Data fusion method of heterogeneous detection equipment

Abstract

The invention provides a data fusion method for heterogeneous detection equipment, and belongs to the field of detection equipment detection, and the method comprises the steps: S1, collecting detection data of two kinds of heterogeneous detection equipment, carrying out time window division and normalized synchronous mapping, and adding the normalized data to obtain a first alarm parameter; s2, performing feature extraction to obtain first feature information and second feature information; s3, performing feature screening, and obtaining a second alarm parameter according to the screened feature weight; s4, calculating a feature fusion weight, and performing dynamic adjustment to obtain a fusion weight; s5, performing weighted fusion on the first feature information and the second feature information according to the fusion weight to generate a fusion feature; and S6, performing early warning grade determination based on the fusion feature, and outputting an early warning signal according to a determination result, the first alarm parameter and the second alarm parameter. According to the method, heterogeneous data are fused, and a two-parameter alarm judgment mechanism is adopted, so that the detection accuracy is improved, and the false alarm rate is reduced.

Description

Technical Field

[0001] The present invention relates to the field of detection equipment technology, and in particular to a data fusion method for heterogeneous detection equipment. Background Art

[0002] The aspiration smoke sensing products on the current market mainly use single detection principles such as laser scattering, cloud chamber, and LED smoke sensing for smoke detection. Among them, the laser scattering type is more sensitive to small smoke particles but is easily interfered by dust; the cloud chamber type has strong anti-interference ability but is not effective in detecting small smoke particles; the LED smoke sensing type has a simple structure and low cost but has a narrow detection range. These products with single detection principles have their own advantages and limitations in actual applications, and it is difficult to meet the comprehensive detection needs in different scenarios.

[0003] However, the detection range of existing suction-type smoke sensors is limited, making it difficult to cover the entire particle size range, and there are "blind spots" for specific types of smoke particles, making it difficult to optimize both the false alarm rate and the missed alarm rate. Even if two different detection devices are used for detection, the signal units output by different detection principles and the physical meanings they represent are different, resulting in long detection times and low detection reliability in complex environments.

[0004] Therefore, finding a method that can both improve the efficiency of smoke detection using two detection devices and improve the detection accuracy is a technical problem that needs to be solved urgently by those skilled in the art. Summary of the invention

[0005] In view of this, the present invention proposes a data fusion method for heterogeneous detection equipment, which can achieve effective fusion of heterogeneous data and adopt a dual-parameter alarm judgment mechanism to improve detection accuracy and effectively reduce the false alarm rate.

[0006] The technical solution of the present invention is implemented as follows: The present invention provides a data fusion method for heterogeneous detection equipment, comprising the following steps:

[0007] S1. Collect detection data of two heterogeneous detection devices to obtain laser smoke detection signal data and cloud chamber detection signal data, divide the laser smoke detection signal data and the cloud chamber detection signal data into time windows and perform normalized synchronous mapping respectively, and add the laser smoke detection signal data after normalized synchronous mapping and the cloud chamber detection signal data after normalized synchronous mapping to obtain a first alarm parameter; wherein the first alarm parameter is used to determine the alarm level range;

[0008] S2, respectively extracting features from the laser smoke detection signal data after normalized synchronous mapping and the cloud chamber detection signal data after normalized synchronous mapping to obtain first feature information and second feature information;

[0009] S3, feature screening is performed on the first feature information and the second feature information to obtain a first feature weight and a second feature weight, and the feature information corresponding to the highest feature weight is selected as a second alarm parameter; wherein the second alarm parameter is used to determine whether to trigger an alarm;

[0010] S4, obtaining historical signal data of two heterogeneous detection devices, calculating feature fusion weights based on the historical signal data and environmental parameters, and dynamically adjusting them to obtain fusion weights;

[0011] S5, performing weighted fusion on the first feature information and the second feature information according to the fusion weight to generate a fusion feature;

[0012] S6. Determine the warning level based on the fusion feature, and output a warning signal according to the determination result, the first alarm parameter and the second alarm parameter.

