Fire-fighting pipe leakage risk early warning system based on big data analysis

Abstract

This invention discloses a fire-fighting pipe fitting leakage risk early warning system based on big data analysis, comprising the following steps: collecting multi-dimensional time-series data such as pressure, flow rate, temperature, and humidity; constructing a data processing and modeling workflow; employing kernel canonical correlation analysis to extract nonlinear correlation features between different monitoring parameters to identify weak correlation changes before leakage; and constructing an anomaly measurement mechanism based on the maximum correlation entropy criterion to quantify the degree of feature shift. By fusing the above correlation features and entropy information, a dynamic risk index is generated and compared with a dynamic threshold to achieve real-time early warning of leakage risk, effectively supporting early fault detection and intelligent assessment of fire-fighting pipe fittings. This invention achieves dynamic perception and intelligent judgment of fire-fighting pipe fitting leakage risk, possessing data-driven early warning capabilities.

Description

Technical Field

[0001] This invention relates to the field of fire safety monitoring technology, and in particular to a fire pipe fitting leakage risk early warning system based on big data analysis. Background Technology

[0002] With the improvement of the intelligence level of urban fire protection systems, IoT-based monitoring of fire protection pipe networks has become an important means to ensure building safety. At present, the monitoring of leaks in fire protection pipe fittings mainly relies on single-point threshold judgment of pressure or flow sensors and simple statistical anomaly detection models, which have limited early warning capabilities in scenarios with significant data fluctuations or frequent changes in system load.

[0003] Existing leak monitoring technologies have significant limitations in multi-source data fusion and weak anomaly identification. On the one hand, traditional methods mainly use linear models or static rules for analysis, which cannot effectively capture the weak nonlinear correlations between variables such as pressure and flow rate before a leak, making it difficult to identify early signs. On the other hand, existing anomaly detection indicators are generally based on fixed error metrics, which are easily affected by extreme operating conditions such as noise and water hammer, resulting in poor stability and adaptability. Furthermore, they lack unified modeling methods for correlation information and entropy changes, making it difficult to build a robust dynamic early warning mechanism.

[0004] Therefore, how to provide a fire-fighting pipe fitting leakage risk early warning system based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a fire-fighting pipe fitting leakage risk early warning system based on big data analysis. This invention uses kernel correlation analysis to extract nonlinear correlation features between variables such as pressure and flow rate, combines the maximum correlation entropy criterion to construct an abnormal offset measurement model, and forms a risk change sequence by fusing correlation features and entropy features, thereby realizing the dynamic generation of leakage risk index and threshold adaptive early warning. It has the advantages of high modeling accuracy, strong anti-interference ability and timely early warning response.

[0006] A fire-fighting pipe fitting leakage risk early warning system based on big data analysis according to an embodiment of the present invention includes the following steps: The data acquisition module is used to collect time-series data on pressure, flow, temperature and humidity of fire-fighting pipe fittings and pipelines, and to construct the original multidimensional dataset according to the time index. The data processing module performs preprocessing on the original multidimensional dataset to generate a structured dataset; The nonlinear correlation analysis module extracts the pressure variable set and the flow variable set from the structured dataset, and performs kernel canonical correlation analysis on the pressure variable set and the flow variable set to generate a correlation feature sequence arranged by time index; The correlation entropy calculation module divides the structured dataset into a reference window and a current window according to the time index, constructs a reference correlation entropy distribution and a current correlation entropy distribution according to the maximum correlation entropy criterion, and calculates the entropy offset sequence from the difference between the two. The feature fusion module receives the associated feature sequence and the entropy offset sequence, aligns them according to the time index, and generates a fused feature sequence through fusion operation; The risk index module is used to receive the fused feature sequence and generate a leakage risk index sequence based on the changing relationship of the fused feature sequence. The early warning module is used to receive the leakage risk index sequence, compare the leakage risk index with the preset dynamic threshold, and output a leakage risk early warning signal.

[0007] Optional preprocessing includes: time synchronization, data completion, anomaly removal, and amplitude normalization.

[0008] Optional, the nonlinear correlation analysis module includes: The pressure variable set construction unit is used to extract the time series channels corresponding to all pressure measurement points from the structured dataset under a unified time index, and combine the values ​​of each pressure channel under each time index into a pressure variable set in a fixed channel order. The pressure variable set is composed of pressure feature vectors under continuous time index, which includes multi-channel pressure values ​​such as pressure change, steady-state pressure and pressure fluctuation from different pressure measurement points. The flow variable set construction unit is used to extract the time series channels corresponding to all flow measurement points from the structured processing dataset under a unified time index, and combine the values ​​of each flow channel under each time index into a flow variable set in a fixed channel order. The flow variable set is composed of flow feature vectors under continuous time index, including multi-channel flow values ​​such as flow change, steady-state flow and flow pulsation from different flow measurement points. The kernel correlation input building unit is used to perform pairwise matching of the pressure variable set and the flow variable set under the same time index, forming input data pairs composed of pressure feature vectors and flow feature vectors, and then performing kernel canonical correlation analysis. The correlation feature output generation unit is used to receive the correlation output corresponding to each time index generated by the kernel canonical correlation analysis, arrange the correlation output in order of time index, and generate a correlation feature sequence arranged by time index.

