Water supply plant water production process unit dosing method, system and equipment based on intelligent sensing and self-adaptive regulation and control and medium

Abstract

The invention relates to a waterworks water production process unit dosing method based on intelligent sensing and self-adaptive regulation, and the method comprises the steps: collecting water quality parameters and environment working condition parameters through a multi-dimensional data sensing network, and building a multi-dimensional data flow; constructing a dosing demand prediction model, establishing a nonlinear mapping relationship between the dynamic change of water quality and the dosing amount through space-time correlation analysis, and outputting a real-time dosing amount prediction value; the predicted value and the working condition parameters are mapped to a fuzzy logic controller, fluctuation characteristics are quantized through a membership function, coupling reasoning is executed based on a fuzzy rule base, and a chemical adding amount adjustment value containing a trend compensation item is generated; after the optimized dosage output by the constructed dynamic adjustment model is applied, model parameter iterative updating, fuzzy rule triggering threshold dynamic calibration and membership function reconstruction are triggered through effluent quality deviation monitoring. According to the invention, through closed-loop linkage of machine learning prediction, fuzzy logic decision and dynamic adjustment, a whole-process intelligent dosing regulation and control system is constructed.

Description

Technical Field

[0001] The present invention relates to the technical field of water treatment processes, and particularly to a method, system, device, and medium for dosing process units in a waterworks based on intelligent sensing and adaptive control for tap water production. Background Art

[0002] In the water production process of a waterworks, the dosing link undertakes the core function of adjusting water quality indicators. Chemical agents such as flocculants and disinfectants need to be added according to the physical and chemical properties of the raw water to remove suspended solids and microbial pollutants. In the prior art, dosing control mostly adopts an empirical decision-making mode based on regular manual sampling. Specifically, operators detect the water sample data collected daily or periodically through laboratory instruments, estimate the dosing amount of the chemical agent in combination with historical operation experience, and manually adjust the parameters of actuators such as metering pumps. Although some waterworks have deployed on-line monitoring devices such as pH meters and residual chlorine detectors, these devices can only provide parameter monitoring functions, and the integration and analysis of their data and dosing decisions still rely on manual experience to complete.

[0003] Due to the dynamic fluctuation characteristics of the raw water quality affected by factors such as seasonal changes and sudden pollution, it is difficult for manual experience judgment to accurately match the dosing requirements in real time, resulting in the situation that the dosing amount often deviates from the optimal value in actual operation. For example, when the turbidity of the raw water suddenly changes, the lag of manual adjustment is likely to cause excessive or insufficient dosing of the flocculant. The former leads to waste of chemical agents and may cause secondary pollution, while the latter affects the flocculation effect and then reduces the filtration efficiency. At the same time, the control of the residual chlorine concentration in the traditional method mainly relies on the ex-post detection results and cannot predict the attenuation trend of the disinfectant in the pipe network. Therefore, there is a risk that the fluctuation range of the residual chlorine concentration exceeds the standard limit value, directly affecting the biological safety of drinking water. In addition, there is a lack of a data collaborative analysis mechanism among existing on-line monitoring devices, and each parameter is displayed independently without establishing a multi-dimensional water quality - dosing correlation model, further restricting the response speed and control accuracy of the dosing system. Summary of the Invention

[0004] (1) Technical Problems to be Solved

[0005] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a method, system, device, and medium for dosing process units in a waterworks based on intelligent sensing and adaptive control, which solves the technical problems in the prior art that due to relying on manual experience judgment and discrete monitoring data, the dosing control has response lag, cannot adapt to the dynamic fluctuation of water quality in real time, and lacks predictive control ability.

[0006] (2) Technical Solutions

[0007] In order to achieve the above object, the main technical solutions adopted by the present invention include:

[0008] In a first aspect, an embodiment of the present invention provides a dosing method for a water treatment process unit of a waterworks based on intelligent sensing and adaptive regulation, including:

[0009] Real-time collect water quality parameters and environmental condition parameters through a multi-dimensional data sensing network pre-deployed in at least part of the water treatment process section to obtain a multi-dimensional data stream with time series characteristics;

[0010] Construct a dosing demand prediction model, perform spatio-temporal correlation analysis on the multi-dimensional data stream and historical dosing data, establish a non-linear mapping relationship between the dynamic change of water quality and the dosing amount, and output a dosing amount prediction value corresponding to the real-time water quality dynamic characteristics;

[0011] Map the dosing amount prediction value and environmental condition parameters together into the fuzzy decision space of the constructed fuzzy logic controller, quantify the fluctuation fuzzy characteristics of the input data through membership functions, and perform coupled reasoning based on a preset fuzzy rule base to output a dosing amount adjustment value with a trend prediction compensation term;

[0012] Construct and train a dynamic regulation model according to the dosing amount adjustment value, water quality parameters and environmental condition parameters. After applying the optimized dosing amount output by the dynamic regulation model, adaptively update the parameters of the dynamic regulation model, adjust the trigger threshold of the fuzzy rule base, and reconstruct the membership function by real-time monitoring the deviation between the water quality parameters at the outlet and the target value.

[0013] Optionally, real-time collect water quality parameters and environmental condition parameters through a multi-dimensional data sensing network pre-deployed in at least part of the water treatment process section to obtain a multi-dimensional data stream with time series characteristics, including:

[0014] According to the water flow direction of each water treatment process section, determine the deployment points of the sensor array at least at the inlet, dosing point, sedimentation tank, filter tank and outlet to form a multi-dimensional data sensing network, where the sensor array is a heterogeneous combination including at least one of a pH sensor, a turbidity sensor, a residual chlorine sensor, a temperature sensor and a flow sensor;

[0015] Generate a synchronous acquisition instruction with a nanosecond-level time stamp, control each sensor to perform parallel data acquisition through a time division multiple access mechanism, and automatically add a time stamp mark to the collected original data stream to obtain a multi-dimensional data vector, where each multi-dimensional data vector is a four-tuple structure including a time stamp, a node ID, a water quality parameter value and a device status code;

[0016] Establish a dynamic confidence interval model based on the historical data distribution characteristics, and perform online marking and isolation processing on the multi-dimensional data vectors that deviate from the confidence interval by more than the set deviation threshold;

[0017] Spatiotemporally align and fuse all multi-dimensional data vectors to generate a multi-dimensional data stream with a unified time reference;

[0018] Dynamically select the 4G / 5G / fiber optic transmission path according to network quality, split and transmit the multi-dimensional data stream to the SCADA system and the cloud platform, and implement dual-path redundant transmission during transmission. When the packet loss rate of the primary transmission channel exceeds the preset threshold, automatically activate the auxiliary transmission channel and reconstruct the lost data segment.

