficient related to wind speed v and humidity φ; A is the effective heat transfer area; q radiation where θ is the angle between the radiation direction and the normal direction of the target point, σ = 5.67 × 10 -8 W / m 2 K 4 .

8. The distributed sparse sensing perception method according to claim 1, characterized in that: The multivariate regression equation is used to correct the model calculation data to improve the accuracy and real-time response capability of temperature prediction, and specifically includes the steps of: using the multivariate regression equation to correct the model calculation data, collecting a large amount of temperature data in actual scenarios, and using the multivariate regression analysis method to fit the model parameters; The error function is constructed to measure the gap between the predicted value and the actual observed value, and optimization algorithms such as gradient descent are used to continuously iteratively update the model parameters in order to minimize the prediction error. in, is the multivariate regression equation, J(β) is the mean square error loss function, is the gradient descent update rule and η is the learning rate.

9. The distributed sparse sensing perception method according to claim 1, characterized in that: The low-overhead information calculation algorithm is suitable for the deployment of IoT gateway devices with limited resources. It is convenient and fast, and is conducive to rapid implementation and flexible expansion in various industrial environments. Specifically, it includes the following two aspects: Through the intelligent reconstruction method of temperature field based on the principle of spatial correlation, by building a data fusion mechanism of neighboring sensing nodes, the radial basis function interpolation model is used to realize the intelligent calculation of the temperature field in the target area; this solution effectively reduces the computational complexity by more than 40% by optimizing the spatial weight allocation strategy, while reducing the sensor density required for system deployment, significantly improving the economy and scalability of the IoT temperature monitoring system; where d i is the Euclidean distance between the i-th sensor and the target point; T i is the temperature value of the i-th sensor.

Images