An Abnormal Detection System and Method for Internet-based Intelligent Fire Extinguishing and Water Intake Hydrants

Abstract

The present invention relates to the technical field of intelligent fire-fighting equipment, and discloses an abnormal detection system and method for an intelligent fire-extinguishing water intake hydrant based on networking. The system includes a multi-source perception module that collects water pressure, flow rate, and environmental state data; an abnormal feature analysis module that generates abnormal feature vectors based on an improved support vector machine; a dynamic state compensation module that outputs state compensation parameters using a wavelet transform algorithm; and a networking collaborative decision-making module that constructs a multi-node optimization model to generate global collaborative control instructions. The multi-source perception module fuses various sensor data and outputs comprehensive detection features through feature fusion and a graph convolutional network. The improved support vector machine integrates a particle swarm optimization algorithm to improve detection accuracy, and the wavelet transform algorithm combines with a Bayesian network to achieve dynamic modeling compensation. The multi-node optimization model adopts a distributed collaborative architecture to achieve collaborative control. The system and method can accurately detect abnormalities, improve the efficiency of the fire-fighting water supply system, and ensure fire safety.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent fire-fighting equipment, and particularly to an abnormal detection system and method for an intelligent fire-extinguishing and water-taking hydrant based on networking. Background Art

[0002] In the field of fire-fighting infrastructure, the fire-extinguishing and water-taking hydrant, as a key fire-fighting water supply device, the reliability of its operating state plays a crucial role in fire safety. However, there are many problems with traditional fire-extinguishing and water-taking hydrants, which seriously affect their effectiveness and reliability in fire-fighting work.

[0003] In terms of detection means, early fire-extinguishing and water-taking hydrants lacked an effective real-time monitoring mechanism. In the past, it relied mainly on manual regular inspections. This method was not only inefficient but also difficult to detect and handle abnormal situations in a timely manner. There were often time intervals in manual inspections. Once an abnormality occurred during the interval, such as abnormal water pressure caused by a pipeline rupture or a change in flow rate due to a damaged valve, it was difficult to detect in a timely manner, which was extremely likely to delay the best opportunity for fire fighting. Even during the inspection process, manual judgment was subjective and limited, and some subtle abnormal signs might not be accurately identified, thus leaving potential safety hazards.

[0004] With the development of technology, although some simple monitoring devices have emerged, their functions are single, and they can only monitor a single parameter such as water pressure or flow rate, and cannot comprehensively consider the influence of various factors on the operating state of the fire-extinguishing and water-taking hydrant. For example, only monitoring water pressure, when the flow rate changes abnormally, an alarm cannot be issued in a timely manner, resulting in an inability to comprehensively and accurately evaluate the actual operating condition of the fire-extinguishing and water-taking hydrant. Moreover, these simple devices are independent of each other and have no networking function, so data sharing and collaborative processing cannot be achieved, and it is difficult to meet the requirements of unified management of large-scale fire-fighting facilities.

[0005] At the level of abnormal detection technology, the accuracy of traditional detection algorithms is poor. Early algorithms were mostly based on simple threshold judgments. When environmental factors interfere or the equipment has complex failures, false alarms and missed alarms occur frequently. For example, when environmental factors such as temperature and humidity change greatly, the water pressure and flow rate data will be affected to a certain extent, and simple threshold judgments may misjudge normal fluctuations as abnormalities, bringing unnecessary troubles to fire-fighting management work; for some gradually developing hidden faults, they may be missed because they do not reach the set threshold, and potential dangers cannot be detected in a timely manner.

[0006] In addition, in large-scale fire protection systems, there is a lack of an effective coordination mechanism among fire hydrants. When a fire breaks out in one place and a large amount of water is needed, it is impossible to rationally allocate water according to the status of surrounding fire hydrants, resulting in insufficient water supply at some fire hydrants while some are idle, causing waste of resources and low utilization efficiency. Moreover, due to the lack of global coordinated control, when dealing with multiple fire points or complex fire scenarios, the entire fire water supply system is difficult to respond effectively and quickly, seriously affecting the fire extinguishing effect. Summary of the Invention

[0007] The purpose of the present invention is to provide an intelligent fire hydrant anomaly detection system and method based on networking to solve the problems raised in the above background technology.

[0008] To achieve the above purpose, the present invention provides the following technical solutions: An intelligent fire hydrant anomaly detection system based on networking, the system includes:

[0009] Multi-source perception module: used to collect water pressure data, flow data and environmental status data of fire hydrants through a variety of sensors;

[0010] Abnormal feature analysis module: based on an improved support vector machine, perform pattern recognition on the water pressure data and flow data to generate abnormal feature vectors;

[0011] Dynamic state compensation module: adopt a wavelet transform algorithm to perform real-time modeling on the mechanical vibration and temperature drift of fire hydrants, and output state compensation parameters;

[0012] Networked collaborative decision-making module: according to the abnormal feature vectors and state compensation parameters, construct a multi-node optimization model and generate global collaborative control instructions.

