A multi-sensor fusion-based monitoring device intelligent early warning and linkage control system
By integrating improved evidence theory and fault repair decision tree algorithm through a multi-sensor fusion intelligent early warning and linkage control system, the problems of low accuracy of multi-source data fusion and delayed fault early warning in the existing monitoring equipment operation and maintenance system are solved. It realizes accurate perception of equipment status in all dimensions and scientific classification and early prediction of faults, thereby improving equipment online rate and operation and maintenance efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 浙江飞至科技有限公司
- Filing Date
- 2026-03-09
- Publication Date
- 2026-06-02
AI Technical Summary
Existing monitoring equipment operation and maintenance systems suffer from a lack of scientific algorithm support for multi-source data fusion, resulting in low fault location accuracy, high rates of missed and false detections, and reliance on experience-based threshold judgments for fault analysis and early warning. They also lack intelligent decision-making algorithms, making it impossible to achieve scientific fault classification and early prediction, and lack effective fault prediction algorithms, which makes it difficult to improve equipment online rates.
The intelligent early warning and linkage control system adopts multi-sensor fusion, integrating improved evidence theory, fault repair decision tree and ARIMA algorithm. Through two-layer data fusion, improved clustering algorithm and fault repair decision tree, it realizes full-dimensional accurate perception of equipment status and scientific classification and early prediction of faults. It is equipped with ARIMA time series prediction algorithm for pre-fault prevention and control.
It significantly improved the online rate of monitoring equipment, reduced operation and maintenance costs, enabled scientific classification and early prediction of faults, improved fault location accuracy and early warning timeliness, increased linkage control efficiency by more than 50%, and increased equipment online rate to more than 98%.
Smart Images
Figure CN122131674A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of alarm systems for abnormal states, and particularly relates to an intelligent early warning and linkage control system for monitoring devices based on multi-sensor fusion, and especially to the improvement and integration of the evidence theory and improvement and a full-life cycle operation and maintenance system for monitoring devices that integrates intelligent algorithms such as fault repair decision trees and ARIMA, and is applicable to the operation and maintenance management of video monitoring devices in scenarios such as雪亮工程, intelligent transportation, sky network projects, and safe cities. Background Technique
[0002] With the rapid development of the security monitoring field, the large-scale implementation of projects such as雪亮工程 and intelligent transportation has led to an exponential increase in the number of deployed video monitoring devices, and their operating stability directly determines the overall effectiveness of the security system. There are many core technical defects in the existing monitoring device operation and maintenance systems: First, the multi-source data fusion lacks scientific algorithm support and only uses simple data splicing methods, which cannot solve the problem of sensor data conflicts, resulting in low fault location accuracy and high rates of missed detection and false detection; Second, fault analysis and early warning rely on empirical threshold judgment, without the ability to extract deep features and perform quantitative clustering, and it is impossible to achieve scientific classification and early prediction of faults; Third, the fault repair strategy matching uses fixed rules without intelligent decision algorithm guidance, with poor adaptability and low linkage control efficiency; Fourth, there is a lack of effective fault prediction algorithms, and it can only passively handle faults, unable to achieve pre-event prevention, and it is difficult to improve the online rate of monitoring devices.
[0003] At the same time, the algorithm implementation details of the existing system are vague, without specific quantitative calculation logic, the model interpretability is poor, and it is difficult to be deeply adapted to embedded hardware, unable to meet the actual operation and maintenance needs of complex outdoor monitoring environments. Therefore, the development of an intelligent early warning and linkage control system for monitoring devices that integrates multiple intelligent algorithms and has clear quantitative calculation logic has become an urgent need in the security monitoring field. Summary of the Invention
[0004] In view of the defects of the existing technology, the present invention provides an intelligent early warning and linkage control system for monitoring devices based on multi-sensor fusion, by integrating and improving the evidence theory and improvement and intelligent algorithms such as fault repair decision trees and ARIMA, and clarifying its specific quantitative calculation logic, to solve the technical problems of the existing system such as lagging fault early warning, low data fusion accuracy, inaccurate fault location, and low linkage control efficiency, and to achieve all-dimensional accurate perception of the operating state of monitoring devices, scientific classification and early prediction of faults, and intelligent matching of repair strategies, greatly improving the online rate of monitoring devices and reducing operation and maintenance costs.
