Internet of Things Intelligent Monitoring and Alarm Method and System for Water Quality Detection

Abstract

The present application discloses an Internet of Things intelligent monitoring and alarm method and system for water quality detection, which relates to the technical field of environmental protection monitoring. The method includes: respectively sampling water quality electrochemical data for corresponding monitored water area blocks based on each Internet of Things node; extracting electrochemical time-series sensing features respectively corresponding to each water quality electrochemical data; extracting pattern time-series dependence features corresponding to the electrochemical time-series sensing features based on a water quality anomaly recognition model to predict water quality anomaly risks; identifying corresponding water quality anomaly risk event types based on an anomaly event recognition model; and performing water area water quality pollution alarm operations according to the block identifiers of each risk monitoring water area block and the corresponding water quality anomaly risk event types. Thus, based on the distributed water quality monitoring of Internet of Things nodes and intelligent risk identification and water quality pollution early warning operations, a significant improvement in the real-time performance, comprehensive coverage, and intelligent management ability of water quality monitoring is achieved.

Description

Technical Field

[0001] This application relates to the technical field of environmental protection monitoring, and particularly to an Internet of Things intelligent monitoring and alarming method and system for water quality detection. Background Art

[0002] With the rapid development of industrialization and urbanization, the problem of water pollution has become increasingly serious, posing a serious threat to the safety of global water resources and the ecological environment. Water quality monitoring, as a key link in water pollution prevention and control, is of great significance for ensuring the ecological health of water bodies and the safety of residents' drinking water.

[0003] Traditional water quality monitoring methods mainly rely on manual sampling and laboratory analysis. Usually, water quality monitoring personnel need to regularly go to the target water area for manual sampling and send the samples to the laboratory for chemical analysis and detection.

[0004] However, the processes of manual sampling and laboratory analysis are time-consuming, usually taking several hours or even days to obtain results. For a dynamically changing water quality environment, especially sudden water pollution incidents (such as toxic chemical leaks, domestic sewage discharges, etc.), this lagging monitoring method is difficult to detect and respond in a timely manner, and it is easy to miss the best treatment opportunity for pollution incidents. In addition, the manual sampling method is limited by the actual operation ability of monitoring personnel and geographical location, and often can only monitor specific points, making it difficult to achieve full coverage of the entire water area and often resulting in a large consumption of human and material resources.

[0005] In response to the above problems, the industry has not yet proposed a better technical solution. Summary of the Invention

[0006] This application provides an Internet of Things intelligent monitoring and alarming method, system, storage medium, computer program product and electronic device for water quality detection, so as to at least solve the problems of insufficient monitoring timeliness and water area coverage in the current related water quality monitoring technology.

[0007] In a first aspect, an embodiment of the present application provides an Internet of Things (IoT) intelligent monitoring and alarm method for water quality detection, including: sampling water quality electrochemical data for corresponding monitored water area blocks based on each IoT node; a plurality of IoT nodes are distributed in different blocks in the monitored water area, and each of the IoT nodes is equipped with an electrochemical sensor; the sensing parameter types of the water quality electrochemical data include the pH value of the water body, the DO content, and the TOC content; extracting the electrochemical time-series sensing features corresponding to each of the water quality electrochemical data; inputting each of the electrochemical time-series sensing features into a water quality anomaly recognition model to extract the pattern time-series dependence features corresponding to the electrochemical time-series sensing features, so as to predict whether there is a risk of water quality anomaly in the corresponding monitored water area block; for each risk-monitored water area block predicted to have a risk of water quality anomaly, inputting the pattern time-series dependence features corresponding to the risk-monitored water area block into an anomaly event recognition model to identify the type of water quality anomaly risk event corresponding to the risk-monitored water area block; and performing a water area water quality pollution alarm operation according to the block identifier of each of the risk-monitored water area blocks and the corresponding type of water quality anomaly risk event.

[0008] In a second aspect, an embodiment of the present application provides an IoT intelligent monitoring and alarm system for water quality detection, including: a water quality parameter sampling unit for sampling water quality electrochemical data for corresponding monitored water area blocks based on each IoT node; a plurality of IoT nodes are distributed in different blocks in the monitored water area, and each of the IoT nodes is equipped with an electrochemical sensor; the sensing parameter types of the water quality electrochemical data include the pH value of the water body, the DO content, and the TOC content; a time-series feature extraction unit for extracting the electrochemical time-series sensing features corresponding to each of the water quality electrochemical data; a water quality anomaly recognition unit for inputting each of the electrochemical time-series sensing features into a water quality anomaly recognition model to extract the pattern time-series dependence features corresponding to the electrochemical time-series sensing features, so as to predict whether there is a risk of water quality anomaly in the corresponding monitored water area block; an anomaly event recognition unit for inputting the pattern time-series dependence features corresponding to each risk-monitored water area block predicted to have a risk of water quality anomaly into an anomaly event recognition model to identify the type of water quality anomaly risk event corresponding to the risk-monitored water area block; and a water area water quality alarm unit for performing a water area water quality pollution alarm operation according to the block identifier of each of the risk-monitored water area blocks and the corresponding type of water quality anomaly risk event.

[0009] In a third aspect, an electronic device is provided, which includes: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the steps of the Internet of Things intelligent monitoring and alarming method for water quality detection according to any embodiment of the present application.

[0010] In a fourth aspect, an embodiment of the present application provides a storage medium, on which a computer program is stored, and characterized in that when the program is executed by a processor, the steps of the Internet of Things intelligent monitoring and alarming method for water quality detection according to any embodiment of the present application are implemented.

[0011] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the Internet of Things intelligent monitoring and alarming method for water quality detection according to any embodiment of the present application are implemented.

[0012] Through an Internet of Things intelligent monitoring and alarming method and system for water quality detection provided by the present application, at least the following technical effects can be achieved:

[0013] (1) By setting a plurality of Internet of Things nodes in the monitored water area, and deploying an electrochemistry sensor at each node to collect the water quality electrochemistry data of the monitored water area block in real time, continuous and real-time water quality data collection and transmission can be realized, enabling the data processing center to timely obtain the change information of the water quality in the water area and quickly make a response, effectively avoiding missing the best opportunity for handling pollution incidents due to monitoring lag. In addition, a plurality of Internet of Things nodes are respectively arranged in different blocks of the water area, realizing fine-grained water quality monitoring coverage of each block of the water area, and being able to accurately locate the specific location where water quality anomalies occur, facilitating the water quality management department to locate and track pollution sources, and improving the refinement level of water quality management.

[0014] (2) By constructing a water quality anomaly recognition model and an anomaly event recognition model, intelligent analysis and pattern recognition can be performed on the electrochemistry time series sensing features collected by different nodes. On the one hand, the water quality anomaly recognition model can predict the water quality anomaly risk of each monitored water area block based on the time series features of the electrochemistry sensing data; on the other hand, the anomaly event recognition model can further identify the specific types of water quality anomaly events (such as chemical pollution, organic matter exceeding the standard, heavy metal pollution, etc.) according to the pattern time series dependence features of the water quality anomaly risk blocks, so as to achieve accurate classification of water quality anomaly events.

[0015] (3) After identifying the water quality anomaly risk and its event type, corresponding water pollution alarm operations are automatically triggered according to the locations and event types of each risk monitoring block, implementing an event type-based response mechanism. This can help the water quality management department quickly locate the pollution source and formulate corresponding treatment measures, thereby controlling the spread of pollution in the shortest time and reducing the harm to the water ecological environment.

[0016] Through this technical solution, through distributed water quality monitoring based on Internet of Things nodes, real-time data collection, intelligent risk identification, and water pollution early warning operations, the real-time nature, comprehensive coverage, and intelligent management capabilities of water quality monitoring have been significantly improved. It can effectively respond to the dynamically changing water quality environment, enhance the detection, location, and response capabilities of water pollution incidents, and contribute to safeguarding the ecological health of water bodies. Description of the Drawings

[0017] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 Shows a flowchart of an example of an Internet of Things intelligent monitoring and alarm method for water quality detection according to an embodiment of the present application;

[0019] Figure 2 Shows an operation flowchart of an example of extracting the electrochemical time series sensing features corresponding to water quality electrochemical data according to an embodiment of the present application;

[0020] Figure 3 Shows a schematic structural connection diagram of an example of a water quality anomaly recognition model according to an embodiment of the present application;

[0021] Figure 4 Shows a schematic structural connection diagram of another example of a water quality anomaly recognition model according to an embodiment of the present application;

[0022] Figure 5 Shows a schematic structural connection diagram of an example of an abnormal event recognition model according to an embodiment of the present application;

[0023] Figure 6 Shows a schematic structural block diagram of an example of an Internet of Things intelligent monitoring and alarm system for water quality detection according to an embodiment of the present application;

[0024] Figure 7 Schematic structural diagram of an embodiment of the electronic device of the present application. Detailed Embodiments

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the following will clearly and completely describe the technical solutions in the embodiments of this application with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in this application without creative efforts belong to the scope of protection of this application.

