PDA water quality intelligent detection system based on edge computing

Abstract

The application discloses a PDA water quality intelligent detection system based on edge computing, comprising: a data acquisition module for acquiring multi-parameter water quality detection data and detection context data; a parameter causal decoupling module for performing causal decoupling mapping to generate a causal decoupling feature sequence; a drift modulation calculation module for performing function modulation to generate a drift modulation feature sequence; an improved KAN network calculation module for performing nonlinear function combination operation to generate a multi-parameter coupling feature tensor; a space-time context coupling module for generating a water quality state judgment feature tensor; a function contribution degree evaluation module for generating an edge lightweight feature tensor; a state judgment module for generating a water quality detection result, a water quality abnormal risk level and a detection data credibility index; and a detection record generation module for generating a water quality detection record and outputting. The application realizes on-site intelligent analysis of water quality detection data, and improves detection efficiency, precision and credibility.

Description

Technical Field

[0001] This invention relates to the field of water quality monitoring technology, and in particular to a PDA-based intelligent water quality detection system based on edge computing. Background Technology

[0002] Water quality monitoring technology plays a vital role in environmental protection, water resource management, and drinking water safety management. The water quality monitoring process involves detecting and analyzing various physicochemical indicators in water bodies, including parameters such as pH, dissolved oxygen, conductivity, turbidity, ammonia nitrogen content, and temperature. Traditional water quality testing methods primarily rely on laboratory testing equipment for sample analysis. After collecting water samples on-site, personnel deliver them to the laboratory, where benchtop testing instruments are used to analyze various indicators. Laboratory testing equipment offers high accuracy and a wide range of testing capabilities. However, the laboratory testing process involves multiple steps, including sample transportation, sample preservation, and experimental analysis, resulting in a relatively long overall testing cycle. With the increasing demand for on-site testing, handheld testing devices are gradually being applied to water quality monitoring scenarios. These devices, combined with mobile terminals, form portable testing tools, allowing data to be recorded and transmitted via mobile devices.

[0003] The application of mobile terminal devices in water quality testing is gradually increasing. Some testing equipment connects to mobile terminals via Bluetooth or serial interfaces, enabling the mobile terminals to complete data collection, storage, and uploading. These mobile terminals possess positioning and data communication capabilities, allowing testing personnel to directly record the testing location and time after completing the test on-site. The water quality monitoring data collection process is increasingly being digitally managed, with some systems transmitting collected data to remote servers for data analysis. These remote servers have high computing power, and water quality data analysis results can be returned to the mobile terminals via the network. However, this network-dependent processing method suffers from data transmission delays in areas with insufficient signal coverage, hindering the rapid generation of on-site testing results.

[0004] In existing technologies, the data processing structure between mobile terminals and water quality testing equipment is relatively simple. Multi-parameter detection data is mainly determined using fixed rule analysis. There are coupling relationships and dynamic changes among the multiple parameters. Traditional rule analysis methods are insufficient in modeling the coupling relationships among multiple parameters. Some systems introduce neural network algorithms for water quality analysis calculations, but the model structure is complex and consumes a lot of computing resources. Mobile terminals cannot directly execute high-complexity calculation tasks. During long-term use, the detection sensors experience drift, resulting in decreased stability of the detection data. Existing systems lack the ability to correct sensor drift. Insufficient multi-parameter coupling modeling ability and drift effect correction ability lead to reduced stability of detection results and insufficient ability to evaluate the reliability of detection results. The real-time judgment and intelligent analysis capabilities in the water quality detection process are limited.

[0005] Therefore, how to provide a PDA-based intelligent water quality detection system based on edge computing is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a PDA-based intelligent water quality detection system based on edge computing. This invention utilizes a PDA terminal to complete multi-parameter water quality data acquisition, and combines an improved KAN network (Kolmogorov-Arnold Network) to perform intelligent analysis and processing of water quality detection data. It realizes multi-parameter coupled modeling and sensor drift adaptive modulation calculation, and has the advantages of high on-site detection efficiency, strong data analysis capabilities, and high reliability of detection results.

[0007] According to an embodiment of the present invention, a PDA-based intelligent water quality detection system based on edge computing includes:

[0008] The data acquisition module is used to collect multi-parameter water quality detection data and detection context data of the water body to be tested, and preprocess it to generate multi-parameter water quality feature sequences and context feature vectors;

[0009] The parameter causal decoupling module is used to receive multi-parameter water quality feature sequences, construct a parameter causal correlation matrix, and perform causal decoupling mapping operations to generate causal decoupling feature sequences.

[0010] The drift modulation calculation module is used to receive the causal decoupling feature sequence, calculate the degree of sensor drift based on the historical measurement sequence of multi-parameter water quality detection data, and perform function modulation operation on the causal decoupling feature sequence to generate a drift modulation feature sequence.

[0011] An improved KAN network computation module is used to receive drift modulation feature sequences and perform nonlinear function combination operations on the node function mapping set and the edge function mapping set to generate multi-parameter coupled feature tensors;

[0012] The spatiotemporal context coupling module is used to receive multi-parameter coupled feature tensors and context feature vectors, and perform spatiotemporal context coupling operations to generate water quality status determination feature tensors.

[0013] The function contribution evaluation module is used to perform function contribution calculation on the water quality status determination feature tensor and perform function pruning to generate edge lightweight feature tensors.

[0014] The state determination module is used to receive edge lightweight feature tensors, perform water quality state determination operations, and generate water quality detection results, water quality anomaly risk levels, and detection data reliability indicators.

[0015] The test record generation module is used to perform structured encapsulation to generate and output water quality test records.

[0016] Optionally, the process of acquiring and preprocessing multi-parameter water quality detection data and detection context data of the water body to be tested in the data acquisition module includes: acquiring detection data of the water body to be tested through a water quality detection digital electrode, transmitting it to a transfer unit via a 485 communication link, converting the 485 communication data into Bluetooth communication data through the transfer unit, sending it to a handheld PDA via a Bluetooth link, generating multi-parameter water quality detection data, obtaining detection location data and detection time data by the positioning unit of the handheld PDA, reading sample identification data, and generating detection context data; performing data verification, outlier filtering, and standardization processing on the multi-parameter water quality detection data to generate a multi-parameter water quality feature sequence, performing encoding processing on the detection context data to generate a context feature vector, and transmitting the multi-parameter water quality detection data, detection context data, and corresponding detection results to the laboratory LIMS system via the network to complete the detection data link connection.