[0019] Optionally, construct a chemical dosing demand prediction model, conduct spatiotemporal correlation analysis on the multi-dimensional data stream and historical chemical dosing data, establish a non-linear mapping relationship between the dynamic changes in water quality and the chemical dosing amount, and output the predicted chemical dosing amount corresponding to the real-time water quality dynamic characteristics, including:

[0020] Dynamically align the real-time multi-dimensional data stream with historical chemical dosing data, adopt a sliding window mechanism to extract the time-delay influence characteristics of the water quality parameters in the previous process section on the current chemical dosing point, and establish a spatiotemporal coupling data set;

[0021] Analyze the non-linear correlation between the dynamic changes in each water quality parameter and the chemical dosing amount N hours ago through an attention network, automatically identify the set fluctuation sensitive periods in the spatiotemporal coupling data set, assign 3-5 times the feature weight to the corresponding time series data, and generate an enhanced training sample set containing process sensitivity markers;

[0022] Construct a chemical dosing demand prediction model through a bidirectional LSTM-TCN hybrid network architecture. Based on the enhanced training sample set, use a temporal convolutional network to capture the local mutation patterns of water quality parameters, combine bidirectional LSTM to extract the long-range dependence characteristics of chemical dosing periodicity, and embed a process knowledge constraint unit in the network hidden layer to achieve process knowledge-driven feature crossing;

[0023] Introduce an adversarial training mechanism, construct a simulated scenario set containing extreme water quality fluctuations through a generative adversarial network, use a discriminator to dynamically identify abnormal chemical dosing patterns, and iteratively update the prediction model parameters to make the prediction error of the model ≤ 2.5% within a 95% confidence interval;

[0024] Implement Monte Carlo Dropout dynamic evaluation. When the standard deviation of 5 consecutive prediction cycles is detected to exceed the set threshold, automatically trigger an online active learning process to guide the SCADA system to collect water quality parameters and environmental condition parameters for model retraining;

[0025] Input the water quality dynamic characteristics of the current and the previous 5 process section data into the trained chemical dosing demand prediction model for prediction, so as to output the predicted value corresponding to the future 15-30 second chemical dosing window, and the predicted value includes the basic chemical dosing amount and its confidence interval.

[0026] Optionally, map the predicted chemical dosing value and environmental operating condition parameters to the fuzzy decision space of the constructed fuzzy logic controller, quantify the fluctuating fuzzy characteristics of the input data through membership functions, and perform coupled reasoning based on a preset fuzzy rule base to output the chemical dosing adjustment value with a trend prediction compensation term, including:

[0027] Map the predicted chemical dosing value and environmental operating condition parameters to a multi-dimensional fuzzy decision space, where the decision space includes the deviation domain of the predicted chemical dosing value, the water quality fluctuation domain, and the environmental operating condition parameter domain;

[0028] Construct an adaptively adjustable trapezoidal membership function for each input dimension to quantify the membership degree of each input data in the fuzzy set, where the turning point parameters of each membership function are determined according to historical operating condition data;

[0029] Call the preset fuzzy rule base, perform three-dimensional space rule matching on the membership degree values activated in each dimension, and obtain the preliminary chemical dosing adjustment amount through weighted calculation;

[0030] Perform sliding time window trend analysis, conduct direction consistency detection on the chemical dosing adjustment amount sequence in the recent N control cycles, and when the same-direction adjustment trend appears in consecutive K cycles, generate a non-linear trend prediction compensation term based on the exponential smoothing algorithm;

[0031] Non-linearly superimpose and fuse the trend prediction compensation term and the preliminary chemical dosing adjustment amount to obtain the chemical dosing adjustment value with a trend prediction compensation term.

[0032] Optionally,

[0033] The deviation domain of the predicted chemical dosing value represents the difference between the machine learning predicted value and the actual chemical dosing amount;

[0034] The water quality fluctuation domain includes a composite index composed of the pH change rate and the second derivative of turbidity;

[0035] The environmental operating condition parameter domain integrates the instantaneous flow rate change rate and the temperature gradient value to form a disturbance intensity index.

[0036] Optionally, perform sliding time window trend analysis, conduct direction consistency detection on the chemical dosing adjustment amount sequence in the recent N control cycles, and when the same-direction adjustment trend appears in consecutive K cycles, generate a non-linear trend prediction compensation term based on the exponential smoothing algorithm, including:

[0037] Perform sliding time window data acquisition, extract the chemical dosing adjustment amount sequence in the recent N control cycles with the current moment as the reference, and record the adjustment direction marker values for each cycle;

[0038] Analyze the adjustment direction marker values in consecutive K cycles, and when the same-direction adjustment trend is detected, trigger the exponential smoothing compensation mechanism:

[0039] Automatically adjust the exponential smoothing coefficient α based on the number of consecutive same-direction trend cycles, where α = basic smoothing coefficient + (number of consecutive same-direction cycles / K) × adaptive gain factor;

[0040] Perform weighted processing on the adjustment amounts of the most recent M same-direction cycles, where the weights decay according to the exponential function of α, and dynamically amplify the calculation result through the non-linear gain coefficient β, and the value of β is determined by the variance σ of the adjustment amount sequence 2 Through β = 1 + log(1 + σ 2 ) is determined;

[0041] Limit the calculated non-linear trend compensation value within a preset safety boundary to generate the final trend prediction compensation term.

[0042] Optionally, construct and train a dynamic adjustment model based on the chemical dosing adjustment value, water quality parameters, and environmental operating condition parameters. After applying the optimized chemical dosing value output by the dynamic adjustment model, adaptively update the parameters of the dynamic adjustment model by real-time monitoring the deviation between the measured water quality parameters at the outlet and the target value. The operations of adjusting the trigger threshold of the fuzzy rule base and reconstructing the membership function include:

[0043] Taking the chemical dosing adjustment value, real-time water quality parameters, and environmental operating condition parameters as inputs, establish a dynamic adjustment model with adjustable regression coefficients through the linear regression algorithm. During the training process, optimize the regression coefficients by minimizing the error between the predicted value and the actual chemical dosing amount, and output the optimized chemical dosing value at the current moment;

[0044] Send the optimized chemical dosing value at the current moment to the chemical dosing mechanism for the chemical dosing process, and real-time monitor the deviation between the measured value of the water quality parameters at the outlet and the target value. When the cumulative deviation amount of N consecutive sampling cycles exceeds the preset threshold, trigger the adaptive adjustment mechanism:

[0045] Adopt a sliding window mechanism to select the latest N sets of operation data, and recalculate the regression coefficients to minimize the mean square error between the predicted chemical dosing amount and the actual chemical dosing amount;

[0046] Recalculate the rule confidence weights based on the statistical distribution characteristics of the deviation amount, and automatically eliminate the redundant rules whose triggering frequency is lower than the set value;

[0047] Based on the distribution change characteristics of historical operating condition data, dynamically reconstruct the membership function parameters of the input dimension, and use the sliding window statistical method to update the turning points of the trapezoidal function to ensure that the fuzzy quantization result matches the current water quality parameters at the outlet;

[0048] Synchronously apply the updated dynamic adjustment model parameters, optimized fuzzy rule base, and reconstructed membership function to the next control cycle to form a closed-loop optimization link.

[0049] Second aspect, an embodiment of the present invention provides a chemical dosing system for a water treatment process unit in a waterworks based on intelligent sensing and adaptive regulation, including:

[0050] A data stream generation module, configured to collect water quality parameters and environmental condition parameters in real time through a multi-dimensional data sensing network pre-deployed in at least part of the water treatment process section, and obtain a multi-dimensional data stream with time series characteristics;

[0051] A chemical dosing amount prediction module, configured to construct a chemical dosing demand prediction model, perform spatio-temporal correlation analysis on the multi-dimensional data stream and historical chemical dosing data, establish a non-linear mapping relationship between the dynamic change of water quality and the chemical dosing amount, and output a chemical dosing amount prediction value corresponding to the real-time water quality dynamic characteristics;

[0052] A chemical dosing amount adjustment module, configured to map the chemical dosing amount prediction value and the environmental condition parameters together into the fuzzy decision space of the constructed fuzzy logic controller, quantify the fluctuation fuzzy characteristics of the input data through a membership function, and perform coupled reasoning based on a preset fuzzy rule base, and output a chemical dosing amount adjustment value with a trend prediction compensation term;

[0053] A chemical dosing amount optimization and parameter update module, configured to construct and train a dynamic regulation model according to the chemical dosing amount adjustment value, water quality parameters and environmental condition parameters, and after applying the chemical dosing amount optimization value output by the dynamic regulation model, adaptively perform parameter update of the dynamic regulation model, trigger threshold adjustment of the fuzzy rule base and reconstruction operation of the membership function by real-time monitoring the deviation between the water quality parameters at the outlet and the target value.

