A multi-sensor fusion environment anomaly detection method under communication restricted conditions

Abstract

The present application relates to the technical field of fire monitoring, and particularly relates to a multi-sensor fusion environment anomaly detection method under communication restricted conditions. The present application collects multi-dimensional environmental parameters such as temperature, humidity, methane, carbon monoxide and carbon dioxide concentration in real time by sensor nodes, fuses multi-source data through Dempster-Shafer evidence theory, solves sensor conflicts and uncertainty, and then identifies fire anomalies based on the fused data by a one-dimensional convolutional neural network (1DCNN) classifier. The present application deploys the DSET filter and 1DCNN classifier on a microcontroller by using edge computing technology. This not only reduces the dependence on cloud computing, but also improves the real-time performance and data security of the system. Through local processing, the system can still maintain efficient operation even in the case of network interruption or delay.

Description

Technical Field

[0001] This invention relates to the field of fire monitoring technology, and in particular to a method for detecting environmental anomalies using multi-sensor fusion under conditions of limited communication. Background Technology

[0002] Traditional fire monitoring methods typically rely on single sensors, such as those for temperature or gas concentration, or rule-based systems based on fixed thresholds. These methods cannot comprehensively utilize multi-source data for cross-validation, resulting in a high false alarm rate. Furthermore, single sensor failures are sensitive and prone to generating false alarms. Additionally, raw data must be transmitted to the cloud for processing, which leads to significant delays in environments with limited communication, making it difficult to respond to abnormal events in a timely manner.

[0003] With the rapid development of IoT technology, sensor networks are widely used for real-time monitoring of environmental parameters to achieve fire monitoring. However, environments such as underground mines, remote areas, and industrial plants often face challenges due to limited communication, including insufficient network bandwidth, signal interference, communication delays, or a complete lack of communication infrastructure. These issues severely affect the real-time performance and reliability of traditional centralized or cloud-based monitoring systems.

[0004] To address these issues, several multi-sensor data fusion methods have been proposed in existing technologies. For example, fuzzy logic, statistical methods, and deep learning-based models have been used to improve the accuracy of data processing. However, existing methods typically only address single anomaly types, lacking the ability to identify complex anomalies, and do not clearly define the specific judgment logic for anomalies. These methods perform poorly when faced with sensor failures or data uncertainty, especially in communication-constrained environments, and cannot effectively meet real-time requirements. Furthermore, existing cloud-based machine learning methods are difficult to implement in communication-constrained environments and suffer from data security and latency issues. Summary of the Invention

[0005] This invention discloses a method for detecting environmental anomalies under communication-constrained conditions using multi-sensor fusion, the specific method of which is as follows:

[0006] The overall monitoring area is divided into n sub-areas, each sub-area is equipped with a sensor node array, each sensor node array includes k identical sensor nodes, and each sensor node is equipped with m different types of sensors.

[0007] Based on the data collected by the sensors, the trust level of each sensor is calculated to determine whether the data collected by each sensor is trustworthy.

[0008] For a sensor node that includes sensors that are not trustworthy, the joint trust level of all sensors within that sensor node is calculated using the Dempster combination rule and then weighted.

[0009] For sensors that are not trustworthy, a spatiotemporal fusion method is used to correct the data they employ.

[0010] Based on the joint trust level, calculate the trust adjustment factor for sensor data that is not trustworthy;

[0011] Data collected by all trustworthy sensors, as well as data corrected from untrustworthy sensors, are input into a 1DCNN classification model to complete fire anomaly identification; the neurons of the 1DCNN classification model have their weights adjusted by a trust level adjustment factor.