[0013] Preferably, the multi-source perception module includes:

[0014] A variety of sensors include pressure transmitters, electromagnetic flowmeters, temperature and humidity sensors, and vibration accelerometers;

[0015] Perform time-domain synchronization processing on the data of pressure transmitters and electromagnetic flowmeters to construct a working condition state model of fire hydrants; perform wavelet packet decomposition on the data of temperature and humidity sensors and vibration accelerometers to extract environmental interference characteristics and mechanical vibration spectra;

[0016] Construct a two-channel feature fusion network. The first channel uses a one-dimensional convolutional neural network to extract the time-domain fluctuation characteristics of the working condition state model, and the second channel uses a frequency-domain attention mechanism to extract the frequency-domain energy distribution characteristics of the mechanical vibration spectrum;

[0017] The time domain fluctuation features and frequency domain energy distribution features are fused through a cross-domain feature alignment strategy to generate a joint feature matrix. The joint feature matrix is modeled with node associations based on a graph convolutional network to output comprehensive detection features including water pressure anomalies, flow mutations, and mechanical wear status.

[0018] Preferably, the improved support vector machine integrated particle swarm optimization algorithm includes:

[0019] Anomaly detection is modeled as a multi-classification problem, where the decision variables include kernel function type, penalty coefficient, and relaxation factor.

[0020] Initialize the support vector machine parameters and calculate the classification hyperplane. The objective function includes maximizing the classification interval, the weight of misclassified samples, and the feature space separability index.

[0021] In the parameter optimization stage, the particle swarm search step size is dynamically adjusted according to the environmental interference characteristics; in the model training stage, the cross-validation strategy is used to screen the optimal kernel function combination;

[0022] The Pareto optimal solution set is introduced to iteratively screen the multi-objective optimization results, and finally the abnormal feature vector is output.

[0023] Preferably, the wavelet transform algorithm is combined with a Bayesian network, including:

[0024] Construct a vibration-temperature coupled state equation, taking the mechanical vibration amplitude, temperature gradient and water pressure fluctuation as observation variables;

[0025] Design a wavelet basis function library to map observed variables into multi-scale wavelet coefficients;

[0026] Dynamically update state transition probabilities through online Bayesian inference and optimize the number of wavelet decomposition layers using maximum a posteriori estimation.

[0027] The output includes state compensation parameters such as vibration suppression coefficient and temperature compensation.

[0028] Preferably, the multi-node optimization model adopts a distributed collaborative architecture, including:

[0029] Construct a node communication topology graph, where nodes represent the control units of fire hydrants and edges represent the data transmission delay constraints between nodes.

[0030] The alternating direction multiplier method is used to decompose the global optimization problem into local sub-problems. Each sub-problem contains a node response time limit, an energy consumption balance term, and a data consistency penalty term.

[0031] Introducing virtual coordination variables at the decision-making layer to balance the conflicts in control instructions of each node, and achieving collaborative iterative solutions through a distributed consensus algorithm;

[0032] Finally, a global collaborative control instruction is generated.

[0033] Preferably, the one-dimensional convolutional neural network adopts a multi-scale residual structure, including:

[0034] The working condition state model is divided into multiple time window sequences, and each time window contains mean, variance, and range statistics;

[0035] At the convolution stage, dilated convolutional kernels with a dilation rate of 2 are used to capture long-term temporal dependence features, and in the residual connection stage, short-term temporal detail features and long-term temporal trend features are fused;

[0036] A temporal attention module is introduced to adaptively weight the feature channels;

[0037] The temporal attention module adopts a frequency-domain - time-domain joint focusing mechanism, including: calculating the energy density weight of the feature map in the frequency domain dimension, screening the dominant frequency components, constructing a sliding window model in the time domain dimension to locate the time period when abnormal events occur, and fusing the frequency-domain weight and the time-domain weight through tensor product operation.

[0038] Preferably, the particle swarm optimization algorithm is implemented based on a dynamic inertia weight mechanism, including: collecting the classification accuracy and calculation time consumption in the historical detection process as optimization factors; constructing a radial basis function network to fit the non-linear relationship between the inertia weight and the optimization factors.

[0039] Preferably, the Bayesian network adopts a two-stage inference structure, including:

[0040] In the forward inference stage, the state transition probability is estimated based on Markov chain Monte Carlo sampling;

[0041] In the backward correction stage, the wavelet basis function parameters are updated through the expectation maximization algorithm;

[0042] A hybrid inference strategy is defined, and the forward probability distribution and the backward parameter correction results are weighted and fused to output the final state compensation parameters.