[0005] To achieve the above objectives, the present invention adopts the following technical solution:
[0006] A monitoring equipment intelligent early warning and linkage control system based on multi-sensor fusion includes a sensing layer, a data fusion processing layer, an intelligent early warning layer and a linkage control layer that are connected in sequence, and also includes an operation and maintenance management platform that communicates bidirectionally with the intelligent early warning layer and the linkage control layer.
[0007] The sensing layer integrates multiple types of sensors to collect data on the operating status of monitoring equipment and data on the surrounding environment of the equipment, and performs standardized processing on the raw collected data.
[0008] The data fusion processing layer is equipped with improvements. Evidence theory algorithm module and The combined algorithm module uses a two-layer fusion strategy to perform fusion analysis on standardized multi-source heterogeneous data, and outputs equipment operation status assessment results and fault location information.
[0009] The intelligent early warning layer is equipped with improvements. The clustering algorithm module optimizes the initial cluster centers to classify faults and generate corresponding alarm information.
[0010] The linkage control layer is equipped with a fault repair decision tree algorithm module, which intelligently matches and executes fault repair strategies based on fault-related features;
[0011] The operation and maintenance management platform is equipped with the ARIMA time series prediction algorithm module, which is used to predict the probability of equipment failure and realize the prevention and control of failure in advance.
[0012] Furthermore, the improvements The evidence theory algorithm module corrects the original basic probability allocation by introducing an evidence credibility coefficient to resolve data conflict issues. Then, it completes the first-level fusion of heterogeneous data in the same dimension through evidence combination rules to obtain a preliminary judgment result of the single-dimensional device status.
[0013] Furthermore, the aforementioned The combined algorithm module extracts spatial features from the first-layer fusion result through a CNN network, then extracts time-series features through a BiLSTM network, and completes the second-layer fusion after feature fusion. It outputs accurate equipment operation status assessment results and fault cause location information, with a feature extraction accuracy of no less than 95% and a fault cause location accuracy of no less than 90%.
[0014] Furthermore, the improvements The clustering algorithm module optimizes the initial cluster centers using the silhouette coefficient method to avoid local optima. It uses the degree of fault impact, fault development trend, and equipment importance as clustering features to cluster the equipment operation status assessment results into level 1, level 2, and level 3 faults, thereby achieving hierarchical early warning.
[0015] Furthermore, the sensing layer includes at least four of the following sensors: power supply monitoring sensor, network status sensor, video fault sensor, temperature and humidity sensor, and anti-theft detection sensor; it may also optionally include one or more of the following: attitude sensor, water leakage sensor, smoke sensor, and lightning strike sensor; the standardization process adopts... Normalization transforms raw data into data with uniform dimensions.
[0016] Furthermore, the fault repair decision tree algorithm module takes fault type, fault level, and equipment operating environment as core decision features, selects the optimal split node through information gain ratio, constructs a decision tree model, and generates repair strategy matching rules. The matching repair strategies include automatic repair strategy, remote control strategy, and linkage capture strategy.
[0017] Furthermore, the automatic repair strategy is applicable to self-recoverable faults such as network failures and abnormal fan temperature control; the remote control strategy is applicable to operations such as adjusting the fill light and remotely restarting the device; and the linkage capture strategy is applicable to security events such as theft and illegal intrusion. Moreover, the linkage control layer has a built-in Beidou / GPS positioning module, which can synchronously collect device location information and bind and upload it with the captured data.
[0018] Furthermore, the ARIMA time series prediction algorithm module is an ARIMA(p,d,q) model, which transforms non-stationary fault time series into stationary series through differencing, and then realizes quantitative prediction of equipment failure probability through an autoregressive-moving average combined model.
[0019] Furthermore, the aforementioned The training dataset for the combined algorithm module consists of historical operational data from hundreds of thousands of video surveillance points, including normal operation data, various fault data, and environmental interference data. The training process iteratively optimizes the model parameters through the mean squared error loss function.
[0020] Furthermore, the sensing layer is integrated into an intelligent monitoring box / cabinet with a protection level of not less than IP55; the algorithm modules of the data fusion processing layer, intelligent early warning layer, and linkage control layer are all deployed in embedded chips, and the algorithm model parameters are updated and optimized online through the gradient descent method.