[0026] In the technical solutions of this application, for the processing of the collection, storage, use, processing, transmission, provision, and disclosure of the user's personal information involved, etc., they all comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0027] Figure 1 The flowchart of an example of the Internet of Things intelligent monitoring and alarm method for water quality detection according to the embodiments of this application is shown.

[0028] Regarding the execution subject of the method in the embodiments of this application, it can be any controller or processor with computing or processing capabilities. Specifically, it can be implemented by a water quality early warning platform server, which receives sensing sampling data from each Internet of Things node and performs corresponding risk analysis and alarm operations. By simultaneously monitoring multiple regions through distributed Internet of Things nodes, frequent manual sampling and laboratory testing are avoided, and the consumption of material resources is reduced. At the same time, the intelligent water quality monitoring and risk identification model can automatically trigger alarm and intervention operations when the water quality is abnormal, improving the overall efficiency and resource utilization rate of water quality monitoring. In this way, through the comprehensive use of Internet of Things technology and intelligent monitoring models, the real-time, comprehensive, intelligent, and efficient nature of water quality monitoring is achieved.

[0029] In some examples, it can be integrated and configured in an electronic device or terminal in a software, hardware, or software-hardware combination manner, and the types of terminals or electronic devices can be diverse, such as mobile phones, tablets, or desktop computers, etc.

[0030] As Figure 1 shown, in step S110, water quality electrochemical data for the corresponding monitored water area block is sampled based on each Internet of Things node.

[0031] In some implementation manners, Internet of Things nodes are respectively arranged in different blocks within the monitored water area to realize the distributed setting of Internet of Things nodes. An electrochemical sensor is deployed in each Internet of Things node, and the electrochemical sensor can collect the water quality electrochemical data of the monitored block in real time, specifically including water quality parameters such as pH value, dissolved oxygen (DO) content, and total organic carbon (TOC) content.

[0032] Specifically, the electrochemical sensors of each Internet of Things node acquire water quality data of the monitoring block at a certain sampling frequency (e.g., once per minute), and the collected data is uploaded to the central data server in real time through the data transmission module in the Internet of Things node. In addition, for complex water environments, the Internet of Things node can dynamically adjust the sampling frequency and sampling time period according to the location, environmental conditions, and water quality change frequency of the monitoring block. For example, for areas in the monitoring block with potential pollution sources (such as water bodies near factories or living areas), the sampling frequency can be appropriately increased to ensure that enough data points are obtained to reflect the water quality change trend.

[0033] In step S120, the electrochemical time-series sensing features corresponding to each water quality electrochemical data are extracted.

[0034] In some embodiments, first, the water quality electrochemical data collected by each Internet of Things node is preprocessed, including noise removal, outlier correction, and data smoothing to improve the reliability of the data. Then, for each type of electrochemical sensing data (such as pH value, DO content, TOC content), time-series features are extracted in the time dimension, including mean, variance, peak value, frequency features, and trend features, etc., so as to reflect the fluctuation law and change trend of each sensing parameter in the time series. Thus, through time-series feature extraction, complex original sensing data can be transformed into feature data with stronger identification ability and time-series dependence features, providing a favorable data basis for the subsequent anomaly recognition of the model.

[0035] In step S130, each electrochemical time-series sensing feature is input into the water quality anomaly recognition model to extract the pattern time-series dependence features corresponding to the electrochemical time-series sensing features, so as to predict whether there is a risk of water quality anomaly in the corresponding monitored water area block.

[0036] In some embodiments, a water quality anomaly recognition model is constructed based on historical water quality data and water quality anomaly sample data. In addition, the model type of the water quality anomaly recognition model can be diversified. For example, the model can adopt machine learning algorithms (such as decision trees, support vector machines) or deep learning models (such as deep neural networks, convolutional neural networks), and ensure the accuracy and generalization ability of the recognition through model training and verification.

[0037] Exemplarily, in the water quality early warning platform server, the model extracts the pattern time series dependence features between the changes of each water quality parameter and the abnormal sample data of the water quality by learning the historical water quality data and the labeled water quality abnormal sample data. Here, the pattern time series dependence features can include the time series dependence relationship and the time series change pattern. The time series dependence relationship refers to the dependent linkage relationship between each parameter type in the time dimension when the water quality is abnormal, which can be the relatively long-term feature information learned. The time series change pattern refers to the association pattern between the parameter fluctuation frequency, amplitude, fluctuation period, etc. and the water quality abnormality, which can be the relatively short-term feature information learned. In this way, the water quality abnormality recognition model can deeply extract the pattern time series dependence features based on the electrochemical time series features collected by each Internet of Things node in the input, analyze the water quality status of the water area block corresponding to each Internet of Things node, and predict whether there is a water quality abnormality risk in this block.

[0038] In step S140, for each risk monitoring water area block predicted to have a water quality abnormality risk, the pattern time series dependence features corresponding to the risk monitoring water area block are input into the abnormal event recognition model to identify the type of water quality abnormal risk event corresponding to the risk monitoring water area block.

[0039] Here, the abnormal event recognition model can adopt the supervised learning method for model training. Based on the labeled water quality abnormal event types (such as chemical pollution, excessive organic matter, heavy metal pollution, etc.), a classification model is constructed. Among them, the chemical pollution event indicates the risk of toxic chemical leakage, the excessive organic matter event indicates the risk of domestic sewage discharge, and the heavy metal pollution indicates the risk of industrial wastewater discharge. In addition, the model type of the abnormal event recognition model can also be diversified. For example, various RNN (Recurrent Neural Network) models, multi-layer perceptrons, ensemble learning models or autoregressive models can be adopted and trained and optimized through a large amount of labeled water quality abnormal event data. By further analyzing the pattern time series dependence features through the abnormal event recognition model, different types of water quality abnormal events can be accurately distinguished, improving the accuracy and discrimination ability of the classification of water quality abnormal risk events, and providing important reference information for tracing the source and fundamental treatment of water pollution.

[0040] In step S150, according to the block identifier of each risk monitoring water area block and the corresponding type of water quality abnormal risk event, a water area water quality pollution alarm operation is performed.

[0041] In some embodiments, after identifying the monitoring water area block with a water quality abnormality risk and its corresponding abnormal event type, the water quality early warning platform server will automatically perform the corresponding water quality pollution alarm operation according to the location of each risk monitoring water area block and its corresponding event type.

[0042] Exemplarily, the water quality early warning platform server maintains corresponding pollutant source event association tables for the block identifiers of different monitored water area blocks. The pollutant source event association tables record multiple types of water quality abnormal risk events and the corresponding pollutant sources. In this way, by querying the pollutant source event association table corresponding to the risk monitoring water area block through the water quality abnormal risk event type, the water quality early warning platform server can effectively locate the corresponding pollutant source, generate an early warning message, and provide it to the water quality management department to quickly locate the pollution source and formulate corresponding treatment measures. For example, for a chemical pollution event, an associated early warning is issued for the chemical plant near the water area block, and for an event of excessive organic matter, an associated early warning is issued for the domestic sewage pipeline near the water area block, and so on.

[0043] Thus, an automated water quality pollution alarm operation is realized, which can send an alarm to the water area manager or block person in charge immediately when a water quality abnormal event is identified, significantly improving the response efficiency of water quality monitoring and management, and effectively reducing the harm of water quality pollution events to the water body ecological environment.

[0044] Figure 2 The operation flowchart of an example for extracting the electrochemical time series sensing features corresponding to the water quality electrochemical data according to an embodiment of the present application is shown.

[0045] As Figure 2 shown, in step S210, the sliding time series data corresponding to the water quality electrochemical data is extracted based on a sliding window, and the size of the sliding window is dynamically adjusted according to the fluctuation characteristics of the time series data within the sliding window.

[0046] In some embodiments, in the data acquisition stage, an initial sliding window size (such as 10 sampling points or a certain time length) is set for each water quality electrochemical parameter (such as pH value, DO content, TOC content). Then, in each sliding window, the fluctuation characteristics of the water quality electrochemical data (such as the mean value, variance, fluctuation amplitude, etc.) are calculated in real time. According to the fluctuation characteristics of the data within the current sliding window, it is judged whether there is an obvious dynamic change in the water quality (such as a sudden rise or fall, etc.). If a significant change in the fluctuation characteristics is detected (such as a sudden change in the mean value or a sharp increase in the variance), the window size is appropriately reduced to more finely capture the details of water quality fluctuations in the next sampling window; on the contrary, when the data fluctuation is stable, the window size is enlarged to reduce the data volume and calculation complexity. In addition, the sliding window slides at a set step size (such as sliding 1 sampling point each time). Each time it slides, the old data within the window is removed, and the data of the new sampling point is added, so as to maintain the timeliness of the data within the window.

[0047] In step S220, for the sliding time-series data within each sliding window, the discrete wavelet transform method is used to decompose the sliding time-series data into multiple periodic amplitude features corresponding to different periodic amplitudes, and the respective periodic amplitude features are weighted and fused to obtain the corresponding window comprehensive amplitude feature.