[0043] Preferably, the virtual coordination variable adopts a lag compensation mechanism, including:

[0044] Measuring the execution time delay of the control instructions of each node, and constructing a time-varying lag coefficient matrix;

[0045] Introducing a lag compensation term in the consensus algorithm, and determining the compensation gain coefficient through stability analysis.

[0046] Preferably, the present invention further includes an abnormal detection method for an intelligent fire-extinguishing water intake hydrant based on networking, including the following steps:

[0047] S1. Collect the water pressure data, flow data, and environmental state data of the fire-extinguishing water intake hydrant through a multi-source sensor network;

[0048] S2. Perform pattern recognition on the water pressure data and flow data based on the improved support vector machine to generate an abnormal feature vector containing abnormal fluctuation characteristics and data correlation.

[0049] S4. Dynamically model the mechanical vibration signal and temperature drift parameter of the fire hydrant using the wavelet transform algorithm to generate real-time state compensation parameters.

[0050] S4. Construct a multi-node optimization decision model based on the abnormal feature vector and state compensation parameters to generate a global collaborative control instruction and a dynamic collaborative resource matching network.

[0051] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0052] In terms of data collection and analysis, the multi-source perception module plays a key role. Through the collaborative work of various sensors such as pressure transmitters, electromagnetic flowmeters, temperature and humidity sensors, and vibration accelerometers, not only can water pressure and flow data be collected in real time, but also environmental state data can be obtained. This makes the condition monitoring of fire hydrants more comprehensive and detailed, avoiding the limitations of single-parameter monitoring. For example, in a humid environment, the data of the temperature and humidity sensor can assist in judging whether there is a risk of equipment corrosion caused by humidity, and combined with the data of the vibration accelerometer, the wear condition of the mechanical components of the equipment can be further analyzed. Through the time-domain synchronization processing, wavelet packet decomposition of the collected data, and the application of a two-channel feature fusion network and a graph convolutional network, the generated comprehensive detection features can accurately reflect water pressure anomalies, flow mutations, and mechanical wear states, greatly improving the accuracy and reliability of detection, and effectively reducing the occurrence of false alarms and missed alarms.

[0053] The abnormal feature analysis module integrates the improved support vector machine of the particle swarm optimization algorithm, models the abnormal detection as a multi-classification problem, and through dynamically adjusting the particle swarm search step size, cross-validation to screen the optimal kernel function combination, and iterative screening of the Pareto optimal solution set, can more accurately identify abnormal patterns in water pressure and flow data. The generated abnormal feature vector contains rich abnormal information, providing a reliable basis for subsequent decision-making. This multi-objective optimization method can better adapt to complex actual application scenarios compared with traditional single-objective detection algorithms, and can stably and accurately detect abnormalities under different environmental interferences.

[0054] The dynamic state compensation module combines the wavelet transform algorithm of Bayesian network to perform real-time modeling on the mechanical vibration and temperature drift of the fire hydrant. By constructing a vibration-temperature coupling state equation, designing a wavelet basis function library, and using online Bayesian inference and maximum a posteriori estimation to optimize the wavelet decomposition layer number, the output state compensation parameters can effectively suppress mechanical vibration and compensate for the influence of temperature drift on the equipment operation state, ensure the accuracy of equipment operation data, extend the service life of the equipment, and improve the stability of equipment operation.

[0055] The networked collaborative decision-making module adopts a multi-node optimization model with a distributed collaborative architecture, constructs a node communication topology map and considers data transmission delay constraints, decomposes the optimization problem by the alternating direction method of multipliers, introduces virtual coordination variables to balance control instruction conflicts, and uses the distributed consensus algorithm to solve iteratively in a coordinated manner, achieving efficient coordination among the control units of each fire hydrant. This enables the system to quickly generate global collaborative control instructions according to the actual states and demands of each hydrant in the face of emergencies such as fires, rationally allocate water resources, avoid the situation of insufficient water supply for some hydrants while some are idle, greatly improve the response speed and water supply efficiency of the fire water supply system, effectively enhance the fire extinguishing effect, and ensure the safety of people's lives and property.

[0056] From the perspective of the overall system architecture, each module cooperates and complements each other, forming an organic whole. The multi-source perception module provides an accurate data basis for the subsequent modules, the abnormal feature analysis module and the dynamic state compensation module deeply analyze and process the data, and the networked collaborative decision-making module realizes global optimization control based on the results of the previous modules. This highly integrated and collaborative design not only improves the performance of abnormal detection of fire hydrants but also provides strong support for the intelligent management of fire protection facilities, with significant economic and social benefits. Brief Description of the Drawings

[0057] Figure 1 It is the working principle diagram of the intelligent fire hydrant abnormal detection system based on networking of the present invention;

[0058] Figure 2 It is the working principle diagram of the multi-source perception module;

[0059] Figure 3 It is the working principle diagram of the one-dimensional convolutional neural network;

[0060] Figure 4 It is the working principle diagram of the particle swarm optimization algorithm. Detailed Embodiments

[0061] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0062] Please refer to Figures 1 - 4 , the present invention provides a technical solution: an intelligent fire-extinguishing water intake hydrant anomaly detection system based on networking, the system includes: a multi-source perception module, an anomaly feature analysis module, a dynamic state compensation module, and a networking collaborative decision-making module.