[0009] Optional, canonical correlation analysis includes: The time-series extended vector construction unit is used to receive the pressure variable set and the flow variable set, combine the pressure feature vector under each time index with the pressure feature vector of the previous time index and the next time index to form a pressure extended vector, and combine the flow feature vector under each time index with the flow feature vector of the previous time index and the next time index to form a flow extended vector, and construct the pressure extended sequence and the flow extended sequence respectively according to the time index order. The time-dependent kernel matrix construction unit is used to receive the pressure spread sequence and the flow spread sequence, and to use the spread vector as the kernel function input. It performs kernel function operation on the spread vectors corresponding to any two time indices in the pressure spread sequence to form the pressure kernel matrix, and performs kernel function operation on the spread vectors corresponding to any two time indices in the flow spread sequence to form the flow kernel matrix. The kernel redundancy suppression decomposition unit is used to receive the pressure kernel matrix and the flow kernel matrix, and generate the deredundant pressure kernel matrix and the deredundant flow kernel matrix by subtracting the mapping component between the pressure kernel matrix and the flow kernel matrix. The regularized covariance construction unit is used to receive the redundant pressure kernel matrix and the redundant flow kernel matrix, construct the pressure autocovariance matrix, the flow autocovariance matrix and the cross-covariance matrix of pressure and flow based on the two, and add regularization terms to the pressure autocovariance matrix and the flow autocovariance matrix to form the regularized covariance matrix. The generalized feature solving unit is used to receive the regularized covariance matrix, construct the generalized feature equation, and solve for the corresponding typical pressure direction coefficient and typical flow direction coefficient. The typical projection generation unit is used to project the pressure variable set and the flow variable set onto the pressure typical direction coefficient and the flow typical direction coefficient, respectively, and to calculate the pressure projection and flow projection corresponding to each time index. The associated output sequence construction unit is used to generate corresponding associated output quantities based on the pressure projection and flow projection quantities under each time index, and arrange them in time index order to form an associated feature sequence.

[0010] Optionally, the relevant entropy calculation module includes: The time window partitioning unit is used to receive the structured processing dataset and the associated feature sequence, and divides the continuous time index into a reference window and a current window with a fixed length. The reference window consists of the continuous time indexes before the current window time index set, and the current window consists of the most recent continuous time index. The window feature extraction unit is used to receive the reference window and the current window, and extract the associated feature vector sequence in the reference window and the associated feature vector sequence in the current window from the associated feature sequence according to the time index; The error generation unit is used to receive the associated feature vector sequence of the reference window and the associated feature vector sequence of the current window, take the associated feature vector corresponding to the time index under each time index of the current window, and calculate the difference with the associated feature vectors arranged in time order in the reference window one by one to generate the error sequence. The kernelized error sequence construction unit is used to receive the error quantity sequence, perform kernel function mapping on each error quantity, convert the error quantity into a kernelized error quantity, and obtain the reference window kernelized error sequence and the current window kernelized error sequence. The reference correlation entropy distribution construction unit is used to receive the reference window kernelization error sequence, sum the kernelization error under the reference window time index according to the maximum correlation entropy criterion, and form the reference correlation entropy distribution; The current correlation entropy distribution construction unit is used to receive the kernelized error sequence of the current window, sum the kernelized error under the current window time index according to the maximum correlation entropy criterion, and form the current correlation entropy distribution; The entropy offset sequence generation unit is used to receive the reference related entropy distribution and the current related entropy distribution, perform difference calculation on the related entropy values ​​under the corresponding time index, generate entropy offset, and arrange them in time index order to form an entropy offset sequence.

[0011] Optionally, the feature fusion module includes: The time index pairing unit is used to receive the associated feature sequence and the entropy offset sequence, compare the time indices of the two sequences, filter out the common time indices that exist in both sequences, and extract the associated feature vector and entropy offset under each common time index to generate feature pairs to be fused arranged in time order. The feature alignment unit is used to receive feature pairs to be fused, and combine the associated feature vectors and the corresponding entropy offsets in the feature pairs to be fused under each common time index to form an aligned feature vector, forming a sequence of aligned feature vectors arranged by time index. The component weighting unit is used to receive the aligned feature vector sequence, apply the corresponding weighting coefficient to each component of the associated feature vector in each aligned feature vector, and apply an independent weighting coefficient to the entropy offset in the aligned feature vector, thereby generating a weighted component set under the corresponding time index. The fusion computing unit is used to receive the weighted component set, perform a term-by-term summation operation on all weighted components under the same time index, and obtain the fusion feature value of the corresponding time index; The fusion sequence construction unit is used to receive the fusion feature values ​​under each time index and arrange them in time index order to form a fusion feature sequence.

[0012] Optional, the risk index module includes: The fusion feature change extraction unit is used to receive the fusion feature sequence, calculate the difference between the fusion feature values ​​between adjacent time indices under a unified time index, generate a difference sequence, and extract the mean, standard deviation and range of the fusion feature values ​​in each time index based on a sliding window of preset length, and construct a set of fusion feature change indicators. The feature change combination unit is used to receive the set of fused feature change indicators. Under each time index, it combines the difference value, mean value, standard deviation and range of the fused feature value corresponding to the time point into a multi-dimensional change vector, and arranges them in the order of the time index to form a fused feature change sequence. The risk index calculation unit is used to receive the fusion feature change sequence, input the change vector to the mapping function model under each time index, and perform linear combination or nonlinear operation on each component in the change vector according to the preset structure to generate the leakage risk index under the current time index. The leakage risk sequence construction unit is used to receive leakage risk indices under each time index, arrange them in time index order, and generate a leakage risk index sequence.