[0054] Third aspect, an embodiment of the present invention provides a chemical dosing device for a water treatment process unit in a waterworks based on intelligent sensing and adaptive regulation, including:

[0055] A multi-dimensional data sensing network, pre-deployed in at least part of the water treatment process section, for multi-source heterogeneous data including water quality parameters and environmental condition parameters;

[0056] A cloud platform, configured to store the collected multi-source heterogeneous data, historical operation data sets and equipment maintenance logs;

[0057] An SCADA control platform, respectively connected to the multi-dimensional data sensing network and the cloud platform, and configured to execute the chemical dosing method for the water treatment process unit in the waterworks based on intelligent sensing and adaptive regulation as described above;

[0058] An edge computing node, deployed in at least part of the water treatment process section, and realizing computing power collaboration with the SCADA control platform through dynamic containerization deployment.

[0059] Fourthly, an embodiment of the present invention provides a computer-readable medium, on which computer-executable instructions are stored. When the executable instructions are executed by a processor, the above-mentioned dosing method for the water treatment process unit of a waterworks based on intelligent sensing and adaptive regulation is implemented.

[0060] (III) Beneficial effects

[0061] The beneficial effects of the present invention are as follows:

[0062] Firstly, the all-time and multi-dimensional data acquisition system constructed based on the multi-dimensional data perception network breaks through the time resolution limitation of traditional discrete monitoring data. Through the fusion analysis of time-series water quality parameters and environmental condition parameters, the continuous capture of the dynamic characteristics of water quality fluctuations is realized, and the problem of response lag caused by manual experience judgment is solved.

[0063] Furthermore, the dosing demand prediction model can autonomously extract dosing decision features from complex water quality change patterns through the non-linear mapping relationship established by machine learning algorithms, replacing the extensive regulation method that relies on the subjective experience of operators. The synergistic effect of this model and the fuzzy logic controller forms a dual-modal decision-making mechanism: on the one hand, machine learning is used to model the deep relationship between water quality parameters and dosing demands, and on the other hand, the membership function is used to quantitatively characterize the fuzzy characteristics of water quality fluctuations. Combining the preset fuzzy rule base, multi-dimensional coupling reasoning of the dosing adjustment value is realized, so as to significantly improve the real-time performance and smoothness of the regulation response when dealing with sudden water quality disturbances.

[0064] At the same time, the adaptive update mechanism of the dynamic regulation model triggers a triple adjustment strategy of iterative optimization of model parameters, dynamic calibration of fuzzy rule triggering thresholds, and reconstruction of the membership function structure by continuously tracking the water quality deviation at the outlet, enabling the system to have the ability to continuously learn the law of water quality evolution. This mechanism can not only immediately correct the current dosing deviation, but also perform forward-looking parameter pre-adjustment based on the prediction of water quality change trends, completely changing the inherent defect of the traditional method lacking predictive regulation ability.

[0065] Therefore, through the closed-loop linkage of the machine learning prediction module, the fuzzy logic decision-making module and the dynamic regulation module, the present invention constructs an intelligent regulation system covering the entire process of "data perception - demand prediction - dynamic decision-making - feedback optimization", realizing the dual improvement of dosing control accuracy and response rate without any manual intervention, and ensuring that the water treatment process unit can still maintain stable dosing control performance in the face of complex working conditions such as sudden changes in source water quality and flow fluctuations. Description of the drawings

[0066] Figure 1 It is a schematic flow chart of the method provided by the embodiment of the present invention;

[0067] Figure 2 Schematic diagram of the specific process of step S1 of the method provided by the embodiment of the present invention;

[0068] Figure 3 Schematic diagram of the specific process of step S2 of the method provided by the embodiment of the present invention;

[0069] Figure 4 Flow chart of machine learning and fuzzy logic control of the method provided by the embodiment of the present invention;

[0070] Figure 5 Schematic diagram of the specific process of step S3 of the method provided by the embodiment of the present invention;

[0071] Figure 6 Schematic diagram of the specific process of step S34 of the method provided by the embodiment of the present invention;

[0072] Figure 7 Schematic diagram of the specific process of step S4 of the method provided by the embodiment of the present invention;

[0073] Figure 8 Process diagram of the dynamic adjustment model of the method provided by the embodiment of the present invention;

[0074] Figure 9 Process diagram of the adaptive adjustment of the method provided by the embodiment of the present invention;

[0075] Figure 10 Schematic diagram of the system composition provided by the embodiment of the present invention;

[0076] Figure 11 Schematic diagram of the overall process of the method provided by the embodiment of the present invention. Detailed implementation manners

[0077] In order to better explain the present invention for easy understanding, the present invention will be described in detail below with reference to the accompanying drawings through specific implementation manners.

[0078] Such as Figure 1As shown in the figure, a dosing method for water treatment process units in a waterworks based on intelligent sensing and adaptive regulation proposed in an embodiment of the present invention includes: real-time collecting water quality parameters and environmental condition parameters through a multi-dimensional data sensing network pre-deployed in at least part of the water treatment process section to obtain a multi-dimensional data stream with time series characteristics; constructing a dosing demand prediction model, performing spatio-temporal correlation analysis on the multi-dimensional data stream and historical dosing data, establishing a non-linear mapping relationship between the dynamic changes in water quality and the dosing amount, and outputting a predicted dosing amount value corresponding to the real-time water quality dynamic characteristics; mapping the predicted dosing amount value and environmental condition parameters together into the fuzzy decision space of the constructed fuzzy logic controller, quantifying the fluctuation fuzzy characteristics of the input data through membership functions, and performing coupled reasoning based on a preset fuzzy rule base to output a dosing amount adjustment value with a trend prediction compensation term; constructing and training a dynamic regulation model according to the dosing amount adjustment value, water quality parameters, and environmental condition parameters, and after applying the optimized dosing amount value output by the dynamic regulation model, adaptively performing parameter update of the dynamic regulation model, trigger threshold adjustment of the fuzzy rule base, and reconstruction operation of the membership function by real-time monitoring the deviation between the water quality parameters at the outlet and the target value.

[0079] First of all, the full-time and multi-dimensional data acquisition system based on the multi-dimensional data sensing network breaks through the time resolution limit of traditional discrete monitoring data. Through the fusion analysis of time-series water quality parameters and environmental condition parameters, the continuous capture of the dynamic characteristics of water quality fluctuations is realized, and the problem of response lag caused by manual experience judgment is solved.

[0080] Furthermore, the non-linear mapping relationship established by the dosing demand prediction model through machine learning algorithms can independently extract dosing decision-making characteristics from complex water quality change patterns, replacing the extensive regulation method that relies on the subjective experience of operators. The synergistic effect of this model and the fuzzy logic controller forms a dual-mode decision-making mechanism: on the one hand, machine learning is used to model the deep relationship between water quality parameters and dosing demand, and on the other hand, the fuzzy characteristics of water quality fluctuations are quantified and characterized through membership functions, and multi-dimensional coupled reasoning of the dosing amount adjustment value is realized in combination with a preset fuzzy rule base, so as to significantly improve the real-time performance and smoothness of the regulation response when dealing with sudden water quality disturbances.

[0081] At the same time, the adaptive update mechanism of the dynamic regulation model triggers a triple adjustment strategy of iterative optimization of model parameters, dynamic calibration of the fuzzy rule trigger threshold, and reconstruction of the membership function structure by real-time tracking the water quality deviation at the outlet, enabling the system to have the ability to continuously learn the law of water quality evolution. This mechanism can not only immediately correct the current dosing amount deviation, but also perform forward-looking parameter pre-adjustment based on the prediction of the water quality change trend, completely changing the inherent defect of the traditional method lacking predictive regulation ability.