[0021] This invention provides an intelligent early warning and linkage control system for monitoring equipment based on multi-sensor fusion, which, compared with existing technologies:
[0022] This invention constructs an "improved" Evidence Theory The dual-layer data fusion system effectively solves the problems of conflict between multi-source heterogeneous sensor data and insufficient feature extraction, and greatly improves the accuracy of equipment status assessment and the accuracy of fault location. The feature extraction accuracy rate is no less than 95%, and the fault location accuracy is no less than 90%.
[0023] This invention improves Clustering algorithms enable fault classification and early warning. By using the silhouette coefficient method to optimize the initial cluster centers and avoid local optima, the algorithm can achieve scientific classification based on the actual impact and development trend of the fault, transforming the traditional "post-event alarm" into "pre-event prediction", which significantly improves the timeliness and scientific nature of fault early warning.
[0024] This invention incorporates a fault repair decision tree algorithm, which selects the optimal decision node through information gain ratio and constructs quantitative repair strategy matching rules. It can quickly match the optimal repair scheme according to fault type, level, and operating environment, breaking through the limitations of traditional fixed rules and improving linkage control efficiency by more than 50%.
[0025] This invention introduces the ARIMA time series prediction algorithm into the operation and maintenance management platform, which realizes the quantitative prediction of the probability of equipment failure. It breaks through the limitation of existing systems that can only passively handle failures, realizes the "prevention and control" of failures, and increases the online rate of monitoring equipment to over 98%.
[0026] All algorithm modules of this invention are deeply adapted to embedded hardware, support online parameter updates and optimization, and can adapt to complex and ever-changing outdoor monitoring environments; at the same time, the sensors adopt a standardized interface design, support flexible selection, and are suitable for various application scenarios such as the Skynet Project, smart transportation, and safe city.
[0027] This invention constructs a complete " "The intelligent operation and maintenance closed loop significantly reduces manual on-site operations, lowers operation and maintenance costs by more than 30%, and achieves refined and standardized management of monitoring equipment operation and maintenance." Attached Figure Description
[0028] Figure 1 The sensor sensing block diagram provided by the present invention;
[0029] Figure 2 This is a block diagram of the linkage control provided by the present invention;
[0030] Figure 3 The intelligent early warning block diagram provided by the present invention. Detailed Implementation
[0031] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0032] Example 1: Operation and Maintenance Application of Smart Traffic Intersection Monitoring Equipment
[0033] The system in this embodiment is used for the operation and maintenance of video surveillance equipment at smart traffic intersections. It is adapted to an integrated cast aluminum intelligent monitoring box with IP66 protection level. The standard sensor layer is equipped with a power supply monitoring sensor, a network status sensor, a video fault sensor, a temperature and humidity sensor, and an anti-theft detection sensor. Optional sensors include a lightning strike sensor and an attitude sensor. The data acquisition frequency is set to 5 seconds / time.
[0034] Data acquisition and processing in the sensing layer
[0035] When the intersection monitoring equipment experiences a video signal interruption, the raw data collected by the sensors is as follows: video signal strength value 0, network packet traffic 0, power supply voltage 220V (normal), and ambient temperature 28℃ (normal). The raw data is processed using the Z-Score standardization formula to eliminate dimensional differences. The formula is as follows:
[0036]
[0037] in, The data is standardized, where x represents the raw data collected by the sensor. This is the average of the sensor's historical data collected over the past three months. The standard deviation is calculated to be -2.3 (severely abnormal) for video signal strength and network packet traffic, and 0.2 (normal) for power supply voltage and ambient temperature. The processed data is then input to the data fusion processing layer.
[0038] Data fusion processing layer, two-layer fusion analysis
[0039] First-level fusion (improved DS evidence theory)
[0040] First, calculate the evidence credibility coefficient for each sensor, using the following formula:
[0041]
[0042] in, Let be the confidence coefficient of the evidence from the i-th sensor. Historical detection accuracy of the sensors (network sensor 0.92, video sensor 0.89, power supply sensor 0.98, temperature and humidity sensor 0.95). The real-time operational stability of the sensor is 0.96 for all values. Let m(A) be the number of sensors participating in the fusion, m'(A) be the original basic probability assignment, m'(A) be the corrected basic probability assignment, and j be the summation variable. The process iterates through all sensors participating in the fusion.