[0048] Here, for the sliding time-series data within each sliding window (such as the time-series data segment composed of 10 sampling points of pH value), discrete wavelet transform (DWT) is performed. Through layer-by-layer decomposition, the discrete wavelet transform decomposes the time-series data into high-frequency and low-frequency components, and describes different characteristics of the data with different periodic amplitudes (i.e., the levels of wavelet decomposition). Among them, each decomposition level corresponds to a feature with a specific period and amplitude, which can capture different periodic changes and fluctuation patterns in the water quality time-series data. For example, the Daubechies wavelet basis function or the Haar wavelet basis function can be selected for decomposition, and the number of decomposition layers (such as 3 layers, 4 layers, etc.) is determined. After each layer of decomposition, the corresponding low-frequency (Approximation) and high-frequency (Detail) sub-band coefficients are obtained.

[0049] In addition, for the low-frequency and high-frequency sub-band coefficients at different decomposition levels, their amplitude features are calculated, including: maximum value, minimum value, mean value, variance, energy distribution (sum of squares), etc., to obtain the corresponding periodic amplitude features. The low-frequency sub-band coefficients mainly characterize the long-period trend changes of the sliding time-series data, and the high-frequency sub-band coefficients mainly characterize the short-period fluctuation characteristics of the sliding time-series data. By extracting multi-level periodic amplitude features, the water quality data can be comprehensively characterized at different time scales. Furthermore, for the high-frequency and low-frequency features obtained by different-level decompositions, the amplitude features of each period are extracted, and according to the target water quality monitoring requirements, different periodic amplitude features are weighted and fused. In addition, the weights can be set based on the importance of the features (such as increasing the weight for high-frequency fluctuation features to focus on data mutations) or determined by machine learning methods (such as automatic weight optimization). Finally, the respective periodic amplitude features are weighted and fused according to the weights to form the comprehensive amplitude feature of this sliding window, which can characterize the multi-level change patterns of the water quality data captured at different periodic amplitudes within the sliding window.

[0050] In step S230, the window comprehensive amplitude features corresponding to each consecutive sliding window are combined to determine the corresponding electrochemistry time-series sensing features.

[0051] In some embodiments, for the water quality electrochemistry parameters collected for each Internet of Things node, a series of window comprehensive amplitude features within a sliding window are obtained (such as multiple comprehensive amplitude feature sequences in chronological order). Furthermore, these consecutive window comprehensive amplitude features are combined in chronological order to form an overall time series feature sequence, which can reflect the fluctuation law and multi-scale amplitude features of the water quality electrochemistry data in the monitoring block over a period of time. Preferably, a feature dimensionality reduction method, such as Principal Component Analysis (PCA) or Linear Discriminant Analysis (LDA), can also be used on the combined time series feature sequence for dimensionality reduction processing to remove redundant information and retain the most representative time series features.

[0052] Through the embodiments of the present application, by adopting a feature extraction strategy that combines a dynamic sliding window with discrete wavelet transform, the fluctuation features and periodic change patterns of water quality data can be accurately extracted under the multi-level periodic amplitude within the window, effectively improving the feature expression ability and time series pattern recognition ability of water quality data. In addition, the dynamic adjustment of the sliding window size can adaptively adjust the window scale according to the fluctuation features of the water quality data, enhancing the sensitivity and robustness to the changes in water quality data, and providing a flexible and reliable technical means for data feature extraction in different water environments. Moreover, through the weighted fusion and feature combination strategy, the periodic amplitude features at multiple levels and different time scales are effectively fused, and the generated electrochemistry time series sensing features can better represent the overall change trend and time series dependence relationship of the water quality data, providing rich and accurate feature input information for the intelligent identification of water quality anomaly risks.

[0053] Regarding the details of the sliding window in step S210, the size of the sliding window directly affects the extraction accuracy and timeliness of the time series features. In some examples of the embodiments of the present application, a sliding window adjustment strategy based on the fluctuation amplitude and change rate of the time series data is provided.

[0054] More specifically, the size of the sliding window is dynamically adjusted in the following manner:

[0055] Based on the fluctuation features of the time series data of the previous sliding window, the initial window size of the current sliding window is determined.

[0056]

[0057] In the formula, w(t) 初始 represents the initial window size of the t-th sliding window, f(t - 1) and a(t - 1) respectively represent the fluctuation frequency and fluctuation amplitude determined through the time series analysis of the water quality parameters of the (t - 1)-th sliding window, and λ1 and λ2 are adaptive adjustment coefficients.

[0058] More specifically, by performing frequency domain analysis (such as Fast Fourier Transform FFT) on the water quality parameter values within the previous time window, the main frequency components are identified, and their average frequency value f(t - 1) is calculated. The specific calculation steps are as follows:

[0059] For the time series data X within the previous time window t-1 ={x t-1,1 ,x t-1,2 ,...,x t-1,w(t-1)}, perform FFT transformation to obtain the spectrum F t-1 ={F t-1,1 ,F t-1,2 ,...,F t-1,k}.

[0060] Calculate the energy values of each frequency component in the spectrum and select the frequency component with the largest energy value as the main frequency component f(t - 1) of the current fluctuation.

[0061] On the other hand, for the calculation of the fluctuation amplitude a(t - 1), the difference between the maximum value X t-1,max and the minimum value X t-1,min of the water quality parameter values within the previous time window can be calculated, which is defined as the fluctuation amplitude a(t - 1)=X t-1,max -X t-1,min .

[0062] In this way, by introducing the fluctuation frequency f(t - 1) and the fluctuation amplitude a(t - 1) of the previous time window, the current window size can be adjusted according to the historical fluctuation characteristics, thereby improving the prediction ability for short-term fluctuations and long-term trends.

[0063] Furthermore, in order to avoid unstable feature extraction caused by frequent fluctuations in the window size, a constraint mechanism for the change rate of the sliding window size is introduced in this paper, that is, if the change in the current sliding window size is too large, then a smooth adjustment strategy is switched to.

[0064] Specifically, calculate the change rate of the window size corresponding to the initial window size of the current sliding window.

[0065]

[0066] In the formula, V w (t) represents the change rate of the window size, and w(t - 1) represents the window size of the (t - 1)-th sliding window.

[0067] If the change rate of the window size of the current sliding window does not exceed the preset change rate threshold, then the initial window size is used as the window size of the current sliding window.

[0068] If the rate of change of the window size of the current sliding window exceeds the change rate threshold, the window size of the current sliding window will be smoothed and calculated by the following formula.

[0069] w(t) = w(t - 1) + η·(w(t) 初始 - w(t - 1)), Equation (3)

[0070] In the formula, w(t) represents the window size of the t-th sliding window, and η is the smoothing coefficient.

[0071] It should be noted that during the actual acquisition process of water quality electrochemical data, water quality parameters (such as pH value, DO content, and TOC content) are often affected by fluctuations in environmental factors and pollutant concentrations, showing complex time series characteristics. Therefore, the sliding window needs to have high sensitivity and response ability to these fluctuation characteristics.

[0072] In the embodiments of the present application, the initial size of the current time window is deduced by calculating the fluctuation frequency and fluctuation amplitude of the previous time window, and on this basis, the window size is further smoothed and adjusted through the control of the change rate, realizing the introduction of a dynamic adjustment mechanism for the window size based on the fluctuation characteristics of the historical window, enabling the current size to smoothly transition to a new state and avoiding sudden changes in the window size caused by fluctuations at a single moment. In addition, by introducing a constraint on the change rate of the window size, when the change rate exceeds the preset threshold, the system can automatically smooth the window size, further ensuring the continuity and stability of the change in the sliding window size, thereby reducing the volatility of the feature extraction results. Thus, by adjusting the current window size according to the fluctuation frequency and fluctuation amplitude of the previous time window, the sliding window can adaptively capture short-term fluctuation characteristics and long-term trend changes, enhancing the response ability of the sliding window to water quality fluctuations, thereby improving the accuracy of feature extraction.

[0073] Regarding the details in step S220, the discrete wavelet transform (DWT) is used to decompose the time series data within the current sliding window, and the time series data is decomposed into feature representations on different periodic scales (i.e., different frequency components). It should be noted that compared with other time-frequency analysis methods, such as the short-time Fourier transform (STFT), the discrete wavelet transform has stronger time-domain and frequency-domain localization capabilities, and can perform more accurate feature extraction on the short-term fluctuations and long-term trends of water quality parameters at different time scales.

[0074] More preferably, in the time series feature extraction of water quality electrochemical data, a time series feature modeling method based on periodic analysis is adopted to perform multi-scale decomposition on the time series data within the adaptive sliding window through the discrete wavelet transform method, so as to extract time series features at different periodic amplitudes.

[0075] Specifically, the sliding time series data of the current sliding window is subjected to discrete wavelet transform to obtain multiple sub-band representations.