[0013] On the basis of the above technical scheme, preferably, the first characteristic information includes the particle size distribution, particle size, particle number and particle concentration of the laser smoke detection signal data after normalized synchronous mapping, and the second characteristic information includes the particle size distribution, particle size, particle number and particle concentration of the cloud chamber detection signal data after normalized synchronous mapping, wherein the particle size distribution range is: 0.02~0.2um, 0.2um-0.4um, 0.4um-0.6um, 0.6um-0.8um, 0.8um-1um, 1um-2um, 2um-4um, 4um-10um, the particle size range is: 0.02um-50um, the particle number range is: 30000-2000000 / cc, and the particle concentration range is: 0.001-20%obs / m.

[0014] On the basis of the above technical solution, preferably, the normalized synchronization mapping specifically includes:

[0015] The laser smoke detection signal data and the cloud chamber detection signal data are normalized respectively to obtain the first normalized feature and the second normalized feature. The calculation formula is:

[0016]

[0017] Among them, X norm1 represents the first normalized feature, X1 represents the laser smoke detection signal data, X min1 Indicates the minimum value of the laser smoke detection signal data, X max1 Indicates the maximum value of the laser smoke detection signal data, X norm2 represents the second normalized feature, X2 represents the cloud chamber detection signal data, X min2 Indicates the minimum value of the cloud chamber detection signal data, Xmax2 Indicates the maximum value of the cloud chamber detection signal data;

[0018] The first normalized feature and the second normalized feature are respectively subjected to Z-score normalization to obtain the laser smoke detection signal data after normalized synchronous mapping and the cloud chamber detection signal data after normalized synchronous mapping. The calculation formula is:

[0019]

[0020] Among them, X std1 represents the laser smoke detection signal data after normalized synchronous mapping, μ1 represents the mean value of the laser smoke detection signal data, σ1 represents the standard deviation of the laser smoke detection signal data, X std2 represents the cloud chamber detection signal data after normalized synchronous mapping, μ2 represents the mean of the cloud chamber detection signal data, and σ2 represents the standard deviation of the cloud chamber detection signal data.

[0021] On the basis of the above technical solution, preferably, the feature screening adopts the ReliefF algorithm, and step S3 specifically includes:

[0022] S31, taking the first feature information and the second feature information as the feature set to be screened, initializing the feature weight vector W to a zero vector, and setting the number of iterations m and the number of neighbors k;

[0023] S32. Randomly select a target sample R from the feature set to be screened i , find the distance R in the feature space based on the standardized Euclidean distance metric i The most recent k samples of the same type and k samples of different types; different types represent different warning levels;

[0024] S33, calculate the target sample R for each feature i The average distance difference between k hit samples and miss samples on the feature dimension, and update the feature weight according to the calculation result;

[0025] S34, repeating step 2 to step 3 m times to obtain a final feature weight vector, and normalizing the final feature weight vector;

[0026] S35, setting a weight threshold, filtering the normalized final feature vector according to the weight threshold, obtaining a filtered feature subset, and taking feature information corresponding to the highest normalized final feature vector as the second alarm parameter.

[0027] Based on the above technical solution, preferably, the environmental parameters include temperature data, humidity data and airflow data.

[0028] On the basis of the above technical solution, preferably, step S4 specifically includes:

[0029] S41, acquiring historical signal data of two heterogeneous detection devices to obtain first historical signal data and second historical signal data, respectively extracting characteristic values ​​of the first historical signal data and the second historical signal data, and establishing a first characteristic anomaly relationship corresponding table and a second characteristic anomaly relationship corresponding table;

[0030] S42, respectively counting the detection rates of anomalies in the first historical signal data and the second historical signal data, generating a first feature reliability score and a second feature reliability score, and generating an initial weight according to the first feature abnormality relationship correspondence table and the second feature abnormality relationship correspondence table and the first feature reliability score and the second feature reliability score;

[0031] S43, collecting environmental parameters of two heterogeneous detection devices in real time, and calculating environmental weights based on the environmental parameters;

[0032] S44. Adjust the initial weight according to the environment weight to obtain the fusion weight.

