Intelligent control method and system of alarm based on internet of things

Abstract

This invention relates to the field of IoT-based intelligent control technology for security systems, specifically to an IoT-based intelligent control method and system for alarms. The method includes: establishing a multi-source sensor node network for public security equipment alarms in an IoT environment; collecting multi-dimensional sensor information to form an environmental perception dataset; extracting a state-linkage feature set through information fusion and feature extraction; and constructing an alarm behavior evolution map. Based on this map, an alarm situation assessment model is established; the situation correlation density of historical triggering events is calculated to determine the early warning parameter set; real-time state-linkage features are extracted to form a situation vector; real-time alarm risk correlation is obtained through situation similarity calculation; and an alarm trigger probability assessment model is then constructed to predict the probability of abnormal triggering and generate control commands. This invention enables the linkage analysis of alarm status and environment, quantifies alarm risk, accurately predicts abnormal triggering, and makes alarm control more intelligent and targeted.

Description

Technical Field

[0001] This invention relates to the field of IoT security intelligent control technology, and in particular to an IoT-based intelligent control method and system for alarms. Background Technology

[0002] Current public security alarm systems mostly rely on the Internet of Things (IoT) for basic data transmission. They typically use single sensor nodes to collect alarm operating parameters and trigger signals, simply storing and thresholding the collected discrete sensor data. They lack a multi-source sensor node network adapted to public security alarm systems and fail to perform information fusion and feature extraction processing on the collected sensor information. Existing technologies cannot extract the linkage characteristics between the alarm's own state and the external environment, nor can they construct a behavioral evolution map that reflects the alarm's behavioral evolution. They can only passively record the alarm's historical trigger data, failing to achieve a deep correlation between the alarm's operating status and environmental information.

[0003] Current alarm status assessments rely solely on fixed thresholds, failing to calculate the corresponding status correlation density based on historical triggering events and thus hindering the creation of a status warning parameter set tailored to the alarm's operational characteristics. Real-time alarm risk detection simply compares real-time sensor data directly with preset thresholds, without quantifying the real-time alarm risk correlation through status similarity calculations. This prevents the construction of an accurate alarm trigger probability assessment model and the prediction of abnormal alarm trigger probabilities.

[0004] It is impossible to form alarm status linkage characteristics and behavior evolution map through the fusion of multi-source sensor information, impossible to quantify alarm risk by relying on situation correlation density calculation and situation similarity comparison, impossible to generate appropriate alarm intelligent control commands based on abnormal trigger probability, and difficult to get rid of the traditional alarm single threshold control operation mode. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing an intelligent control method and system for alarms based on the Internet of Things.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent control method for alarms based on the Internet of Things, comprising: A multi-source sensor node network for public security equipment alarms is established in an Internet of Things environment, and multi-dimensional sensor information of the alarms is collected through the multi-source sensor node network during a preset monitoring phase to generate an alarm environmental perception dataset. Information fusion and feature extraction processing are performed on the alarm environmental perception dataset to extract a state linkage feature set that reflects the linkage between the alarm state and the environment, and an alarm behavior evolution map is constructed based on the state linkage feature set. Based on the alarm behavior evolution map, an alarm situation assessment model is established, the situation correlation density of historical triggering events of the alarm in the alarm behavior evolution map is calculated, and the alarm status warning parameter set is determined according to the situation correlation density. Based on the real-time sensing information collected by the multi-source sensor node network, the real-time status linkage features of the alarm are extracted to form the real-time situation vector of the alarm. The real-time situation vector of the alarm is then imported into the alarm situation assessment model, and the situation similarity is calculated with the alarm status warning parameter set to output the real-time alarm risk correlation. By combining the real-time alarm risk correlation, the state linkage feature set, and the alarm behavior evolution map, an alarm trigger probability assessment model is generated to predict the abnormal trigger probability of the alarm, and alarm control commands are generated based on the prediction results.

[0007] As a further aspect of the present invention, the establishment of a multi-source sensor node network for public security equipment alarms in an Internet of Things environment, and the collection of multi-dimensional sensor information of the alarms through the multi-source sensor node network during a preset monitoring phase to generate an alarm environmental perception dataset, specifically includes: Deploy multi-source sensor nodes, including vibration sensors, acoustic sensors, ambient light sensors, and positioning beacons, on public security equipment alarm devices; At least one preset monitoring phase is set, and the preset monitoring phase includes at least two consecutive data sampling periods; The multi-source sensing nodes are controlled to collect data sampling cycles during the preset monitoring phase, synchronously collecting vibration intensity of the alarm body, ambient background sound patterns, changes in light intensity, and alarm location information to generate multi-dimensional sensing information. The collected multidimensional sensor information is timestamped and packaged to form an alarm environmental perception dataset containing vibration intensity sequence, background acoustic waveform sequence, light intensity sequence and location information sequence.

[0008] As a further aspect of the present invention, information fusion and feature extraction processing are performed on the alarm environmental perception dataset to extract a state linkage feature set reflecting the interaction between the alarm state and the environment, and an alarm behavior evolution map is constructed based on the state linkage feature set, specifically as follows: Time-domain and frequency-domain analyses were performed on the vibration intensity sequence, background acoustic waveform sequence, light intensity sequence, and location information sequence in the environmental perception dataset of the alarm device. From the time-domain and frequency-domain analysis results, the vibration intensity characteristics of the alarm body, the energy entropy characteristics of the environmental sound pattern, the frequency characteristics of the sudden change in the ambient light, and the location movement trajectory characteristics are extracted to form a state linkage feature set; Different feature parameters in the state linkage feature set are used as graph nodes, and the temporal correlation and statistical dependence between feature parameters are used as edges between nodes. The edges are assigned corresponding weight values ​​to construct an alarm behavior evolution graph in the form of a directed weighted graph. The alarm behavior evolution map reflects the dynamic correlation and evolution relationship between alarm features under different states.