[0013] Optionally, the early warning module includes: The threshold generation unit is used to receive the leakage risk index sequence and, in combination with the mean, standard deviation and trend of the leakage risk index sequence within a preset time window, generate a corresponding dynamic early warning threshold under each time index according to the threshold control rules, forming a dynamic threshold sequence that corresponds one-to-one with the time index. The threshold pairing unit is used to align the leakage risk index sequence with the dynamic threshold sequence according to the time index. Under each time index, the corresponding risk index value and dynamic threshold are extracted to form a pairing comparison set arranged by time index. The risk comparison unit is used to receive the paired comparison set, compare the risk index value with the corresponding dynamic threshold at each time index, determine whether the condition that the risk index is greater than the dynamic threshold is met, and record the comparison result in Boolean label form to generate a Boolean label sequence. The early warning tag generation unit is used to receive a Boolean tag sequence and generate a corresponding early warning tag under each time index. The tag 1 indicates that the early warning is triggered when the threshold is higher than the dynamic threshold, and the tag 0 indicates that the early warning is not triggered. All tags are arranged in chronological order and a leakage risk early warning signal is generated. The early warning output unit is used to receive the early warning status sequence and output the early warning mark under each time index and its corresponding risk index to the display module, remote platform or control interface for triggering alarm response, logging or linkage control.

[0014] The beneficial effects of this invention are: (1) By introducing canonical correlation analysis and maximum correlation entropy criterion, this invention realizes accurate modeling and measurement of nonlinear weak correlation shifts between multi-source sensor signals such as pressure and flow in fire protection pipelines, overcomes the lag and misjudgment problems of traditional linear statistical methods in early leakage detection, and improves the sensitivity and accuracy of leakage risk identification.

[0015] (2) By constructing a fusion mechanism of correlation features and entropy features and introducing a risk index calculation model based on weighted strategy, this invention effectively realizes dynamic trend modeling and quantitative expression of leakage risk. With the adaptively adjustable dynamic threshold, the system can flexibly respond to complex environmental changes under different working conditions, different seasons and different life cycle stages, thereby enhancing the stability and adaptability of the early warning mechanism.

[0016] (3) The present invention adopts a structured modular design, which connects the entire process of data acquisition, preprocessing, feature extraction, risk index generation and early warning output, forming a lightweight early warning solution that does not rely on deep learning. While ensuring real-time performance and interpretability, it has high portability and practical engineering deployment capabilities, and is suitable for the intelligent transformation needs of various urban fire protection infrastructures. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a structural framework diagram of a fire-fighting pipe fitting leakage risk early warning system based on big data analysis proposed in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figure 1 A fire-fighting pipe fitting leakage risk early warning system based on big data analysis includes the following steps: The data acquisition module is used to collect time-series data on pressure, flow, temperature and humidity of fire-fighting pipe fittings and pipelines, and to construct the original multidimensional dataset according to the time index. The data processing module performs preprocessing on the original multidimensional dataset to generate a structured dataset; The nonlinear correlation analysis module extracts the pressure variable set and the flow variable set from the structured dataset, and performs kernel canonical correlation analysis on the pressure variable set and the flow variable set to generate a correlation feature sequence arranged by time index; The correlation entropy calculation module divides the structured dataset into a reference window and a current window according to the time index, constructs a reference correlation entropy distribution and a current correlation entropy distribution according to the maximum correlation entropy criterion, and calculates the entropy offset sequence from the difference between the two. The feature fusion module receives the associated feature sequence and the entropy offset sequence, aligns them according to the time index, and generates a fused feature sequence through fusion operation; The risk index module is used to receive the fused feature sequence and generate a leakage risk index sequence based on the changing relationship of the fused feature sequence. The early warning module is used to receive the leakage risk index sequence, compare the leakage risk index with the preset dynamic threshold, and output a leakage risk early warning signal.

[0020] In this embodiment, preprocessing includes: time synchronization, data completion, anomaly removal, and amplitude normalization. Time synchronization aligns time series collected by different sensors with a unified time index, ensuring all sampling points have consistent time markers on the same time axis. Data completion fills in missing sampling points on the time axis; short-term missing points are filled with interpolation results from adjacent valid data, while continuous gaps are filled with the average or median of the sequence over adjacent time periods. Anomaly removal identifies and removes sampling points that are inconsistent with the overall range or trend of the sequence, including abrupt amplitude changes, invalid jumps, or measurement results that cannot be derived from adjacent variation patterns, and replaces them with alternative values ​​calculated from adjacent normal sampling points. Amplitude normalization maps values ​​generated by different sensors to a unified scale, linearly compressing the time series of each variable according to its central trend and dispersion within a selected time period, so that each physical quantity is represented within the same dimensional range.

[0021] In this embodiment, the nonlinear correlation analysis module includes: The pressure variable set construction unit is used to extract the time series channels corresponding to all pressure measurement points from the structured dataset under a unified time index, and combine the values ​​of each pressure channel under each time index into a pressure variable set in a fixed channel order. The pressure variable set is composed of pressure feature vectors under continuous time index, which includes multi-channel pressure values ​​such as pressure change, steady-state pressure and pressure fluctuation from different pressure measurement points. The flow variable set construction unit is used to extract the time series channels corresponding to all flow measurement points from the structured processing dataset under a unified time index, and combine the values ​​of each flow channel under each time index into a flow variable set in a fixed channel order. The flow variable set is composed of flow feature vectors under continuous time index, including multi-channel flow values ​​such as flow change, steady-state flow and flow pulsation from different flow measurement points. The kernel correlation input building unit is used to perform pairwise matching of the pressure variable set and the flow variable set under the same time index, forming input data pairs composed of pressure feature vectors and flow feature vectors, and then performing kernel canonical correlation analysis. Pair matching refers to combining each pressure feature vector in the pressure variable set with a flow feature vector in the flow variable set with the same time index under a unified time index, so that the two feature vectors form a pair of input data at the same time point. Specifically, under each time index, the corresponding pressure feature vector and flow feature vector are read, and the two are combined into a pair of inputs in a fixed order, and then arranged continuously according to the time index order to form the input sequence for subsequent kernel canonical correlation analysis. The correlation feature output generation unit is used to receive the correlation output corresponding to each time index generated by the kernel canonical correlation analysis, arrange the correlation output in order of time index, and generate a correlation feature sequence arranged by time index.