[0017] Optionally, the generation of the causal decoupling feature sequence in the parameter causal decoupling module includes:

[0018] A parameter index set is constructed based on the categories of water quality detection parameters in the multi-parameter water quality feature sequence. A parameter causal association table is established based on the physical correlation between water quality parameters. A parameter causal association matrix is ​​generated based on the parameter causal association table, where each matrix element represents the causal association strength between two corresponding water quality detection parameters.

[0019] The multi-parameter water quality feature sequence is arranged according to the parameter index set to generate a parameter feature vector. The parameter feature vector is then input into the parameter causal correlation matrix to perform matrix mapping operation, thereby generating an initial causal mapping feature vector.

[0020] Based on the correlation strength of each matrix element in the parameter causal correlation matrix, a causal weight reconstruction operation is performed on the initial causal mapping feature vector to generate a parameter coupling feature vector;

[0021] Perform sequence rearrangement on the parameter-coupled feature vectors to generate causal decoupled feature sequences.

[0022] Optionally, the process of calculating the sensor drift degree in the drift modulation calculation module includes: receiving a causal decoupling feature sequence; performing parameter grouping processing on the causal decoupling feature sequence according to the water quality detection parameter category to generate a parameter detection subsequence; reading the historical measurement sequence of the corresponding water quality detection parameter stored in the PDA terminal; calculating the historical mean and historical fluctuation amplitude of each water quality detection parameter based on the historical measurement sequence; performing difference calculation between the current detection value and the historical mean in the parameter detection subsequence to generate a detection deviation value; calculating the parameter drift degree index based on the detection deviation value and the historical fluctuation amplitude; performing drift intensity evaluation operation on each water quality detection parameter based on the parameter drift degree index to generate a sensor drift degree vector; and calculating the drift modulation factor.

[0023] Optionally, the process of performing function modulation operations in the drift modulation calculation module includes:

[0024] The system receives a causal decoupling feature sequence and a drift modulation factor. Based on the parameter index set, it performs position alignment processing on each parameter feature in the causal decoupling feature sequence and the corresponding drift modulation factor to generate a parameter modulation pairing set. Based on the drift modulation factor in the parameter modulation pairing set, it determines the function modulation intensity, function modulation interval and function modulation order for each parameter feature to generate a parameter modulation configuration set.

[0025] Based on the parameter modulation configuration set, piecewise function modulation operation is performed on each parameter feature, amplitude compression processing is performed on the high drift segment of each parameter feature, and feature preservation processing is performed on the low drift segment of each parameter feature to generate parameter modulation subsequence;

[0026] Based on the parameter causal correlation matrix, correlation consistency correction is performed on each parameter modulation subsequence to generate parameter consistent modulation subsequence. Boundary smoothing is then performed on the parameter consistent modulation subsequence. Sequence recombination is then performed according to the order of water quality detection parameters corresponding to the parameter index set to generate drift modulation feature sequence.

[0027] Optionally, the process of generating multi-parameter coupled feature tensors in the improved KAN network computation module includes:

[0028] The system receives a drift modulation feature sequence, maps the drift modulation feature sequence to the corresponding network node input channel according to the parameter index set, and generates a node input feature vector; it assigns a corresponding node function mapping to each network node according to a preset node function library, and performs adaptive adjustment on the function shape parameters of the node function according to the drift modulation factor, thereby generating a node function mapping set.

[0029] Construct edge connections between network nodes based on the parameter causal correlation matrix, configure corresponding edge function mappings for each edge connection, and generate a set of edge function mappings.

[0030] The node input feature vector is input into the node function mapping set to perform nonlinear function mapping operation to generate node mapping features; based on the edge connection relationship between network nodes, the node mapping features are input into the edge function mapping set to perform edge function combination operation to generate edge function coupling features; the node mapping features and edge function coupling features are then combined to generate a multi-parameter coupling feature tensor.

[0031] Optionally, the process of generating a water quality status determination feature tensor by performing spatiotemporal context coupling operation in the spatiotemporal context coupling module includes: receiving a multi-parameter coupled feature tensor and a context feature vector; reading the detection time data and detection location data from the context feature vector; determining the corresponding time modulation weight based on the detection time data; determining the corresponding spatial modulation weight based on the detection location data; performing weight modulation operation on each parameter feature in the multi-parameter coupled feature tensor according to the time modulation weight and the spatial modulation weight; adjusting the parameter features under different detection time and detection location conditions according to the corresponding weights; and performing feature aggregation operation on the modulated parameter features to generate a water quality status determination feature tensor.

[0032] Optionally, the process of performing function contribution calculation in the function contribution evaluation module includes: receiving the water quality status determination feature tensor, establishing a function index set based on the function identification information recorded in the node function mapping set and the edge function mapping set, reading the output values ​​of each node function and each edge function that participated in generating the water quality status determination feature tensor, and extracting the corresponding function output feature sequence according to the function correspondence.

[0033] For each function output feature sequence, calculate the feature contribution value in the water quality status determination feature tensor. Determine the function contribution index by calculating the feature change between the function output feature sequence and the water quality status determination feature tensor. Sort the function contribution of each node function and each edge function according to the function contribution index to generate a function contribution sequence. Output the function contribution sequence to the function pruning process to perform function pruning operation.