[0009] As a further aspect of the present invention, based on the alarm behavior evolution map, an alarm situation assessment model is established, the situation correlation density of historical triggering events of the alarm in the alarm behavior evolution map is calculated, and the alarm status warning parameter set is determined according to the situation correlation density, specifically as follows: Using the alarm behavior evolution map as the underlying graph structure, an alarm situation assessment model is established. The input of the alarm situation assessment model is the state linkage feature set, and the output of the alarm situation assessment model is the alarm situation risk value. In the alarm behavior evolution graph, the positions of the corresponding graph nodes when the alarm's historical triggering events occurred are marked; The clustering degree of the graph node set corresponding to all historical triggering events in the alarm behavior evolution graph is calculated to obtain the situation correlation density; Set a situation correlation density threshold and filter out the set of graph nodes whose situation correlation density exceeds the threshold; Extract the state linkage feature parameters corresponding to the graph node set to form the alarm state warning parameter set.

[0010] As a further aspect of the present invention, based on the real-time sensing information collected by the multi-source sensor node network, the real-time status linkage features of the alarm are extracted to form a real-time alarm situation vector. This real-time alarm situation vector is then imported into the alarm situation assessment model, and a situation similarity calculation is performed with the alarm status warning parameter set to output the real-time alarm risk correlation. Specifically: The sensor information stream of the alarm is collected in real time through the multi-source sensor node network; The sensor information stream is preprocessed and features are extracted to obtain real-time alarm body vibration intensity features, environmental sound pattern energy entropy features, ambient light sudden change frequency features, and position movement trajectory features, which are combined to form the alarm real-time situation vector. Input the real-time situation vector of the alarm device into the alarm situation assessment model; In the alarm situation assessment model, the graph theory distance between the graph node position corresponding to the real-time situation vector of the alarm and the graph node position corresponding to each warning parameter in the alarm status warning parameter set is calculated, and the situation similarity is calculated based on the graph theory distance. The quantification result of the situational similarity is output as the real-time alarm risk correlation degree.

[0011] As a further aspect of the present invention, by combining the real-time alarm risk correlation degree, the state linkage feature set, and the alarm behavior evolution map, an alarm trigger probability assessment model is generated to predict the abnormal trigger probability of the alarm, and alarm control commands are generated based on the prediction results, specifically as follows: Using the real-time alarm risk correlation degree, the state linkage feature set, and the graph structure features of the alarm behavior evolution map as inputs, an alarm trigger probability evaluation model is constructed. Using the alarm trigger probability evaluation model, the evolution path of the alarm's real-time situation vector in the graph structure is analyzed, and combined with the distribution pattern of historical trigger events in the graph, the probability value of the alarm being abnormally triggered within a future preset time window is calculated and generated. The probability value is compared with a preset alarm trigger threshold and a false alarm suppression threshold; Based on the comparison results, generate alarm control commands that include alarm trigger enable, alarm sensitivity adjustment, alarm silence control, or alarm signal reporting.

[0012] As a further aspect of the present invention, it also includes the steps of data augmentation and anomaly cleaning of the alarm environmental perception dataset, specifically: Detect missing segments of sensor information in the environmental perception dataset of the alarm device; An interpolation algorithm based on the spatiotemporal correlation of sensor information is used to reconstruct and fill in the missing segments of the sensor information. Data points in the environmental perception dataset of the alarm device that deviate from a preset fluctuation threshold are identified as noise points and then removed. After removing noise points, the dataset is processed using a sliding window smoothing algorithm to obtain a cleaned and standardized alarm environmental perception dataset, which is then used for subsequent information fusion and feature extraction.

[0013] As a further aspect of the present invention, it also includes dynamically optimizing the behavior evolution map of the alarm device, specifically: Periodically acquire multi-dimensional sensor information collected by the multi-source sensor node network during the new monitoring phase; The newly added multidimensional sensing information is processed to extract the newly added state linkage features and update the state linkage feature set. Based on the updated state linkage feature set, add or delete graph nodes in the alarm behavior evolution graph, and recalculate the weights of the edges between graph nodes; Based on the recalculated edge weights, the connection structure of the alarm behavior evolution graph is adjusted to generate an optimized alarm behavior evolution graph, which is used to update the alarm situation assessment model.

[0014] As a further aspect of the present invention, it also includes establishing an alarm control strategy knowledge base, specifically: Record the real-time alarm risk correlation degree generated each time, the predicted probability value of the alarm trigger probability evaluation model, and the final alarm control command executed; Record the actual status of the alarm and environmental changes fed back by the multi-source sensor node network after the alarm control command is executed; The real-time alarm risk correlation, predicted probability value, alarm control command and the actual status information fed back are associated and stored in the knowledge base as a control strategy case. In the alarm control strategy knowledge base, control strategy cases are classified, summarized, and rules are extracted to form control strategy rules, which are used to assist in generating or modifying the alarm control commands.

[0015] As a further aspect of the present invention, the present invention also includes an intelligent control system for an alarm based on the Internet of Things (IoT), the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the intelligent control method for an alarm based on the IoT described above.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: A multi-source sensor node network for public security equipment alarms is established in an Internet of Things (IoT) environment. During the pre-set monitoring phase, multi-dimensional sensor information of the alarms is collected and an environmental perception dataset is generated. Information fusion and feature extraction processing are performed on the perception dataset to extract a set of state linkage features that reflect the interaction between the alarm state and the environment. Based on the set of state linkage features, an alarm behavior evolution map is constructed. This map can integrate the correlation between multi-dimensional alarm data, intuitively present the linkage logic between the alarm state and the environment, and fully demonstrate the dynamic evolution process of alarm behavior.