[0033] On the basis of the above technical solution, preferably, the calculation formula of the fusion weight is:

[0034] W(t)=Normalize(min(|λ×(W base ×(1-E i (t)))+(1-λ)×W(t-

[0035] 1)-W(t-1)|,δ);

[0036]

[0037] Where W(t) represents the fusion weight at time t, Normalize(·) represents the normalization function, min(·) represents the minimization function, λ represents the smoothing factor, and E i (t) represents the environment weight, W(t-1) represents the fusion weight at time t-1, δ represents the change rate threshold, α T represents the temperature influence weight, β H represents the humidity impact weight, γ F represents the airflow influence weight, T(t) represents the temperature at time t, T0 represents the calibration temperature, H(t) represents the humidity at time t, H0 represents the calibration humidity, F(t) represents the airflow at time t, F0 represents the calibration airflow, ΔT represents the allowable temperature deviation, ΔH represents the allowable humidity deviation, ΔF represents the allowable airflow deviation, R i represents the reliability score of the i-th feature, W base represents the initial weight, C iRepresents the correlation coefficient between the ith feature and the anomaly.

[0038] On the basis of the above technical solution, preferably, the feature fusion includes feature-level fusion and decision-level fusion, the feature-level fusion is a weighted combination of the first feature information and the second feature information, and the decision-level fusion is a synthesis of the feature discrimination results.

[0039] More preferably, step S5 specifically includes:

[0040] Performing weighted fusion on the first feature information and the second feature information according to the fusion weight to obtain a first fusion feature;

[0041] Performing threshold discrimination and pattern recognition on the first feature information and the second feature information respectively to obtain a decision result;

[0042] Based on the decision results, DS evidence theory is used to fuse and obtain the second fusion feature;

[0043] A joint feature matrix is ​​established according to the first fusion feature and the second fusion feature, and feature correlation analysis is performed to obtain a fusion feature.

[0044] Based on the above technical solution, preferably, step S6 specifically includes:

[0045] S61, setting a multi-level warning threshold based on historical signal data and environmental parameters, and establishing a determination threshold of a first alarm parameter and a second alarm parameter, wherein the determination threshold of the first alarm parameter corresponds to a different warning level range, and the determination threshold of the second alarm parameter is used to determine whether to trigger an alarm;

[0046] S62, respectively calculating the deviation value between the fusion feature and the warning threshold, and determining the warning level range of the first alarm parameter and whether the second alarm parameter exceeds the alarm triggering threshold, wherein the warning level range of the first alarm parameter determines the highest level at which the device can alarm, and the second alarm parameter is allowed to output an alarm signal only when it exceeds the threshold;

[0047] S63. Preliminarily determine the warning level according to the deviation value of the fused feature, and output a warning signal according to the warning level.

[0048] The data fusion method provided by the present invention realizes the effective fusion of heterogeneous data by dividing the data of heterogeneous detection devices into time windows and normalizing and synchronously mapping them, combining feature screening and a multi-level fusion mechanism, and adopts a dual-parameter alarm judgment mechanism to improve the detection accuracy of the scheme and effectively reduce the false alarm rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 A three-dimensional diagram of the data fusion method of the present invention;

[0051] Figure 2 This is a technical implementation diagram of the data fusion method of the present invention. DETAILED DESCRIPTION

[0052] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] like Figure 1 and Figure 2 As shown, the present invention provides a data fusion method for heterogeneous detection devices, comprising the following steps:

[0054] S1. Collect detection data from two heterogeneous detection devices to obtain laser smoke detection signal data and cloud chamber detection signal data, divide the laser smoke detection signal data and the cloud chamber detection signal data into time windows and perform normalized synchronous mapping respectively, and add the laser smoke detection signal data after normalized synchronous mapping and the cloud chamber detection signal data after normalized synchronous mapping to obtain a first alarm parameter; wherein the first alarm parameter is used to determine the alarm level range.

[0055] In one embodiment of the present invention, the laser smoke detection signal data and the cloud chamber detection signal data are taken as a cycle of 6 seconds, and the laser smoke detection signal data and the cloud chamber detection signal data are synchronously mapped into the range of 0-1000 through normalization.