[0082] Thus, through the closed-loop linkage of the machine learning prediction module, the fuzzy logic decision-making module, and the dynamic adjustment module, the present invention constructs an intelligent control system covering the entire process of "data perception - demand prediction - dynamic decision-making - feedback optimization", achieving a double improvement in the dosing control accuracy and response rate without any manual intervention, and ensuring that the water treatment process unit can still maintain stable dosing control performance in the face of complex working conditions such as sudden changes in source water quality and flow fluctuations.

[0083] To better understand the above technical solution, the exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0084] Specifically, an embodiment of the present invention provides a dosing method for a water treatment process unit of a waterworks based on intelligent sensing and adaptive control, which is characterized by including:

[0085] S1. Real-time collect water quality parameters and environmental condition parameters through a multi-dimensional data perception network pre-deployed in at least part of the water treatment process section to obtain a multi-dimensional data stream with time series characteristics.

[0086] Further, as Figure 2 shown, step S1 includes:

[0087] S11. According to the water flow direction of each water treatment process section, determine the deployment points of the sensor array at least at the water inlet, dosing point, sedimentation tank, filter tank, and water outlet to form a multi-dimensional data perception network, where each process section includes at least one water treatment process unit, and the sensor array is a heterogeneous combination including at least one of a pH sensor, a turbidity sensor, a residual chlorine sensor, a temperature sensor, and a flow sensor.

[0088] In this step, according to the water treatment process characteristics, the water quality monitoring sensor array (pH, residual chlorine, turbidity) and the environmental monitoring nodes (temperature, flow, pressure, etc.) are collaboratively deployed at the key nodes of the process units such as the water inlet, dosing reaction area, sedimentation tank, and water outlet to construct a distributed perception architecture with spatial topological association, ensuring the high-precision, long-term stability, and efficient operation of the sensors.

[0089] S12. Generate a synchronous acquisition instruction with a nanosecond-level time stamp, control each sensor to perform parallel data acquisition through a time division multiple access mechanism, and automatically add a time stamp mark to the collected original data stream to obtain a multi-dimensional data vector.

[0090] In this step, a time-division multiple access (TDMA) mechanism is adopted to divide the acquisition time slots. Each node completes data sampling within the allocated 500 μs time slot window and adds four-dimensional metadata tags to the original data stream to form the following structured data packet: [timestamp (IEEE 1588 precise clock protocol), node ID (SHA-256 encrypted encoding), water quality parameter value (floating-point double precision), device status code (16-bit binary status word)].

[0091] S13. Establish a dynamic confidence interval model based on the historical data distribution characteristics, and perform online marking and isolation processing on the multi-dimensional data vectors that deviate from the confidence interval by more than the set deviation threshold.

[0092] Specifically, set the dynamic confidence interval as [μ - 2.5κ, μ + 3κ]. When the real-time data deviates from the interval, a three-level alarm is triggered: Level 1 alarm (deviation of 2.5 - 3κ): Mark the anomaly but do not discard the data; Level 2 alarm (3 - 4κ): Isolate the data and start calibrating the backup sensor; Level 3 alarm (>4κ): Automatically switch to the redundant sensor channel.

[0093] S14. Align and fuse all multi-dimensional data vectors in space and time to generate a multi-dimensional data stream with a unified time reference.

[0094] S15. Dynamically select the 4G / 5G / fiber optic transmission path according to the network quality, split and transmit the multi-dimensional data stream to the SCADA system and the cloud platform, and implement dual-path redundant transmission during transmission. When the packet loss rate of the main transmission channel exceeds the preset threshold, automatically activate the auxiliary transmission channel and reconstruct the lost data segment.

[0095] It should be noted that the present invention can also establish a circular buffer locally. When the network delay is detected to exceed 200 ms, the local circular buffer is automatically enabled to store the multi-dimensional data stream of the most recent 60 seconds, and perform data retransmission with priority after the network is restored.

[0096] S2. Construct a chemical dosing demand prediction model, perform spatio-temporal correlation analysis on the multi-dimensional data stream and the historical chemical dosing data, establish a non-linear mapping relationship between the dynamic change of water quality and the chemical dosing amount, and output the chemical dosing amount prediction value corresponding to the real-time water quality dynamic characteristics.

[0097] Further, as Figure 3 shown, step S2 includes:

[0098] S21. Dynamically align the real-time multi-dimensional data stream with the historical chemical dosing data, adopt a sliding window mechanism to extract the time-delay influence characteristics of the water quality parameters in the previous process section on the current chemical dosing point, and establish a spatio-temporal coupling data set including spatial propagation delay and chemical reaction kinetics.

[0099] Fuse the following data through a sliding window mechanism (window length: 30 seconds, step size: 5 seconds): real-time multi-dimensional data stream: extract dynamic features such as the second derivative of pH value, chlorine decay rate (ΔC / Δt), and turbidity time series variation coefficient; historical chemical dosing data: analyze the chemical concentration gradient curve within the working cycle of the dosing pump (sampling rate: 10 Hz); process kinetics compensation: introduce the hydraulic retention time (HRT) of the reaction tank as a spatial propagation delay correction factor. Finally, establish a spatio-temporal coupling matrix containing reaction kinetics, with the dimension of [timestamp × process node × feature channel].

[0100] S22. Analyze the non-linear correlation between the dynamic changes of various water quality parameters and the chemical dosage N hours ago through an attention network, automatically identify the highly fluctuating and sensitive periods of the spatio-temporal coupling data set (such as when the pH mutation exceeds ±0.5 / minute), assign 3-5 times the feature weight to the corresponding time series data, and generate an enhanced training sample set containing process sensitivity markers.

[0101] S23. Construct a chemical dosing demand prediction model through a bidirectional LSTM-TCN hybrid network architecture. Based on the enhanced training sample set, use the temporal convolutional network to capture the local mutation patterns of water quality parameters, combine the bidirectional LSTM to extract the long-range dependence features of chemical dosing periodicity, and embed process knowledge constraint units in the network hidden layer to achieve process knowledge-driven feature crossing (such as the exponential relationship constraint between chlorine decay rate and pH value).

[0102] Among them, the bidirectional LSTM-TCN coupling model includes:

[0103] Temporal convolutional module: Configure 8 layers of dilated causal convolution (dilation factor d = 1, 2, 4, 8, 16, 32, 64, 128), with 64 filters in each layer, to capture local features such as turbidity mutation and sudden chlorine drop.

[0104] Bidirectional LSTM module: The hidden layer dimension is 256, which extracts the long-range dependence of chemical dosing periodicity (period range: 5 minutes - 2 hours).

[0105] Process knowledge constraint unit: Embed in the fully connected layer constraints such as the exponential relationship between chlorine decay rate and pH value: f(pH,Cl 2 ) = η·e -λ(pH-0.7) ·Cl 2 1.5 , where η and λ are learnable parameters, forcing the network to conform to the chemical kinetics law of chlorine decay.

[0106] S24. Introduce an adversarial training mechanism, construct a simulation scenario set containing extreme water quality fluctuations (such as ammonia nitrogen concentration exceeding the standard by 500%) through a generative adversarial network (GAN), use a discriminator to dynamically identify abnormal dosing patterns (samples with an error rate > 15%), and iteratively update the prediction model parameters to make the prediction error of the model within a 95% confidence interval ≤ 2.5%.

[0107] S25. Implement Monte Carlo Dropout dynamic evaluation. When the standard deviation of 5 consecutive prediction cycles exceeds the set threshold, automatically trigger an online active learning process to guide the SCADA system to collect water quality parameters and environmental condition parameters for model retraining, update the weights of the dosing demand prediction model without changing the topology structure, and ensure the dynamic evolution of the model.