[0017] An alarm situation assessment model is established based on the alarm behavior evolution map. The situation correlation density of historical triggering events in the behavior evolution map is calculated. The alarm status warning parameter set is determined based on the situation correlation density. Real-time status linkage features are extracted to form a real-time situation vector. The real-time situation vector is imported into the alarm situation assessment model to complete the situation similarity calculation and output the real-time alarm risk correlation. Combining the real-time alarm risk correlation, the status linkage feature set and the alarm behavior evolution map, an alarm trigger probability assessment model is generated to predict the alarm abnormal trigger probability and generate alarm control commands. The situation correlation characteristics of historical triggering events can be quantified to achieve accurate quantification of real-time alarm risk. Intelligent control of the alarm is achieved through abnormal trigger probability prediction. Attached Figure Description

[0018] Figure 1 This is a flowchart of the IoT-based intelligent control method for alarms as described in this invention; Figure 2 A flowchart for establishing a multi-source sensor node network and generating an alarm environmental perception dataset; Figure 3 A flowchart for extracting real-time status linkage features and outputting real-time alarm risk correlation; Figure 4 A time-series diagram for extracting state-linked features; Figure 5 This is a time sequence diagram for alarm risk assessment and trigger probability. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] See Figure 1 This invention provides an intelligent control method for alarms based on the Internet of Things (IoT), the overall implementation of which is as follows: A multi-source sensor node network for public security equipment alarms is established under an IoT environment. During a pre-defined monitoring phase, this network collects multi-dimensional sensor information from the alarms, generating an alarm environment perception dataset. Information fusion and feature extraction are performed on this dataset to extract a state-linkage feature set reflecting the interaction between the alarm's state and its environment. Based on this feature set, an alarm behavior evolution map is constructed. An alarm situation assessment model is established based on this map, calculating the situation correlation density of historical triggering events within the map. The alarm state warning parameter set is then determined based on this density. Real-time sensor information collected by the multi-source sensor node network is used to extract real-time state linkage features, forming a real-time alarm situation vector. This vector is imported into the alarm situation assessment model, and its similarity to the alarm state warning parameter set is calculated to output the real-time alarm risk correlation. Combining the real-time alarm risk correlation, the state linkage feature set, and the alarm behavior evolution map, an alarm trigger probability assessment model is generated to predict the probability of abnormal alarm triggering. Based on the prediction results, alarm control commands are generated.

[0022] In one embodiment of the present invention, see [reference] Figure 2 Multi-source sensor nodes, including vibration sensors, acoustic sensors, ambient light sensors, and positioning beacons, are deployed on public security equipment alarms. At least one preset monitoring phase is established, comprising at least two consecutive data sampling cycles. The multi-source sensor nodes are controlled to synchronously collect vibration intensity of the alarm body, ambient background acoustic waves, changes in light intensity, and alarm location information during each data sampling cycle within the preset monitoring phase, generating multi-dimensional sensor information. The collected multi-dimensional sensor information is timestamped and packaged to form an alarm environmental perception dataset containing vibration intensity sequences, background acoustic wave sequences, light intensity sequences, and location information sequences.

[0023] Time-domain and frequency-domain analyses were performed on the vibration intensity sequence, background acoustic waveform sequence, light intensity sequence, and location information sequence from the alarm's environmental perception dataset. From the time-domain and frequency-domain analysis results, the alarm's vibration intensity characteristics, environmental acoustic energy entropy characteristics, ambient light abrupt change frequency characteristics, and location movement trajectory characteristics were extracted to construct a state-linkage feature set. Different feature parameters in the state-linkage feature set were used as graph nodes, and the temporal correlation and statistical dependence between feature parameters were used as edges between nodes, with corresponding weight values ​​assigned to the edges, to construct an alarm behavior evolution graph in the form of a directed weighted graph. The alarm behavior evolution graph reflects the dynamic correlation and evolution relationship between alarm features under different states.

[0024] In practical implementation, the IoT-based intelligent control method for alarms achieves data acquisition by deploying a multi-source sensor node network including vibration sensors, acoustic sensors, ambient light sensors, and positioning beacons. This multi-source sensor node network is integrated into the public security equipment alarm to monitor its operating environment. At least one preset monitoring phase is set, comprising at least two consecutive data sampling cycles. The multi-source sensor node network is controlled to synchronously collect the alarm's vibration intensity, ambient background acoustic signature, light intensity changes, and alarm location information during each data sampling cycle within the preset monitoring phase, generating multi-dimensional sensor information. The collected multi-dimensional sensor information is then timestamped and packaged to form an alarm environmental perception dataset containing vibration intensity sequences, background acoustic signature waveform sequences, light intensity sequences, and location information sequences. In some embodiments, vibration sensors record the mechanical vibration amplitude of the alarm body at a fixed sampling frequency, acoustic signature sensors capture audio waveforms in the environment, ambient light sensors measure the surrounding light level, and positioning beacons provide the alarm's coordinate information via a Global Positioning System (GPS) or indoor positioning technology. All these sensor data are time-aligned and combined to form the alarm environmental perception dataset. Optionally, the data sampling period can be adjusted to millisecond or second intervals according to monitoring needs to ensure that the alarm's environmental perception dataset can capture rapidly changing environmental events.

[0025] It is understandable that setting at least two consecutive data sampling periods is to collect multiple data points required to form a time series within the preset monitoring phase. The resulting alarm environmental perception dataset, including vibration intensity sequence, background sound waveform sequence, light intensity sequence, and location information sequence, serves as the data foundation for subsequent time-domain and frequency-domain analysis, extraction of state linkage feature sets, and construction of an alarm behavior evolution map that reflects the dynamic correlation and evolution relationship between features.