[0076]

[0077] Among them, the time series data within the t-th sliding window is represented as x = {x1, x2,..., x w(t)}, and x is subjected to discrete wavelet transform to obtain multiple sub-band representations X s = {X s1 , X s2 ,..., X sm}, where m is the number of sub-bands, and each sub-band X si represents the characteristic component under the i-th cycle amplitude s i ; x j represents the value of the water quality electrochemistry parameter at the j-th time step in x, and X si (t) represents the characteristic component of the t-th sliding window under the cycle amplitude s i ; ψ si (j) represents the wavelet basis function of the t-th sliding window under the cycle amplitude s i ; the wavelet basis functions under different cycle amplitudes can capture the characteristics of the time series data at different frequencies; j represents the time step offset, which is used to describe the weight of the characteristic value at the j-th time step of the t-th sliding window.

[0078] Then, calculate the energy values of the characteristic components of each cycle amplitude. Here, in order to measure the significance of the characteristics of each cycle scale, the energy value is introduced as an important metric for the cycle characteristics.

[0079]

[0080] In the formula, E si (t) represents the energy value of the sub-band characteristics under the cycle amplitude s i within the t-th sliding window, which can reflect the importance of the characteristic components under the current cycle amplitude; represents the sum of the squares of the characteristic values of the sub-band X si at all time steps within the t-th sliding window.

[0081] Furthermore, based on the significance metric, calculate the comprehensive cycle significance of the characteristic components of each cycle amplitude. The significance metric includes the periodic intensity and the relative energy ratio.

[0082] In the embodiments of the present application, on the basis of the basic energy value measurement, a multi-dimensional measurement index based on the significance of periodic characteristics is introduced, including periodic intensity and relative energy proportion. Here, the periodic intensity represents the significance of the periodic characteristics within the current time window, and is defined as the product of the feature fluctuation amplitude and the energy value of the frequency component. In addition, the relative energy proportion represents the relative importance of the periodic characteristics in all periodic scales.

[0083] θ si (t) = l1·P si (t) + l2·REP si (t), Equation (6)

[0084] P si (t) = A si (t)·E si (t), Equation (7)

[0085] A si (t) = max{X si (j)} - min{X si (j)}, Equation (8)

[0086]

[0087] In the formula, P si (t) represents the periodic intensity under the periodic amplitude s in the t-th sliding window; A i (t) represents the fluctuation amplitude of the sub-band characteristics under the periodic amplitude s in the t-th sliding window, which is defined as the difference between the maximum value max{X si (j)} and the minimum value min{X i (j)} of the sub-band characteristic values; REP si (t) represents the relative energy proportion of the feature components under the periodic amplitude s in the t-th sliding window; si (t) represents the sum of the energy values of the sub-band characteristics of all periodic amplitudes in the t-th sliding window, which is a normalization term; θ si (t) represents the comprehensive periodic significance of the periodic amplitude s in the t-th sliding window, and l1 and l2 respectively represent the weight coefficients of the corresponding significance measurement indexes. i It should be noted that when the energy value of a certain periodic amplitude accounts for a large proportion in the overall energy value, it indicates that the periodic characteristic has a high importance in the current time window. si (t) represents the comprehensive periodic significance of the periodic amplitude s in the t-th sliding window; i

[0088] Through multi-scale decomposition of the time series data by discrete wavelet transform, feature components can be extracted at different periodic scales, and through the periodic intensity P si

[0089] si ​(t) and Relative Energy Proportion (REP) si (t) measures the significance of periodic features, thus accurately capturing the complex fluctuation patterns of water quality parameters within the current sliding window.

[0090] Furthermore, based on the comprehensive periodic significance, the amplitude features of each period are weighted and fused to obtain the corresponding window comprehensive amplitude feature.

[0091]

[0092] In the formula, F(t) represents the window comprehensive amplitude feature corresponding to the t-th sliding window.

[0093] Here, after fusing the feature components at all period scales, the generated comprehensive feature representation F(t) can reflect the overall fluctuation and periodic change characteristics of water quality parameters within the current sliding window, enhancing the feature representation ability. Thus, it can not only reflect the main fluctuation patterns in water quality data but also reveal the periodic change laws hidden in the data, enabling more accurate identification of potential water quality abnormal states during anomaly detection.

[0094] It should be noted that the temporal variation of water quality parameters is often affected by environmental conditions (such as water temperature, water flow velocity, external pollutant concentration, etc.) and internal system states (such as monitoring equipment drift, data sampling frequency change, etc.). Traditional fixed window or single-scale feature extraction methods are difficult to adapt to the dynamic changes in these complex environments.

[0095] Through the embodiments of this application, an adjustment strategy of an adaptive sliding window is introduced, and the size of the current sliding window is dynamically determined according to the fluctuation characteristics of the previous time window, so that the sliding window can adaptively adjust following the fluctuation of water quality parameters. Furthermore, DWT is used to perform multi-scale decomposition on the temporal data within the sliding window, and through the energy value calculation and significance weight assignment strategy, the temporal features can be modeled and optimized at different period scales. The comprehensive scale feature representation of each sliding window can simultaneously reflect short-period features (such as instantaneous fluctuations, sudden anomalies) and long-period features (such as overall trend changes, long-term pollutant accumulation effects), enhancing the system's adaptability to various feature patterns in complex environments.

[0096] Figure 3 Shows a schematic structural connection diagram of an example of a water quality anomaly recognition model according to an embodiment of this application.

[0097] As Figure 3 shown, the water quality anomaly recognition model 300 includes a cascaded temporal feature modeling module 310 and a feature anomaly recognition module 320.

[0098] The timing feature modeling module 310 is used to model the input electrochemistry timing sensing features, extract the dynamic change patterns and time-dependent relationships of the features for the monitored water area block, so as to obtain the corresponding pattern timing-dependent features.

[0099] In some embodiments, the timing feature modeling module 310 first performs normalization processing on the input data (such as normalizing each feature value to the interval of 0 to 1) to eliminate the dimensional differences between the electrochemistry features. Then, various neural network models suitable for timing data modeling, such as LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), or ARIMA (Autoregressive Integrated Moving Average), can be used to model the input electrochemistry timing sensing features. Exemplarily, LSTM and GRU have the ability to process long timing-dependent relationships and can effectively capture the dynamic patterns and internal dependencies of the water quality electrochemistry features changing with time. For example, through the LSTM network, the lag effect of the change in pH value on the change in DO content, or whether the increase in TOC content will cause the decrease in DO content can be identified. Furthermore, the hidden layer features (such as the hidden state vector of the neural network) extracted during the timing feature modeling process can be regarded as pattern timing-dependent features to reflect the dynamic change trend and periodic features of the water quality features in the current monitored block.

[0100] The feature anomaly recognition module 320 is used to process the pattern timing-dependent features to predict whether there is a water quality anomaly risk in the corresponding monitored water area block.

[0101] Here, an anomaly recognition model suitable for processing pattern temporal dependence features is constructed. For example, traditional machine learning algorithms (such as SVM (Support Vector Machine), RF (Random Forest)) or deep learning models (such as MLP (Multilayer Perceptron), CNN (Convolutional Neural Network)) can be used. Exemplarily, the labeled data used in the data sample set of the anomaly recognition model comes from the water quality anomaly records of historical monitoring blocks, and supervised learning or semi-supervised learning (if some data is unlabeled) is performed according to the actual data of various water quality anomaly records. When training the model, special attention is paid to the relationship between temporal dependence features and water quality anomaly records. For example, when there is a sudden sharp fluctuation in water quality features (such as a sudden drop in pH), whether there is a similar pattern to a certain type of water quality anomaly in historical data. In this way, in the inference application stage, inputting the pattern temporal dependence features into the trained anomaly recognition model can better predict whether there is a risk of water quality anomaly in the current monitoring block.

[0102] Preferably, in some examples of the embodiments of the present application, the temporal feature modeling module adopts a combination of a deep neural network and time series analysis, so as to effectively handle water quality anomaly situations in complex environments.

[0103] Figure 4 FIG. shows a schematic structural connection diagram of another example of the water quality anomaly recognition model according to the embodiments of the present application.

[0104] As Figure 4 shown, the water quality anomaly recognition model 400 includes a temporal feature modeling module 410 and a feature anomaly recognition module 420. The temporal feature modeling module 410 includes an LSTM layer 411, a TCN (Temporal Convolutional Network) layer 412, and a feature fusion layer 413. The LSTM layer 411 and the TCN layer 412 are connected in parallel to the feature fusion layer 413. The feature anomaly recognition module 420 includes an MLP layer 421, a Softmax layer 422, and an anomaly classification layer 423.

[0105] The LSTM layer 411 is used to extract long-term trend features in the electrochemistry time series sensing features through a gating structure to capture the trend changes and long-term dependence relationships of water quality parameters.