[0063] During the operation of the system, the multi-source perception module uses a variety of sensors, such as pressure transmitters, electromagnetic flowmeters, temperature and humidity sensors, and vibration accelerometers, to collect the water pressure data, flow data, and environmental state data of the fire-extinguishing water intake hydrant. These sensors obtain information from different dimensions, providing a rich data basis for subsequent anomaly detection.

[0064] After collecting the data, the anomaly feature analysis module performs pattern recognition on the water pressure data and flow data based on an improved support vector machine. The improved support vector machine integrates the particle swarm optimization algorithm. Through a series of operations, such as modeling anomaly detection as a multi-classification problem, reasonably setting decision variables, initializing the parameters of the support vector machine, calculating the classification hyperplane, and performing parameter optimization and model training at different stages, an anomaly feature vector is finally generated. This vector contains important information such as the hidden anomaly fluctuations in the water pressure and flow data and the correlation between the data.

[0065] The dynamic state compensation module uses the wavelet transform algorithm to perform real-time modeling on the mechanical vibration and temperature drift of the fire-extinguishing water intake hydrant. This algorithm combines the Bayesian network. By constructing a vibration-temperature coupling state equation, designing a wavelet basis function library, and performing online Bayesian inference and other steps, state compensation parameters including vibration suppression coefficients and temperature compensation amounts are output. These parameters can effectively compensate for the errors that may be caused by mechanical vibration and temperature drift of the fire-extinguishing water intake hydrant, improving the accuracy of detection.

[0066] The networking collaborative decision-making module constructs a multi-node optimization model based on the anomaly feature vector and state compensation parameters. This model adopts a distributed collaborative architecture. By constructing a node communication topology graph, decomposing the global optimization problem into local sub-problems, and introducing virtual coordination variables at the decision-making layer and other operations, global collaborative control instructions are generated. These instructions can achieve the collaborative control of multiple fire-extinguishing water intake hydrants, ensuring that the entire system can quickly respond in case of anomalies and guaranteeing the normal supply of fire-fighting water.

[0067] The following further illustrates the present invention in conjunction with Embodiments 1 to 5:

[0068] Example 1: This example mainly focuses on the multi-source perception module. In actual application scenarios, various sensors of the multi-source perception module are installed at appropriate positions on and around the fire hydrant. The pressure transmitter is used to accurately measure the water pressure data inside the fire hydrant, the electromagnetic flowmeter is responsible for collecting the flow data, the temperature and humidity sensor monitors the temperature and humidity of the surrounding environment, and the vibration accelerometer obtains the mechanical vibration information of the fire hydrant.

[0069] After collecting the data from the pressure transmitter and the electromagnetic flowmeter, time-domain synchronization processing is required. This is because there may be differences in the data collection times of different sensors. If synchronization is not performed, the subsequent constructed working condition state model will be inaccurate. Through a specific time-domain synchronization algorithm, these two types of data are made consistent in time, and then the working condition state model of the fire hydrant is constructed. This model comprehensively reflects the changes in water pressure and flow over time and is an important basis for subsequent analysis.

[0070] For the data of the temperature and humidity sensor and the vibration accelerometer, the method of wavelet packet decomposition is used for processing. Wavelet packet decomposition can decompose the signal in detail in different frequency bands, thereby extracting the environmental interference characteristics and the mechanical vibration spectrum. The environmental interference characteristics can reflect the influence of surrounding environmental factors on the fire hydrant, while the mechanical vibration spectrum contains information about the operating state of the mechanical components of the fire hydrant itself.

[0071] To make more effective use of these features, a two-channel feature fusion network is constructed. The first channel uses a one-dimensional convolutional neural network (1D-CNN). The working condition state model is divided into multi-time window sequences, and each time window contains mean, variance, and range statistics. In the convolution stage, a dilated convolution kernel with a dilation rate of 2 is used. The dilated convolution kernel can expand the receptive field of the convolution kernel without increasing the number of parameters, capturing long-time series dependence features. In the residual connection stage, short-time series details and long-time series trend features are fused, enabling the network to better learn various features of the data. At the same time, a time series attention module is introduced to adaptively weight the feature channels. The time series attention module adopts a frequency-domain-time-domain joint focusing mechanism, calculating the energy density weights of the feature map in the frequency domain dimension , screening out the dominant frequency components; constructing a sliding window model in the time domain dimension to locate the time period when abnormal events occur, and fusing the frequency-domain weights and time-domain weights through tensor product operations , obtaining the comprehensive weights . This can highlight the features that are important for anomaly detection and improve the detection accuracy.