[0108] S26. Input the water quality dynamic characteristics of the current and the previous 5 process sections into the trained dosing demand prediction model for prediction, so as to output the predicted values corresponding to the dosing window in the future 15 - 30 seconds. The predicted values include the basic dosing amount and its confidence interval (±2.5%).

[0109] In addition, the trained dosing demand prediction model can be converted into the TensorRT format through layer fusion and quantization compression and deployed on the edge computing unit to achieve a real-time inference response at the 50ms level, while retaining a prediction accuracy of more than 98.7% of the original model.

[0110] S3. Map the dosing prediction value and the environmental condition parameters to the fuzzy decision space of the constructed fuzzy logic controller together, quantify the fluctuating fuzzy characteristics of the input data through the membership function, and perform coupled reasoning based on the preset fuzzy rule base to output the dosing adjustment value with a trend prediction compensation term.

[0111] As Figure 4 shown, after predicting the dosing demand, fuzzy logic control (FLC) is used to further process these predicted values to flexibly adjust the dosing amount. Fuzzy logic can effectively handle the uncertainty and fuzziness in water quality fluctuations. For example, when the water quality fluctuates greatly or changes rapidly, fuzzy control can smoothly adjust the dosing amount to avoid problems such as over-dosing or under-dosing. The inputs of the fuzzy logic control system include factors such as the dosing prediction value output by the prediction model and the real-time water quality fluctuations, and the dosing adjustment value is output through fuzzy rule reasoning.

[0112] Furthermore, as Figure 5 shown, step S3 includes:

[0113] S31. Map the dosing prediction value and the environmental condition parameters (instantaneous flow rate change rate, temperature gradient value) to a multi-dimensional fuzzy decision space together. The decision space consists of a dosing prediction value deviation domain, a water quality fluctuation domain, and an environmental condition parameter domain.

[0114] Specifically, the deviation range of the predicted chemical dosage value characterizes the difference between the machine learning predicted value and the actual chemical dosage; the water quality fluctuation range includes a composite index composed of the pH change rate and the second derivative of turbidity; the environmental condition parameter range integrates the instantaneous flow rate change rate and the temperature gradient value to form a disturbance intensity index.

[0115] S32. Construct an adaptively adjustable trapezoidal membership function for each input dimension to quantify the membership degree of each input data in the fuzzy set.

[0116] In one embodiment, the membership function is used to map precise input values (such as water quality parameter changes, chemical dosage requirements, etc.) to fuzzy sets (such as "small", "medium", "large", etc.), so as to determine the membership degree of the input value in each fuzzy set. In this way, the fuzzy logic control system can handle the uncertainty and ambiguity of input data and provide a basis for subsequent fuzzy reasoning and decision-making;

[0117] For inputs related to the degree of water quality fluctuation, a trapezoidal membership function can be used. The trapezoidal membership function can better describe fuzzy concepts with a certain transition range. For example, the situation of "small" water quality fluctuation may not be an exact numerical range, but gradually transitions to the situation of "medium" fluctuation within an interval, and the trapezoidal membership function can accurately express this ambiguity.

[0118] It should be noted that the turning point parameters of each membership function are determined according to historical operating condition data. The turning points of the membership function of the deviation range of the predicted chemical dosage value are dynamically calibrated according to the distribution characteristics of historical prediction errors. The turning points of its trapezoidal membership function are dynamically calibrated based on the 95% quantile of the prediction errors in the previous 30 days. For example, when the historical maximum error is 12%, the turning points are set to [8%, 15%]; the parameters of the membership function of the water quality fluctuation range are automatically adjusted based on the extreme value fluctuation range of turbidity in the previous N hours. The base width of its membership function is automatically expanded according to the turbidity range difference in the previous 6 hours. If the turbidity fluctuation range > 50 NTU / h, the base width is expanded by 1.8 times; the base width of the membership function of the environmental condition parameter range is positively correlated with the historical mutation frequency of the influent flow rate, and the shape parameter of its membership function is positively correlated with the influent flow rate mutation frequency. When the flow rate mutation frequency > 5 times / hour, the right deviation slope of the function increases by 40%.

[0119] S33. Call the preset fuzzy rule base, perform three-dimensional space rule matching on the membership degree values activated in each dimension, and obtain the preliminary chemical dosage adjustment amount through weighted calculation.

[0120] Taking the "small" fuzzy set of ΔpH t (the change amount of pH value at the current moment t) as an example:

[0121] Let ΔpHt The trapezoidal membership function of the "small" fuzzy set is as follows:

[0122] f(ΔpH t ) = 1 when ΔpH t ≤ a;

[0123]

[0124] f(ΔpH t ) = 0 when ΔpH t > b;

[0125] where a and b are parameters. a represents the lower limit value at which ΔpH t begins to deviate from the "small" range, and b represents the upper limit value that completely does not belong to the "small" range. For example, when a = 0.1 and b = 0.3, if ΔpH t = 0.2, then ΔpH t = 0.2 indicates that the membership degree in the "small" fuzzy set is 0.5. By collecting a large amount of historical water quality data and corresponding chemical dosing adjustment data, analyze the variation ranges of different water quality parameters and the actual chemical dosing adjustment situations. For the membership function parameters related to the degree of water quality fluctuation, such as the parameters of the "small", "medium", "large", etc. fuzzy sets of ΔpH t , initially determine them by statistically analyzing the distribution in historical data.

[0126] When real-time data is input into the fuzzy logic controller, first calculate the membership degree of each input value in the corresponding fuzzy set according to the membership function. For example, for the value at the current moment, calculate its membership degrees in the "small", "medium", "large", etc. fuzzy sets through the above trapezoidal membership function. Then, the fuzzy inference engine performs inference calculations based on the preset fuzzy rules and the membership degrees of the input values. During the inference process, the membership degrees participate in the calculation as weights, and finally obtain a fuzzy output, and then convert the fuzzy output into an accurate chemical dosing adjustment value to achieve precise control of the chemical dosing operation.

[0127] S34. Perform a sliding time window trend analysis, conduct a direction consistency detection on the chemical dosing adjustment amount sequence of the most recent N control cycles. When K consecutive cycles show a same-direction adjustment trend, generate a non-linear trend prediction compensation term based on the exponential smoothing algorithm.

[0128] Furthermore, as Figure 6 shown, step S34 includes:

[0129] S341. Perform sliding time window data acquisition, extract the chemical dosing adjustment amount sequence of the most recent N control cycles with the current moment as the reference, and record the adjustment direction marker values of each cycle.

[0130] S342. Analyze the adjustment direction marker values for consecutive K periods. When a same-direction adjustment trend is detected, trigger the exponential smoothing compensation mechanism.

[0131] S343. Automatically adjust the exponential smoothing coefficient α based on the number of consecutive same-direction trend periods; where α = basic smoothing coefficient + (number of consecutive same-direction periods / K) × adaptive gain factor.

[0132] S344. Perform weighted processing on the adjustment amounts for the most recent M same-direction periods, where the weights decay according to the exponential function of α, and dynamically amplify the calculation result through the non-linear gain coefficient β; where the value of β is determined by the variance σ of the adjustment amount sequence 2 through β = 1 + log(1 + σ 2 ) is determined.

[0133] S345. Limit the calculated non-linear trend compensation value within a preset safety boundary to generate the final trend prediction compensation term.

[0134] S35. Non-linearly superimpose and fuse the trend prediction compensation term with the preliminary chemical dosing adjustment amount to obtain the chemical dosing adjustment value with the trend prediction compensation term.