[0034] Optionally, the process of performing water quality status determination calculation in the status determination module includes: receiving an edge lightweight feature tensor and reading the standard threshold intervals corresponding to each water quality detection parameter in the preset water quality evaluation standard; extracting the feature values ​​corresponding to each water quality detection parameter from the edge lightweight feature tensor according to the parameter index set, performing interval comparison operation between each feature value and the corresponding standard threshold interval to generate parameter status determination results; performing status aggregation operation based on the parameter status determination results of each water quality detection parameter, determining an abnormal water quality state when a parameter status determination result exceeds a preset abnormal threshold, and determining a normal water quality state when all parameter status determination results are within the standard threshold interval; calculating the water quality abnormality risk level based on the deviation degree of each water quality detection parameter, and generating the corresponding water quality detection result and water quality abnormality risk level.

[0035] The beneficial effects of this invention are:

[0036] (1) The PDA terminal and the digital electrode for water quality testing form an integrated testing structure. The testing personnel complete the collection of multi-parameter water quality testing data, the recording of testing time and the recording of testing location on the water body site. The testing data is processed and calculated in the terminal device. The testing process reduces the sample transportation links and reliance on remote servers. The water quality testing results are generated directly on the site, and the efficiency of water quality monitoring is improved.

[0037] (2) The improved Kolmogorov-Arnold Network structure introduces node function mapping and edge function mapping calculation mechanism. Multi-parameter water quality detection data are coupled and calculated between network nodes through function mapping. The correlation between water quality detection parameters is expressed in the network structure, the multi-parameter coupling modeling capability is enhanced, and the water quality status determination process has a higher ability to identify complex water quality change characteristics.

[0038] (3) The sensor drift degree calculation mechanism is combined with function modulation operation to form a drift modulation calculation structure. The detection data completes drift correction processing before entering the network calculation process. The parameter features with large drift influence are suppressed in the calculation process, and the stable parameter features maintain their original expression ability. The detection data credibility assessment mechanism generates the detection data credibility index simultaneously, and the stability and reliability of water quality detection results are improved. Attached Figure Description

[0039] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0040] Figure 1 This is a module structure diagram of a PDA-based intelligent water quality detection system based on edge computing proposed in this invention;

[0041] Figure 2 This is a schematic diagram of an improved KAN network computing module for a PDA-based intelligent water quality detection system based on edge computing, as proposed in this invention.

[0042] Figure 3 This is a schematic diagram of the parameter causal correlation matrix construction and causal decoupling calculation process based on edge computing proposed in this invention. Detailed Implementation

[0043] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0044] refer to Figures 1-3 A PDA-based intelligent water quality detection system based on edge computing includes:

[0045] The data acquisition module is used to collect multi-parameter water quality detection data and detection context data of the water body to be tested, and preprocess it to generate multi-parameter water quality feature sequences and context feature vectors;

[0046] The parameter causal decoupling module is used to receive multi-parameter water quality feature sequences, construct a parameter causal correlation matrix, and perform causal decoupling mapping operations to generate causal decoupling feature sequences.

[0047] The drift modulation calculation module is used to receive the causal decoupling feature sequence, calculate the degree of sensor drift based on the historical measurement sequence of multi-parameter water quality detection data, and perform function modulation operation on the causal decoupling feature sequence to generate a drift modulation feature sequence.

[0048] An improved KAN network computation module is used to receive drift modulation feature sequences and perform nonlinear function combination operations on the node function mapping set and the edge function mapping set to generate a multi-parameter coupled feature tensor. The node function mapping set is composed of learnable functions and is used to replace fixed weight mapping. The edge function mapping set is dynamically adjusted according to the drift modulation factor to perform function shape parameters.

[0049] The spatiotemporal context coupling module is used to receive multi-parameter coupled feature tensors and context feature vectors, and perform spatiotemporal context coupling operations to generate water quality status determination feature tensors.

[0050] The function contribution evaluation module is used to perform function contribution calculation on the water quality status determination feature tensor and perform function pruning to generate edge lightweight feature tensors.

[0051] The state determination module is used to receive edge lightweight feature tensors, perform water quality state determination operations, and generate water quality detection results, water quality anomaly risk levels, and detection data credibility indicators.

[0052] The test record generation module is used to perform structured encapsulation to generate and output water quality test records.

[0053] In this embodiment, the process of acquiring and preprocessing multi-parameter water quality detection data and detection context data of the water body to be tested in the data acquisition module includes: acquiring detection data of the water body to be tested through a water quality detection digital electrode, transmitting it to a transfer unit via a 485 communication link, converting the 485 communication data into Bluetooth communication data through the transfer unit, sending it to a handheld PDA via a Bluetooth link, generating multi-parameter water quality detection data, obtaining detection location data and detection time data by the positioning unit of the handheld PDA, reading sample identification data, and generating detection context data; performing data verification, outlier filtering, and standardization processing on the multi-parameter water quality detection data to generate a multi-parameter water quality feature sequence, performing encoding processing on the detection context data to generate a context feature vector, and transmitting the multi-parameter water quality detection data, detection context data, and corresponding detection results to the laboratory LIMS system via the network to complete the connection of the detection data link.

[0054] In this embodiment, the generation of the causal decoupling feature sequence in the parameter causal decoupling module includes:

[0055] A parameter index set is constructed based on the categories of water quality detection parameters in the multi-parameter water quality feature sequence. A parameter causal association table is established based on the physical correlation between water quality parameters. A parameter causal association matrix is ​​generated based on the parameter causal association table, where each matrix element represents the causal association strength between two corresponding water quality detection parameters.

[0056] The parameter causal correlation matrix is ​​constructed based on the physical interaction relationships between water quality detection parameters. Specifically, it involves: establishing a set of correlation rules between parameters according to their influence relationships in the field of water quality monitoring; determining the degree of influence of each water quality detection parameter on other water quality detection parameters based on the set of correlation rules; constructing the corresponding matrix structure according to the parameter index set order, so that the parameter causal correlation matrix can reflect the physical coupling relationship between water quality detection parameters; arranging the multi-parameter water quality feature sequence according to the parameter index set to generate parameter feature vectors; inputting the parameter feature vectors into the parameter causal correlation matrix to perform matrix mapping operations to generate an initial causal mapping feature vector.