[0056] Furthermore, the normalized synchronization mapping specifically includes:

[0057] The laser smoke detection signal data and the cloud chamber detection signal data are normalized respectively to obtain the first normalized feature and the second normalized feature. The calculation formula is:

[0058]

[0059] Among them, X norm1represents the first normalized feature, X1 represents the laser smoke detection signal data, X min1 Indicates the minimum value of the laser smoke detection signal data, X max1 Indicates the maximum value of the laser smoke detection signal data, X norm2 represents the second normalized feature, X2 represents the cloud chamber detection signal data, X min2 Indicates the minimum value of the cloud chamber detection signal data, X max2 Indicates the maximum value of the cloud chamber detection signal data;

[0060] The first normalized feature and the second normalized feature are respectively subjected to Z-score normalization to obtain the laser smoke detection signal data after normalized synchronous mapping and the cloud chamber detection signal data after normalized synchronous mapping. The calculation formula is:

[0061]

[0062] Among them, X std1 represents the laser smoke detection signal data after normalized synchronous mapping, μ1 represents the mean value of the laser smoke detection signal data, σ1 represents the standard deviation of the laser smoke detection signal data, X std2 represents the cloud chamber detection signal data after normalized synchronous mapping, μ2 represents the mean of the cloud chamber detection signal data, and σ2 represents the standard deviation of the cloud chamber detection signal data.

[0063] A specific embodiment is used to illustrate:

[0064] In a 6-second time window, the laser smoke detection signal data is [0.8, 1.2, 1.5, 1.8, 2.0], and the cloud chamber detection signal data is [0.3, 0.5, 0.8, 1.0, 1.2];

[0065] The two sets of data were processed by normalized synchronous mapping. The laser smoke detection signal data after normalized synchronous mapping was -1.87, and the cloud chamber detection signal data after normalized synchronous mapping was -0.57.

[0066] The absolute values ​​of the laser smoke detection signal data after normalized synchronous mapping and the cloud chamber detection signal data after normalized synchronous mapping are summed, and the first alarm parameter is 2.44.

[0067] In one embodiment of the present invention, a fixed sampling time window is set to implement a sliding window mechanism, and data cache queues are established according to the laser smoke detection signal data and the cloud chamber detection signal data to obtain a first data cache queue and a second data cache queue;

[0068] Timestamps are added to the first data cache queue and the second data cache queue, and data alignment operations are performed to obtain first synchronization detection data and second synchronization detection data.

[0069] Furthermore, when there is data with inconsistent sampling frequencies in the first data cache queue and the second data cache queue, interpolation compensation processing is first performed on the first data cache queue and the second data cache queue, and then a normalized synchronous mapping operation is performed to obtain laser smoke detection signal data after normalized synchronous mapping and cloud chamber detection signal data after normalized synchronous mapping.

[0070] The present invention adopts a synchronous mapping method combining normalization processing with Z-score standardization, which solves the problem of inconsistent data dimensions of different detection devices, while maintaining the relative change relationship of the data, and improving the comparability and fusion effect of heterogeneous data.

[0071] S2. Perform feature extraction on the laser smoke detection signal data after normalized synchronous mapping and the cloud chamber detection signal data after normalized synchronous mapping to obtain first feature information and second feature information.

[0072] In a preferred embodiment of the present invention, the first characteristic information includes the particle size distribution, particle size, particle number and particle concentration of the laser smoke detection signal data after normalized synchronous mapping, and the second characteristic information includes the particle size distribution, particle size, particle number and particle concentration of the cloud chamber detection signal data after normalized synchronous mapping, wherein the particle size distribution range is: 0.02~0.2um, 0.2um-0.4um, 0.4um-0.6um, 0.6um-0.8um, 0.8um-1um, 1um-2um, 2um-4um, 4um-10um, the particle size range is: 0.02um-50um, the particle number range is: 30000-2000000 / cc, and the particle concentration range is: 0.001-20%obs / m.

[0073] S3. Perform feature screening on the first feature information and the second feature information to obtain a first feature weight and a second feature weight, and select the feature information corresponding to the highest feature weight as the second alarm parameter; wherein the second alarm parameter is used to determine whether to trigger an alarm.