[0106] h g,t =σ1(W h ·[h g,t-1 ,V g,t +b h ), Equation (11)

[0107] H g = [h g,1 , h g,2 ,..., h g,T , Equation (12)

[0108] In the formula, h g,t represents the hidden state output by the LSTM layer for the monitored water area block at the t-th sliding window, W h represents the weight matrix of the LSTM layer, h g,t-1 represents the hidden state output by the LSTM layer for the monitored water area block at the (t - 1)-th sliding window; V g,t represents the electrochemical input feature vector of the monitored water area block at the t-th sliding window, b h represents the bias term, σ1(·) represents the activation function of the LSTM layer; T represents the total number of time steps corresponding to the electrochemical time series sensing features, H g represents the long-term trend feature extracted by the LSTM layer for the monitored water area block .

[0109] The LSTM layer can effectively capture the long-term dependence relationships in time series data through its internal gating structure and is suitable for extracting the trend changes in time series features. Therefore, for each monitored water area block, the hidden state outputs of the LSTM at all time steps constitute its time series feature representation sequence, which contains the long-term dependence relationships in the electrochemical time series data and can reflect the trend changes and long-term fluctuation patterns of water quality parameters.

[0110] The TCN layer 412 extracts the local fluctuation pattern features in the electrochemical time series sensing features by stacking multiple convolutional operations to capture the short-term fluctuations and local change patterns of water quality parameters.

[0111] y g,t = W p * V g,t + b p , Equation (13)

[0112] Y g = [y g,1 , y g,2 ,..., y g,T , Equation (14)

[0113] In the formula, y g,t represents the convolutional output feature of the TCN layer for the monitored water area block at the t-th sliding window, W p and b prespectively represent the convolutional kernel weights and bias terms of the TCN layer, and * represents the one-dimensional convolution operation; Y g represents the local fluctuation pattern features extracted by the LSTM layer for the monitored water area block extracted by the LSTM layer for the monitored water area block

[0114] Here, the TCN extracts multi-scale features of the features through multiple convolutional operations. Further, by stacking multiple convolutional operations, the TCN can gradually extract multi-scale features in the time series data and splice the outputs of all convolutional layers into a local feature representation.

[0115] The feature fusion layer 413 is used to perform weighted fusion on the output results of the LSTM layer and the output results of the TCN layer to obtain the corresponding pattern time series dependence features.

[0116] F g = α·H g +(1 - α)·Y g , Equation (15)

[0117] In the formula, F g represents the pattern time series dependence features for the monitored water area block , and α is the feature fusion coefficient.

[0118] In this way, based on the weighted fusion feature fusion strategy, the long-term feature representation output by the LSTM and the local feature representation output by the TCN are weighted and fused to realize the long-term and short-term modeling of the time series features, and can effectively extract the dynamic change patterns and time dependence relationships of each monitored block.

[0119] On the other hand, the feature anomaly recognition module 420 is used to process the pattern time series dependence features through the first MLP layer 421 and predict the probability value of the risk of water quality anomaly in the monitored water area block through the Softmax layer 422.

[0120]

[0121] In the formula,[[]] represents the output feature representation of the f-th hidden layer for the monitored water area block , W (f) and b (f) represent the weight matrix and bias term of the f-th hidden layer; represents the output feature representation of the (f - 1)-th hidden layer for the monitored water area block β g , and in the calculation of the first hidden layer P g represents the anomaly probability of the risk of water quality anomaly in the monitored water area block , L1 is the total number of hidden layers of the first MLP layer, and σ2(·) represents the activation function of the hidden layer, and respectively represent the weight matrix and bias term of the Softmax layer, represent the feature representation output by the last hidden layer of the first MLP layer for the monitored water area block where Softmax(·) represents the Softmax activation function.

[0122] The anomaly classification layer 423 is used to compare the predicted probability value of water quality anomaly risk with a preset anomaly probability threshold, and determine whether there is a water quality anomaly risk in the corresponding monitored water area block according to the comparison result.

[0123] Here, the extracted features are further integrated through the feature anomaly recognition module, and the anomaly risk probability of each monitored block is output. By comparing with the threshold, it is judged whether there is a water quality anomaly risk in the monitored block. Through the combined model based on deep temporal feature modeling and feature anomaly recognition module provided by the embodiments of the present application, it focuses on modeling the temporal features of each monitored block and identifying the abnormal state, which belongs to the detection stage. Through the deep temporal feature modeling module (LSTM and TCN), dynamic feature extraction is performed on the temporal data to identify the long-term dependence relationship and local fluctuation pattern in the data; then, through the feature anomaly recognition module, feature fusion and anomaly detection are performed to judge whether there is a water quality anomaly state in each monitored water area block. Thus, it can accurately capture the abnormal change pattern in the temporal data and output the abnormal state (i.e., there is an anomaly or no anomaly) of each monitored block.

[0124] It should be noted that in the actual water quality monitoring scenario, the chemical components of water bodies (such as pH value, DO content, and TOC content) are affected by various complex factors, such as water flow changes, pollution source diffusion, temperature fluctuations, etc. These factors often cause the water quality parameters to show different scale change patterns in time, including two characteristics of long-term trends (such as seasonal changes) and short-term fluctuations (such as sudden pollutant injection).

[0125] Through the embodiments of the present application, an LSTM layer and a TCN layer are used for parallel modeling in the water quality anomaly recognition model. The LSTM layer can effectively capture the long-term trend characteristics in the electrochemical temporal sensing features and extract the long-term dependence relationship of water quality parameters changing with time, such as the gradual change trend of the water body pH value over multiple days. In addition, through the TCN layer, local fluctuation patterns in the electrochemical temporal sensing features can be extracted through multi-layer convolution operations, and short-term fluctuations (such as sudden increase in TOC content within a certain time period) and local change patterns in water quality parameters can be captured. Thus, through the parallel temporal feature extraction method, not only can the model better understand the complex changes of water quality data, but also the problem of insufficient feature representation caused by only using a single model can be avoided, significantly improving the sensitivity and detection accuracy of the water quality anomaly state.

[0126] In addition, in actual water quality monitoring, different types of water quality anomalies (such as sudden changes in pH value, decrease in DO content, excessive TOC content, etc.) will exhibit different temporal characteristic patterns. By using the MLP layer in the water quality anomaly recognition model to perform non-linear mapping and transformation on the fused features, and by using the Softmax layer and the classification layer to refine the judgment of the water quality anomaly probability, the recognition ability and sensitivity for various types of water quality anomalies can be significantly improved.

[0127] Figure 5 A schematic structural connection diagram of an example of an anomaly event recognition model according to an embodiment of the present application is shown.

[0128] As Figure 5 shown, the anomaly event recognition model 500 includes a cascaded multi-scale convolution module 510 and an event type classification module 520. The event type classification module 520 includes an attention layer 521, a second MLP layer 522, a confidence calculation layer 523, and an anomaly event classification layer 524.

[0129] The multi-scale convolution module 510 is used to extract multi-scale temporal features corresponding to the pattern temporal dependence features of the risk monitoring water area block.

[0130] Here, the multi-scale convolution module is used to process the input pattern temporal dependence features, and by designing convolution kernels of multiple time scales, multi-scale convolution operations are used to model and fuse the features with different time lengths or scales, so as to obtain a richer multi-scale temporal feature representation, capture the fluctuations and trend information within different time lengths in the features, and thus improve the model's recognition ability for complex water quality anomaly event types.

[0131]

[0132] In the formula, represents the convolution output feature representation of the monitoring block at the e-th convolution kernel scale, which contains the local fluctuation and trend features at this convolution kernel scale; and respectively represent the convolution kernel weight matrix and the bias term at the e-th convolution kernel scale; F D represents the pattern temporal dependence feature of the risk monitoring water area block ; σ3(·) represents the activation function of the multi-scale convolution module; M D represents the comprehensive multi-scale feature representation of the risk monitoring water area block ; μ e represents the scale fusion weight at the e-th convolution kernel scale, and U represents the total number of convolution kernel scales in the multi-scale convolution module.

[0133] It should be noted that in actual water quality monitoring scenarios, water quality anomaly events (such as chemical pollution, excessive organic matter, heavy metal pollution, etc.) are often accompanied by the interference of various complex factors, such as water flow changes, external pollution source diffusion, etc. These factors make the anomaly features in water quality data present various temporal variation patterns and spatial distribution characteristics. By adopting a multi-scale convolutional feature extraction and fusion strategy, the model can dynamically model the temporal dependence features of the input patterns at different time scales, generate more accurate multi-scale feature representations, and thus significantly improve the recognition accuracy of the model for various types of water quality anomaly events.

[0134] The attention layer 521 is used to calculate the weight assignment matrix for the multi-scale temporal features to determine the corresponding attention features.

[0135]

[0136] In the formula, represents the attention weight for the e-th convolutional kernel scale of the risk monitoring water area block ; and represent the attention weight matrix and bias term for the e-th convolutional kernel scale; represents the attention features output by the attention layer for the risk monitoring water area block .