[0057] The matrix mapping operation is performed by matrix multiplication. The parameter feature vector is multiplied by the parameter causal correlation matrix, so that each parameter feature in the parameter feature vector is simultaneously integrated with other water quality detection parameter features that have a causal relationship with that parameter, resulting in an initial causal mapping feature vector containing multi-parameter correlation information. Based on the correlation strength of each matrix element in the parameter causal correlation matrix, a causal weight reconstruction operation is performed on the initial causal mapping feature vector to generate a parameter coupling feature vector.

[0058] The causal weight reconstruction operation reads the correlation strength values ​​from the parameter causal correlation matrix and performs weight adjustment processing on each parameter feature in the initial causal mapping feature vector. This process gives higher weights to parameter features with higher correlation strengths during the coupling calculation and reduces the weights of parameter features with lower correlation strengths, generating a parameter coupling feature vector that can characterize the multi-parameter causal coupling relationship. The sequence rearrangement process is then performed on the parameter coupling feature vector to generate a causal decoupling feature sequence. The sequence rearrangement process reorders the positions of each parameter feature in the parameter coupling feature vector according to the parameter position index recorded in the parameter index set, restoring each parameter feature to the position of the corresponding water quality detection parameter, generating a causal decoupling feature sequence consistent with the structure of the multi-parameter water quality feature sequence.

[0059] In this embodiment, the process of calculating the sensor drift degree in the drift modulation calculation module includes: receiving the causal decoupling feature sequence, performing parameter grouping processing on the causal decoupling feature sequence according to the water quality detection parameter category to generate parameter detection sub-sequences; reading the historical measurement sequence of the corresponding water quality detection parameter stored in the PDA terminal, calculating the historical mean and historical fluctuation amplitude of each water quality detection parameter based on the historical measurement sequence; performing difference calculation between the current detection value and the historical mean in the parameter detection sub-sequence to generate a detection deviation value, calculating the parameter drift degree index based on the detection deviation value and the historical fluctuation amplitude; performing drift intensity evaluation calculation on each water quality detection parameter based on the parameter drift degree index to generate a sensor drift degree vector, and calculating the drift modulation factor.

[0060] In this embodiment, the process of performing drift intensity assessment is as follows: Each water quality detection parameter is graded according to its corresponding drift index. The drift index is compared with a preset drift level threshold set. When the drift index is less than the first drift threshold, it is determined to be in a stable state. When the drift index is between the first and second drift thresholds, it is determined to be in a slightly drifting state. When the drift index is greater than the second drift threshold, it is determined to be in a significantly drifting state. Based on the determination results, drift weight values ​​are assigned to the corresponding water quality detection parameters to obtain a sensor drift degree vector representing the drift state of each water quality detection parameter.

[0061] The process of calculating the drift modulation factor is as follows: based on the drift weight values ​​in the sensor drift degree vector, the corresponding parameter features in the causal decoupling feature sequence are processed by modulation coefficient calculation, the drift weight values ​​are mapped to modulation coefficients, and the corresponding parameter features are processed by scaling based on the modulation coefficients to generate the drift modulation factor, thereby reducing the influence weight of water quality detection parameters with large drift degrees in feature calculation.

[0062] In this embodiment, the process of performing function modulation calculation in the drift modulation calculation module includes:

[0063] Receive the causal decoupling feature sequence and the drift modulation factor, perform position alignment processing on each parameter feature in the causal decoupling feature sequence and the corresponding drift modulation factor according to the parameter index set, and generate a parameter modulation pairing set; according to the drift modulation factor in the parameter modulation pairing set, determine the function modulation intensity, function modulation interval and function modulation order for each parameter feature, and generate a parameter modulation configuration set;

[0064] Specifically, the process involves reading the drift modulation factor corresponding to each parameter feature, calculating the corresponding function modulation intensity based on the magnitude of the drift modulation factor, where a larger drift modulation factor indicates a higher degree of sensor drift, thus allocating a higher function modulation intensity to enhance the modulation operation's ability to correct for that parameter feature; based on the numerical distribution range of each parameter feature in the causal decoupling feature sequence, segmentation processing is performed on the value range of each parameter feature, dividing the value range of the parameter feature into multiple continuous segments, determining the segment requiring modulation processing based on the magnitude of the drift modulation factor, and obtaining the corresponding function modulation interval; after completing function modulation... After interval division, the modulation order of each parameter feature is sorted according to the parameter correlation strength recorded in the parameter causal correlation matrix, so that water quality detection parameters with higher correlation strength are given priority for function modulation operation to ensure the consistency of modulation results among related parameters. Piecewise function modulation operation is performed on each parameter feature according to the parameter modulation configuration set. Amplitude compression is performed on high-drift segments of each parameter feature, and feature preservation processing is performed on low-drift segments of each parameter feature, generating parameter modulation subsequences. Specifically, the historical measurement sequences of the corresponding water quality detection parameters are read, the historical mean and historical standard deviation are calculated, and the current parameter features are calculated. The deviation value from the historical mean is calculated. This deviation is then compared to a preset drift threshold, which is the product of the historical standard deviation and a preset drift coefficient. If the deviation is greater than the threshold, the corresponding parameter feature segment is classified as a high-drift segment and proportionally compressed based on the drift modulation factor. If the deviation is less than or equal to the threshold, the segment is classified as a low-drift segment, and the original parameter feature value is preserved. Based on the parameter causal correlation matrix, correlation consistency correction is performed on each parameter modulation subsequence to generate a parameter-consistent modulation subsequence. The consistency modulation subsequence undergoes boundary smoothing processing, and sequence recombination processing is performed according to the order of water quality detection parameters corresponding to the parameter index set to generate a drift modulation feature sequence. The consistency adjustment operation is specifically as follows: for parameter pairs in the parameter causal correlation matrix whose correlation strength is greater than a preset correlation threshold, the feature values ​​of the corresponding parameter modulation subsequences at the same position are read, the difference between the two feature values ​​is calculated, and the larger feature value is reduced and the smaller feature value is increased according to a preset correction ratio, so that the difference between the two corrected feature values ​​is reduced, and the feature changes of the parameter modulation subsequences with correlation are kept consistent at the same position.