[0026] In practical implementation, information fusion and feature extraction are performed on the alarm environmental perception dataset. Time-domain and frequency-domain analyses are conducted on the vibration intensity sequence, background acoustic waveform sequence, light intensity sequence, and location information sequence within the dataset. Time-domain analysis involves calculating the root mean square value of the vibration intensity sequence to reflect the vibration intensity, while frequency-domain analysis uses Fast Fourier Transform to convert the background acoustic waveform sequence into a spectrum and calculates the energy entropy. From the time-domain and frequency-domain analysis results, the alarm body vibration intensity features, environmental acoustic energy entropy features, ambient light abrupt change frequency features, and location movement trajectory features are extracted to form a state-linkage feature set. The alarm body vibration intensity features represent the intensity of the external force acting on the alarm; the environmental acoustic energy entropy features quantify the complexity of the ambient sound; the ambient light abrupt change frequency features record the rate of light change; and the location movement trajectory features describe the alarm's displacement path in space. It can be understood that the state-linkage feature set reflects the linkage relationship between the alarm's state and environmental parameters, providing a foundation for constructing an alarm behavior evolution map. In specific implementation, different feature parameters in the state linkage feature set are used as graph nodes, and the temporal correlation and statistical dependence between feature parameters are used as edges between nodes. Each edge is assigned a corresponding weight value, constructing an alarm behavior evolution graph in the form of a directed weighted graph. This graph visualizes the dynamic correlation and evolution relationships between different feature parameters. In some embodiments, the edge weights are calculated based on the Pearson correlation coefficient and temporal lag mutual information between feature parameters. The weight calculation formula is expressed as: ; in: It is a characteristic parameter With characteristic parameters The weight values ​​of the edges between them. It is a characteristic parameter With characteristic parameters Pearson correlation coefficient, It is a characteristic parameter With characteristic parameters In time lag The mutual information value below, and These are preset weighting coefficients used to balance the contributions of correlation coefficients and mutual information. This formula ensures that the weights of the edges in the alarm behavior evolution graph simultaneously capture both the linear correlation and nonlinear temporal dependence of the feature parameters. Optionally, a temporal lag... The data sampling period can be adjusted to optimize the alarm behavior evolution map's ability to represent the alarm's state evolution. The alarm behavior evolution map is ultimately stored in the form of an adjacency matrix or graph database for subsequent alarm situation assessment model calculations.

[0027] In one embodiment of the present invention, an alarm situation assessment model is established using an alarm behavior evolution graph as the underlying graph structure. The input to the alarm situation assessment model is a set of state linkage features, and the output is an alarm situation risk value. In the alarm behavior evolution graph, the positions of the graph nodes corresponding to historical triggering events of the alarm are marked. The clustering degree of the graph node sets corresponding to all historical triggering events in the alarm behavior evolution graph is calculated to obtain the situation correlation density. A situation correlation density threshold is set, and graph node sets with situation correlation densities exceeding the threshold are selected. The state linkage feature parameters corresponding to the graph node sets are extracted to form an alarm state warning parameter set.

[0028] In practical implementation, an alarm situation assessment model is established using the alarm behavior evolution graph as the underlying graph structure. The input of the alarm situation assessment model is the state linkage feature set, and the output is the alarm situation risk value. In the alarm behavior evolution graph, the positions of the graph nodes corresponding to the historical triggering events of the alarm are marked. These historical triggering events are manually labeled or extracted from alarm logs. Each event corresponds to a time point, and the graph node mapped to the alarm's real-time situation vector at that time point is marked as the event node. In some embodiments, the alarm situation assessment model can be a graph neural network model, whose input is the numerical values ​​of each feature parameter in the state linkage feature set, and whose output is a scalar value representing the alarm situation risk value. The structural information of the alarm behavior evolution graph is used as the message passing path of the graph neural network. It can be understood that marking the graph nodes corresponding to historical triggering events is the basis for calculating the situation correlation density.

[0029] In practical implementation, the situational correlation density is obtained by calculating the clustering degree of the graph node set corresponding to all historical triggering events in the alarm behavior evolution graph. The clustering degree is quantified by analyzing the spatial distribution of all marked event nodes in the graph structure. The formula for calculating the situational correlation density D can be expressed as: ; in: Indicates the situational correlation density. This represents the total number of graph nodes corresponding to historical triggering events. Represents graph nodes With graph nodes In the shortest path distance of the alarm behavior evolution graph, the distance is 1 when two graph nodes are directly connected. The reciprocal form of the formula means that nodes that are closer together contribute more, thus increasing the situational correlation density. A higher value indicates a greater clustering of event nodes in the graph. This can be understood as the situational correlation density reflecting the concentration pattern of historical alarm events in the graph feature space. In some embodiments, the shortest path distance... The situation correlation density is calculated based on the weights of the edges in the alarm behavior evolution graph. Edges with larger weights are considered to be closer. The weighted shortest path distance between all event node pairs is calculated and then substituted into the situation correlation density formula. Optionally, the calculation cycle of the situation correlation density can be synchronized with the model update cycle. Whenever a new historical trigger event is recorded, the situation correlation density is recalculated.

[0030] In practical implementation, a situational association density threshold is set, and a set of graph nodes whose situational association density exceeds the threshold is selected. The situational association density threshold is a preset value used to determine whether the clustering degree of the graph node set is significant. The calculated situational association density is then used to... Compared with the situation correlation density threshold, if the situation correlation density If the situational correlation density threshold is exceeded, the set of all current event nodes is determined to be a highly clustered set. Optionally, the situational correlation density threshold can be set by statistically analyzing historical data from periods without events to obtain a baseline value. State linkage feature parameters corresponding to the graph node set are extracted to form an alarm status warning parameter set. After the graph node set is selected, each graph node in the set is traversed. Each graph node corresponds to one or more specific feature parameters in the state linkage feature set, such as "vibration intensity_high frequency" or "acoustic energy entropy_nighttime". These feature parameters are then summarized to form the alarm status warning parameter set. The alarm status warning parameter set serves as the direct comparison benchmark for subsequent real-time situational similarity calculations.

[0031] In one embodiment of the present invention, see [reference] Figure 3 The system collects real-time sensor information streams from the alarm via a multi-source sensor node network. The sensor information streams are preprocessed and feature extracted to obtain real-time alarm vibration intensity characteristics, environmental sound signature energy entropy characteristics, ambient light abrupt change frequency characteristics, and position movement trajectory characteristics. These are combined to form a real-time alarm situation vector. This real-time alarm situation vector is then input into an alarm situation assessment model. Within the model, the graph theory distance between the graph node positions corresponding to the alarm's real-time situation vector and the graph node positions corresponding to each warning parameter in the alarm's state warning parameter set is calculated. Situation similarity is then calculated based on the graph theory distance. The quantified situation similarity result is output as the real-time alarm risk correlation degree.