[0072] The second channel uses a frequency-domain attention mechanism to extract the frequency-domain energy distribution characteristics of the mechanical vibration spectrum. The frequency-domain attention mechanism analyzes the energy distribution of different frequency components in the mechanical vibration spectrum, finds the frequency bands that contribute more to anomaly detection, and assigns corresponding weights.

[0073] Finally, through the cross-domain feature alignment strategy, the time-domain fluctuation characteristics and the frequency-domain energy distribution characteristics are fused to generate a joint feature matrix. Based on the graph convolutional network, node association modeling is performed on the joint feature matrix, fully considering the mutual relationships between different features, and comprehensive detection features including water pressure anomalies, flow rate mutations, and mechanical wear states are output. These comprehensive detection features provide comprehensive and valuable information for subsequent anomaly analysis.

[0074] Embodiment 2: This embodiment focuses on the specific implementation of improving the support vector machine integrated particle swarm optimization algorithm in the anomaly feature analysis module.

[0075] Model the anomaly detection as a multi-classification problem, and the decision variables include the kernel function type , the penalty coefficient and the relaxation factor . The kernel function type determines the mapping method of the data in the feature space, and different kernel functions are suitable for different data distributions; the penalty coefficient is used to balance the maximization of the classification margin and the penalty degree of misclassified samples; the relaxation factor allows a certain degree of misclassification to adapt to the possible noise and complex distributions in the actual data.

[0076] Initialize the support vector machine parameters, including selecting appropriate initial kernel function type, penalty coefficient, and relaxation factor, and calculate the classification hyperplane. The objective function is , where, represents the second norm of the weight vector , and its role is to maximize the classification margin; is the penalty coefficient, as mentioned above, used to balance the classification margin and the penalty of misclassified samples; is the sum of the relaxation factors of all samples, is the number of samples, which reflects the tolerance for misclassified samples. The design of this objective function comprehensively considers the maximization of the classification margin, the weight of misclassified samples, and the separability index of the feature space.

[0077] In the parameter optimization stage, dynamically adjust the particle swarm search step size according to the environmental interference characteristics. Collect the classification accuracy and the calculation time during the historical detection process as optimization factors, and construct a radial basis function network to fit the non-linear relationship between the inertia weight and the optimization factors. Assume that the output of the radial basis function network is ,in, is the number of radial basis functions, It is The weights of the radial basis functions, is the input vector, and the classification accuracy is and computation time composition, It is The center of the radial basis function, It is In this way, the search step size of the particle swarm is dynamically adjusted according to different environmental interference conditions, thereby improving the search efficiency and accuracy of the algorithm.

[0078] During the model training phase, a cross-validation strategy is used to screen the optimal kernel function combination. The dataset is divided into multiple subsets, and training and validation are performed on different subsets. The model performance under different kernel function combinations is compared, and the kernel function combination with the best performance is selected.

[0079] Finally, the Pareto optimal solution set is introduced to iteratively screen the multi-objective optimization results. This Pareto optimal solution set contains a set of solutions that strike a balance between multiple objectives. Through continuous iterative screening, an anomaly feature vector is ultimately output. This anomaly feature vector comprehensively considers multiple factors and can more accurately reflect anomalies in the water pressure and flow data.

[0080] Example 3: This example describes in detail the wavelet transform algorithm combined with the Bayesian network in the dynamic state compensation module.

[0081] First, the vibration-temperature coupling state equation is constructed, and the mechanical vibration amplitude is , temperature gradient and water pressure fluctuations As observation variables. These observation variables can fully reflect the changes in the operating state of the fire hydrant. The vibration-temperature coupling state equation can be expressed in the form of a state space model, such as ,in, yes The state vector at the moment contains information such as the mechanical vibration amplitude, temperature change gradient, and water pressure fluctuation; is the state transition matrix, which describes the relationship between the state and time; is the control input matrix (in this scenario, it may be related to factors such as external interference); yes Control input at each moment; yes The process noise at the moment.

[0082] Design a wavelet basis function library to map the observed variables into multi-scale wavelet coefficients. The selection of wavelet basis functions is crucial, as different wavelet basis functions have different decomposition effects on signals. According to the characteristics of the fire hydrant data, select appropriate wavelet basis functions, such as Daubechies wavelets, etc. Through wavelet transform, decompose the observed variables at different scales to obtain multi-scale wavelet coefficients, which contain information about the signal at different frequencies and time scales.

[0083] During the processing, dynamically update the state transition probability through online Bayesian inference. The Bayesian network adopts a two-stage inference structure. In the forward inference stage, estimate the state transition probability based on Markov chain Monte Carlo sampling. Assume the state transition probability is , and Markov chain Monte Carlo sampling constructs a Markov chain whose stationary distribution is the target distribution . By sampling on the Markov chain, obtain the estimated value of the state transition probability.