[0074] In an embodiment of the present application, the feature screening adopts the ReliefF algorithm, and step S3 specifically includes:

[0075] S31, taking the first feature information and the second feature information as a feature set to be screened, initializing the feature weight vector to a zero vector, and setting the number of iterations m and the number of neighbors k;

[0076] S32. Randomly select a target sample R from the feature set to be screened i , find the distance R in the feature space based on the standardized Euclidean distance metric i The most recent k samples of the same type and k samples of different types; different types represent different warning levels;

[0077] S33, calculate the target sample R for each feature i The average distance difference between k hit samples and miss samples on the feature dimension, and update the feature weight vector according to the calculation result;

[0078] S34, repeating step 2 to step 3 m times to obtain a final feature weight vector, and normalizing the final feature weight vector;

[0079] S35, setting a weight threshold, screening the normalized final feature weight vector according to the weight threshold, obtaining a screened feature subset, and taking the feature information corresponding to the maximum normalized final feature vector as the second alarm parameter.

[0080] The present invention screens features through the ReliefF algorithm and combines the weight calculation of time domain features and frequency domain features to effectively identify feature parameters that have an important impact on early warning judgment, reduces data redundancy, and improves the efficiency of feature extraction.

[0081] S4. Obtain historical signal data of two heterogeneous detection devices, calculate feature fusion weights based on the historical signal data and environmental parameters, and dynamically adjust to obtain fusion weights. The environmental parameters include temperature data, humidity data, and airflow data.

[0082] Specifically, step S4 specifically includes:

[0083] S41, acquiring historical signal data of two heterogeneous detection devices to obtain first historical signal data and second historical signal data, respectively extracting characteristic values ​​of the first historical signal data and the second historical signal data, and establishing a first characteristic anomaly relationship corresponding table and a second characteristic anomaly relationship corresponding table;

[0084] S42, respectively counting the detection rates of anomalies in the first historical signal data and the second historical signal data, generating a first feature reliability score and a second feature reliability score, and generating an initial weight according to the first feature abnormality relationship correspondence table and the second feature abnormality relationship correspondence table and the first feature reliability score and the second feature reliability score; wherein the calculation formula for the reliability score is:

[0085]

[0086] R i represents the reliable score of the i-th feature;

[0087] The calculation formula for the initial weight is:

[0088]

[0089] Among them, W base represents the initial weight, C i Represents the correlation coefficient between the i-th feature and the anomaly, that is, C i =|Conv(feature i, anomaly label)|.

[0090] S43, collecting environmental parameters of two heterogeneous detection devices in real time, and calculating environmental weights based on the environmental parameters;

[0091] S44. Adjust the initial weight according to the environment weight to obtain the fusion weight.

[0092] Furthermore, the calculation formula of the fusion weight is:

[0093] W(t)=Normalize(min(|λ×(W base ×(1-E i (t)))+(1-λ)×W(t-

[0094] 1)-W(t-1)|,δ);

[0095]

[0096] Where W(t) represents the fusion weight at time t, Normalize(·) represents the normalization function, min(·) represents the minimization function, λ represents the smoothing factor, and E i (t) represents the environment weight, W(t-1) represents the fusion weight at time t-1, δ represents the change rate threshold, α T represents the temperature influence weight, β H represents the humidity impact weight, γ F represents the airflow influence weight, T(t) represents the temperature at time t, T0 represents the calibration temperature, H(t) represents the humidity at time t, H0 represents the calibration humidity, F(t) represents the airflow at time t, F0 represents the calibration airflow, ΔT represents the allowable temperature deviation, ΔH represents the allowable humidity deviation, and ΔF represents the allowable airflow deviation.

[0097] As can be understood by those skilled in the art, the range of the fusion weight W(t) is [0,1], α T +β H +γ F =1, the range of the smoothing factor λ is [0,1].

[0098] The present invention establishes a dynamic weight adjustment mechanism based on historical signal data and environmental parameters, and adjusts the initial weight in real time through the environmental weight, so that the fusion weight can be adaptively adjusted as the environment changes, thereby improving the adaptability of the scheme under different environmental conditions.