[0022] In this embodiment, the nuclear canonical correlation analysis includes: The time-series extended vector construction unit is used to receive the pressure variable set and the flow variable set, combine the pressure feature vector under each time index with the pressure feature vector of the previous time index and the next time index to form a pressure extended vector, and combine the flow feature vector under each time index with the flow feature vector of the previous time index and the next time index to form a flow extended vector, and construct the pressure extended sequence and the flow extended sequence respectively according to the time index order. The time-dependent kernel matrix construction unit is used to receive the pressure spread sequence and the flow spread sequence, and to use the spread vector as the kernel function input. It performs kernel function operation on the spread vectors corresponding to any two time indices in the pressure spread sequence to form the pressure kernel matrix, and performs kernel function operation on the spread vectors corresponding to any two time indices in the flow spread sequence to form the flow kernel matrix. A kernel function is a functional form used to map original feature vectors to a high-dimensional feature space to calculate the nonlinear similarity relationship between samples. In this method, the kernel function takes two extended feature vectors as input and outputs their similarity measurement results in the high-dimensional space. It is often in the form of radial basis function or polynomial kernel function to ensure that the kernel matrix is ​​symmetric, positive semi-definite, and corresponds to the time index. The kernel redundancy suppression decomposition unit is used to receive the pressure kernel matrix and the flow kernel matrix, and generate the deredundant pressure kernel matrix and the deredundant flow kernel matrix by subtracting the mapping component between the pressure kernel matrix and the flow kernel matrix. The regularized covariance construction unit is used to receive the redundant pressure kernel matrix and the redundant flow kernel matrix, construct the pressure autocovariance matrix, the flow autocovariance matrix and the cross-covariance matrix of pressure and flow based on the two, and add regularization terms to the pressure autocovariance matrix and the flow autocovariance matrix to form the regularized covariance matrix. The generalized feature solving unit is used to receive the regularized covariance matrix, construct the generalized feature equation, and solve for the corresponding typical pressure direction coefficient and typical flow direction coefficient. The typical projection generation unit is used to project the pressure variable set and the flow variable set onto the pressure typical direction coefficient and the flow typical direction coefficient, respectively, and to calculate the pressure projection and flow projection corresponding to each time index. The associated output sequence construction unit is used to generate corresponding associated output quantities based on the pressure projection and flow projection quantities under each time index, and arrange them in time index order to form an associated feature sequence.

[0023] In this embodiment, the relevant entropy calculation module includes: The time window partitioning unit is used to receive the structured processing dataset and the associated feature sequence, and divides the continuous time index into a reference window and a current window with a fixed length. The reference window consists of the continuous time indexes before the current window time index set, and the current window consists of the most recent continuous time index. The window feature extraction unit is used to receive the reference window and the current window, and extract the associated feature vector sequence in the reference window and the associated feature vector sequence in the current window from the associated feature sequence according to the time index; The error generation unit is used to receive the associated feature vector sequence of the reference window and the associated feature vector sequence of the current window, take the associated feature vector corresponding to the time index under each time index of the current window, and calculate the difference with the associated feature vectors arranged in time order in the reference window one by one to generate the error sequence. The kernelized error sequence construction unit is used to receive the error quantity sequence, perform kernel function mapping on each error quantity, convert the error quantity into a kernelized error quantity, and obtain the reference window kernelized error sequence and the current window kernelized error sequence. Kernel mapping refers to performing a nonlinear transformation on the error quantity to obtain its similarity representation in a high-dimensional feature space. This process takes the error quantity at each time index as input, calculates the kernel value between the error quantity and the zero vector through a kernel function, often in the form of a Gaussian radial basis kernel function, and maps the magnitude difference of the error vector to a similarity value between [0,1], thereby constructing a kernelized error sequence, which serves as the basic input for calculating the correlation entropy in the maximum correlation entropy criterion; The reference correlation entropy distribution construction unit is used to receive the reference window kernelization error sequence, sum the kernelization error under the reference window time index according to the maximum correlation entropy criterion, and form the reference correlation entropy distribution; The maximum correlation entropy criterion is an information theory-based method used to measure the dependency between variables in a nonlinear feature space. Its core idea is to measure the degree of deviation by minimizing the correlation entropy between the kernelized error and the zero vector. In practice, the error is mapped using a kernel function to obtain the kernelized error value, and then a probability density estimate or empirical distribution is constructed based on this value. The correlation entropy is then calculated based on this value. This correlation entropy reflects the degree of aggregation of the error distribution; the smaller the correlation entropy, the more concentrated the error and the smaller the deviation between variables. The maximum correlation entropy criterion extracts a stability index from the kernelized error in this way and is widely used for anomaly detection and structural change perception in complex systems. Summation refers to the process of accumulating the kernelized error values ​​within a certain window after mapping by the kernel function in the calculation of correlation entropy. Specifically, the kernelized error values ​​generated under each time index in the reference window or the current window are arranged in chronological order and then summed sequentially with weights. The calculation process can use methods such as weighted average, sliding stack, or normalized integration to obtain a single correlation entropy value, which is used to represent the degree of clustering of errors in the high-dimensional space within that time window. The current correlation entropy distribution construction unit is used to receive the kernelized error sequence of the current window, sum the kernelized error under the current window time index according to the maximum correlation entropy criterion, and form the current correlation entropy distribution; The entropy offset sequence generation unit is used to receive the reference related entropy distribution and the current related entropy distribution, perform difference calculation on the related entropy values ​​under the corresponding time index, generate entropy offset, and arrange them in time index order to form an entropy offset sequence.