[0084] In the backward correction stage, update the wavelet basis function parameters through the expectation-maximization algorithm. The expectation-maximization algorithm is an iterative algorithm, which is divided into an expectation step and a maximization step. In the expectation step, calculate the expected value of the log-likelihood function under the current parameter estimation; in the maximization step, find the parameter value that maximizes the expected value, and iterate continuously until the parameters converge.

[0085] Define a hybrid inference strategy, and fuse the forward probability distribution and the backward parameter correction result by weighting to output the final state compensation parameter. Let the forward probability distribution be , the backward parameter correction result be , and the weights be and , then the final state compensation parameter . In this way, comprehensively consider the forward and backward information, improve the accuracy of the state compensation parameter, output the state compensation parameter including the vibration suppression coefficient and the temperature compensation amount, and achieve effective compensation for the mechanical vibration and temperature drift of the fire hydrant.

[0086] Example 4: This example focuses on the distributed collaborative architecture adopted by the multi-node optimization model in the networked collaborative decision-making module.

[0087] Construct a node communication topology graph. The nodes represent the control units of each fire hydrant, and each node is responsible for collecting and processing the relevant data of the fire hydrant where it is located. The edges represent the data transmission delay constraints between nodes. The data transmission delay is an important factor affecting the collaborative performance of the system. Assume that the data transmission delay between node and node is , and this delay may be affected by various factors such as network bandwidth and transmission distance.

[0088] The global optimization problem is decomposed into local sub - problems by the alternating direction method of multipliers. Each sub - problem includes the node response time limit, the energy consumption balance term, and the data consistency penalty term. Taking the sub - problem of node as an example, its objective function can be expressed as , where is the part of the objective function related to node itself, for example, related to the anomaly detection accuracy of the fire hydrant controlled by this node; is the term related to data consistency, represents the local variable of node ; is the penalty parameter used to adjust the intensity of the data consistency penalty; is the set of neighbor nodes of node ; is the decision variable of node , is the dual variable between node and node . The node response time limit ensures that each node can respond to abnormal situations within the specified time, the energy consumption balance term ensures that the energy consumption of each node is relatively balanced, avoiding some nodes consuming excessive energy, and the data consistency penalty term ensures the data consistency between different nodes.

[0089] At the decision - making level, a virtual coordination variable is introduced to balance the conflicts of control instructions of each node. The virtual coordination variable adopts a lag compensation mechanism to measure the execution delay of each node's control instruction and constructs a time - varying lag coefficient matrix . Suppose is the time - varying lag coefficient from node to node , which changes with time and reflects the difference in the execution delay of control instructions between different nodes. A lag compensation term is introduced in the consensus algorithm, and the compensation gain coefficient is determined through stability analysis. Stability analysis is usually based on methods such as Lyapunov stability theory to ensure that the system still operates stably after introducing the lag compensation term. Through the distributed consensus algorithm, collaborative iterative solution is achieved. Each node continuously exchanges information and adjusts its decision variables, and finally generates global collaborative control instructions. These instructions can coordinate the work of each fire hydrant and achieve the efficient operation of the entire system.

[0090] Example 5: This example further illustrates the comprehensive application and advantages of the system from multiple aspects. In a large - scale fire - fighting water supply network, multiple fire hydrants equipped with the intelligent fire hydrant anomaly detection system of the present invention are deployed.

[0091] The multi-source perception module continuously collects the water pressure data, flow rate data, temperature and humidity data, and vibration data of each fire hydrant. These data are transmitted to the subsequent processing module in real time through the network. For example, within a certain period of time, the environmental temperature near one of the fire hydrants changes significantly, and the temperature and humidity sensor promptly captures this information and transmits it together with other data.

[0092] The abnormal feature analysis module deeply analyzes the collected water pressure and flow rate data. By improving the support vector machine integrated particle swarm optimization algorithm, it accurately identifies the abnormal fluctuation features in the data. Suppose at a certain moment, the water pressure of this fire hydrant drops abnormally and the flow rate also shows a sudden change. The abnormal feature analysis module quickly generates an abnormal feature vector containing this abnormal information.

[0093] The dynamic state compensation module performs real-time modeling for mechanical vibration and temperature drift. Since temperature changes may affect the performance of the internal components of the fire hydrant, the wavelet transform algorithm combined with the Bayesian network comes into play. It processes the observed variables such as the mechanical vibration amplitude, temperature change gradient, and water pressure fluctuation amount collected according to the vibration-temperature coupling state equation. For example, through online Bayesian inference, it timely adjusts the state transition probability and accurately outputs state compensation parameters such as vibration suppression coefficients and temperature compensation amounts to compensate for the errors that may be caused by temperature and vibration.