[0065] In this embodiment, the process of generating multi-parameter coupled feature tensors in the improved KAN network computation module includes:

[0066] The system receives a drift modulation feature sequence, maps the drift modulation feature sequence to the corresponding network node input channel according to the parameter index set, and generates a node input feature vector; it assigns a corresponding node function mapping to each network node according to a preset node function library, and performs adaptive adjustment on the function shape parameters of the node function according to the drift modulation factor, thereby generating a node function mapping set.

[0067] In this embodiment, the adaptive adjustment specifically involves: reading the drift modulation factor of the input channel of the corresponding network node; updating the function shape parameters of the node function according to the magnitude of the drift modulation factor, where the function shape parameters include the function amplitude coefficient and the function curvature coefficient; when the drift modulation factor increases, reducing the function amplitude coefficient and the function curvature coefficient makes the change in the node function output smoother, reducing the impact of high drift parameter features on the network calculation results; when the drift modulation factor is small, keeping the node function amplitude coefficient and function curvature coefficient unchanged, so that the stable parameter features retain their original feature expression ability during the function mapping process, generating a drift adaptive node function mapping; constructing the edge connection relationship between network nodes based on the parameter causal correlation matrix, configuring a corresponding edge function mapping for each edge connection relationship, and generating an edge function mapping set; specifically, reading each matrix element in the parameter causal correlation matrix, determining parameter pairs with a correlation strength greater than a preset correlation threshold as node pairs with a function connection relationship, and establishing an edge connection between the corresponding network nodes. Next, the edge connection weights are determined based on the correlation strength of matrix elements, so that parameter pairs with higher correlation strength form preferential connections in the network structure, constructing a network edge connection topology that reflects the physical correlation structure of water quality detection parameters; the node input feature vectors are input into the node function mapping set to perform nonlinear function mapping operations, generating node mapping features; based on the edge connection relationships between network nodes, the node mapping features are input into the edge function mapping set to perform edge function combination operations, the corresponding node mapping features are read based on the edge connection relationships between network nodes, the node mapping features are input into the corresponding edge function mapping to perform function calculations, and the output features of several edge functions are superimposed to generate edge function coupling features; the node mapping features and edge function coupling features are superimposed to generate a multi-parameter coupling feature tensor; specifically, the node mapping features and edge function coupling features are read, the corresponding feature values ​​are extracted at the same parameter index positions, the corresponding feature values ​​are summed element-wise to generate superimposed feature values, and the superimposed feature values ​​are reorganized into tensors according to the parameter order corresponding to the parameter index set.

[0068] In this embodiment, the process of generating a water quality state determination feature tensor by performing spatiotemporal context coupling operation in the spatiotemporal context coupling module includes: receiving a multi-parameter coupled feature tensor and a context feature vector, reading the detection time data and detection location data in the context feature vector, determining the corresponding time modulation weight based on the detection time data, and determining the corresponding spatial modulation weight based on the detection location data.

[0069] Specifically, the system reads the detection time data, maps the detection time to the corresponding time interval according to preset time division rules, and assigns corresponding time weight coefficients to different time intervals. The time intervals are divided based on common time variation patterns in water quality monitoring, including diurnal variation intervals and periodic variation intervals. Corresponding time weight tables are pre-stored in the system, and the corresponding time modulation weights are obtained through table lookup. The system also reads the detection location data, determines the water area number to which the detection point belongs, and retrieves the corresponding spatial weight coefficients from the pre-established area weight table. The area weight table is set based on the water quality stability and historical monitoring variation amplitude of different water areas to obtain the corresponding spatial modulation weights. Finally, the system applies time modulation weights and spatial modulation weights to the water quality monitoring data. Each parameter feature in the multi-parameter coupled feature tensor undergoes weighted modulation operation. The parameter features under different detection times and locations are adjusted according to their corresponding weights. Feature aggregation operation is then performed on the modulated parameter features to generate a water quality status determination feature tensor. Specifically, the weighted modulation operation includes: calculating a comprehensive modulation weight based on the temporal modulation weight and the spatial modulation weight, which is obtained by multiplying the temporal modulation weight and the spatial modulation weight; reading the parameter feature values ​​in the multi-parameter coupled feature tensor; performing element-wise multiplication of the corresponding parameter feature values ​​with the comprehensive modulation weight to obtain the modulated parameter feature values; and recombining the modulated parameter feature values ​​according to the original parameter order to generate a water quality status determination feature tensor that has undergone time-varying modulation processing.

[0070] In this embodiment, the process of performing function contribution calculation in the function contribution evaluation module includes: receiving the water quality status determination feature tensor, establishing a function index set based on the function identification information recorded in the node function mapping set and the edge function mapping set, reading the output values ​​of each node function and each edge function that participated in generating the water quality status determination feature tensor, and extracting the corresponding function output feature sequence according to the function correspondence.

[0071] For each function output feature sequence, the feature contribution value in the water quality state determination feature tensor is calculated. The function contribution index is determined by calculating the feature change between the function output feature sequence and the water quality state determination feature tensor. Specifically, the function output feature sequence corresponding to a certain function is read. Using the current water quality state determination feature tensor as the reference feature tensor, the function output feature sequence is removed while keeping the output feature sequences of other functions unchanged, and a new determination feature tensor is recalculated. The difference between the new determination feature tensor and the reference feature tensor is calculated. The absolute value of the difference is summed to obtain the feature change. The feature change is proportional to the sum of the feature values ​​of the reference feature tensor to obtain the function contribution index. The degree of function contribution is determined based on the comparison between the function contribution index and the preset contribution threshold. The function contribution index is used to sort the contribution of each node function and each edge function to generate a function contribution sequence. The function contribution sequence is output to the function pruning process to perform function pruning operations.