[0043] Calculations show that network sensors and video sensors... The values are 0.253 and 0.240 respectively, for the power supply and temperature / humidity sensor. All values were 0.253. After correction, the evidence was fused using the DS evidence combination formula, as follows:
[0044]
[0045] Where K is the conflict coefficient, in this embodiment (Minor data conflict). The fusion yielded the following preliminary single-dimensional assessment results: network anomaly (confidence 0.91), video anomaly (confidence 0.88), power supply normal (confidence 0.99), temperature and humidity normal (confidence 0.99).
[0046] Where m(A): the basic probability assignment of proposition A after fusion, reflecting the comprehensive support for "device state is A" after multi-sensor fusion;
[0047] B: The proposition corrected by the first sensor (e.g., "network anomaly");
[0048] C: The proposition corrected by the second sensor (e.g., "video anomaly");
[0049] B∩C=A: The intersection of propositions B and C is proposition A (i.e., the device state that both sensor data point to).
[0050] ∑B∩C=A: Sum of the products of the basic probability assignments of all B and C whose intersection is A;
[0051] m1′(B): The basic probability assignment of proposition B after correction by the first sensor; m2′(C): The basic probability assignment of proposition C after correction by the second sensor;
[0052] K: Conflict coefficient, ranging from 0 to 1. The larger the value, the more severe the conflict between data from different sensors.
[0053] B∩C= Propositions B and C have no overlap (i.e., the data from the two sensors point to completely opposite device states).
[0054] ∑B∩C= : Summing the products of the basic probability assignments of all non-intersecting B and C.
[0055] Second layer fusion (CNN-BiLSTM combined algorithm)
[0056] Using the preliminary judgment result as input, spatial features are extracted through the CNN convolution formula, as follows:
[0057]
[0058] Where f is the ReLU activation function ( ), The weights are 3×3 convolution kernel weights. For the first layer of fusion input data, This is the convolution bias term.
[0059] Then, time series features are extracted using the BiLSTM forward / backward hidden layer formula, as follows:
[0060]
[0061]
[0062] in, It is the Sigmoid activation function. The weights from the input layer to the hidden layer. The weights from one hidden layer to another. This is a bias term.
[0063] Finally, feature fusion is completed using a feature fusion formula, as follows:
[0064]
[0065] Here, `text{Concat}` represents the concatenation operation. For the weights of the fusion layer, This is the bias term for the fusion layer.
[0066] in, Output features of the BiLSTM forward hidden layer at time t
[0067] Output features of the BiLSTM inverse hidden layer at time t
[0068] The weight matrix from the input layer to the hidden layer in the feedforward hidden layer.
[0069] The hidden layer weight matrix from the previous time step to the current time step in the forward hidden layer.
[0070] The hidden layer weight matrix from the next time step to the current time step in the reverse hidden layer.
[0071] : Forward hidden layer bias term;
[0072] The model training uses the mean squared error loss function to optimize the parameters, as shown in the following formula:
[0073]
[0074] Wherein, Loss: mean squared error loss value, reflects the degree of deviation between the model's prediction results and the actual results; the smaller the value, the higher the model accuracy.
[0075] i is the training sample index; N is the number of training samples. For real labels, Predict labels for the model;
[0076] For all N samples ( 2. Summation.
[0077] Based on the fusion analysis, the output assessment result is: Level 2 fault, the video signal interruption is caused by the network link interruption, and the fault location accuracy is 92%.
[0078] The intelligent early warning layer fault classification adopts an improved K-means clustering algorithm.
[0079] First, the initial cluster centers are selected using the silhouette coefficient formula, as follows:
[0080]
[0081] Where S(i) is the silhouette coefficient of sample i, a(i) is the average distance from sample i to other samples in the same cluster, and b(i) is the average distance from sample i to the nearest heterogeneous cluster. max(a(i),b(i)) is to take the maximum value of a(i) and b(i) to normalize the silhouette coefficient; the three samples corresponding to the maximum value of S(i) are selected as the initial cluster centers (corresponding to level one, two, and three faults).