[0137] It should be noted that since different types of water quality anomaly events often present different distribution patterns in the feature space, in view of this, through the embodiments of the present application, the feature weight assignment strategy based on the attention mechanism can perform weighted assignment on the feature representations of each time scale to highlight the feature representations with high discrimination for specific event types. The weight assignment matrix calculated through the attention mechanism can dynamically adjust the importance of each feature in the feature representations of different time scales, thereby enhancing the recognition ability of the model for different types of water quality anomaly events.

[0138] In addition, by adopting a multi-scale convolutional feature extraction and feature fusion strategy, the model can independently model the features at different time scales respectively in the feature extraction stage, and then generate a comprehensive feature representation through the dynamic weighted combination of feature fusion and the attention mechanism. Through the model structure design of hierarchical modeling and feature combination, the complexity of the model's feature representation can be effectively reduced, thereby improving the model's feature representation ability and training efficiency.

[0139] The second MLP layer 522 is used to perform non-linear mapping and feature transformation processing on the attention features, and determine the confidence matrix of the risk monitoring water area block for the preset risk event type set through the confidence calculation layer 523.

[0140]

[0141] In the formula, represents the output feature representation of the r-th hidden layer for the risk monitoring water area block ; W (r) and b (r) represent the weight matrix and bias term of the r-th hidden layer; represents the output feature representation of the (r - 1)-th hidden layer for the risk monitoring water area block , and in the calculation of the first hidden layer L2 is the total number of hidden layers of the second MLP layer, and σ4(·) represents the activation function of the hidden layer, represents the comprehensive risk feature representation output by the last hidden layer of the second MLP layer for the risk monitoring water area block ; and respectively represent the weight matrix and bias term of the confidence calculation layer, and P D represents the confidence matrix of the risk monitoring water area block , which contains the confidence scores for each risk event type in the risk event type set.

[0142] Here, the feature representation based on the attention mechanism is input into the second MLP layer. Through the layer-by-layer calculation of the second MLP layer, the feature representation can be mapped into a more compact feature representation to achieve the classification and judgment of event types.

[0143] The abnormal event classification layer 524 is used to process the confidence matrix to identify the water quality abnormal risk event type corresponding to the risk monitoring water area block.

[0144] Here, the processing method for the confidence matrix can be diversified and can be selected or adjusted according to business requirements. For example, the highest confidence selection strategy for outputting a single event and the confidence threshold comparison strategy for allowing the output of multiple types of events, etc.

[0145] Through the embodiments of the present application, based on the principle of the distribution pattern differences of different water quality abnormal event types in the feature space, by introducing the attention mechanism, the model can effectively strengthen the selection and weighting of features at specific time scales, thereby improving the discrimination ability and classification accuracy for different event types. In addition, in the face of the situation that the water quality data of different monitoring blocks may show a highly unbalanced and complex distribution pattern, by introducing multi-scale feature extraction and the attention mechanism, the model can dynamically adapt to different data distribution patterns and refine the prediction results in combination with the confidence classification module, effectively reducing the probability of false alarms and missed reports.

[0146] It should be noted that in the embodiments of the present application, by using two independent intelligent models to identify water quality anomalies, the water quality anomaly identification model focuses on the anomaly risk detection task, while the anomaly event identification model focuses on the prediction task of specific anomaly risk events. They can be refined and optimized respectively according to their specific tasks, enabling more efficient design, optimization, and maintenance of the models. In addition, different algorithms and strategies are adopted in the detection stage and the prediction stage respectively to improve the adaptability and flexibility of the models. For example, different time series modeling networks are used in the detection stage, while different classification and time series prediction algorithms are used in the prediction stage to achieve algorithm decoupling of detection and prediction and improve the overall system performance.

[0147] It should be understood that in the embodiments of the present application, the types of water quality anomaly risk events included in the risk event type set can be diverse. In addition to chemical pollution, organic matter exceeding the standard, and heavy metal pollution described above, other risk event types can also be included, such as trend-based water quality deterioration events, sudden pollution events, and long-term stability deviation events.

[0148] Specifically, a trend-based water quality deterioration event refers to a continuous deterioration trend of water quality characteristics over a relatively long time period (such as several weeks or months), rather than a sudden anomaly. Such events usually reflect chronic pollution of the water environment or a gradually deteriorating trend of water quality. A sudden pollution event refers to a sudden and significant change in a certain water quality characteristic (or multiple characteristics) within a short time (such as several hours or one day), and the change significantly exceeds the normal fluctuation range. Such events are usually caused by sudden pollution sources (such as factory emissions, sudden leakage accidents, etc.). A long-term stability deviation event refers to a certain water quality characteristic (such as pH value, DO content) continuously and stably deviating from the normal range over a long time, but without an obvious trend-based change. Such events are usually related to stable pollution sources (such as groundwater pollution) or insufficient water self-purification ability.

[0149] By adopting modules such as multi-layer feature modeling, multi-scale time series feature fusion, and attention mechanism, the anomaly event identification model can accurately identify various types of water quality anomaly risk events in complex water quality monitoring scenarios, effectively solving the limitations of traditional models that cannot handle multi-dimensional complex features and cannot identify compound and trend-based water quality events, enabling the model to comprehensively analyze the complex relationships and dynamic change patterns among multiple water quality parameters, thereby identifying more complex and diverse risk event types and achieving a more comprehensive and accurate risk event identification ability.

[0150] In some examples of the embodiments of the present application, the loss function of the abnormal event recognition model is used to measure the performance of the model when processing data samples of multiple risk monitoring water area blocks in a complex water quality monitoring scenario, so as to improve its classification accuracy, time series feature modeling ability and feature allocation ability in a complex environment through optimized training of the model.

[0151] Here, the data sample set contains multiple data samples for risk monitoring water area blocks, and each data sample contains feature data and corresponding labels. Each data sample contains multiple water quality electrochemistry parameter features (such as pH value, DO content, TOC content, etc.), and may be provided with spatio-temporal information or identifiers related to the monitoring block. The label of each data sample represents the type of water quality abnormal risk event to which it belongs.

[0152] Specifically, the abnormal event recognition model adopts a comprehensive loss function including classification cross-entropy loss, adaptive balance loss term and time series consistency loss term. Specifically, the loss function is:

[0153]

[0154]

[0155] In the formula, represents the loss function of the abnormal event recognition model, represents the classification cross-entropy loss term, represents the adaptive balance loss term, and represents the time series consistency loss term. β1, β2, and β3 respectively represent the weight coefficients of the corresponding loss terms; N represents the total number of samples in the risk data sample set for the risk monitoring water area, C represents the total number of water quality abnormal risk event types in the risk event type set, and y d,c represents the actual label of the d-th risk data sample on the c-th risk event type. If the block belongs to the c-th event type, then y d,c =1, otherwise y d,c =0; p d,c represents the confidence that the d-th risk data sample is predicted by the abnormal event recognition model as the c-th risk event type; γ is an adaptive balance coefficient used to adjust the balance relationship between feature weights and classification confidence; w d,c represents the feature weight allocation value of the d-th risk data sample on the c-th risk event type; Q represents the total number of time steps of the multi-scale time series feature, represents the feature representation of the d-th risk data sample at the q-th time step on the e-th convolution kernel scale; represents the feature representation of the d-th risk data sample at the q-1-th time step on the e-th convolution kernel scale.

[0156] In the comprehensive loss function provided by the embodiments of the present application, the categorical cross-entropy loss can measure the classification accuracy of the model for each event type by comparing with the label data of the data sample set, and improve the fitting ability of the model to the data sample set through optimization, so as to improve the classification performance of the model for different risk event types in practical applications. The adaptive balance loss based on feature weights and confidence can measure the balance relationship between feature weight allocation and classification confidence. By combining the feature data and label data in the data sample set, it can improve the dynamic weight allocation ability of the model for each data sample, thus enhancing the dynamic allocation ability of the model for sample features. The temporal consistency loss based on multi-scale temporal feature modeling measures the feature consistency of the model at different time scales. By combining with the temporal feature data in the data sample set, it ensures that the model's judgments on the same risk event at multiple time steps are consistent, and can help the model better understand the feature relationships at different time steps.

[0157] It should be noted that different from the design of traditional single loss functions, the comprehensive loss function provided by the embodiments of the present application introduces an adaptive balance mechanism and temporal consistency constraints. The model can maintain a high generalization ability in a complex and changeable water quality monitoring environment. When dealing with complex risk events (such as compound pollution events and long-term trend events), the model can adaptively adjust feature weights and temporal consistency according to the distribution characteristics of multi-dimensional water quality features, helping the model to still maintain a high classification accuracy and stability in different monitoring environments (such as environments with large fluctuations in water quality parameters and areas with complex and changeable pollution sources), thereby effectively improving the generalization ability and robustness of the model in complex environments.

[0158] Through the embodiments of the present application, the comprehensive loss function can, on the basis of ensuring classification accuracy, simultaneously optimize classification accuracy, feature weight allocation and temporal consistency in the same training process, avoid the problem of unbalanced model performance that may be brought by a single loss term, improve the comprehensive performance and overall performance of the model, and can significantly enhance the comprehensive ability of the model in multi-type event recognition, composite feature representation and multi-time scale modeling, providing more comprehensive and powerful technical support for the risk warning of the intelligent water quality monitoring system.