[0032] In specific implementation, based on real-time sensing information collected by a multi-source sensor node network, the alarm's sensing information stream is acquired in real time through the multi-source sensor node network. This sensing information stream includes raw data sequences generated by vibration sensors, acoustic sensors, ambient light sensors, and positioning beacons at continuous time points. The sensing information stream undergoes preprocessing and feature extraction. Preprocessing includes filtering the raw data sequences to eliminate high-frequency interference. Feature extraction applies the same time-domain and frequency-domain analysis methods used when constructing the state-linkage feature set to obtain real-time alarm body vibration intensity characteristics, ambient acoustic energy entropy characteristics, ambient light abrupt change frequency characteristics, and position movement trajectory characteristics, which are combined to form the alarm's real-time situation vector. The alarm's real-time situation vector is an array containing fixed-dimensional feature values, and its structure is consistent with the feature dimensions of individual samples in the state-linkage feature set. In some embodiments, real-time feature extraction is performed using a sliding window approach, calculating feature values ​​and updating the alarm's real-time situation vector based on the latest window data at regular intervals. Optionally, the preprocessing step may also include time synchronization calibration of different sensor data in the sensing information stream to ensure that the feature values ​​in the alarm's real-time situation vector are aligned in time.

[0033] In practical implementation, the real-time situation vector of the alarm is input into the alarm situation assessment model. Within this model, the graph theory distance between the graph node positions corresponding to the real-time situation vector and the graph node positions corresponding to each warning parameter in the alarm state warning parameter set is calculated. Graph theory distance refers to the shortest path length in the alarm behavior evolution graph from the source graph node mapped from the real-time situation vector to the target graph node corresponding to a certain warning parameter in the alarm state warning parameter set. The path length is the sum of the inverses of the weights of all edges on the path. It can be understood that the smaller the graph theory distance, the closer the real-time state is to the historical warning state in the feature association graph. Situation similarity. Based on the calculated graph theory distance, its calculation formula can be expressed as: ; in: Indicates the similarity of situations. This indicates the total number of warning parameters included in the alarm status warning parameter set. This represents the graph node corresponding to the real-time situation vector of the alarm device up to the 1st... Each warning parameter corresponds to a graph theory distance between graph nodes. This is a preset positive scaling factor used to control the rate at which the influence of distance on similarity decays. This formula calculates the average measure of proximity between the real-time state and all warning states. This can be understood as situational similarity. The value ranges from 0 to 1, with higher values ​​indicating a greater overall similarity between the real-time situation and historical early warning patterns. This represents the situational similarity. The quantified results are output as the real-time alarm risk correlation, which is a numerical value indicating the alarm risk at the current moment. Refer to Table 1, which shows an example alarm's real-time situation vector, its corresponding graph theory distance, and intermediate calculation results. .

[0034] Table 1. Examples of real-time situation vector, graph distance, and situation similarity calculation: Note: Scaling factors are assumed in the table. Total number of warning parameters Calculate the situation similarity according to the formula. This value serves as the output of the real-time alarm risk correlation. In some embodiments, graph theory distance... The calculations are performed using graph search algorithms within the alarm behavior evolution graph, such as Dijkstra's algorithm to find the weighted shortest path. Optional scaling factor. The value can be learned from historical data to optimize the ability of situation similarity to distinguish real alarm events.

[0035] In one embodiment of the present invention, an alarm trigger probability assessment model is constructed using the graph structure features of real-time alarm risk correlation, state linkage feature set, and alarm behavior evolution map as input. Using this model, the evolution path of the alarm's real-time situation vector in the graph structure is analyzed, and combined with the distribution patterns of historical triggering events in the graph, a probability value for abnormal alarm triggering within a preset future time window is calculated. This probability value is compared with preset alarm trigger thresholds and false alarm suppression thresholds. Based on the comparison results, alarm control commands are generated, including alarm trigger enable, alarm sensitivity adjustment, alarm silence control, or alarm signal reporting.

[0036] Missing segments of sensor information in the alarm environmental perception dataset are detected. An interpolation algorithm based on the spatiotemporal correlation of sensor information is used to reconstruct and fill in the missing segments. Data points in the alarm environmental perception dataset whose deviation exceeds a preset fluctuation threshold are identified as noise points and removed. The dataset after noise removal is processed using a sliding window smoothing algorithm to obtain a cleaned and standardized alarm environmental perception dataset, which is used for subsequent information fusion and feature extraction.

[0037] In practical implementation, an alarm trigger probability assessment model is generated by combining real-time alarm risk correlation, state linkage feature set, and alarm behavior evolution graph. The model uses the graph structure features of the real-time alarm risk correlation, state linkage feature set, and alarm behavior evolution graph as input. This model can be a machine learning model, with its input feature vector composed of the real-time alarm risk correlation value, the current state linkage feature set value, and graph structure features extracted from the alarm behavior evolution graph. The graph structure features include the degree centrality or clustering coefficient of the graph nodes corresponding to the current alarm real-time situation vector. The alarm trigger probability assessment model is used to analyze the evolution path of the alarm real-time situation vector in the graph structure. Combined with the distribution patterns of historical trigger events in the graph, the probability value of abnormal triggering of the alarm within a preset time window is calculated. Analyzing the evolution path refers to predicting the graph nodes that the alarm real-time situation vector may reach in subsequent time steps based on the weights and directions of the edges in the alarm behavior evolution graph. The distribution patterns of historical trigger events in the graph are represented by the statistical features of the positions of historical event nodes in the alarm behavior evolution graph. The probability P of the alarm being abnormally triggered within a preset time window in the future can be expressed as: ; in: This indicates the probability that the alarm will be triggered abnormally within a preset time window in the future. Indicates the correlation between real-time alarm risks. This represents a comprehensive feature index calculated based on the current state's linked feature set. This represents a graph evolution risk index calculated based on graph structure characteristics. , , These are the corresponding values ​​learned by the model. , , The weighting coefficients, It is an activation function that maps linear combinations to the probability interval [0,1]. This can be understood as the probability value... It integrates real-time risk, current feature state, and graph structure evolution information. In some embodiments, the integrated feature index... It can be a weighted sum or nonlinear combination of multiple feature values ​​in the state linkage feature set, a graph evolution risk indicator. It can be obtained by calculating the reciprocal of the weighted graph distance from the real-time situation node to the nearest historical event node.