[0024] In this embodiment, the feature fusion module includes: The time index pairing unit is used to receive the associated feature sequence and the entropy offset sequence, compare the time indices of the two sequences, filter out the common time indices that exist in both sequences, and extract the associated feature vector and entropy offset under each common time index to generate feature pairs to be fused arranged in time order. Filtering out common time indices that exist in two sequences means extracting the complete time index sets contained in the associated feature sequence and the entropy offset sequence respectively, and performing an intersection operation on the two sets to obtain a set of time indices that exist in both. In the specific implementation, the time indices of the two sequences are first unified in format and sorted. Then, one time index set is traversed, and each time index is checked to see if it exists in the other set. If it exists, it is recorded as a common time index. Finally, a list of common time indices is output. The feature alignment unit is used to receive feature pairs to be fused, and combine the associated feature vectors and the corresponding entropy offsets in the feature pairs to be fused under each common time index to form an aligned feature vector, forming a sequence of aligned feature vectors arranged by time index. The component weighting unit is used to receive the aligned feature vector sequence, apply the corresponding weighting coefficient to each component of the associated feature vector in each aligned feature vector, and apply an independent weighting coefficient to the entropy offset in the aligned feature vector, thereby generating a weighted component set under the corresponding time index. Weighting coefficients refer to the numerical weights assigned to feature components from different sources during the fusion calculation process. They are used to adjust the relative contribution of each component to the overall fused features. Each component in the associated feature vector corresponds to a weighting coefficient, which is assigned a value according to its importance in the fusion. Independent weighting coefficients are separate weights specifically set for entropy offset, independent of the weighting coefficients of the associated feature components. They are used to control the proportion of entropy offset in the fused feature values. All weighting coefficients can be obtained through configuration settings or offline calibration before calculation. During fusion, they are multiplied with their respective corresponding components and accumulated to obtain the fusion result. The fusion computing unit is used to receive the weighted component set, perform a term-by-term summation operation on all weighted components under the same time index, and obtain the fusion feature value of the corresponding time index; Performing a term-by-term summation operation means multiplying the associated feature components with their corresponding weighting coefficients and the entropy offset with the independent weighting coefficients under the same time index, and then summing all the product results in the order of the components to obtain the fusion feature value corresponding to that time index. The fusion sequence construction unit is used to receive the fusion feature values ​​under each time index and arrange them in time index order to form a fusion feature sequence.

[0025] In this embodiment, the risk index module includes: The fusion feature change extraction unit is used to receive the fusion feature sequence, calculate the difference between the fusion feature values ​​between adjacent time indices under a unified time index, generate a difference sequence, and extract the mean, standard deviation and range of the fusion feature values ​​in each time index based on a sliding window of preset length, and construct a set of fusion feature change indicators. A preset-length sliding window refers to a window region composed of a fixed number of consecutive time indices selected in a time series. This region is used to extract fused feature values ​​within the window's coverage area at each time index. Specifically, several consecutive time indices are selected backward from the current time index to form the data set within the window. The window's length remains constant as it moves, updating only as the time indices slide forward. The feature change combination unit is used to receive the set of fused feature change indicators. Under each time index, it combines the difference value, mean value, standard deviation and range of the fused feature value corresponding to the time point into a multi-dimensional change vector, and arranges them in the order of the time index to form a fused feature change sequence. The risk index calculation unit is used to receive the fusion feature change sequence, input the change vector to the mapping function model under each time index, and perform linear combination or nonlinear operation on each component in the change vector according to the preset structure to generate the leakage risk index under the current time index. The mapping function model is a functional structure used to transform the changing relationships of fused features into a leakage risk index. It typically consists of a set of input variables, corresponding mapping coefficients, and a mapping method. Input variables include statistical features such as the difference, moving mean, moving standard deviation, and range for each time index in the fused feature sequence. Mapping coefficients are used to assign weights to each variable. The mapping method can be a linear weighted summation or a nonlinear combination performed through a preset activation function. This model generates a complete risk index by weighting each input variable and summing the results to output the risk value at the current time index. The leakage risk sequence construction unit is used to receive leakage risk indices under each time index, arrange them in time index order, and generate a leakage risk index sequence.