[0099] S5. Perform weighted fusion on the first feature information and the second feature information according to the fusion weight to generate a fusion feature. The feature fusion includes feature-level fusion and decision-level fusion. The feature-level fusion is a weighted combination of the first feature information and the second feature information, and the decision-level fusion is a synthesis of the discrimination results of the features.

[0100] Specifically, step S5 specifically includes:

[0101] The first feature information and the second feature information are weightedly fused according to the fusion weight to obtain the first fusion feature; wherein the calculation formula of the first fusion feature F1 is:

[0102] F1=W(t)×X std1 +(1-W(t))×X std2 ;

[0103] Performing threshold discrimination and pattern recognition on the first characteristic information and the second characteristic information respectively to obtain a decision result; wherein the decision result includes a fault probability distribution and a fault mode probability;

[0104] Based on the decision results, DS evidence theory is used to fuse and obtain the second fusion feature;

[0105] A joint feature matrix is ​​established based on the first fusion feature and the second fusion feature, and feature correlation analysis is performed to obtain the fusion feature, where the calculation formula of the fusion feature is:

[0106] F = optimize(M);

[0107] M = [F1, F2];

[0108] Among them, F represents the fusion feature, M represents the joint feature matrix, F2 represents the second fusion feature, and optimize(·) represents the feature optimization function.

[0109] In one embodiment of the present invention, fusion is performed based on the decision results using DS evidence theory, specifically including:

[0110] Construct a basic probability assignment matrix based on the fault probability distribution and the fault mode probability;

[0111] Apply the DS combination rule to the basic probability assignment matrix to obtain the second fusion matrix.

[0112] The present invention adopts a multi-level fusion mechanism combining feature-level fusion and decision-level fusion, obtains the first fusion feature through weighted fusion, obtains the second fusion feature through DS evidence theory fusion, and performs feature correlation analysis, thereby improving the reliability and robustness of the fusion feature.

[0113] S6. Determine the warning level based on the fusion feature, and output a warning signal according to the determination result, the first alarm parameter and the second alarm parameter.

[0114] Specifically, step S6 specifically includes:

[0115] S61, setting a multi-level warning threshold based on historical signal data and environmental parameters, and establishing a determination threshold of a first alarm parameter and a second alarm parameter, wherein the determination threshold of the first alarm parameter corresponds to a different warning level range, and the determination threshold of the second alarm parameter is used to determine whether to trigger an alarm;

[0116] S62, respectively calculating the deviation value between the fusion feature and the warning threshold, and determining the warning level range of the first alarm parameter and whether the second alarm parameter exceeds the alarm triggering threshold, wherein the warning level range of the first alarm parameter determines the highest level at which the device can alarm, and the second alarm parameter is allowed to output an alarm signal only when it exceeds the threshold;

[0117] S63. Preliminarily determine the warning level according to the deviation value of the fused feature, and output a warning signal according to the warning level.

[0118] It can be understood that the warning threshold and the warning level can be set according to actual usage, and the present invention does not impose specific limitations on this.

[0119] The present invention realizes the effective fusion of heterogeneous data by dividing the data of heterogeneous detection equipment into time windows and normalizing and synchronously mapping them, combining feature screening and multi-level fusion mechanism, and adopts a dual-parameter alarm judgment mechanism, in which the first alarm parameter determines the alarm level range and the second alarm parameter determines whether to trigger the alarm, thereby improving the detection accuracy of the scheme and effectively reducing the false alarm rate.

[0120] A specific embodiment is used to illustrate:

[0121] The judgment threshold of the first alarm parameter is divided into four intervals: a value between 0 and 1.0 indicates no alarm state, between 1.0 and 2.0 corresponds to a low-level warning, between 2.0 and 3.0 corresponds to a medium-level warning, and greater than 3.0 indicates a high-level warning;

[0122] Set the trigger threshold of the second alarm parameter to 0.8;

[0123] When the first alarm parameter is 2.44 (falling within the intermediate warning range), the second alarm parameter is 0.85 (exceeding the alarm trigger threshold), and the fusion feature value is 0.92, the first alarm parameter determines that the highest possible alarm level is the intermediate warning, and the second alarm parameter exceeds the trigger threshold, indicating that an alarm needs to be issued, and finally determines to output an intermediate warning signal.