[0072] In this embodiment, the process of performing water quality status determination calculation in the status determination module includes: receiving an edge lightweight feature tensor and reading the standard threshold range corresponding to each water quality detection parameter in the preset water quality evaluation standard; extracting the feature values ​​corresponding to each water quality detection parameter from the edge lightweight feature tensor according to the parameter index set, performing interval comparison operation between each feature value and the corresponding standard threshold range to generate parameter status determination results; performing status aggregation operation based on the parameter status determination results of each water quality detection parameter, determining an abnormal water quality state when a parameter status determination result exceeds a preset abnormal threshold, and determining a normal water quality state when all parameter status determination results are within the standard threshold range; calculating the water quality abnormality risk level based on the deviation degree of each water quality detection parameter, and generating the corresponding water quality detection result and water quality abnormality risk level.

[0073] Specifically, the water quality test results are categorized as follows: The status of each water quality test parameter is determined as follows: when all water quality test parameters are within their corresponding standard threshold ranges, the water quality is considered normal; when at least one water quality test parameter exceeds the standard threshold range but the deviation is less than 10% of the upper limit of the standard threshold, the water quality is considered slightly abnormal; when at least one water quality test parameter has a deviation between 10% and 30% of the upper limit of the standard threshold, the water quality is considered moderately abnormal; and when at least one water quality test parameter has a deviation greater than 30% of the upper limit of the standard threshold, the water quality is considered severely abnormal.

[0074] The water quality anomaly risk level is calculated as follows: the deviation value of each water quality detection parameter is calculated, and a weighted summation operation is performed on each deviation value according to the preset weight to obtain the comprehensive deviation value; the water quality anomaly risk level is determined according to the correspondence between the comprehensive deviation value and the risk level threshold range, where a comprehensive deviation value less than 0.1 is judged as Level 1 risk, a comprehensive deviation value between 0.1 and 0.3 is judged as Level 2 risk, a comprehensive deviation value between 0.3 and 0.6 is judged as Level 3 risk, and a comprehensive deviation value greater than 0.6 is judged as Level 4 risk.

[0075] Example 1: To verify the feasibility of this invention in practice, it was applied to a routine water body monitoring scenario. In water environment management, testing personnel need to rapidly test and analyze samples from different water bodies. Traditional testing methods mainly rely on sending samples to a laboratory for analysis. The testing process involves multiple stages, including sample preservation, transportation, and testing of experimental equipment, resulting in a long testing cycle. Testing personnel cannot obtain real-time results on-site. When abnormal changes occur in the water body, management personnel cannot promptly grasp the water quality status. Although on-site testing equipment can complete basic testing, the data analysis process relies on remote server calculations. Real-time data processing is difficult in areas with unstable network signals. Water quality testing equipment experiences sensor drift over long-term use, leading to decreased stability of test results. Complex correlations exist between multiple water quality data parameters, making it difficult for traditional rule-based judgment methods to accurately identify water quality change trends.

[0076] In actual monitoring, the testing personnel carry a PDA terminal equipped with the system of this invention and a digital electrode device for water quality testing to the water body testing location. After the digital electrode is inserted into the water body to be tested, the digital electrode begins to collect multi-parameter detection data in the water body in real time, including pH, conductivity, dissolved oxygen, turbidity, and water temperature data. The PDA terminal reads the detection data through the digital electrode interface and automatically records the detection time data and detection location data. The collected multi-parameter water quality detection data enters the data acquisition module to perform data verification and standardization processing. After the outlier filtering process is completed, a multi-parameter water quality feature sequence is generated. The detection data enters the parameter causal decoupling module for processing. The system constructs a parameter causal correlation matrix based on the physical correlation between water quality detection parameters. The multi-parameter water quality feature sequence performs mapping operations in the parameter causal correlation matrix to generate a causal decoupling feature sequence. This processing process expresses the correlation between different water quality parameters.

[0077] The causal decoupling feature sequence enters the drift modulation calculation module to calculate the drift degree. The system reads historical measurement sequences and calculates the historical mean and historical standard deviation. The deviation between the current detection parameter value and the historical mean is used to calculate the drift degree index. Detection parameters with higher drift degrees undergo function modulation operations in subsequent calculations. Piecewise function modulation operations divide the detection parameters into high-drift and low-drift segments. The high-drift segment undergoes amplitude compression processing, while the low-drift segment retains its original features. After the drift modulation processing is completed, a drift modulation feature sequence is generated. The drift modulation feature sequence is then input into the improved Kolmogorov-Arnold converter. The Network computing module performs nonlinear function combination operations on the set of function mappings for network nodes and the set of function mappings for edges. Multi-parameter water quality features are coupled and calculated within the network structure. The node mapping features and edge function coupling features are superimposed to generate a multi-parameter coupled feature tensor. The coupled features are then processed in the spatiotemporal context coupling module. The detection time data and detection location data form a spatiotemporal modulation weight. The multi-parameter coupled features are weighted and modulated under different time and location conditions. The spatiotemporally modulated features are then processed in the function contribution evaluation module. Functions with low contribution are pruned. The system generates a lightweight edge feature tensor, which is then processed in the state determination module to determine the water quality state. The system performs threshold comparison calculations on each detection parameter according to the water quality evaluation standards, generating water quality detection results, water quality anomaly risk levels, and detection data reliability indicators. The detection results are displayed in real time on the PDA terminal and saved as a water quality detection record.

[0078] During continuous monitoring, the system of this invention completes multiple water body tests in the field environment. Personnel repeatedly test the same water body multiple times, and the test data is used to verify the system's accuracy and stability. Experimental data shows that traditional handheld testing devices have high errors in multi-parameter analysis, and sensor drift significantly affects the test results. The improved Kolmogorov-Arnold Network demonstrates higher recognition ability in multi-parameter coupled modeling, and the drift modulation computation mechanism effectively suppresses errors caused by sensor drift. The system completes the detection and analysis process in the field environment, and personnel obtain water quality test results within seconds. Multiple test results show that the system of this invention exhibits significant improvements in detection accuracy, result stability, and data reliability. Statistical analysis of the test data indicates that the multi-parameter coupled computation mechanism can accurately identify water quality change trends, the function contribution evaluation mechanism reduces the interference of low-contribution functions on the results, and the test result reliability index reflects the reliability of the test data.