[0026] In this embodiment, the early warning module includes: The threshold generation unit is used to receive the leakage risk index sequence and, in combination with the mean, standard deviation and trend of the leakage risk index sequence within a preset time window, generate a corresponding dynamic early warning threshold under each time index according to the threshold control rules, forming a dynamic threshold sequence that corresponds one-to-one with the time index. The mean, standard deviation, and trend within the preset time window refer to the statistical characteristics of the leakage risk index sequence within a fixed time range before the current time index during the dynamic threshold generation process. The mean represents the average value of all risk indices within the window, the standard deviation represents the degree of dispersion of the risk index around the mean within the window, and the trend reflects the direction of the risk index's rise or fall during the time period by calculating the difference between the mean of the second half and the mean of the first half within the window. This is used to dynamically adjust the warning threshold of the current time index. Threshold control rules refer to calculating the dynamic threshold under the current time index based on the mean, standard deviation and trend of the leakage risk index within a preset time window. It is often generated in the form of mean weighted adjustment, that is, the current threshold is equal to the weighted sum of the mean and standard deviation of the window, and an adjustable offset term can be added or removed according to the trend. The threshold pairing unit is used to align the leakage risk index sequence with the dynamic threshold sequence according to the time index. Under each time index, the corresponding risk index value and dynamic threshold are extracted to form a pairing comparison set arranged by time index. The risk comparison unit is used to receive the paired comparison set, compare the risk index value with the corresponding dynamic threshold at each time index, determine whether the condition that the risk index is greater than the dynamic threshold is met, and record the comparison result in Boolean label form to generate a Boolean label sequence. The early warning tag generation unit is used to receive a Boolean tag sequence and generate a corresponding early warning tag under each time index. The tag 1 indicates that the early warning is triggered when the threshold is higher than the dynamic threshold, and the tag 0 indicates that the early warning is not triggered. All tags are arranged in chronological order and a leakage risk early warning signal is generated. The early warning output unit is used to receive the early warning status sequence and output the early warning mark under each time index and its corresponding risk index to the display module, remote platform or control interface for triggering alarm response, logging or linkage control.

[0027] Example 1: To verify the feasibility of this invention in practice, it was applied to the underground fire protection pipe network system of a large public building. The specific deployment method, data processing flow, and early warning output process of the proposed big data analysis-based fire protection pipe fitting leakage risk early warning system in this engineering environment are described. In this scenario, after long-term operation, the fire protection system suffers from problems such as aging of local pipe fittings and loosening of interfaces. Traditional leakage detection methods based on single-point thresholds are difficult to identify minor leaks in their early stages and are easily affected by instantaneous pressure fluctuations, water hammer, and other interferences, leading to false alarms. Therefore, they cannot meet the accurate early warning requirements of high-density public spaces.

[0028] In this project, the system is deployed according to the overall structure of this invention, including a data acquisition module, a data processing module, a nonlinear correlation analysis module, a correlation entropy calculation module, a feature fusion module, a risk index module, and an early warning module, and is operated in conjunction with each other through a unified control platform.

[0029] The data acquisition module is equipped with multi-channel sensing units, deploying over a hundred pressure, flow, temperature, and humidity sensors along the main pipelines and key branches, with a data acquisition cycle set to 10 seconds. The acquired raw data undergoes preprocessing operations such as time alignment, missing data completion, anomaly removal, filtering, smoothing, and normalization by the data processing module to form a structured dataset for direct use by subsequent modules.

[0030] The nonlinear correlation analysis module selects the processed set of pressure and flow variables, extracts the nonlinear correlation features between them based on a kernel canonical correlation model, introduces a time-series extension structure to enhance time dependence, and improves feature effectiveness through redundancy suppression. The processing results are output as a correlation feature sequence in units of time index.

[0031] The correlation entropy calculation module performs window partitioning on the associated feature sequence, constructing a reference window and the current window at each index point, and calculating the kernelization error value based on the maximum correlation entropy criterion. The correlation entropy distribution is solved separately in the two windows, and the difference is calculated to generate an entropy offset sequence reflecting changes in variable structure.

[0032] The feature fusion module receives the associated feature sequence and the entropy offset sequence, extracts common time points through a time index alignment mechanism, and concatenates the associated features and entropy offset values ​​at each time point to form an aligned feature vector. The fusion process is based on a weighted strategy, performing linear superposition by multiplying each component by a set weighting coefficient to form a fused feature sequence.

[0033] The risk index module performs statistical analysis on the fused feature sequence, extracts the difference, moving average, standard deviation and range at each time point, forms the fused feature change vector, and inputs it into the risk mapping function model to calculate a continuous leakage risk index sequence.

[0034] The early warning module receives a risk index sequence, compares the risk values ​​generated at each time point using a dynamic threshold mechanism, outputs a Boolean early warning result, and triggers an alarm device. The threshold is adaptively adjusted based on the mean, standard deviation, and trend of the risk values ​​within a sliding window.

[0035] During continuous operation, the system identified a slow shift in a flow-pressure nonlinear coupling structure at a certain stage, accompanied by a continuous increase in entropy. The following are some key data recorded by the system during the detection period: Table 1: Data Records for Key Time Periods When Warnings Are Triggered ; The data above shows that the system can continuously track potential anomalies based on nonlinear structural changes and entropy offset indicators before a significant pressure drop occurs, and trigger an early warning at time T2. This is at least two sampling cycles earlier than the traditional system, which may only trigger an alarm due to threshold exceeding the limit in the T4-T5 period.

[0036] Further comparison of the system's performance with that of a traditional fixed-threshold system in the same scenario over a full month yielded the following results: Table 2: Performance Comparison of Different Systems During Long-Term Operation ; As can be seen from the key time period data in Table 1, at time index T1, the system monitored a pressure of 0.438 MPa and a flow rate of 13.8 L / min. Although the physical parameters were still within the normal range, the correlation feature value had reached 0.756, the entropy offset value was 0.0147, the fusion feature value was 0.581, and the corresponding risk index was 0.615, which had not yet exceeded the dynamic threshold of 0.638, so no warning was triggered. However, starting from T2, as the pressure continued to decrease and the flow rate gradually increased, the correlation feature value and the entropy offset value increased synchronously. The risk index exceeded the threshold for the first time at T2, and the system immediately issued a warning. Subsequently, the risk index continued to rise at each time point from T3 to T5 and remained stable above the dynamic threshold. The system continuously maintained a warning status, indicating that the system has the ability to track subtle trend changes.