[0124] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A data fusion method for heterogeneous detection equipment, characterized in that: The following steps are involved: S1. Collect detection data of two heterogeneous detection devices to obtain laser smoke detection signal data and cloud chamber detection signal data, divide the laser smoke detection signal data and the cloud chamber detection signal data into time windows and perform normalized synchronous mapping respectively, and add the laser smoke detection signal data after normalized synchronous mapping and the cloud chamber detection signal data after normalized synchronous mapping to obtain a first alarm parameter; wherein the first alarm parameter is used to determine the alarm level range; S2, respectively extracting features from the laser smoke detection signal data after normalized synchronous mapping and the cloud chamber detection signal data after normalized synchronous mapping to obtain first feature information and second feature information; S3, feature screening is performed on the first feature information and the second feature information to obtain a first feature weight and a second feature weight, and the feature information corresponding to the highest feature weight is selected as a second alarm parameter; wherein the second alarm parameter is used to determine whether to trigger an alarm; S4, obtaining historical signal data of two heterogeneous detection devices, calculating feature fusion weights based on the historical signal data and environmental parameters, and dynamically adjusting them to obtain fusion weights; S5, performing weighted fusion on the first feature information and the second feature information according to the fusion weight to generate a fusion feature; S6. Determine the warning level based on the fusion feature, and output a warning signal according to the determination result, the first alarm parameter and the second alarm parameter.

2. The data fusion method of heterogeneous detection equipment according to claim 1, characterized in that: The first characteristic information includes the particle size distribution, particle size, particle number and particle concentration of the laser smoke detection signal data after normalized synchronous mapping, and the second characteristic information includes the particle size distribution, particle size, particle number and particle concentration of the cloud chamber detection signal data after normalized synchronous mapping, wherein the particle size distribution range is: 0.02~0.2um, 0.2um-0.4um, 0.4um-0.6um, 0.6um-0.8um, 0.8um-1um, 1um-2um, 2um-4um, 4um-10um, the particle size range is: 0.02um-50um, the particle number range is: 30000-2000000 / cc, and the particle concentration range is: 0.001-20%obs / m.

3. The data fusion method of heterogeneous detection equipment according to claim 2, characterized in that: The normalized synchronization mapping specifically includes: The laser smoke detection signal data and the cloud chamber detection signal data are normalized respectively to obtain the first normalized feature and the second normalized feature. The calculation formula is: Among them, X norm1 represents the first normalized feature, X1 represents the laser smoke detection signal data, X min1 Indicates the minimum value of the laser smoke detection signal data, X max1 Indicates the maximum value of the laser smoke detection signal data, X norm2 represents the second normalized feature, X2 represents the cloud chamber detection signal data, X min2 Indicates the minimum value of the cloud chamber detection signal data, X max2 Indicates the maximum value of the cloud chamber detection signal data; The first normalized feature and the second normalized feature are respectively subjected to Z-score normalization to obtain the laser smoke detection signal data after normalized synchronous mapping and the cloud chamber detection signal data after normalized synchronous mapping. The calculation formula is: Among them, X std1 represents the laser smoke detection signal data after normalized synchronous mapping, μ1 represents the mean value of the laser smoke detection signal data, σ1 represents the standard deviation of the laser smoke detection signal data, X std2 represents the cloud chamber detection signal data after normalized synchronous mapping, μ2 represents the mean of the cloud chamber detection signal data, and σ2 represents the standard deviation of the cloud chamber detection signal data.

4. The data fusion method of heterogeneous detection equipment according to claim 3, characterized in that: The feature screening adopts the ReliefF algorithm, and step S3 specifically includes: S31, taking the first feature information and the second feature information as a feature set to be screened, initializing the feature weight vector to a zero vector, and setting the number of iterations m and the number of neighbors k; S32. Randomly select a target sample R from the feature set to be screened i , find the distance R in the feature space based on the standardized Euclidean distance metric i The most recent k samples of the same type and k samples of different types; different types represent different warning levels; S33, calculate the target sample R for each feature i The average distance difference between k hit samples and miss samples on the feature dimension, and update the feature weight vector according to the calculation result; S34, repeating step 2 to step 3 m times to obtain a final feature weight vector, and normalizing the final feature weight vector; S35, setting a weight threshold, filtering the normalized final feature weight vector according to the weight threshold, obtaining a filtered feature subset, and taking feature information corresponding to the maximum normalized final feature vector as the second alarm parameter.