[0012] Furthermore, the confidence level of each sensor is calculated using the following method:

[0013] Let the sensor node array A1~A n Each sensor node is N1 to N2. k Each sensor node is equipped with sensors G1 to G1. m ;

[0014] For any sensor node x∈[G1,...,G1] in a sensor array, m The framework for distinguishing regional environmental states is defined as follows:

[0015] FOD x ={{H},{N}}

[0016] Where H represents a dangerous state and N represents a non-dangerous state;

[0017] Then its event set POW x for:

[0018]

[0019] Assignment function m based on basic beliefs x (e) POW event set x The elements in the array are mapped to the interval [0,1], and satisfy the following conditions:

[0020]

[0021] The basic belief assignment function m for each sensor x (e) Depends on the set alarm threshold Ref x With sensor readings Obs x Differences between Diff x =Ref x -Obs x The specific definition is as follows:

[0022]

[0023] Where, m x (N),m x (H),m x (H,N) represent the confidence levels of sensor x in being assigned to the non-dangerous state N, the dangerous state H, and the uncertain state, respectively. This indicates that the sensor reading is less than the alarm threshold. x <Ref x The time difference factor reflects the degree to which the sensor reading is lower than the alarm threshold; This indicates that the sensor reading is greater than the alarm threshold OBs. x >Ref x The time difference factor reflects the degree to which the sensor reading exceeds the alarm threshold; SF1 x and SF2 x Two scaling factors are used to adjust the difference factor DF1. x and DF2 x The value of the sensor reading Obs x With alarm threshold Ref x The differences between them are linearly mapped to a standardized interval;

[0024] Define a trust threshold τ, for sensor x if its trust level m x (N) or m x If (H) is less than the trust threshold τ, then the sensor's data is considered low-trust data.

[0025] Furthermore, the joint trust level is calculated using the following method:

[0026] For each of the k nodes in the sensor node array, there are k belief assignment functions for each type of data. The joint belief degree of the sensor x is calculated using the Dempster combination rule, and the specific formula is as follows:

[0027]

[0028] in, It is the conflict factor, representing the degree of inconsistency between different nodes; M x {H} and M x {N} represents the overall trust level of the sensor array for hazardous and non-hazardous states after fusing all sensor data; m1 x m2 x ,...,mk x These represent the sensor x in other relevant nodes N1, N2, ..., N, respectively. k Trust level; weight w x =max(m x (H),m x(N)) Dynamically adjust based on the sensor's level of trust.

[0029] Furthermore, the trust adjustment factor γ for untrustworthy sensor data is calculated using the following method:

[0030] Compare M x {H} and M x {N} determines the decision result of sensor x, and the specific formula is as follows:

[0031]

[0032] Define a trust adjustment factor γ, which is used when the trust level is less than a trust threshold, i.e., M. x {H} < τ or M x If {N} < τ, apply this factor to adjust the update magnitude of 1DCNN neuron weights:

[0033] γ=min(max(m x (H),m x (N)),τ).

[0034] Furthermore, the data used was revised, specifically as follows:

[0035] When the data from sensor node x is determined to have low trust, i.e., M x When {N} < τ, perform time domain supplementary value correction, spatial domain supplementary value correction, spatiotemporal fusion supplementation, and reliability adjustment of supplementary data.

[0036] Furthermore, the time-domain supplementary value is calculated using the following formula:

[0037]

[0038] in, As a time-domain complementary value, X t-1 and X t-2 These are the two most recent valid historical data from the sensor, where α is the time weighting coefficient.

[0039] Furthermore, the spatial domain is supplemented with values, specifically as follows:

[0040] Calculate distance-based weighting coefficients:

[0041] S j =exp(-γd ij )

[0042] Where, d ij γ is the actual physical distance between node i and its neighboring node j, and γ is the distance decay coefficient, used to control the degree of influence of distance on weight.

[0043] Calculate the spatial domain supplementary value based on the weighting coefficient:

[0044]

[0045] in, As a supplementary value based on the spatial domain, X j S represents the current data of adjacent, normally functioning sensor nodes. j For distance-based weighting coefficients, N i Let i be the set of neighboring nodes of node i.