[0082] Then, the clustering calculation is performed using the Euclidean distance formula, as follows:
[0083]
[0084] Where d(x,y) is the Euclidean distance between sample x and cluster center y, x is the equipment status assessment sample to be clustered; y is the cluster center sample; k is the cluster feature dimension number; p is the total dimension of cluster features (in this embodiment, p=3, corresponding to the degree of fault impact, fault development trend, and equipment importance). Let x be the value of the sample in the k-th dimension feature; Let y be the value of the cluster center on the k-th dimension feature.
[0085] The fault was clustered into a level 2 fault, triggering a local audible and visual alarm in the monitoring box and pushing the alarm information to the mobile terminal of traffic maintenance personnel, automatically generating a maintenance work order.
[0086] Linkage control layer repair strategy matching and execution
[0087] The fault repair decision tree algorithm is used. First, the uncertainty of the dataset is calculated using the information entropy formula, as follows:
[0088]
[0089] Wherein, Ent(D) is the information entropy of dataset D, with a value ≥ 0. The smaller the value, the lower the uncertainty of the dataset; D is the total dataset for fault repair strategy matching (containing samples of fault features and corresponding repair strategies); k is the repair strategy category number; K is the total number of repair strategy categories (in this embodiment, K=3, corresponding to automatic repair, remote control, and linked capture); Ck is the subset of samples belonging to the k-th repair strategy in dataset D; |Ck| is the number of samples in the subset Ck; |D| is the total number of samples in dataset D.
[0090] The optimal split node is then selected using the information gain formula and the information gain ratio formula, as follows:
[0091]
[0092]
[0093]
[0094] in:
[0095] Gain(D,A): Information gain of feature A on dataset D. The larger the value, the stronger the effect of feature A in reducing the uncertainty of the dataset.
[0096] A: Decision characteristics (such as "fault type", "fault level", "operating environment");
[0097] v: The value number of feature A (e.g., the value of feature "fault level" v=1 corresponds to level 1 fault, v=2 corresponds to level 2 fault).
[0098] V: The total number of values for feature A (e.g., V=3 for feature "fault level");
[0099] Dv: A subset of samples in dataset D that takes the v-th value of feature A;
[0100] |Dv|: The number of samples in the sample subset Dv;
[0101] Ent(Dv): Information entropy of the sample subset Dv;
[0102] Gain_ratio(D,A): Information gain ratio of feature A, used to correct the information gain bias for features with multiple values;
[0103] IV(A): The intrinsic value of feature A, reflecting the degree of dispersion of the value of feature A.
[0104] Using "video fault, secondary fault, and outdoor environment at the intersection" as decision features, the maximum information gain ratio was calculated to be 0.87, matching the automatic repair strategy. The system automatically performed network link repair operations, restoring the network link within 10 seconds, and the video signal returned to normal.
[0105] The fault prediction of the operation and maintenance management platform adopts the ARIMA(2,1,1) model for fault prediction.
[0106] First, the non-stationary fault sequence is transformed into a stationary sequence using the difference formula, as follows:
[0107]
[0108] in:
[0109] It is a d-order difference operator used to transform a non-stationary sequence into a stationary sequence;
[0110] d: Difference order (in this embodiment, d=1, i.e., first-order difference);
[0111] yt: The value of the original fault time series at time t (e.g., t=7 corresponds to the number of faults on the 7th day);
[0112] The original fault time series is in The value at time.
[0113] Then, a prediction model is constructed using a combination of autoregressive and moving average formulas, as follows:
[0114]
[0115] in, It is a stationary sequence after differencing. The mean of the sequence. These are the autoregressive coefficients. For the stationary sequence after difference, The value at time, The moving average coefficient is... Let i be the random error term and i be the order index of the autoregressive term. for The random error term (residual) at time step reflects the random factors that the model did not fit.
[0116] Calculations show that the probability of network failure at this intersection is 8% in the next 7 days. Therefore, no prevention and control alerts will be triggered at this time. The platform will automatically complete the work order closure and record all maintenance data.