[0079] To verify the feasibility of this invention in practice, it was applied to a multi-parameter water quality testing scenario in a certain water body. By comparing the detection results of traditional handheld testing devices and the system of this invention in the same water sample testing process, statistical analysis was conducted on the detection accuracy, detection stability, and detection time, and the following experimental results were obtained.

[0080] Table 1: Performance Comparison Analysis of PDA Edge Computing Intelligent Water Quality Detection Systems

[0081] Detection parameters Standard reference value Detection values ​​of traditional handheld testing equipment The detection value of the system of the present invention Traditional equipment error rate Error rate of the present invention pH 7.20 7.46 7.24 3.61% 0.55% Electrical conductivity (μS / cm) 512 538 517 5.08% 0.97% Dissolved oxygen (mg / L) 6.80 6.23 6.74 8.38% 0.88% Turbidity (NTU) 3.10 3.72 3.18 20.00% 2.58% Water temperature (°C) 18.5 19.2 18.7 3.78% 1.08% Average time per test — 38 seconds 11 seconds — — Continuous detection stability deviation — 7.6% 1.9% — — Reliability of test data — 0.63 0.91 — —

[0082] Table 1 shows the performance differences between traditional handheld testing devices and the system of this invention in multi-parameter water quality testing. The pH test results show that the standard reference value is 7.20, while the traditional handheld device's value is 7.46, with an error rate of 3.61%, indicating a significant deviation. The system of this invention's value is 7.24, with an error rate of 0.55%, close to the standard reference value, demonstrating higher measurement accuracy in pH testing. The conductivity test results show a similar trend: the traditional device's value is 538 μS / cm, with an error rate of 5.08%, while the system of this invention's value is 517 μS / cm, with an error rate controlled within 1%. The multi-parameter coupling calculation mechanism demonstrates high stability in conductivity data analysis.

[0083] The difference in dissolved oxygen detection results is even more significant. Traditional detection equipment yielded a value of 6.23 mg / L with an error rate of 8.38%, indicating a large deviation in the result. The system of this invention yielded a value of 6.74 mg / L, reducing the error rate to 0.88%. Dissolved oxygen data in water quality monitoring is greatly affected by environmental changes and sensor drift. The drift modulation calculation mechanism plays a crucial role in this system; the high drift range characteristics are processed through amplitude compression, significantly reducing the interference of outliers on the detection results. The turbidity detection results further demonstrate the advantages of the multi-parameter coupled analysis structure. Traditional detection equipment yielded a value of 3.72 NTU with an error rate of 20.00%, showing significant fluctuations in the result. The system of this invention yielded a value of 3.18 NTU with an error rate controlled at 2.58%, resulting in a more stable detection result.

[0084] Water temperature detection results showed that the traditional device detected a value of 19.2℃ with an error rate of 3.78%, while the system of this invention detected a value of 18.7℃, reducing the error rate to 1.08%. Detection time data indicated that the traditional handheld device took an average of 38 seconds to complete one detection, while the system of this invention took an average of 11 seconds, improving detection efficiency by over 70%. Continuous detection stability deviation data showed that the traditional device had a stability deviation of 7.6%, while the system of this invention had a stability deviation of 1.9%, significantly enhancing the repeatability of the detection results. Regarding the reliability index of the detection data, the traditional device had an average reliability of 0.63, while the system of this invention achieved a reliability of 0.91, demonstrating that the function contribution evaluation mechanism and the edge lightweight computation structure improved the stability of the detection results.

[0085] Overall data shows that the system of this invention exhibits significant advantages in detection accuracy, detection stability, and detection efficiency. The multi-parameter causal decoupling computational structure enhances the correlation modeling capability between different water quality detection parameters, the drift modulation computation mechanism effectively suppresses the influence of sensor drift, and the improved Kolmogorov-Arnold Network structure improves multi-parameter data analysis capabilities, enabling the rapid generation of reliable water quality detection results in on-site testing environments.

[0086] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A PDA-based intelligent water quality detection system based on edge computing, characterized in that, include: The data acquisition module is used to collect multi-parameter water quality detection data and detection context data of the water body to be tested, and preprocess it to generate multi-parameter water quality feature sequences and context feature vectors; The parameter causal decoupling module is used to receive multi-parameter water quality feature sequences, construct a parameter causal correlation matrix, and perform causal decoupling mapping operations to generate causal decoupling feature sequences. The drift modulation calculation module is used to receive the causal decoupling feature sequence, calculate the degree of sensor drift based on the historical measurement sequence of multi-parameter water quality detection data, and perform function modulation operation on the causal decoupling feature sequence to generate a drift modulation feature sequence. The process of calculating the sensor drift degree specifically includes: receiving a causal decoupling feature sequence; performing parameter grouping processing on the causal decoupling feature sequence according to the water quality detection parameter category to generate a parameter detection sub-sequence; reading the historical measurement sequence of the corresponding water quality detection parameter stored in the PDA terminal; calculating the historical mean and historical fluctuation amplitude of each water quality detection parameter based on the historical measurement sequence; calculating the difference between the current detection value and the historical mean in the parameter detection sub-sequence to generate a detection deviation value; calculating the parameter drift degree index based on the detection deviation value and the historical fluctuation amplitude; performing a drift intensity evaluation operation on each water quality detection parameter based on the parameter drift degree index to generate a sensor drift degree vector; and calculating the drift modulation factor. The improved KAN network computation module receives drift-modulated feature sequences and performs nonlinear function combination operations on the node function mapping set and the edge function mapping set to generate a multi-parameter coupled feature tensor; specifically, it includes: The system receives a drift modulation feature sequence, maps the drift modulation feature sequence to the corresponding network node input channel according to the parameter index set, and generates a node input feature vector; it assigns a corresponding node function mapping to each network node according to a preset node function library, and performs adaptive adjustment on the function shape parameters of the node function according to the drift modulation factor, thereby generating a node function mapping set. Construct edge connections between network nodes based on the parameter causal correlation matrix, configure corresponding edge function mappings for each edge connection, and generate a set of edge function mappings. The node input feature vector is input into the node function mapping set to perform nonlinear function mapping operation to generate node mapping features; based on the edge connection relationship between network nodes, the node mapping features are input into the edge function mapping set to perform edge function combination operation to generate edge function coupling features; the node mapping features and edge function coupling features are then combined to generate a multi-parameter coupling feature tensor. The spatiotemporal context coupling module is used to receive multi-parameter coupled feature tensors and context feature vectors, and perform spatiotemporal context coupling operations to generate water quality status determination feature tensors. The function contribution evaluation module is used to perform function contribution calculation on the water quality status determination feature tensor and perform function pruning operation to generate edge lightweight feature tensors. Specifically, it includes: receiving the water quality status determination feature tensor, establishing a function index set based on the function identification information recorded in the node function mapping set and the edge function mapping set, reading the output values ​​of each node function and each edge function involved in generating the water quality status determination feature tensor, and extracting the corresponding function output feature sequence according to the function correspondence. For each function output feature sequence, calculate the feature contribution value in the water quality status determination feature tensor. Determine the function contribution index by calculating the feature change between the function output feature sequence and the water quality status determination feature tensor. Sort the functions of each node and each edge according to the function contribution index to generate a function contribution sequence. Output the function contribution sequence to the function pruning process to perform function pruning operation. The state determination module is used to receive edge lightweight feature tensors, perform water quality state determination operations, and generate water quality detection results, water quality anomaly risk levels, and detection data credibility indicators. The test record generation module is used to perform structured encapsulation to generate and output water quality test records.