[0135] In this embodiment, through the dynamically adjusted exponential smoothing coefficient and non-linear amplification mechanism, the trend compensation amount can avoid overshoot while maintaining the response speed. The safety boundary function dynamically expands the compensation range according to the current working conditions, improving the adjustment sensitivity on the premise of ensuring system stability. At the same time, by dynamically balancing the immediate control demand and trend prediction correction, the system can not only quickly eliminate the current water quality deviation, but also compensate for potential disturbances in the next 3 - 5 control periods in advance based on trend prediction, ultimately achieving a smooth transition and accurate tracking of the chemical dosing adjustment.

[0136] S4. Construct and train a dynamic adjustment model based on the chemical dosing adjustment value, water quality parameters, and environmental condition parameters. After applying the optimized chemical dosing value output by the dynamic adjustment model, adaptively update the parameters of the dynamic adjustment model, adjust the trigger threshold of the fuzzy rule base, and reconstruct the membership function by real-time monitoring the deviation between the water quality parameters at the outlet and the target value.

[0137] The present invention utilizes the above-provided chemical dosing demand prediction and fuzzy logic control strategies, combined with the real-time monitored water quality changes (including the speed, amplitude, and change trend of water quality fluctuations, etc.), to dynamically adjust the chemical dosing in real time. It can not only respond promptly to the immediate fluctuations of water quality, but also predict the future chemical dosing demand in advance according to the water quality change trend, realizing true adaptive control. This dynamic adjustment process is based on the above chemical dosing demand prediction model and fuzzy logic control strategy, and is further refined and optimized in real time on this basis to ensure that the chemical dosing accuracy always meets the requirements of water quality stability.

[0138] Further, as Figure 7 shown, step S4 includes:

[0139] S41. Taking the chemical addition amount adjustment value, real-time water quality parameters, and environmental condition parameters as inputs, establishing a dynamic adjustment model with adjustable regression coefficients through a linear regression algorithm, and optimizing the regression coefficients by minimizing the error between the predicted value and the actual chemical addition amount during the training process, and outputting the optimized value of the chemical addition amount at the current moment.

[0140] In a specific embodiment, as Figure 8 shown, to achieve the goal of precise chemical addition, a dynamic adjustment model considering multiple factors is constructed. The model inputs include real-time water quality parameters (assuming the pH value is pH t ., the turbidity is Turbidity t , the ammonia nitrogen content is Ammonia t , etc., where t represents the current moment), the flow rate Flow t , the temperature Temp t , and the chemical addition amount demand Dose r eq output by the machine learning model in step two. The dynamic adjustment model formula:

[0141] Dose t = β 0 + β 1 ×pH t + β 2 ×Turbidity t + β 3 ×Ammonia t + β 4 ×Flow t + β 5 ×Temp t + β 6 ×Dose r eq;

[0142] In the formula, β 0 - β 6 are the regression coefficients of the model. The above formula is constructed through training with a large amount of historical data to establish a quantitative relationship between water quality parameters, flow rate, temperature, chemical addition amount demand, and actual chemical addition amount, enabling the model to accurately predict the chemical addition amount based on real-time input data. During the training process, the model continuously adjusts the coefficients to minimize the error between the predicted chemical addition amount and the actual chemical addition amount, thereby achieving precise chemical addition control;

[0143] For example: at a certain moment, the real-time data collected is pH t = 7.5, Turbidity t = 5 NTU, Ammonia t= 0.5 mg / L, Flow t = 1000 m 3 / h, Temp t = 20 °C, Dose output by the machine learning model in Step 2 r eq = 50 mg / L. At this time, if the regression coefficient β obtained through training 0 = 10, β 1 = 2, β 2 = 3, β 3 = 1, β 4 = 0.5, β 5 = 0.2, β 6 = 0.8, then according to the above formula, the actual required chemical dosage at the current moment can be calculated as Dose t :

[0144] Dose t = 10 + 2×7.5 + 3×5 + 1×0.5 + 0.5×1000 + 0.2×20 + 0.8×50

[0145] = 10 + 15 + 15 + 0.5 + 500 + 4 + 40

[0146] = 584.5 mg / L;

[0147] Through the above calculation process, the model can accurately predict the required chemical dosage at the current moment based on various real-time input data and the regression coefficients obtained through training, providing precise guidance for subsequent chemical dosing operations. At the same time, as the system continues to run and new data accumulates continuously, the model will continuously update the regression coefficients to adapt to changes in water quality and operating conditions, further improving the accuracy of chemical dosing prediction.

[0148] S42. Feed the optimized value of the chemical dosage at the current moment to the chemical dosing mechanism for the chemical dosing process, and continuously monitor the deviation between the measured value and the target value of the water quality parameters at the outlet. When the cumulative deviation over N consecutive sampling periods exceeds the preset threshold, trigger the adaptive adjustment mechanism.

[0149] In one embodiment, the following adaptive adjustment mechanism is triggered: The sliding window mechanism is adopted to select the latest N groups of operation data, and the regression coefficients are recalculated to minimize the mean square error between the predicted chemical dosage and the actual chemical dosage. The rule confidence weights are recalculated based on the statistical distribution characteristics of the deviation amount, and the redundant rules with a triggering frequency lower than the set value are automatically eliminated. In addition, supplementary rule entries are generated according to the current working conditions (such as the combination mode of high turbidity and low pH), and the new rules need to pass the process safety verification (such as the water quality stability test after simulating a ±20% mutation in the chemical dosage). And. Based on the change characteristics of the historical working condition data distribution, the membership function parameters of the input dimension are dynamically reconstructed, and the turning points of the trapezoidal function are updated by using the sliding window statistical method to ensure that the fuzzy quantization result matches the current water quality parameters at the water outlet.

[0150] In another embodiment, as Figure 9 shown, to enhance the system adaptability, an adaptive adjustment mechanism is introduced. The model is evaluated based on the historical chemical dosing effect data (such as the water quality compliance rate, chemical consumption cost, etc.), and evaluation indicators are defined;

[0151] For example:

[0152] Water Quality Compliance Index and Chemical Consumption Cost Index

[0153] When the QDI is lower than the set threshold or the CDI is higher than the set threshold, the model adaptive adjustment is started. The adjustment strategies include dynamically adjusting the parameter update of the model, adjusting the triggering threshold of the fuzzy rule base, and reconstructing the membership function. Specifically, it is to adjust the weights of the input variables of the dynamic adjustment model, optimize the fuzzy rules of the fuzzy logic control (refine or modify the fuzzy rules according to the actual operation conditions), and adjust the learning rate (such as appropriately increasing the learning rate when the water quality fluctuates greatly to accelerate the model convergence speed), etc. Through the adaptive adjustment, the system continuously optimizes its performance, adapts to different water quality conditions and operation requirements, and realizes long-term stable and accurate chemical dosing control. The comprehensive evaluation system with QDI as the core realizes the upgrade from "passive correction" to "active prevention", enabling the system to achieve the optimal balance between chemical consumption cost and water quality compliance.

[0154] S46. Synchronously apply the updated dynamic adjustment model parameters, the optimized fuzzy rule base, and the reconstructed membership function to the next control cycle to form a closed-loop optimization link.

[0155] In addition, an embodiment of the present invention provides a chemical dosing system for a water treatment process unit of a waterworks based on intelligent sensing and adaptive regulation, including:

[0156] A data stream generation module, configured to collect water quality parameters and environmental working condition parameters in real time through a multi-dimensional data sensing network pre-deployed in at least part of the water treatment process sections to obtain a multi-dimensional data stream with time series characteristics.