[0038] In practice, the probability value is compared with preset alarm trigger thresholds and false alarm suppression thresholds. The alarm trigger threshold is a high probability threshold used to determine whether an alarm should be triggered, while the false alarm suppression threshold is a low probability threshold used to determine whether possible false alarms should be suppressed. Based on the comparison result, alarm control commands are generated, including alarm trigger enable, alarm sensitivity adjustment, alarm silence control, or alarm signal reporting. The comparison logic follows preset rules. For example, when the probability value... When the probability value is greater than or equal to the alarm trigger threshold, an alarm trigger enable or alarm signal reporting command is generated; when the probability value is greater than or equal to the alarm trigger threshold, an alarm trigger enable or alarm signal reporting command is generated. When the false alarm suppression threshold is below, an alarm silence control command is generated; when the probability value... When the threshold falls between the false alarm suppression threshold and the alarm trigger threshold, an alarm sensitivity adjustment command can be generated to dynamically adjust the sensor threshold for more cautious monitoring. Refer to Table 2 for an example of comparison logic and control command generation.

[0039] Table 2 Comparison of Alarm Trigger Probability and Threshold, and Example of Control Command Generation: In some embodiments, the alarm trigger threshold and false alarm suppression threshold can be dynamically adjusted according to different time periods or alarm deployment environments. Optionally, control commands are sent to the control unit of the public security equipment alarm via an Internet of Things (IoT) communication protocol for execution.

[0040] In practical implementation, data augmentation and anomaly cleaning are performed on the alarm environmental perception dataset. Missing sensor information segments are detected within the dataset, referring to periods where one or more sensor data sequences exhibit consecutive null or invalid values ​​due to temporary sensor malfunctions or communication interruptions. An interpolation algorithm based on the spatiotemporal correlation of sensor information is used to reconstruct and fill these missing segments. This algorithm utilizes valid data from the same sensor before and after the missing period (temporal correlation) and valid data from other types of sensors at the same time (spatial or parameter correlation) for joint estimation to fill the missing values. Data points in the alarm environmental perception dataset whose deviation exceeds a preset fluctuation threshold are identified as noise points and removed. The preset fluctuation threshold is an objective numerical benchmark that can be pre-set based on sensor characteristics and the specific monitoring environment or obtained through historical data analysis. By statistically analyzing data representing normal conditions in the alarm environmental perception dataset, the normal fluctuation range of each sensor data sequence can be determined, and a threshold for judging whether a data point deviates abnormally can be set accordingly. This threshold serves as an objective judgment standard for subsequent data cleaning, distinguishing valid data from noise. By using statistical methods such as range determination based on standard deviation or interquartile range, or by identifying outliers using the isolated forest algorithm, identified noise points are removed from the data sequence. The dataset after noise removal is then processed using a sliding window smoothing algorithm to obtain a cleaned and standardized alarm environmental perception dataset. The sliding window smoothing algorithm takes a fixed-size window centered on each data point, replacing the original data point with the statistics of all valid data within the window, which are used for subsequent information fusion and feature extraction. It can be understood that the data cleaning step improves the quality of the alarm environmental perception dataset. In some embodiments, the size of the sliding window is determined based on the data sampling frequency and the expected noise frequency to be filtered out. Optionally, the standardization process also includes scaling the data values ​​from different sensors to the same numerical range to eliminate the influence of units.

[0041] See Figure 4This is a time-series diagram of state-linkage feature extraction, fully presenting the dynamic changes of multi-source sensor features. The vibration intensity feature shows significant peaks in the 6th and 9th sampling periods, corresponding to high vibration intensity scenarios; it rapidly declines in the 7th period, corresponding to the disappearance of vibration. This directly reflects whether the alarm has been subjected to real triggering sources such as physical impact, shaking, or knocking, and is one of the core indicators for judging alarm triggering. The overall fluctuation amplitude of the environmental sound pattern energy entropy feature is smaller than the other two types of features, and its trend is highly synchronized with the vibration intensity feature, verifying the core logic of "state linkage." It is used to assist in verifying the authenticity of alarm triggering, distinguishing between "real knocking sounds" and "ambient noise," and effectively reducing the false alarm rate. The ambient light sudden change frequency feature has the largest fluctuation amplitude and is a highly sensitive indicator of alarm state changes. It reaches its peak synchronously with the vibration intensity feature in the 6th and 9th periods, forming a high-risk linkage signal. It identifies environmental changes that may interfere with the sensor, such as obstruction, strong light illumination, and rapid switching of lights, and is used to supplement the judgment of the authenticity of the triggering scenario.

[0042] In one embodiment of the present invention, multi-dimensional sensor information collected by a multi-source sensor node network during the new monitoring phase is periodically acquired. The newly acquired multi-dimensional sensor information is processed to extract new state linkage features and update the state linkage feature set. Based on the updated state linkage feature set, graph nodes are added or deleted in the alarm behavior evolution graph, and the weights of the edges between graph nodes are recalculated. Based on the recalculated edge weights, the connection structure of the alarm behavior evolution graph is adjusted to generate an optimized alarm behavior evolution graph, which is used to update the alarm situation assessment model.

[0043] Record the real-time alarm risk correlation, the predicted probability value of the alarm trigger probability assessment model, and the final alarm control command executed for each generated alarm. Record the actual alarm status and environmental changes fed back by the multi-source sensor node network after the alarm control command is executed. Associate the real-time alarm risk correlation, predicted probability value, alarm control command, and the feedback actual status information, storing this as a control strategy case in the knowledge base. In the alarm control strategy knowledge base, classify, summarize, and extract rules from the control strategy cases to form control strategy rules, which are used to assist in generating or modifying alarm control commands.