[0046] Furthermore, spatiotemporal fusion is supplemented, and the specific methods are as follows:

[0047]

[0048] in, For the final supplementary data, β is the spatiotemporal fusion weighting coefficient.

[0049] Furthermore, the credibility of the supplementary data is adjusted, and the specific methods are as follows:

[0050] To reduce the impact of supplementary data on model decisions, the confidence level of the supplementary data is attenuated, using the following formula:

[0051]

[0052] in, To supplement the confidence level of the data, μ is the confidence decay coefficient.

[0053] Furthermore, the neurons in the 1DCNN classification model adjust their weights using a confidence adjustment factor, as shown in the following formula:

[0054]

[0055] Where η is the learning rate. It is the gradient of the loss function L with respect to the weights W.

[0056] Due to the adoption of the above technical solutions, the present invention has the following beneficial effects:

[0057] 1. Effectively address the uncertainty and faults of sensor data through the Dempster-Shafer evidence theory, assess the reliability of sensor data, and filter out low-reliability or faulty data.

[0058] 2. Utilizing the automatic feature extraction capability of 1DCNN, the filtered data can be efficiently classified, which is suitable for processing time series data and can accurately detect abnormal events in the environment.

[0059] 3. Deploy DSET, 1DCNN models, and spatiotemporal fusion-based data supplementation algorithms on edge devices to reduce reliance on cloud computing and ensure real-time performance and data security in communication-constrained environments.

[0060] 4. For low-trust sensor data, a data supplementation strategy based on spatiotemporal fusion is proposed to ensure the consistency of model input dimensions and improve the reliability of the system in the event of sensor failure.

[0061] 5. This invention employs edge computing technology, deploying the DSET filter and 1DCNN classifier on a microcontroller. This not only reduces reliance on cloud computing but also improves the system's real-time performance and data security. Through local processing, the system can maintain high efficiency even in the event of network interruptions or delays.

[0062] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0063] The accompanying drawings of this invention are described below.

[0064] Figure 1 This is a schematic diagram of the overall process of the present invention.

[0065] Figure 2 This is a schematic diagram of the 1DCNN model structure.

[0066] Figure 3 This is a schematic diagram of the hardware structure. Detailed Implementation

[0067] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0068] A multi-sensor fusion method for environmental anomaly detection under communication-constrained conditions, such as Figure 1 As shown, the specific steps are as follows:

[0069] S1. Deploy sensors

[0070] To achieve fire anomaly detection under communication-constrained conditions, multi-sensor data monitoring was employed in the monitored area, with the following deployment structure: Figure 3 As shown.

[0071] The overall monitoring area is divided into n sub-regions, each equipped with a sensor node array (A1~A2). n Each sensor node array contains k identical sensor nodes (N1 to N2).k Each node is equipped with m different types of sensors (G1~G...). m This system is used to monitor changes in the local environment in real time. By setting up multiple sensor nodes in each area, the system can not only improve the accuracy of data acquisition, but also effectively address issues such as single sensor failure or data uncertainty.

[0072] S2. Trust level assessment using the DSET filter.

[0073] In this step, DSET assesses the reliability of each sensor observation by defining a basic belief assignment and uses the Dempster combination rule to make a comprehensive decision on data from multiple sensors. The specific steps are as follows:

[0074] S21. Basic Belief Assignment. For any sensor node x∈[G1,...,G1] in any sensor array... m First, we define a framework for distinguishing regional environmental states:

[0075] FOD x ={{H},{N}}

[0076] Where H represents a dangerous state and N represents a non-dangerous state;

[0077] Then its event set POW x for:

[0078]

[0079] Assignment function m based on basic beliefs x (e) Event set POW x The elements in the array are mapped to the interval [0,1], and satisfy the following conditions:

[0080]

[0081] The basic belief assignment function m for each sensor x (e) Depends on the set alarm threshold Ref x With sensor readings Obs x Differences between Diff x =Ref x -Obs x The specific definition is as follows:

[0082]

[0083]