[0159] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of actions combined. However, those skilled in the art should know that this application is not limited by the described action sequence, because according to this application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application. In the above embodiments, the descriptions of each embodiment have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0160] Figure 6 FIG. 4 shows a structural block diagram of an example of an Internet of Things intelligent monitoring and alarm system for water quality detection according to an embodiment of the present application.

[0161] As Figure 6 shown, the Internet of Things intelligent monitoring and alarm system 600 for water quality detection includes a water quality parameter sampling unit 610, a timing feature extraction unit 620, a water quality anomaly identification unit 630, an anomaly event identification unit 640, and a water area water quality alarm unit 650.

[0162] The water quality parameter sampling unit 610 is used to sample the water quality electrochemical data for the corresponding monitored water area block based on each Internet of Things node; a plurality of Internet of Things nodes are distributed in different blocks in the monitored water area, and each of the Internet of Things nodes is equipped with an electrochemical sensor; the sensing parameter types of the water quality electrochemical data include the pH value, DO content, and TOC content of the water body.

[0163] The timing feature extraction unit 620 is used to extract the electrochemical timing sensing features corresponding to each of the water quality electrochemical data.

[0164] The water quality anomaly identification unit 630 is used to input each of the electrochemical timing sensing features into the water quality anomaly identification model, extract the pattern timing dependence features corresponding to the electrochemical timing sensing features, so as to predict whether there is a water quality anomaly risk in the corresponding monitored water area block.

[0165] The anomaly event identification unit 640 is used to input the pattern timing dependence features corresponding to the risk monitoring water area block into the anomaly event identification model for each risk monitoring water area block predicted to have a water quality anomaly risk, so as to identify the type of water quality anomaly risk event corresponding to the risk monitoring water area block.

[0166] The water area water quality alarm unit 650 is used to perform a water area water quality pollution alarm operation according to the block position of each of the risk monitoring water area blocks and the corresponding type of water quality anomaly risk event.

[0167] In some embodiments, the embodiments of the present application provide a non-volatile computer-readable storage medium, in which one or more programs including execution instructions are stored, and the execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to perform the steps of any one of the above-mentioned Internet of Things intelligent monitoring and alarm methods for water quality detection of the present application.

[0168] In some embodiments, the embodiments of the present application further provide a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium. The computer program includes program instructions, and when the program instructions are executed by a computer, the computer is enabled to execute the steps of any one of the above-mentioned Internet of Things intelligent monitoring and alarm methods for water quality detection.

[0169] In some embodiments, the embodiments of the present application further provide an electronic device, which includes: at least one processor, and a memory communicatively connected to the at least one processor. Among them, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the steps of the Internet of Things intelligent monitoring and alarm method for water quality detection.

[0170] Figure 7 FIG. [ID] is a schematic hardware structure diagram of an electronic device for executing the Internet of Things intelligent monitoring and alarm method for water quality detection provided by another embodiment of the present application. As Figure 7 shown, the device includes:

[0171] One or more processors 710 and a memory 720. Figure 7 Here, one processor 710 is taken as an example.

[0172] The device for executing the Internet of Things intelligent monitoring and alarm method for water quality detection may further include: an input device 730 and an output device 740.

[0173] The processor 710, the memory 720, the input device 730, and the output device 740 may be connected through a bus or other means. Figure 7 Here, taking the connection through a bus as an example.

[0174] The memory 720, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the Internet of Things intelligent monitoring and alarm method for water quality detection in the embodiments of the present application. The processor 710 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions, and modules stored in the memory 720, that is, implements the Internet of Things intelligent monitoring and alarm method for water quality detection in the above method embodiments.

[0175] The memory 720 may include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the electronic device, etc. In addition, the memory 720 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 720 may optionally include a memory remotely provided relative to the processor 710, and these remote memories can be connected to the electronic device through a network. Examples of the above network include but are not limited to the Internet, enterprise intranets, local area networks, mobile communication networks, and their combinations.

[0176] The input device 730 can receive input digital or character information, and generate signals related to the user settings and function controls of the electronic device. The output device 740 may include a display device such as a display screen.

[0177] The one or more modules are stored in the memory 720, and when executed by the one or more processors 710, implement the Internet of Things intelligent monitoring and alarm method for water quality detection in any of the above method embodiments.

[0178] The above product can execute the method provided by the embodiments of the present application, and has the corresponding functional modules and beneficial effects for executing the method. For technical details not described in detail in this embodiment, reference can be made to the method provided by the embodiments of the present application.

[0179] The electronic devices in the embodiments of the present application exist in various forms, including but not limited to:

[0180] (1) Mobile communication devices: These devices are characterized by having mobile communication functions and mainly aiming to provide voice and data communication. Such terminals include: smart phones, multimedia phones, functional phones, and low-end phones, etc.

[0181] (2) Ultra-mobile personal computer devices: These devices belong to the category of personal computers, have computing and processing functions, and generally also have the characteristic of mobile Internet access. Such terminals include: PDAs, MIDs, and UMPC devices, etc.

[0182] (3) Portable entertainment devices: Such devices can display and play multimedia content. This type of device includes: audio and video players, handheld game consoles, e-books, as well as smart toys and portable in-vehicle navigation devices.

[0183] (4) Other airborne electronic devices with data interaction functions, such as in-vehicle device installed on a vehicle.

[0184] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0185] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the above technical solution, in essence, or the part that contributes to the related technology can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0186] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present application.

Claims

1. An IoT intelligent monitoring and alarm method for water quality detection, comprising: Based on each IoT node, water quality electrochemical data for the corresponding monitored water area is sampled; A plurality of IoT nodes are distributedly arranged in different blocks in the monitored water area, and each IoT node is deployed with an electrochemical sensor; the sensing parameter types of the water quality electrochemical data include pH value, DO content and TOC content of the water body; Extracting the electrochemical time-series sensing features corresponding to each of the water quality electrochemical data; Input each of the electrochemical time-series sensing features into a water quality anomaly recognition model, extract the pattern time-series dependency features corresponding to the electrochemical time-series sensing features, and predict whether there is a risk of water quality anomaly in the corresponding monitored water area; For each risk monitoring water area block predicted to have abnormal water quality risk, the pattern time-series dependency feature corresponding to the risk monitoring water area block is input into the abnormal event recognition model to identify the type of abnormal water quality risk event corresponding to the risk monitoring water area block; Execute water quality pollution alarm operation in the water area according to the block identification of each risk monitoring water area block and the corresponding water quality abnormal risk event type; The step of extracting the electrochemical time-series sensing features corresponding to each of the water quality electrochemical data comprises: Extracting sliding time series data corresponding to the water quality electrochemical data based on a sliding window; wherein the size of the sliding window is dynamically adjusted according to the fluctuation characteristics of the time series data within the sliding window; For the sliding time series data in each sliding window, the sliding time series data is decomposed into a plurality of period amplitude features corresponding to different period amplitudes by using a discrete wavelet transform method, and each of the period amplitude features is weightedly fused based on the comprehensive period significance to obtain the corresponding window comprehensive amplitude feature; wherein, based on the significance measurement index, the comprehensive period significance of each period amplitude feature is calculated, and the significance measurement index includes periodic intensity and relative energy proportion; The window comprehensive amplitude characteristics corresponding to each continuous sliding window are combined to determine the corresponding electrochemical timing sensing characteristics.

2. The method according to claim 1, wherein: The size of the sliding window is dynamically adjusted in the following way: According to the fluctuation characteristics of the time series data of the previous sliding window, determine the initial window size of the current sliding window: ， In the formula, Indicates The initial window size of the sliding window, and Respectively, through The fluctuation frequency and amplitude determined by the time series analysis of water quality parameters in a sliding window, and is the adaptive adjustment coefficient; Calculate the window size change rate corresponding to the initial window size of the current sliding window: ， In the formula, Indicates the rate at which the window size changes. Indicates The window size of the sliding window; If the window size change rate of the current sliding window does not exceed the preset change rate threshold, the initial window size is used as the window size of the current sliding window; as well as If the window size change rate of the current sliding window exceeds the change rate threshold, the window size of the current sliding window will be smoothly calculated by the following formula: ， In the formula, Indicates The window size of the sliding window, is the smoothing coefficient.