[0094] The networked collaborative decision-making module constructs a multi-node optimization model based on the abnormal feature vector and the state compensation parameter. In this large-scale fire water supply network, each fire hydrant control unit serves as a node and is interconnected through a node communication topology diagram. When an abnormality occurs in a certain fire hydrant, the multi-node optimization model adopts a distributed collaborative architecture for processing. By using the alternating direction method of multipliers to decompose the global optimization problem, each node calculates according to its own sub-problem. At the same time, a virtual coordination variable is introduced to balance the conflicts of control instructions among nodes. Suppose there may be conflicts in the control instructions of different nodes during the process of adjusting the water pressure. The virtual coordination variable is adjusted according to the time-varying lag coefficient matrix and the compensation gain coefficient through a lag compensation mechanism, and finally, a collaborative iterative solution is achieved through a distributed consensus algorithm to generate global collaborative control instructions. These instructions can coordinate the work of each fire hydrant to ensure that the entire fire water supply network can still operate stably and efficiently in case of abnormalities and guarantee the normal supply of fire water. It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0095] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made in these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An abnormal detection system for intelligent fire - fighting water intake hydrants based on networking, characterized in that, Including: Multi-source perception module: used to collect the water pressure data, flow data and environmental status data of the fire hydrant through multiple sensors; Abnormal feature analysis module: based on the improved support vector machine, perform pattern recognition on the water pressure data and flow data to generate abnormal feature vectors; Dynamic state compensation module: adopt the wavelet transform algorithm to perform real-time modeling on the mechanical vibration and temperature drift of the fire hydrant, and output state compensation parameters; Networked collaborative decision-making module: construct a multi-node optimization model according to the abnormal feature vectors and state compensation parameters, and generate global collaborative control instructions; The wavelet transform algorithm combined with the Bayesian network includes: Construct a vibration-temperature coupling state equation, and use the mechanical vibration amplitude, temperature change gradient and water pressure fluctuation as observation variables; Design a wavelet basis function library to map the observation variables into multi-scale wavelet coefficients; Dynamically update the state transition probability through online Bayesian inference, and optimize the wavelet decomposition layer number by using the maximum a posteriori estimation; Output state compensation parameters including vibration suppression coefficients and temperature compensation amounts; The multi-source perception module includes: Multiple sensors include pressure transmitters, electromagnetic flow meters, temperature and humidity sensors, and vibration accelerometers; Perform time-domain synchronization processing on the pressure transmitter data and electromagnetic flow meter data to construct a working condition state model of the fire hydrant; perform wavelet packet decomposition on the temperature and humidity sensor data and vibration accelerometer data to extract environmental interference features and mechanical vibration spectra; Construct a dual-channel feature fusion network. The first channel uses a one-dimensional convolutional neural network to extract the time-domain fluctuation features of the working condition state model, and the second channel uses a frequency-domain attention mechanism to extract the frequency-domain energy distribution features of the mechanical vibration spectrum; Fuse the time-domain fluctuation features and frequency-domain energy distribution features through a cross-domain feature alignment strategy to generate a joint feature matrix; perform node association modeling on the joint feature matrix based on a graph convolutional network, and output comprehensive detection features including water pressure anomalies, flow rate mutations, and mechanical wear states; The one-dimensional convolutional neural network adopts a multi-scale residual structure, including: Divide the working condition state model into multi-time window sequences, and each time window contains mean, variance, and range statistics; In the convolution stage, use a dilated convolution kernel with a dilation rate of 2 to capture long-time sequence dependence features, and in the residual connection stage, fuse short-time sequence details and long-time sequence trend features; Introduce a time-series attention module to adaptively weight the feature channels; The time-series attention module adopts a frequency-domain-time-domain joint focusing mechanism, including: calculating the energy density weight of the feature map in the frequency domain dimension, screening the dominant frequency components, constructing a sliding window model in the time domain dimension to locate the time period when abnormal events occur, and fusing the frequency domain weight and time domain weight through tensor product operations.

2. The abnormal detection system for an intelligent fire-extinguishing water intake hydrant based on networking according to claim 1, characterized in that, The improved support vector machine integrates the particle swarm optimization algorithm, including: Model the anomaly detection as a multi-classification problem, and the decision variables include the kernel function type, penalty coefficient, and relaxation factor; Initialize the support vector machine parameters and calculate the classification hyperplane. The objective function includes maximizing the classification margin, the weight of misclassified samples, and the separability index of the feature space; In the parameter optimization stage, the particle swarm search step size is dynamically adjusted according to the environmental interference characteristics; in the model training stage, a cross-validation strategy is adopted to screen the optimal kernel function combination; The Pareto optimal solution set is introduced to iteratively screen the multi-objective optimization results, and finally an abnormal feature vector is output.