[0084] Where, m x (N),m x (H),mx (H,N) represent the confidence levels of sensor x in being assigned to the non-dangerous state N, the dangerous state H, and the uncertain state, respectively. This indicates that the sensor reading is less than the alarm threshold. x <Ref x The time difference factor reflects the degree to which the sensor reading is below the alarm threshold. This indicates that the sensor reading is greater than the alarm threshold OBs. x >Ref x The time difference factor reflects the degree to which the sensor reading exceeds the alarm threshold; SF1 x and SF2 x Two scaling factors are used to adjust the difference factor DF1. x and DF2 x The value of the sensor reading Obs x With alarm threshold Ref x The differences between them are linearly mapped to a standardized interval to facilitate subsequent belief assignment calculations.

[0085] Define a trust threshold τ, for sensor x if its trust level m x (N) or m x If (H) is less than the confidence threshold τ, the sensor data is considered low-confidence data and requires special processing. This helps to filter out high-confidence data and reduce the impact of noise and outliers.

[0086] S21. Dempster Combination Rule Based on Trust Weighting. For each of the k nodes in the sensor node array, there are k belief assignment functions for each type of data. The joint trust of sensor x is calculated using the Dempster combination rule, and a weighting mechanism is introduced to make high-trust sensors have a greater influence on the final decision.

[0087]

[0088] in, It is the conflict factor, representing the degree of inconsistency between different nodes; M x {H} and M x {N} represents the overall trust level of the sensor array for hazardous and non-hazardous states after fusing all sensor data; m1 x m2 x ,...,mk x These represent the sensor x in other relevant nodes N1, N2, ..., N, respectively. k Trust level; weight w x =max(m x (H),m x(N)) Dynamically adjust based on the sensor's level of trust.

[0089] S22, Fault Node Filtering and Trust Adjustment. Joint trust is used to determine the combined decision result, and a trust adjustment mechanism is introduced. For low-trust sensors, their impact on subsequent 1DCNN training is reduced.

[0090] First, compare M. x {H} and M x {N} determines the decision outcome for sensor x:

[0091]

[0092] Define a trust adjustment factor γ, which is used when the trust level is less than a trust threshold, i.e., M. x {H} < τ or M x If {N} < τ, apply this factor to adjust the update magnitude of 1DCNN neuron weights:

[0093] γ=min(max(m x (H),m x (N)),τ)

[0094] For sensor data with low trust levels, its impact will be significantly reduced in subsequent 1DCNN training, thus avoiding the negative impact of low-quality data on model training.

[0095] Based on the above decision results, nodes that do not comply with the decision by sensor x are identified as faulty nodes, and the erroneous observations of the faulty nodes will be removed.

[0096] S3. Correct data using spatiotemporal fusion.

[0097] When the data from sensor node x is determined to have low trust (M x {H} < τ or M x When {N} < τ), the following method is used to supplement missing data:

[0098] S31, Calculation of time domain supplementary values.

[0099]

[0100] in, As a time-domain complementary value, X t-1 and X t-2 These are the two most recent valid historical data of the sensor, where α is the time weighting coefficient, and 0 < α < 1.

[0101] S32, Spatial Domain Supplementary Value Calculation.

[0102] Calculate distance-based weighting coefficients:

[0103] S j =exp(-γd ij )

[0104] Where, d ij Let be the actual physical distance between node i and its neighboring node j, and γ be the distance decay coefficient, used to control the degree of influence of distance on weight. The greater the distance, the more exponentially the weight decays.

[0105] Calculate the spatial domain supplementary value based on the weighting coefficient:

[0106]

[0107] in, As a supplementary value based on the spatial domain, X j S represents the current data of adjacent, normally functioning sensor nodes. j For distance-based weighting coefficients, N i Let i be the set of neighboring nodes of node i.

[0108] S33, Spacetime Fusion Supplement.

[0109]

[0110] in, For the final supplementary data, β is the spatiotemporal fusion weighting coefficient, 0 < β < 1.