3. The method according to claim 2, wherein: The sliding time series data in each sliding window is decomposed into a plurality of period amplitude features corresponding to different period amplitudes by using a discrete wavelet transform method, and each of the period amplitude features is weightedly fused to obtain a corresponding window comprehensive amplitude feature, including: Perform discrete wavelet transform on the sliding time series data of the current sliding window to obtain multiple sub-band representations: ， Among them, The time series data in a sliding window is expressed as ,Will Perform discrete wavelet transform to obtain multiple sub-band representations , is the number of subbands, each subband Indicates i Cycle Amplitude The characteristic component below: express Middle The electrochemical parameter values ​​of water quality in time steps, Indicates A sliding window with a period amplitude The characteristic component below; Indicates The periodic amplitude of the sliding window Wavelet basis functions under ; Represents the time step offset, used to describe the Sliding window The weight of the feature value at each time step; Calculate the energy value of each period amplitude feature: ， In the formula, Indicates The period amplitude in the sliding window Energy value of the lower subband feature; Indicates subband In the The sum of squares of eigenvalues ​​at all time steps within a sliding window; Based on the significance metric, the comprehensive period significance of each period amplitude feature is calculated. The significance metric includes period strength and relative energy proportion: ， ， ， ， In the formula, Indicates The period amplitude in the sliding window The periodic intensity under Indicates The period amplitude in the sliding window The fluctuation amplitude of the lower subband characteristic, which is defined as the maximum value of the subband characteristic value With minimum difference; Indicates The period amplitude in the sliding window The relative energy proportion of the characteristic components under ; is the normalization term, indicating the The sum of the energy values ​​of the sub-band features of all periodic amplitudes within a sliding window; Indicates The period amplitude in the sliding window The comprehensive period significance of and Respectively represent the weight coefficients of the corresponding significance measurement indicators; Based on the comprehensive period significance, each of the period amplitude features is weighted and fused to obtain the corresponding window comprehensive amplitude feature: ， In the formula, Indicates The window comprehensive amplitude characteristics corresponding to the sliding window.

4. The method according to claim 2, wherein: The water quality anomaly identification model comprises a cascaded temporal feature modeling module and a feature anomaly identification module; The time series feature modeling module is used to model the input electrochemical time series sensing features, extract the dynamic change pattern and time dependency of the features of the monitored water area block, and obtain the corresponding pattern time series dependency features; The feature anomaly identification module is used to process the pattern time-dependent characteristics to predict whether there is a risk of water quality anomaly in the corresponding monitored water area.

5. The method according to claim 4, wherein: The temporal feature modeling module comprises an LSTM layer, a TCN layer and a feature fusion layer, wherein the LSTM layer and the TCN layer are connected in parallel to the feature fusion layer; The feature anomaly recognition module includes an MLP layer, a Softmax layer and an anomaly classification layer; The LSTM layer is used to extract long-term trend features in electrochemical time-series sensing features through a gating structure to capture the trend changes and long-term dependencies of water quality parameters: ， ， In the formula, Indicates the LSTM layer monitors the water area In the The hidden state of the sliding window output, represents the weight matrix of the LSTM layer, Indicates the LSTM layer monitors the water area In the The hidden state of the sliding window output; Indicates monitored water area In the The electrochemical input feature vector of sliding windows, represents the bias term, represents the activation function of the LSTM layer; represents the total number of time steps corresponding to the electrochemical sequential sensing characteristics, Indicates that the LSTM layer is used to monitor the water area The extracted long-term trend features; The TCN layer extracts local fluctuation pattern features in electrochemical time-series sensing features by stacking multiple layers of convolution operations to capture short-term fluctuations and local change patterns of water quality parameters: ， ， In the formula, Indicates the TCN layer for monitoring water area blocks In the The convolution output features of the sliding window, and Respectively represent the convolution kernel weight and bias term of the TCN layer, Represents a one-dimensional convolution operation; Indicates that the LSTM layer is used to monitor the water area The extracted local fluctuation pattern features; The feature fusion layer is used to perform weighted fusion on the output result of the LSTM layer and the output result of the TCN layer to obtain the corresponding pattern time-dependent features: ， In the formula, Indicates monitoring water area The temporal dependence characteristics of the pattern, is the feature fusion coefficient; The feature anomaly recognition module is used to process the pattern time-dependent features through the first MLP layer, and predict the probability value of the risk of water quality anomaly in the monitored water area through the Softmax layer: ， ， In the formula, Indicates monitoring water area No. The output feature representation of the hidden layer is and Indicates The weight matrix and bias term of the hidden layer; Indicates monitoring water area No. The output feature representation of the hidden layer is shown in the calculation of the first hidden layer. ; Indicates monitored water area The probability of abnormality with risk of abnormal water quality; is the total number of hidden layers in the first MLP layer, represents the activation function of the hidden layer, and Represent the weight matrix and bias term of the Softmax layer respectively, The last hidden layer of the first MLP layer is used to monitor the water area. The output feature representation is Represents the Softmax activation function; The anomaly classification layer is used to compare the predicted probability value of the water quality anomaly risk with a preset anomaly probability threshold, and determine whether the corresponding monitored water area has a water quality anomaly risk based on the comparison result.

6. The method according to claim 1, wherein: The abnormal event recognition model includes a cascaded multi-scale convolution module and an event type classification module; The multi-scale convolution module is used to extract the multi-scale time series features corresponding to the pattern time series dependency features of the risk monitoring water area block: ， ， In the formula, Indicates monitoring block In the The convolution output feature representation at the convolution kernel scale includes the local fluctuation and trend characteristics at the convolution kernel scale; and Respectively represent The convolution kernel weight matrix and bias term of the convolution kernel scale; Indicates risk monitoring waters The temporal dependence characteristics of the patterns; Represents the activation function of the multi-scale convolution module; Indicates risk monitoring waters The comprehensive multi-scale feature representation of Indicates The scale fusion weight of the convolution kernel scale, Represents the total number of convolution kernel scales in the multi-scale convolution module; The event type classification module includes an attention layer, a second MLP layer, a confidence calculation layer, and an abnormal event classification layer; the attention layer is used to calculate a weight distribution matrix for the multi-scale temporal features to determine the corresponding attention features: ， ， In the formula, Indicates risk monitoring waters No. The attention weight of the convolution kernel scale; and Indicates The attention weight matrix and bias term of the convolution kernel scale; Indicates that the attention layer monitors the water area for risks Output attention features; The second MLP layer is used to perform nonlinear mapping and feature conversion processing on the attention features, and determine the confidence matrix of the risk monitoring water area block for the preset risk event type set through the confidence calculation layer: ， ， In the formula, Indicates risk monitoring waters No. The output feature representation of the hidden layer; and Indicates The weight matrix and bias term of the hidden layer; Indicates risk monitoring waters No. The output feature representation of the hidden layer is shown in the calculation of the first hidden layer. ; is the total number of hidden layers in the second MLP layer, represents the activation function of the hidden layer, The last hidden layer of the second MLP layer is used to monitor the risk water area. The output comprehensive risk characterization; and Represent the weight matrix and bias term of the confidence calculation layer respectively, Indicates risk monitoring water area The confidence matrix includes the confidence scores for each risk event type in the risk event type set; The abnormal event classification layer is used to process the confidence matrix to identify the water quality abnormal risk event type corresponding to the risk monitoring water area block.

7. The method according to claim 6, wherein: The loss function of the abnormal event recognition model is: ， ， ， ， In the formula, represents the loss function of the abnormal event recognition model, represents the classification cross entropy loss term, represents the adaptive balancing loss term, and represents the temporal consistency loss term, Respectively represent the weight coefficients of each corresponding loss term; represents the total number of samples in the risk data sample set for risk monitoring waters, Indicates the total number of abnormal water quality risk event types in the risk event type set, Indicates The risk data sample is in The actual label on the risk event type. If the block belongs to event type, then ,otherwise ; Indicates The risk data sample is predicted by the abnormal event recognition model as The confidence level of each risk event type; is the adaptive balance coefficient, which is used to adjust the balance between feature weight and classification confidence; Indicates The risk data sample is in The characteristic weight distribution value on each risk event type; Represents the total number of time steps of multi-scale temporal features, Indicates The risk data sample is in e The convolution kernel scale Feature representation of time steps; Indicates The risk data sample is in e The convolution kernel scale The feature representation of each time step.

8. An Internet of Things intelligent monitoring and alarm system for water quality detection, used to implement the method according to any one of claims 1 to 7, the system comprising: A water quality parameter sampling unit is used to sample the water quality electrochemical data of the corresponding monitored water area based on each IoT node; A plurality of IoT nodes are distributedly arranged in different blocks in the monitored water area, and each IoT node is deployed with an electrochemical sensor; the sensing parameter types of the water quality electrochemical data include pH value, DO content and TOC content of the water body; A time series feature extraction unit, used to extract the electrochemical time series sensing features corresponding to each of the water quality electrochemical data; A water quality anomaly identification unit, used to input each of the electrochemical time-series sensing features into a water quality anomaly identification model, extract the mode time-series dependency features corresponding to the electrochemical time-series sensing features, so as to predict whether there is a risk of water quality anomaly in the corresponding monitored water area; An abnormal event identification unit is used to input the pattern time-series dependency characteristics corresponding to each risk monitoring water area block predicted to have abnormal water quality risk into an abnormal event identification model to identify the type of abnormal water quality risk event corresponding to the risk monitoring water area block; The water quality alarm unit of the water area is used to perform the water quality pollution alarm operation of the water area according to the block identification of each risk monitoring water area block and the corresponding water quality abnormal risk event type.

Images