[0157] The chemical dosing amount prediction module is used to construct a chemical dosing demand prediction model, conduct spatio-temporal correlation analysis on multi-dimensional data streams and historical chemical dosing data, establish a non-linear mapping relationship between the dynamic change of water quality and the chemical dosing amount, and output the chemical dosing amount prediction value corresponding to the real-time water quality dynamic characteristics.

[0158] The chemical dosing amount adjustment module is used to jointly map the chemical dosing amount prediction value and environmental condition parameters to the fuzzy decision space of the constructed fuzzy logic controller, quantify the fluctuation fuzzy characteristics of the input data through membership functions, and perform coupled reasoning based on a preset fuzzy rule base, and output the chemical dosing amount adjustment value with a trend prediction compensation term.

[0159] The chemical dosing amount optimization and parameter update module is used to construct and train a dynamic adjustment model according to the chemical dosing amount adjustment value, water quality parameters and environmental condition parameters. After applying the chemical dosing amount optimization value output by the dynamic adjustment model, the parameters of the dynamic adjustment model are updated adaptively by real-time monitoring the deviation between the water quality parameters at the water outlet and the target value, the trigger threshold of the fuzzy rule base is adjusted, and the membership function is reconstructed.

[0160] Furthermore, the embodiment of the present invention also provides a chemical dosing device for a water treatment process unit of a waterworks based on intelligent sensing and adaptive regulation, including: a multi-dimensional data perception network, which is pre-deployed in at least part of the water treatment process section and is used for multi-source heterogeneous data including water quality parameters and environmental condition parameters; a cloud platform, which is used to store the collected multi-source heterogeneous data, historical operation data sets and equipment maintenance logs; an SCADA control platform, which is respectively connected to the multi-dimensional data perception network and the cloud platform and is used to execute the above-mentioned chemical dosing method for the water treatment process unit of the waterworks based on intelligent sensing and adaptive regulation; an edge computing node, which is deployed in at least part of the water treatment process section and realizes computing power cooperation with the SCADA control platform through dynamic containerization deployment.

[0161] Such as Figure 10As shown, in this embodiment, the present invention constructs a comprehensive, intelligent, and adaptive precise dosing system through a series of steps such as data collection and upload, dosing strategy setting, dynamic adjustment algorithm setting, and cloud platform construction and data warehouse building. The data collection and upload link ensures the accuracy, integrity, and real-time nature of the data, providing a solid data foundation for subsequent decision-making. The combination of machine learning and fuzzy logic control in the dosing strategy setting realizes the precise prediction and flexible adjustment of the dosing amount, effectively coping with the uncertainty of water quality fluctuations. Subsequently, a dynamic adjustment algorithm is adopted to further optimize the dosing process, enabling it to dynamically adjust according to real-time situations and ensuring that the dosing accuracy always meets the requirements of water quality stability. The construction of the cloud platform and data warehouse provides strong support for data storage, analysis, and system optimization, realizing the effective utilization of resources and the continuous improvement of system performance. At the same time, the automatic fault handling analysis and result feedback mechanism ensure the reliability and stability of the system. Thus, this system can significantly improve the dosing accuracy, intelligent level, and water supply quality of water treatment plants, reduce the chemical consumption cost, enhance the overall performance of the water supply system, and has broad application prospects and important practical significance.

[0162] Then, the embodiment of the present invention also provides a computer-readable medium, on which computer-executable instructions are stored, and when the executable instructions are executed by a processor, the dosing method for the water treatment process unit of a water treatment plant based on intelligent sensing and adaptive regulation as described above is realized.

[0163] In summary, the embodiment of the present invention provides a dosing method, system, device, and medium for the water treatment process unit of a water treatment plant based on intelligent sensing and adaptive regulation. Refer to Figure 11 , first, the present invention collects a large amount of historical water quality data (such as pH, turbidity, ammonia nitrogen, residual chlorine, etc.) and corresponding dosing amount data, trains a model using machine learning algorithms, and deeply learns the complex non-linear relationship between water quality changes and dosing requirements, so as to be able to accurately predict the dosing amount requirements based on real-time water quality data and provide a scientific basis for dosing control. Based on the dosing amount requirements output by the machine learning model, a fuzzy logic control system is introduced, and its unique membership function and fuzzy rules are used to handle uncertain factors such as water quality fluctuations. For example, a suitable trapezoidal membership function is defined for the change amount of water quality parameters (such as ΔpH t ), mapping the precise input to a fuzzy set, and reasoning and calculating according to preset fuzzy rules (such as the corresponding rules for different water quality fluctuation situations and dosing amount adjustments), smoothly adjusting the dosing amount, realizing precise and flexible control of the dosing operation, and effectively avoiding problems such as over-dosing or under-dosing. Introducing a fuzzy logic control system based on the dosing amount requirements output by the machine learning model is the key point of this solution. At the same time, the combination of machine learning and fuzzy logic control is the technical means to achieve precise dosing in this solution and is also the protection point of the solution.

[0164] Secondly, an adaptive adjustment mechanism is established to automatically adjust the weights of the input variables of the dynamic adjustment model, optimize the fuzzy rules of the fuzzy logic control, and adjust the learning rate, so as to continuously optimize the system performance and ensure long-term stable and accurate dosing control. This part is also the key point of this solution.

[0165] Therefore, the present invention significantly improves the dosing accuracy through a series of technical means. In the data collection and upload link, high-precision and long-term stable water quality monitoring sensors (such as pH, turbidity, residual chlorine, etc.) and environmental monitoring devices (such as temperature, flow rate, etc.) are deployed at the key control points of the waterworks, and a variety of data are collected in real time by means of the SCADA system, transmitted using industrial protocols, and data integrity and synchronization mechanisms, redundancy and backup are set to ensure the accuracy and integrity of the data. Based on this, in the dosing strategy setting, the machine learning model is trained using a large amount of historical water quality and dosing data to learn the relationship between water quality changes and dosing requirements, and predict the dosing amount under different water quality conditions; the fuzzy logic control system further processes uncertainties, and through membership functions and fuzzy rules, smoothly adjusts the dosing amount according to factors such as real-time water quality fluctuations, avoiding over-dosing or under-dosing. The dynamic adjustment algorithm optimizes the model parameters according to the real-time feedback data, continuously adapts to the dynamic changes of water quality, and makes the dosing accuracy always meet the requirements of water quality stability, thus effectively ensuring the stable compliance of the effluent water quality and reducing water quality problems caused by inaccurate dosing.

[0166] Subsequently, by combining the machine learning algorithm with the fuzzy logic control, the system can automatically identify the non-linear complex patterns of dosing requirements and water quality fluctuations. During operation, according to the real-time monitoring data (including water quality parameters, flow rate, temperature, etc.) and the dosing amount requirements output by the machine learning model, the fuzzy logic control system dynamically adjusts the dosing amount in real time according to the preset fuzzy rules and membership functions. For example, when the speed and amplitude of water quality fluctuations change, the system can not only respond to the immediate fluctuations in a timely manner to adjust the dosing, but also predict the future dosing requirements according to the change trend and adjust in advance to adapt to the actual situations such as water source changes and flow rate fluctuations in different waterworks, without frequent manual intervention, improving the timeliness and accuracy of control, enhancing the adaptability of the system to complex working conditions, and ensuring the scientificity and effectiveness of the dosing strategy throughout the water treatment process.

[0167] Therefore, by storing the collected historical and real-time data with the help of the cloud platform and data warehouse, deeply understanding the relationship between water quality change laws and dosing effects, and further optimizing the dosing strategy, the drug consumption is minimized on the premise of ensuring water quality compliance.

[0168] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) that contain computer-usable program code.

[0169] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can also be implemented by computer program instructions.

[0170] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications after learning the basic creative concepts. Therefore, the claims should be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present invention.

[0171] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention should also include these modifications and variations.