3. The abnormal detection system for intelligent fire extinguishing water intake hydrants based on networking according to claim 1, wherein, The multi-node optimization model adopts a distributed cooperative architecture, including: Construct a node communication topology graph, where nodes represent each fire hydrant control unit, and edges represent data transmission delay constraints between nodes; The alternating direction multiplier method is used to decompose the global optimization problem into local sub-problems. Each sub-problem includes node response time limits, energy consumption balance terms, and data consistency penalty terms; A virtual coordination variable is introduced at the decision-making layer to balance the control instruction conflicts of each node, and a distributed consensus algorithm is used to achieve collaborative iterative solution; Finally, a global collaborative control instruction is generated.

4. The abnormal detection system for intelligent fire-extinguishing water intake hydrants based on networking according to claim 2, wherein The particle swarm optimization algorithm is implemented based on a dynamic inertia weight mechanism, including: collecting the classification accuracy and calculation time-consuming in the historical detection process as optimization factors; constructing a radial basis function network to fit the non-linear relationship between the inertia weight and the optimization factors.

5. The abnormal detection system for intelligent fire extinguishing and water intake hydrants based on networking according to claim 1, characterized in that, The Bayesian network adopts a two-stage inference structure, including: In the forward inference stage, the state transition probability is estimated based on Markov chain Monte Carlo sampling; In the backward correction stage, the wavelet basis function parameters are updated through the expectation maximization algorithm; Define a hybrid inference strategy, and the forward probability distribution and the backward parameter correction results are weighted and fused to output the final state compensation parameters.

6. The abnormal detection system for intelligent fire-extinguishing water intake hydrants based on networking according to claim 3, wherein, The virtual coordination variable adopts a lag compensation mechanism, including: Measure the execution time delay of each node's control instruction, and construct a time-varying lag coefficient matrix; Introduce a lag compensation term in the consensus algorithm, and determine the compensation gain coefficient through stability analysis.

7. An abnormal detection method for an intelligent fire - fighting water intake hydrant based on networking, characterized in that, Including the following steps: S1. Collect the water pressure data, flow data, and environmental state data of the fire hydrant through a multi-source sensor network; S2. Based on the improved support vector machine, perform pattern recognition on the water pressure data and flow data to generate an abnormal feature vector containing abnormal fluctuation characteristics and data correlation; S3. Adopt the wavelet transform algorithm to dynamically model the mechanical vibration signal and temperature drift parameter of the fire hydrant, and generate real-time state compensation parameters; S4. According to the abnormal feature vector and state compensation parameters, construct a multi-node optimization decision model to generate a global collaborative control instruction and a dynamic collaborative resource matching network; The wavelet transform algorithm combines with the Bayesian network, including: Construct a vibration-temperature coupling state equation, and use the mechanical vibration amplitude, temperature change gradient, and water pressure fluctuation as observation variables; Design a wavelet basis function library to map the observation variables into multi-scale wavelet coefficients; Dynamically update the state transition probability through online Bayesian inference, and optimize the wavelet decomposition layer number using the maximum a posteriori estimation; Output state compensation parameters including vibration suppression coefficients and temperature compensation amounts; The multi-source sensor includes a pressure transmitter, an electromagnetic flowmeter, a temperature and humidity sensor, and a vibration accelerometer; Perform time-domain synchronization processing on the pressure transmitter data and the electromagnetic flowmeter data to construct a working condition state model of the fire hydrant; perform wavelet packet decomposition on the temperature and humidity sensor data and the vibration accelerometer data to extract environmental interference characteristics and mechanical vibration spectra; Construct a dual-channel feature fusion network. The first channel uses a one-dimensional convolutional neural network to extract the time-domain fluctuation features of the working condition state model, and the second channel uses a frequency-domain attention mechanism to extract the frequency-domain energy distribution features of the mechanical vibration spectrum; Fuse the time-domain fluctuation features and the frequency-domain energy distribution features through a cross-domain feature alignment strategy to generate a joint feature matrix; perform node correlation modeling on the joint feature matrix based on a graph convolutional network, and output comprehensive detection features including abnormal water pressure, flow rate mutation, and mechanical wear state; The one-dimensional convolutional neural network adopts a multi-scale residual structure, including: Divide the working condition state model into multi-time window sequences, and each time window contains mean, variance, and range statistics; Use a dilated convolutional kernel with a dilation rate of 2 in the convolutional stage to capture long-time sequence dependence features, and fuse short-time sequence details and long-time sequence trend features in the residual connection stage; Introduce a time series attention module to adaptively weight the feature channels; The time series attention module adopts a frequency-domain-time-domain joint focusing mechanism, including: calculating the energy density weight of the feature map in the frequency domain dimension, screening the dominant frequency components, constructing a sliding window model in the time domain dimension to locate the time period when abnormal events occur, and fusing the frequency-domain weight and the time-domain weight through tensor product operation.

Images