[0037] Table 2 further compares the data, showing that during the 30-day monitoring period, the system of this invention processed an average of 90,000 data entries per day, more than four times that of the traditional system, demonstrating significant data carrying capacity. Simultaneously, this system can provide an average of 2.5 minutes' advance warning before a leak occurs and can identify leaks as small as 0.08 L / min, while the traditional system's detection threshold is 0.40 L / min. Furthermore, this system only experienced one false alarm and no missed alarms, while the traditional system had multiple false alarms and missed alarms, indicating that the present invention has significant advantages in accuracy and stability, verifying its high reliability and engineering applicability under actual working conditions.

[0038] This embodiment verifies that the present invention possesses highly interconnected data-driven capabilities and dynamic decision-making capabilities when facing nonlinear variable structure changes, weak signal leakage characteristics, and complex operating condition interference. By combining kernel canonical correlation analysis with the maximum correlation entropy criterion, this system can achieve joint modeling of pressure-flow relationship and operating entropy, thereby constructing a fused feature sequence, and dynamically calculating the risk index by leveraging the changing trend, ultimately realizing an intelligent early warning mechanism with variable thresholds and adaptive response.

[0039] The implementation of this invention does not rely on deep learning models or large-scale training data. The algorithm structure is lightweight and suitable for the rapid access and upgrading of existing urban old fire protection pipe network systems. While ensuring detection accuracy, it improves deployment flexibility and system interpretability.

[0040] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A fire-fighting pipe fitting leakage risk early warning system based on big data analysis, characterized in that, include: The data acquisition module is used to collect time-series data on pressure, flow, temperature and humidity of fire-fighting pipe fittings and pipelines, and to construct the original multidimensional dataset according to the time index. The data processing module performs preprocessing on the original multidimensional dataset to generate a structured dataset; The nonlinear correlation analysis module extracts the pressure variable set and the flow variable set from the structured dataset, and performs kernel canonical correlation analysis on the pressure variable set and the flow variable set to generate a correlation feature sequence arranged by time index; The correlation entropy calculation module divides the structured dataset into a reference window and a current window according to the time index, constructs a reference correlation entropy distribution and a current correlation entropy distribution according to the maximum correlation entropy criterion, and calculates the entropy offset sequence from the difference between the two. The feature fusion module receives the associated feature sequence and the entropy offset sequence, aligns them according to the time index, and generates a fused feature sequence through fusion operation; The risk index module is used to receive the fused feature sequence and generate a leakage risk index sequence based on the changing relationship of the fused feature sequence. The early warning module is used to receive the leakage risk index sequence, compare the leakage risk index with the preset dynamic threshold, and output a leakage risk early warning signal.

2. The fire-fighting pipe fitting leakage risk early warning system based on big data analysis according to claim 1, characterized in that, Preprocessing includes: Time synchronization, data completion, anomaly removal, and amplitude normalization.

3. The fire-fighting pipe fitting leakage risk early warning system based on big data analysis according to claim 1, characterized in that, The nonlinear correlation analysis module includes: The pressure variable set construction unit is used to extract the time series channels corresponding to all pressure measurement points from the structured dataset under a unified time index, and combine the values ​​of each pressure channel under each time index into a pressure variable set according to a fixed channel order. The flow variable set construction unit is used to extract the time series channels corresponding to all flow measurement points from the structured dataset under a unified time index, and combine the values ​​of each flow channel under each time index into a flow variable set according to a fixed channel order. The kernel correlation input building unit is used to perform pairwise matching of the pressure variable set and the flow variable set under the same time index, forming input data pairs composed of pressure feature vectors and flow feature vectors, and then performing kernel canonical correlation analysis. The correlation feature output generation unit is used to receive the correlation output corresponding to each time index generated by the kernel canonical correlation analysis, arrange the correlation output in order of time index, and generate a correlation feature sequence arranged by time index.

4. The fire-fighting pipe fitting leakage risk early warning system based on big data analysis according to claim 3, characterized in that, Nuclear canonical correlation analysis includes: The time-series extended vector construction unit is used to receive the pressure variable set and the flow variable set, combine the pressure feature vector under each time index with the pressure feature vector of the previous time index and the next time index to form a pressure extended vector, and combine the flow feature vector under each time index with the flow feature vector of the previous time index and the next time index to form a flow extended vector, and construct the pressure extended sequence and the flow extended sequence respectively according to the time index order. The time-dependent kernel matrix construction unit is used to receive the pressure spread sequence and the flow spread sequence, and to use the spread vector as the kernel function input. It performs kernel function operation on the spread vectors corresponding to any two time indices in the pressure spread sequence to form the pressure kernel matrix, and performs kernel function operation on the spread vectors corresponding to any two time indices in the flow spread sequence to form the flow kernel matrix. The kernel redundancy suppression decomposition unit is used to receive the pressure kernel matrix and the flow kernel matrix, and generate the deredundant pressure kernel matrix and the deredundant flow kernel matrix by subtracting the mapping component between the pressure kernel matrix and the flow kernel matrix. The regularized covariance construction unit is used to receive the redundant pressure kernel matrix and the redundant flow kernel matrix, construct the pressure autocovariance matrix, the flow autocovariance matrix and the cross-covariance matrix of pressure and flow based on the two, and add regularization terms to the pressure autocovariance matrix and the flow autocovariance matrix to form the regularized covariance matrix. The generalized feature solving unit is used to receive the regularized covariance matrix, construct the generalized feature equation, and solve for the corresponding typical pressure direction coefficient and typical flow direction coefficient. The typical projection generation unit is used to project the pressure variable set and the flow variable set onto the pressure typical direction coefficient and the flow typical direction coefficient, respectively, and to calculate the pressure projection and flow projection corresponding to each time index. The associated output sequence construction unit is used to generate corresponding associated output quantities based on the pressure projection and flow projection quantities under each time index, and arrange them in time index order to form an associated feature sequence.