[0117] Example 2: Operation and Maintenance Application of Safe City Community Monitoring Equipment
[0118] The system in this embodiment is used for the operation and maintenance of video surveillance equipment in safe city communities. It is compatible with a pole-mounted intelligent monitoring box with IP55 protection level. The standard sensor layer is equipped with power supply monitoring sensors, network status sensors, video fault sensors, temperature and humidity sensors, and anti-theft detection sensors. Optional sensors include water leakage sensors and smoke sensors. The data acquisition frequency is set to 10 seconds / time.
[0119] Data acquisition and processing in the sensing layer
[0120] When the community monitoring box is illegally opened and accompanied by excessive smoke concentration, the raw data collected by the sensor is as follows: anti-theft detection switch quantity 1 (triggered), smoke concentration 85ppm (exceeding the standard), ambient temperature 65℃ (exceeding the standard), power supply voltage 220V (normal), and network packet traffic 120kb / s (normal). After processing using the Z-Score normalization formula in Example 1, the normalized values for anti-theft detection, smoke concentration, and ambient temperature are 2.1 (severely abnormal), while the normalized values for power supply and network are 0.3 (normal). The processed data is then input to the data fusion processing layer.
[0121] Data fusion processing layer, two-layer fusion analysis
[0122] First-level fusion (improved DS evidence theory)
[0123] The evidence credibility coefficient formula and DS evidence combination formula in Example 1 are used for calculation of the burglar sensor, smoke sensor, and temperature and humidity sensor. The values are 0.258, 0.250, and 0.245 respectively, for power supply and network sensors. Both are 0.247. (Conflict coefficient of the corrected evidence combination) After fusion, the preliminary judgment results are as follows: illegal opening of the box (confidence level 0.95), fire hazard (confidence level 0.91), excessive temperature and humidity (confidence level 0.88), normal power supply (confidence level 0.99), and normal network (confidence level 0.99).
[0124] Second layer fusion (CNN-BiLSTM combined algorithm)
[0125] The CNN convolution formula, BiLSTM hidden layer formula, feature fusion formula and loss function formula in Example 1 are used for fusion analysis, and the output evaluation result is: Level 3 fault, the monitoring box is illegally opened and accompanied by fire hazard, and the fault location accuracy is 94%.
[0126] The intelligent early warning layer fault classification uses the contour coefficient formula and Euclidean distance formula from Example 1 for cluster calculation, and the fault is clustered into a level three fault. The system synchronously triggers local audible and visual alarms on the monitoring box, pushes alarm information to the community security platform and the mobile terminal of the local police, and automatically generates emergency maintenance work orders and dispatch work orders.
[0127] Linkage control layer repair strategy matching and execution
[0128] The information entropy formula, information gain formula, and information gain ratio formula from Example 1 are used for decision-making. With "theft prevention + fire fault, level 3 fault, and community outdoor environment" as decision features, the maximum information gain ratio is calculated to be 0.91. A linkage capture strategy and a remote control strategy are then applied. The system triggers the capture module to collect on-site images and uses the BeiDou / GPS positioning module to collect precise location information of the monitoring points. The images and location information are then bound together and uploaded to the operation and maintenance management platform. Simultaneously, the monitoring box's opening lights are remotely activated to facilitate on-site personnel operations.
[0129] The operation and maintenance management platform uses the ARIMA(3,1,2) model for fault prediction. The calculation is performed using the difference formula and autoregressive-moving average combination formula from Example 1, predicting a 25% probability of security-related faults occurring in the community's equipment over the next 7 days. The platform then sends prevention and control tips to community security personnel, suggesting increased patrols in key areas to reduce security risks at the source.
[0130] Industrial applicability
[0131] The intelligent early warning and linkage control system for monitoring equipment based on multi-sensor fusion of the present invention constructs a complete algorithm system with specific quantitative calculation formulas as the core. All algorithm modules can be modularly deployed in embedded chips, and online parameter updates and optimizations are achieved through gradient descent method. It has strong hardware compatibility with existing video surveillance equipment.
[0132] This system can be directly applied to the upgrading and transformation of existing security monitoring projects such as the Skynet Project, Smart Transportation, Skynet Project, and Safe City. It can also be used for the deployment of operation and maintenance systems for newly built monitoring projects. It can achieve precise operation and maintenance of monitoring equipment and prevent and control faults in advance, greatly improve the online rate of equipment, and reduce operation and maintenance costs. It has broad industrial application prospects and significant market value.