[0044] In practice, the alarm behavior evolution graph is dynamically optimized. Multi-dimensional sensor information collected by the multi-source sensor node network during newly added monitoring phases is periodically acquired. A newly added monitoring phase refers to a continuous monitoring period set after the initial preset monitoring phase. The newly added multi-dimensional sensor information is processed in the same way as when generating the initial alarm environmental perception dataset, including timestamp alignment, data cleaning, and feature extraction. New state-linkage features are extracted, and the state-linkage feature set is updated by appending the new feature data to the existing state-linkage feature set data sequence. Based on the updated state-linkage feature set, graph nodes are added or deleted in the alarm behavior evolution graph. When a new feature pattern not covered by the original features appears in the data of a newly added monitoring phase, a new graph node is created for it. If certain original features no longer appear over a long period or have extremely small variance, their corresponding graph nodes are considered for deletion from the alarm behavior evolution graph. The weights of the edges between graph nodes are recalculated based on the updated state linkage feature set. The same method used to construct the initial alarm behavior evolution graph is employed to calculate the temporal correlation and statistical dependency between all feature parameter pairs to update the weights. Based on the recalculated edge weights, the connection structure of the alarm behavior evolution graph is adjusted to generate an optimized alarm behavior evolution graph, which is used to update the alarm situation assessment model. Adjustments to the connection structure include removing edges with weights below a specific threshold and adding new high-weight connections. The threshold for updating edge weights between graph nodes is also specified. The calculation formula can be expressed as: ; in: This represents the edge weight update threshold, used to determine whether to keep or remove an edge. This represents the average of all recalculated edge weights. This represents the standard deviation of all edge weights. It is a preset positive coefficient used to adjust the strictness of the threshold, with a weight lower than... The edges will be removed from the optimized alarm behavior evolution graph. This formula achieves dynamic sparsity reduction of the graph connections. In some embodiments, the dynamic optimization period can be set to be performed weekly or monthly. Optionally, a preset positive coefficient... The settings can be adjusted based on the stability of the alarm deployment environment; a smaller setting should be used when the environment changes frequently. Values ​​are set to retain more connections.

[0045] In practical implementation, an alarm control strategy knowledge base is established to record the real-time alarm risk correlation, the predicted probability value of the alarm trigger probability assessment model, and the final executed alarm control command for each generated alarm. After executing the alarm control command, the actual state of the alarm and environmental changes fed back by the multi-source sensor node network are recorded. The actual state information fed back includes whether the sensors detected a real alarm event and whether there were significant changes in environmental characteristics within a certain period after the command execution. The real-time alarm risk correlation, predicted probability value, alarm control command, and actual state information fed back are associated and stored in the knowledge base as a control strategy case. The association is based on a timestamp to ensure that data from the same decision-making cycle are correctly bound. In some embodiments, control strategy cases are stored in the form of structured data records, with each case containing fields such as timestamp, input data, decision output, and execution feedback. In the alarm control strategy knowledge base, control strategy cases are classified, summarized, and rules extracted to form control strategy rules. Classification can be based on the type of alarm control command or the final actual state feedback; summarization involves identifying common conditions from a large number of similar cases; and rule extraction formalizes these conditions into "if-then" logical statements.

[0046] See Figure 5 This is a time-series graph showing the risk assessment and trigger probability of an alarm system, fully illustrating the entire logic from risk calculation to probability prediction and threshold determination. The risk correlation peaks above 0.8 in the 2nd and 9th hours, corresponding to high-risk periods; extreme low values ​​below 0 appear in the 3rd, 6th, and 14th hours, corresponding to completely risk-free silent periods. Higher values ​​indicate that the current alarm state is closer to historical trigger scenarios, and the higher the alarm risk. The trigger probability is highly synchronized with the overall trend of the risk correlation, but with smoother fluctuations, reflecting the model's smoothing and filtering effect on risk, avoiding misjudgments caused by instantaneous noise. The probability of the alarm system triggering abnormally within a preset time window is the core basis for generating control commands. The fluctuation trends of the two curves are highly synchronized; the peak risk correlation during high-risk periods directly drives a synchronous increase in trigger probability, forming a high-probability trigger signal. During low-risk periods, the risk correlation drops sharply, and the trigger probability synchronously falls below 0, triggering silent control.