2. The PDA-based intelligent water quality detection system according to claim 1, characterized in that, The data acquisition module's process of acquiring and preprocessing multi-parameter water quality detection data and detection context data of the water body under test includes: acquiring detection data of the water body under test through a digital electrode for water quality detection, transmitting it to a transfer unit via a 485 communication link, converting the 485 communication data into Bluetooth communication data through the transfer unit, sending it to a handheld PDA via a Bluetooth link, generating multi-parameter water quality detection data, obtaining detection location data, obtaining detection time data, reading sample identification data through the positioning unit of the handheld PDA, and generating detection context data; performing data verification, outlier filtering, and standardization processing on the multi-parameter water quality detection data to generate a multi-parameter water quality feature sequence, performing encoding processing on the detection context data to generate a context feature vector, and transmitting the multi-parameter water quality detection data, detection context data, and corresponding detection results to the laboratory LIMS system via the network to complete the detection data link connection.

3. The PDA-based intelligent water quality detection system according to claim 1, characterized in that, The generation of the causal decoupling feature sequence in the parameter causal decoupling module includes: A parameter index set is constructed based on the categories of water quality detection parameters in the multi-parameter water quality feature sequence. A parameter causal association table is established based on the physical correlation between water quality parameters. A parameter causal association matrix is ​​generated based on the parameter causal association table, where each matrix element represents the causal association strength between two corresponding water quality detection parameters. The multi-parameter water quality feature sequence is arranged according to the parameter index set to generate a parameter feature vector. The parameter feature vector is then input into the parameter causal correlation matrix to perform matrix mapping operation, thereby generating an initial causal mapping feature vector. Based on the correlation strength of each matrix element in the parameter causal correlation matrix, a causal weight reconstruction operation is performed on the initial causal mapping feature vector to generate a parameter coupling feature vector; Perform sequence rearrangement on the parameter-coupled feature vectors to generate causal decoupled feature sequences.

4. The PDA-based intelligent water quality detection system according to claim 1, characterized in that, The process of performing function modulation operations in the drift modulation calculation module includes: Receive the causal decoupling feature sequence and the drift modulation factor, perform position alignment processing on each parameter feature in the causal decoupling feature sequence and the corresponding drift modulation factor according to the parameter index set, and generate a parameter modulation pairing set; according to the drift modulation factor in the parameter modulation pairing set, determine the function modulation intensity, function modulation interval and function modulation order for each parameter feature, and generate a parameter modulation configuration set; Based on the parameter modulation configuration set, piecewise function modulation operation is performed on each parameter feature, amplitude compression processing is performed on the high drift segment of each parameter feature, and feature preservation processing is performed on the low drift segment of each parameter feature to generate parameter modulation subsequence; Based on the parameter causal correlation matrix, correlation consistency correction is performed on each parameter modulation subsequence to generate parameter consistent modulation subsequence. Boundary smoothing is then performed on the parameter consistent modulation subsequence. Sequence recombination is then performed according to the order of water quality detection parameters corresponding to the parameter index set to generate drift modulation feature sequence.

5. The PDA-based intelligent water quality detection system according to claim 1, characterized in that, The process of generating a water quality status determination feature tensor by performing spatiotemporal context coupling operation in the spatiotemporal context coupling module includes: receiving a multi-parameter coupled feature tensor and a context feature vector; reading the detection time data and detection location data from the context feature vector; determining the corresponding time modulation weight based on the detection time data; determining the corresponding spatial modulation weight based on the detection location data; performing weight modulation operation on each parameter feature in the multi-parameter coupled feature tensor according to the time modulation weight and the spatial modulation weight; adjusting the parameter features under different detection time and detection location conditions according to the corresponding weights; and performing feature aggregation operation on each modulated parameter feature to generate a water quality status determination feature tensor.

6. The PDA-based intelligent water quality detection system according to claim 1, characterized in that, The process of performing water quality status determination calculation in the status determination module includes: receiving an edge lightweight feature tensor and reading the standard threshold range corresponding to each water quality detection parameter in the preset water quality evaluation standard; extracting the feature values ​​corresponding to each water quality detection parameter from the edge lightweight feature tensor according to the parameter index set, performing interval comparison operation between each feature value and the corresponding standard threshold range to generate parameter status determination results; performing status aggregation operation based on the parameter status determination results of each water quality detection parameter, determining an abnormal water quality state when a parameter status determination result exceeds a preset abnormal threshold, and determining a normal water quality state when all parameter status determination results are within the standard threshold range; calculating the water quality abnormality risk level based on the deviation degree of each water quality detection parameter, and generating the corresponding water quality detection result and water quality abnormality risk level.

Images