[0133] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A smart early warning and linkage control system for monitoring equipment based on multi-sensor fusion, characterized in that, It includes a sensing layer, a data fusion processing layer, an intelligent early warning layer, and a linkage control layer that are connected in sequence, and also includes an operation and maintenance management platform that communicates bidirectionally with the intelligent early warning layer and the linkage control layer; The sensing layer integrates multiple types of sensors to collect data on the operating status of monitoring equipment and data on the surrounding environment of the equipment, and performs standardized processing on the raw collected data. The data fusion processing layer is equipped with improvements. Evidence theory algorithm module and The combined algorithm module uses a two-layer fusion strategy to perform fusion analysis on standardized multi-source heterogeneous data, and outputs equipment operation status assessment results and fault location information. The intelligent early warning layer is equipped with improvements. The clustering algorithm module optimizes the initial cluster centers to classify faults and generate corresponding alarm information. The linkage control layer is equipped with a fault repair decision tree algorithm module, which intelligently matches and executes fault repair strategies based on fault-related features; The operation and maintenance management platform is equipped with the ARIMA time series prediction algorithm module, which is used to predict the probability of equipment failure and realize the prevention and control of failure in advance.
2. The system according to claim 1, characterized in that, The improvements The evidence theory algorithm module corrects the original basic probability allocation by introducing an evidence credibility coefficient to resolve data conflict issues. Then, it completes the first-level fusion of heterogeneous data in the same dimension through evidence combination rules to obtain a preliminary judgment result of the single-dimensional device status.
3. The system according to claim 1, characterized in that, The The combined algorithm module extracts spatial features from the first-layer fusion result through a CNN network, then extracts time-series features through a BiLSTM network, and completes the second-layer fusion after feature fusion. It outputs accurate equipment operation status assessment results and fault cause location information, with a feature extraction accuracy of no less than 95% and a fault cause location accuracy of no less than 90%.
4. The system according to claim 1, characterized in that, The improvements The clustering algorithm module optimizes the initial cluster centers using the silhouette coefficient method to avoid local optima. It uses the degree of fault impact, fault development trend, and equipment importance as clustering features to cluster the equipment operation status assessment results into level 1, level 2, and level 3 faults, thereby achieving hierarchical early warning.
5. The system according to claim 1, characterized in that, The sensing layer includes at least four of the following sensors: power supply monitoring sensor, network status sensor, video fault sensor, temperature and humidity sensor, and anti-theft detection sensor. It may also optionally include one or more of the following: attitude sensor, water leakage sensor, smoke sensor, and lightning strike sensor. The standardization process adopts... Normalization transforms raw data into data with uniform dimensions.
6. The system according to claim 1, characterized in that, The fault repair decision tree algorithm module takes fault type, fault level, and equipment operating environment as core decision features, selects the optimal split node through information gain ratio, constructs a decision tree model, and generates repair strategy matching rules. The matching repair strategies include automatic repair strategy, remote control strategy, and linkage capture strategy.
7. The system according to claim 6, characterized in that, The automatic repair strategy is applicable to self-recoverable faults such as network failures and abnormal fan temperature control. The remote control strategy is applicable to operations such as adjusting the fill light and remotely restarting the device. The linkage capture strategy is applicable to security events such as theft and illegal intrusion. The linkage control layer has a built-in Beidou / GPS positioning module, which can synchronously collect device location information and bind and upload it with the captured data.
8. The system according to claim 1, characterized in that, The ARIMA time series prediction algorithm module is an ARIMA(p,d,q) model. It transforms non-stationary fault time series into stationary series through differencing, and then uses an autoregressive-moving average combined model to quantitatively predict the probability of equipment failure.
9. The system according to claim 3, characterized in that, The The training dataset for the combined algorithm module consists of historical operational data from hundreds of thousands of video surveillance points, including normal operation data, various fault data, and environmental interference data. The training process iteratively optimizes the model parameters through the mean squared error loss function.
10. The system according to claim 1, characterized in that, The sensing layer is integrated into an intelligent monitoring box / cabinet with a protection level of not less than IP55; the algorithm modules of the data fusion processing layer, intelligent early warning layer, and linkage control layer are all deployed in embedded chips, and the algorithm model parameters are updated and optimized online through the gradient descent method.