[0047] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for intelligent control of an alarm based on the Internet of Things, characterized by, include: A multi-source sensor node network for public security equipment alarms is established in an Internet of Things environment, and multi-dimensional sensor information of the alarms is collected through the multi-source sensor node network during a preset monitoring phase to generate an alarm environmental perception dataset. Information fusion and feature extraction processing are performed on the alarm environmental perception dataset to extract a state linkage feature set that reflects the linkage between the alarm state and the environment, and an alarm behavior evolution map is constructed based on the state linkage feature set. Based on the alarm behavior evolution map, an alarm situation assessment model is established, the situation correlation density of historical triggering events of the alarm in the alarm behavior evolution map is calculated, and the alarm status warning parameter set is determined according to the situation correlation density. Based on the real-time sensing information collected by the multi-source sensor node network, the real-time status linkage features of the alarm are extracted to form the real-time situation vector of the alarm. The real-time situation vector of the alarm is then imported into the alarm situation assessment model, and the situation similarity is calculated with the alarm status warning parameter set to output the real-time alarm risk correlation. By combining the real-time alarm risk correlation, the state linkage feature set, and the alarm behavior evolution map, an alarm trigger probability assessment model is generated to predict the abnormal trigger probability of the alarm, and alarm control commands are generated based on the prediction results. 2.The IoT-based alarm intelligent control method according to claim 1, characterized in that, The establishment of a multi-source sensor node network for public security equipment alarms in an Internet of Things (IoT) environment, and the collection of multi-dimensional sensor information from the alarms through the multi-source sensor node network during a preset monitoring phase to generate an alarm environmental perception dataset, specifically involves: Deploy multi-source sensor nodes, including vibration sensors, acoustic sensors, ambient light sensors, and positioning beacons, on public security equipment alarm devices; At least one preset monitoring phase is set, and the preset monitoring phase includes at least two consecutive data sampling periods; The multi-source sensing nodes are controlled to collect data sampling cycles during the preset monitoring phase, synchronously collecting vibration intensity of the alarm body, ambient background sound patterns, changes in light intensity, and alarm location information to generate multi-dimensional sensing information. The collected multidimensional sensor information is timestamped and packaged to form an alarm environmental perception dataset containing vibration intensity sequence, background acoustic waveform sequence, light intensity sequence and location information sequence. 3.The IoT-based alarm intelligent control method according to claim 2, characterized in that, Information fusion and feature extraction processing are performed on the alarm environmental perception dataset to extract a state linkage feature set reflecting the interaction between the alarm state and the environment. Based on this state linkage feature set, an alarm behavior evolution map is constructed, specifically as follows: Time-domain and frequency-domain analyses were performed on the vibration intensity sequence, background acoustic waveform sequence, light intensity sequence, and location information sequence in the environmental perception dataset of the alarm device. From the time-domain and frequency-domain analysis results, the vibration intensity characteristics of the alarm body, the energy entropy characteristics of the environmental sound pattern, the frequency characteristics of the sudden change in the ambient light, and the location movement trajectory characteristics are extracted to form a state linkage feature set; Different feature parameters in the state linkage feature set are used as graph nodes, and the temporal correlation and statistical dependence between feature parameters are used as edges between nodes. The edges are assigned corresponding weight values ​​to construct an alarm behavior evolution graph in the form of a directed weighted graph. The alarm behavior evolution map reflects the dynamic correlation and evolution relationship between alarm features under different states. 4.The IoT-based alarm intelligent control method according to claim 3, characterized in that, Based on the alarm behavior evolution map, an alarm situation assessment model is established. The situation correlation density of historical triggering events of the alarm in the alarm behavior evolution map is calculated. The alarm status warning parameter set is determined according to the situation correlation density, specifically: Using the alarm behavior evolution map as the underlying graph structure, an alarm situation assessment model is established. The input of the alarm situation assessment model is the state linkage feature set, and the output of the alarm situation assessment model is the alarm situation risk value. In the alarm behavior evolution graph, the positions of the corresponding graph nodes when the alarm's historical triggering events occurred are marked; The clustering degree of the graph node set corresponding to all historical triggering events in the alarm behavior evolution graph is calculated to obtain the situation correlation density; Set a situation correlation density threshold and filter out the set of graph nodes whose situation correlation density exceeds the threshold; Extract the state linkage feature parameters corresponding to the graph node set to form the alarm state warning parameter set. 5.The IoT-based alarm intelligent control method according to claim 4, characterized in that, Based on the real-time sensing information collected by the multi-source sensor node network, the real-time status linkage features of the alarm are extracted to form a real-time alarm situation vector. This real-time alarm situation vector is then imported into the alarm situation assessment model, and a situation similarity calculation is performed with the alarm status warning parameter set to output the real-time alarm risk correlation. Specifically: The sensor information stream of the alarm is collected in real time through the multi-source sensor node network; The sensor information stream is preprocessed and features are extracted to obtain real-time alarm body vibration intensity features, environmental sound pattern energy entropy features, ambient light sudden change frequency features, and position movement trajectory features, which are combined to form the alarm real-time situation vector. Input the real-time situation vector of the alarm device into the alarm situation assessment model; In the alarm situation assessment model, the graph theory distance between the graph node position corresponding to the real-time situation vector of the alarm and the graph node position corresponding to each warning parameter in the alarm status warning parameter set is calculated, and the situation similarity is calculated based on the graph theory distance. The quantification result of the situational similarity is output as the real-time alarm risk correlation degree. 6.The IoT-based alarm intelligent control method according to claim 5, characterized in that, Combining the real-time alarm risk correlation, the state linkage feature set, and the alarm behavior evolution map, an alarm trigger probability assessment model is generated to predict the abnormal trigger probability of the alarm. Based on the prediction results, alarm control commands are generated, specifically: Using the real-time alarm risk correlation degree, the state linkage feature set, and the graph structure features of the alarm behavior evolution map as inputs, an alarm trigger probability evaluation model is constructed. Using the alarm trigger probability evaluation model, the evolution path of the alarm's real-time situation vector in the graph structure is analyzed, and combined with the distribution pattern of historical trigger events in the graph, the probability value of the alarm being abnormally triggered within a future preset time window is calculated and generated. The probability value is compared with a preset alarm trigger threshold and a false alarm suppression threshold; Based on the comparison results, generate alarm control commands that include alarm trigger enable, alarm sensitivity adjustment, alarm silence control, or alarm signal reporting. 7.The IoT-based alarm intelligent control method according to claim 6, wherein, It also includes steps for data augmentation and anomaly cleaning of the alarm environmental perception dataset, specifically: Detect missing segments of sensor information in the environmental perception dataset of the alarm device; An interpolation algorithm based on the spatiotemporal correlation of sensor information is used to reconstruct and fill in the missing segments of the sensor information. Data points in the environmental perception dataset of the alarm device that deviate from a preset fluctuation threshold are identified as noise points and then removed. After removing noise points, the dataset is processed using a sliding window smoothing algorithm to obtain a cleaned and standardized alarm environmental perception dataset, which is then used for subsequent information fusion and feature extraction. 8.The IoT-based alarm intelligent control method according to claim 7, wherein, It also includes dynamically optimizing the behavior evolution map of the alarm, specifically: Periodically acquire multi-dimensional sensor information collected by the multi-source sensor node network during the new monitoring phase; The newly added multidimensional sensing information is processed to extract the newly added state linkage features and update the state linkage feature set. Based on the updated state linkage feature set, add or delete graph nodes in the alarm behavior evolution graph, and recalculate the weights of the edges between graph nodes; Based on the recalculated edge weights, the connection structure of the alarm behavior evolution graph is adjusted to generate an optimized alarm behavior evolution graph, which is used to update the alarm situation assessment model. 9.The IoT-based alarm intelligent control method according to claim 8, wherein, This also includes establishing a knowledge base for alarm control strategies, specifically: Record the real-time alarm risk correlation degree generated each time, the predicted probability value of the alarm trigger probability evaluation model, and the final alarm control command executed; Record the actual status of the alarm and environmental changes fed back by the multi-source sensor node network after the alarm control command is executed; The real-time alarm risk correlation, predicted probability value, alarm control command and the actual status information fed back are associated and stored in the knowledge base as a control strategy case. In the alarm control strategy knowledge base, control strategy cases are classified, summarized, and rules are extracted to form control strategy rules, which are used to assist in generating or modifying the alarm control commands.

10. An alarm intelligent control system based on Internet of Things, comprising a memory, a processor and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the IoT-based intelligent control method for alarms as described in any one of claims 1 to 9.

Images