5. The data fusion method of heterogeneous detection equipment according to claim 1, characterized in that: The environmental parameters include temperature data, humidity data and air flow data.

6. The data fusion method of heterogeneous detection equipment according to claim 5, characterized in that: Step S4 specifically includes: S41, acquiring historical signal data of two heterogeneous detection devices to obtain first historical signal data and second historical signal data, respectively extracting characteristic values ​​of the first historical signal data and the second historical signal data, and establishing a first characteristic anomaly relationship corresponding table and a second characteristic anomaly relationship corresponding table; S42, respectively counting the detection rates of anomalies in the first historical signal data and the second historical signal data, generating a first feature reliability score and a second feature reliability score, and generating an initial weight according to the first feature abnormality relationship correspondence table and the second feature abnormality relationship correspondence table and the first feature reliability score and the second feature reliability score; S43, collecting environmental parameters of two heterogeneous detection devices in real time, and calculating environmental weights based on the environmental parameters; S44. Adjust the initial weight according to the environment weight to obtain the fusion weight.

7. The data fusion method of heterogeneous detection equipment according to claim 6, characterized in that: The calculation formula of the fusion weight is: W(t)=Normalize(min(|λ×(W base ×(1-E i (t)))+(1-λ)×W(t- 1)-W(t-1)|,δ); Where W(t) represents the fusion weight at time t, Normalize(·) represents the normalization function, min(·) represents the minimization function, λ represents the smoothing factor, and E i (t) represents the environment weight, W(t-1) represents the fusion weight at time t-1, δ represents the change rate threshold, α T represents the temperature influence weight, β H represents the humidity impact weight, γ F represents the airflow influence weight, T(t) represents the temperature at time t, T0 represents the calibration temperature, H(t) represents the humidity at time t, H0 represents the calibration humidity, F(t) represents the airflow at time t, F0 represents the calibration airflow, ΔT represents the allowable temperature deviation, ΔH represents the allowable humidity deviation, ΔF represents the allowable airflow deviation, R i represents the reliability score of the i-th feature, W base represents the initial weight, C i Represents the correlation coefficient between the ith feature and the anomaly.

8. The data fusion method of heterogeneous detection equipment according to claim 1, characterized in that: The feature fusion includes feature level Fusion and decision-level fusion, the feature-level fusion is a weighted combination of the first feature information and the second feature information, and the decision-level fusion is a synthesis of the feature discrimination results.

9. The data fusion method of heterogeneous detection equipment according to claim 8, characterized in that: Step S5 specifically includes: Performing weighted fusion on the first feature information and the second feature information according to the fusion weight to obtain a first fusion feature; Performing threshold discrimination and pattern recognition on the first feature information and the second feature information respectively to obtain a decision result; Based on the decision results, DS evidence theory is used to fuse and obtain the second fusion feature; A joint feature matrix is ​​established according to the first fusion feature and the second fusion feature, and feature correlation analysis is performed to obtain a fusion feature.

10. The data fusion method of heterogeneous detection equipment according to claim 9, characterized in that: Step S6 specifically includes: S61, setting a multi-level warning threshold based on historical signal data and environmental parameters, and establishing a determination threshold of a first alarm parameter and a second alarm parameter, wherein the determination threshold of the first alarm parameter corresponds to a different warning level range, and the determination threshold of the second alarm parameter is used to determine whether to trigger an alarm; S62, respectively calculating the deviation value between the fusion feature and the warning threshold, and determining the warning level range of the first alarm parameter and whether the second alarm parameter exceeds the alarm triggering threshold, wherein the warning level range of the first alarm parameter determines the highest level at which the device can alarm, and the second alarm parameter is allowed to output an alarm signal only when it exceeds the threshold; S63. Preliminarily determine the warning level according to the deviation value of the fused feature, and output a warning signal according to the warning level.

Images