[0111] S34. Adjustment of the credibility of supplementary data.

[0112] To reduce the impact of supplementary data on model decisions, the confidence level of the supplementary data is attenuated:

[0113]

[0114] in, To supplement the confidence level of the data, μ is the confidence decay coefficient, 0 < μ < 1.

[0115] The S4 and 1DCNN classification models were used to identify fire anomalies.

[0116] 1DCNN is a deep learning model specifically designed for processing sequential data, suitable for time series data and signal processing tasks, such as... Figure 2 As shown, 1DCNN is used to classify healthy node data after DSET filtering to identify anomalous events in the environment. The 1DCNN model adopts a multiple-input multiple-output architecture, including 1DCNN layers, max-pooling layers, Dropout layers, and fully connected layers.

[0117] Each input sample in the model input layer contains m×k data points within T seconds, i.e., each node has m sensors. These data points are organized into a matrix of shape T×m×k to represent the m sensor readings per second within T seconds.

[0118] 1DCNN layers use one-dimensional convolutional kernels to automatically extract features. Assume the kernel size is k. f If the stride is s, then the convolution operation of the l-th layer can be expressed as:

[0119] O l =σ(W l *X+b l )

[0120] Among them, O l W is the output feature map of the l-th layer. l b is the convolution kernel weight matrix of the l-th layer. l σ is the bias term, * represents the convolution operation, and σ is the activation function. Multiple convolutional layers can be stacked together to capture more complex features.

[0121] After each convolutional layer, a max-pooling layer is added to reduce data dimensionality while preserving important features. Assume the pooling window size is k. p The step size is s p The max pooling operation can then be expressed as:

[0122] P l =MaxPool(O l ,k p ,s p )

[0123] Finally, based on the extracted features, a fully connected layer is used to make a classification decision. Assume the weight matrix of the fully connected layer is W. f The bias term is b f The operation of a fully connected layer can then be expressed as:

[0124] Y = softmax(W) f ·D l +b f )

[0125] The confidence adjustment factor γ is used to dynamically adjust the magnitude of gradient updates during backpropagation, thereby reducing the negative impact of low-quality data on model training.

[0126]

[0127] Where η is the learning rate. It is the gradient of the loss function L with respect to the weights W.

[0128] In this way, the confidence adjustment factor γ dynamically adjusts the magnitude of gradient updates, ensuring that high-quality data contributes more to model training, while the impact of low-quality data is weakened, thereby improving the robustness and accuracy of the model.

[0129] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for detecting environmental anomalies under communication-constrained conditions using multi-sensor fusion, characterized in that, The specific method is as follows: The overall monitoring area is divided into Each sub-region is equipped with a sensor node array, and each sensor node array includes... There are multiple identical sensor nodes, and each sensor node is equipped with... A variety of different types of sensors; Based on the data collected by the sensors, the trust level of each sensor is calculated to determine whether the data collected by each sensor is trustworthy. For a sensor node that includes sensors that are not trustworthy, the joint trust level of all sensors within that sensor node is calculated using the Dempster combination rule and then weighted. For sensors that are not trustworthy, a spatiotemporal fusion method is used to correct the data they employ. Based on the joint trust level, calculate the trust adjustment factor for sensor data that is not trustworthy; Data collected by all trustworthy sensors, as well as data corrected from untrustworthy sensors, are input into a 1DCNN classification model to complete fire anomaly identification; the neurons of the 1DCNN classification model have their weights adjusted by a trust level adjustment factor. Calculate the trust adjustment factor for untrustworthy sensor data. The specific method is as follows: Compare and Determine the sensor The decision result is given by the following formula: ； Define a trust adjustment factor When the trust level is less than the trust threshold, that is... or This factor is applied to adjust the update magnitude of 1DCNN neuron weights: ； in, Indicates a dangerous situation. Indicates a non-dangerous state; and These represent the overall confidence level of the sensor array in hazardous and non-hazardous conditions after fusing all sensor data.