5. A fire-fighting pipe fitting leakage risk early warning system based on big data analysis according to claim 1, characterized in that, The relevant entropy calculation module includes: The time window partitioning unit is used to receive the structured processing dataset and the associated feature sequence, and divides the continuous time index into a reference window and a current window with a fixed length; The window feature extraction unit is used to receive the reference window and the current window, and extract the associated feature vector sequence in the reference window and the associated feature vector sequence in the current window from the associated feature sequence according to the time index; The error generation unit is used to receive the associated feature vector sequence of the reference window and the associated feature vector sequence of the current window, take the associated feature vector corresponding to the time index under each time index of the current window, and calculate the difference with the associated feature vectors arranged in time order in the reference window one by one to generate the error sequence. The kernelized error sequence construction unit is used to receive the error quantity sequence, perform kernel function mapping on each error quantity, convert the error quantity into a kernelized error quantity, and obtain the reference window kernelized error sequence and the current window kernelized error sequence. The reference correlation entropy distribution construction unit is used to receive the reference window kernelization error sequence, sum the kernelization error under the reference window time index according to the maximum correlation entropy criterion, and form the reference correlation entropy distribution; The current correlation entropy distribution construction unit is used to receive the kernelized error sequence of the current window, sum the kernelized error under the current window time index according to the maximum correlation entropy criterion, and form the current correlation entropy distribution; The entropy offset sequence generation unit is used to receive the reference related entropy distribution and the current related entropy distribution, perform difference calculation on the related entropy values ​​under the corresponding time index, generate entropy offset, and arrange them in time index order to form an entropy offset sequence.

6. The fire-fighting pipe fitting leakage risk early warning system based on big data analysis according to claim 1, characterized in that, The feature fusion module includes: The time index pairing unit is used to receive the associated feature sequence and the entropy offset sequence, compare the time indices of the two sequences, filter out the common time indices that exist in both sequences, and extract the associated feature vector and entropy offset under each common time index to generate feature pairs to be fused arranged in time order. The feature alignment unit is used to receive feature pairs to be fused, and combine the associated feature vectors and the corresponding entropy offsets in the feature pairs to be fused under each common time index to form an aligned feature vector, forming a sequence of aligned feature vectors arranged by time index. The component weighting unit is used to receive the aligned feature vector sequence, apply the corresponding weighting coefficient to each component of the associated feature vector in each aligned feature vector, and apply an independent weighting coefficient to the entropy offset in the aligned feature vector, thereby generating a weighted component set under the corresponding time index. The fusion computing unit is used to receive the weighted component set, perform a term-by-term summation operation on all weighted components under the same time index, and obtain the fusion feature value of the corresponding time index; The fusion sequence construction unit is used to receive the fusion feature values ​​under each time index and arrange them in time index order to form a fusion feature sequence.

7. The fire-fighting pipe fitting leakage risk early warning system based on big data analysis according to claim 1, characterized in that, The risk index module includes: The fusion feature change extraction unit is used to receive the fusion feature sequence, calculate the difference between the fusion feature values ​​between adjacent time indices under a unified time index, generate a difference sequence, and extract the mean, standard deviation and range of the fusion feature values ​​in each time index based on a sliding window of preset length, and construct a set of fusion feature change indicators. The feature change combination unit is used to receive the set of fused feature change indicators. Under each time index, it combines the difference value, mean value, standard deviation and range of the fused feature value corresponding to the time point into a multi-dimensional change vector, and arranges them in the order of the time index to form a fused feature change sequence. The risk index calculation unit is used to receive the fusion feature change sequence, input the change vector into the mapping function model under each time index, and generate the leakage risk index under the current time index. The leakage risk sequence construction unit is used to receive leakage risk indices under each time index, arrange them in time index order, and generate a leakage risk index sequence.

8. A fire-fighting pipe fitting leakage risk early warning system based on big data analysis according to claim 1, characterized in that, The early warning module includes: The threshold generation unit is used to receive the leakage risk index sequence and, in combination with the mean, standard deviation and trend of the leakage risk index sequence within a preset time window, generate a corresponding dynamic early warning threshold under each time index according to the threshold control rules, forming a dynamic threshold sequence that corresponds one-to-one with the time index. The threshold pairing unit is used to align the leakage risk index sequence with the dynamic threshold sequence according to the time index. Under each time index, the corresponding risk index value and dynamic threshold are extracted to form a pairing comparison set arranged by time index. The risk comparison unit is used to receive the paired comparison set, compare the risk index value with the corresponding dynamic threshold at each time index, determine whether the condition that the risk index is greater than the dynamic threshold is met, and record the comparison result in Boolean label form to generate a Boolean label sequence. The early warning tag generation unit is used to receive a Boolean tag sequence and generate a corresponding early warning tag under each time index. The tag 1 indicates that the early warning is triggered when the threshold is higher than the dynamic threshold, and the tag 0 indicates that the early warning is not triggered. All tags are arranged in chronological order and a leakage risk early warning signal is generated. The early warning output unit is used to receive the early warning status sequence and output the early warning mark under each time index and its corresponding risk index to the display module, remote platform or control interface for triggering alarm response, logging or linkage control.

Images