2. The multi-sensor fusion environmental anomaly detection method under communication-constrained conditions as described in claim 1, characterized in that, The confidence level for each sensor is calculated using the following method: Let the sensor node array Each sensor node is respectively Each sensor node is equipped with the following sensors: ; For any sensor node in a sensor array The framework for distinguishing regional environmental states is defined as follows: Then its event set for: Assignment function based on basic beliefs event set Elements in the range are mapped to the range In, and satisfy: Basic belief assignment function for each sensor Depends on the set alarm threshold With sensor readings Differences between The specific definition is as follows: in, These represent the sensors. Assigned to a non-dangerous state Dangerous conditions And the degree of trust in uncertain states; This indicates that the sensor reading is less than the alarm threshold. The time difference factor reflects the degree to which the sensor reading is lower than the alarm threshold; This indicates that the sensor reading is greater than the alarm threshold. The time difference factor reflects the degree to which the sensor reading exceeds the alarm threshold; and Two scaling factors are used to adjust the difference factor. and The value of the sensor reading With alarm threshold The differences between them are linearly mapped to a standardized interval; Define a trust threshold For sensors If its trust level or Less than the trust threshold If so, the sensor's data is considered low-trust data.

3. The multi-sensor fusion environmental anomaly detection method under communication-constrained conditions as described in claim 2, characterized in that, The joint trust level is calculated using the following method: For each sensor node array There are nodes, and each type of data has [number] nodes. A belief assignment function is used to calculate the sensor's belief assignment using Dempster's combination rule. The formula for joint trust level is as follows: in, It is a conflict factor, representing the degree of inconsistency between different nodes; These represent the sensors. Other relevant nodes Trust level; weight The system dynamically adjusts based on the level of trust in the sensors.

4. The multi-sensor fusion environmental anomaly detection method under communication-constrained conditions as described in claim 1, characterized in that, The data used was corrected using the following method: When sensor node The data was deemed to have low trust level, i.e. At that time, time domain supplementary value correction, spatial domain supplementary value correction, spatiotemporal fusion supplementation, and credibility adjustment of supplementary data are performed.

5. The multi-sensor fusion environmental anomaly detection method under communication-constrained conditions as described in claim 4, characterized in that, The time-domain supplementary value is calculated using the following formula: in, These are supplementary values ​​based on the time domain. and These are the two most recent valid historical data from the sensor. This is the time weighting coefficient.

6. The multi-sensor fusion environmental anomaly detection method under communication-constrained conditions as described in claim 4, characterized in that, Spatial domain supplementary values ​​are provided using the following method: Calculate distance-based weighting coefficients: in, For nodes With neighboring nodes The actual physical distance between them This is the distance decay coefficient, used to control the degree of influence of distance on the weights; Calculate the spatial domain supplementary value based on the weighting coefficient: in, As a supplementary value based on the spatial domain, This represents the current data of adjacent, normally functioning sensor nodes. For distance-based weighting coefficients, For nodes The set of adjacent nodes.

7. The multi-sensor fusion environmental anomaly detection method under communication-constrained conditions as described in claim 4, characterized in that, The spatiotemporal fusion supplement is performed using the following methods: in, For final supplementary data, This represents the spatiotemporal fusion weighting coefficient.

8. The multi-sensor fusion environmental anomaly detection method under communication-constrained conditions as described in claim 4, characterized in that, The credibility of supplementary data is adjusted using the following methods: To reduce the impact of supplementary data on model decisions, the confidence level of the supplementary data is attenuated, using the following formula: in, To enhance the credibility of the data, This is the trust decay coefficient.

9. The multi-sensor fusion environmental anomaly detection method under communication-constrained conditions as described in claim 4, characterized in that, The neurons in a 1DCNN classification model adjust their weights using a confidence adjustment factor, as shown in the following formula: in, It's the learning rate. It is a loss function Relative to weight The gradient.

Images