Claims

1. A method for adding medicine to a water production process unit in a water plant based on intelligent sensing and adaptive regulation, characterized in that: include: Water quality parameters and environmental condition parameters are collected in real time through a multidimensional data sensing network pre-deployed in at least part of the water production process section to obtain a multidimensional data stream with time series characteristics; Construct a dosing demand prediction model, conduct spatiotemporal correlation analysis on multidimensional data streams and historical dosing data, establish a nonlinear mapping relationship between dynamic changes in water quality and dosing amount, and output the dosing amount prediction value corresponding to the real-time dynamic characteristics of water quality; The dosage prediction value and environmental condition parameters are mapped to the fuzzy decision space of the constructed fuzzy logic controller. The fluctuation fuzzy characteristics of the input data are quantified through the membership function, and coupled reasoning is performed based on the preset fuzzy rule base to output the dosage adjustment value with trend prediction compensation items. A dynamic adjustment model is constructed and trained according to the dosage adjustment value, water quality parameters and environmental operating parameters. After applying the dosage optimization value output by the dynamic adjustment model, the deviation between the outlet water quality parameters and the target value is monitored in real time, and the parameters of the dynamic adjustment model are adaptively updated, the trigger threshold of the fuzzy rule base is adjusted, and the membership function is reconstructed.

2. The method for adding medicine to a water production process unit of a water plant based on intelligent sensing and adaptive regulation as claimed in claim 1, characterized in that: By pre-deploying a multi-dimensional data sensing network in at least part of the water production process to collect water quality parameters and environmental condition parameters in real time, a multi-dimensional data stream with time series characteristics is obtained, including: According to the water flow direction of each water production process section, the deployment points of the sensor array are determined at least at the water inlet, the dosing point, the sedimentation tank, the filtration tank and the water outlet to form a multi-dimensional data perception network, wherein the sensor array is a heterogeneous combination including at least one of a pH sensor, a turbidity sensor, a residual chlorine sensor, a temperature sensor and a flow sensor; Generate synchronous acquisition instructions with nanosecond timestamps, control each sensor to perform parallel data acquisition through the time division multiple access mechanism, automatically add timestamps to the collected raw data streams, and obtain multidimensional data vectors, where each multidimensional data vector is a four-tuple structure containing timestamp, node ID, water quality parameter value, and device status code; A dynamic confidence interval model is established based on the distribution characteristics of historical data, and online marking and isolation processing is performed on multidimensional data vectors that deviate from the confidence interval by more than the set deviation threshold. All multi-dimensional data vectors are aligned and fused in time and space to generate a multi-dimensional data stream with a unified time base; Dynamically select 4G / 5G / fiber transmission paths based on network quality, segment and transmit multi-dimensional data streams to the SCADA system and cloud platform. At the same time, implement dual-path redundant transmission during transmission. When the packet loss rate of the main transmission channel exceeds the preset threshold, the auxiliary transmission channel is automatically activated and the lost data segments are reconstructed.

3. The method for adding medicine to a water production process unit of a water plant based on intelligent sensing and adaptive regulation as claimed in claim 1, characterized in that: Construct a dosing demand prediction model, conduct spatiotemporal correlation analysis on multidimensional data streams and historical dosing data, establish a nonlinear mapping relationship between dynamic changes in water quality and dosing amount, and output dosing amount prediction values ​​corresponding to real-time water quality dynamic characteristics, including: Dynamically align the real-time multi-dimensional data stream with the historical dosing data, use the sliding window mechanism to extract the time-lag effect of the water quality parameters of the previous process section on the current dosing point, and establish a spatiotemporal coupled data set; The attention network is used to analyze the nonlinear correlation between the dynamic changes of various water quality parameters and the dosage N hours ago, automatically identify the set fluctuation sensitive period of the spatiotemporal coupling data set, assign 3-5 times the feature weight to the corresponding time series data, and generate an enhanced training sample set containing process sensitivity markers; A dosing demand prediction model is constructed through a bidirectional LSTM-TCN hybrid network architecture. Based on the enhanced training sample set, a temporal convolutional network is used to capture the local mutation pattern of water quality parameters. The bidirectional LSTM is combined to extract the long-range dependency characteristics of dosing periodicity, and a process knowledge constraint unit is embedded in the hidden layer of the network to realize process knowledge-driven feature crossover. An adversarial training mechanism was introduced to construct a set of simulated scenarios containing extreme water quality fluctuations by generating adversarial networks, and the discriminator was used to dynamically identify abnormal dosing patterns and iteratively update the prediction model parameters so that the prediction error of the model within the 95% confidence interval was ≤2.5%; Implement Monte Carlo Dropout dynamic evaluation. When it is detected that the standard deviation of 5 consecutive prediction cycles exceeds the set threshold, the online active learning process is automatically triggered to guide the SCADA system to collect water quality parameters and environmental condition parameters for model retraining. The dynamic characteristics of water quality of the current and previous five process stage data are input into the trained dosing demand prediction model for prediction to output the predicted value corresponding to the future 15-30 second dosing window, which includes the basic dosing amount and its confidence interval.

4. The method for adding medicine to a water production process unit of a water plant based on intelligent sensing and adaptive regulation as claimed in claim 1, characterized in that: The dosage prediction value and environmental condition parameters are mapped to the fuzzy decision space of the constructed fuzzy logic controller. The fluctuation fuzzy characteristics of the input data are quantified through the membership function, and coupled reasoning is performed based on the preset fuzzy rule base. The dosage adjustment value with trend prediction compensation item is output, including: The dosage prediction value and the environmental condition parameters are mapped together to a multi-dimensional fuzzy decision space, which includes the dosage prediction value deviation domain, the water quality fluctuation domain and the environmental condition parameter domain; An adaptively adjustable trapezoidal membership function is constructed for each input dimension to quantify the membership of each input data in the fuzzy set, where the turning point parameters of each membership function are determined based on historical operating data; The preset fuzzy rule library is called to match the membership values ​​activated in each dimension with three-dimensional space rules, and the preliminary dosage adjustment amount is obtained through weighted calculation; Perform sliding time window trend analysis, perform directional consistency detection on the dosing adjustment amount sequence of the most recent N control cycles, and generate nonlinear trend prediction compensation items based on the exponential smoothing algorithm when K consecutive cycles show the same adjustment trend; The trend prediction compensation item and the preliminary dosage adjustment amount are nonlinearly superimposed and fused to obtain the dosage adjustment value with the trend prediction compensation item.

5. The method for adding medicine to water production process unit of a water plant based on intelligent sensing and adaptive regulation as claimed in claim 4, characterized in that: The deviation domain of the dosage prediction value represents the difference between the machine learning prediction value and the actual dosage; The water quality fluctuation domain includes a composite index consisting of the pH change rate and the second-order derivative of turbidity; In the environmental condition parameter domain, the instantaneous flow rate change rate and temperature gradient value are integrated to form the disturbance intensity index.

6. The method for adding medicine to a water production process unit of a water plant based on intelligent sensing and adaptive regulation as claimed in claim 4, characterized in that: Perform sliding time window trend analysis and perform directional consistency detection on the dosing adjustment amount sequence of the most recent N control cycles. When K consecutive cycles show the same adjustment trend, the nonlinear trend prediction compensation items generated based on the exponential smoothing algorithm include: Perform sliding time window data collection, extract the dosage adjustment sequence of the latest N control cycles based on the current time, and record the adjustment direction mark value of each cycle; Analyze the adjustment direction mark values ​​of K consecutive periods. When the same-direction adjustment trend is detected, the exponential smoothing compensation mechanism is triggered: Automatically adjust the exponential smoothing coefficient α based on the number of consecutive same-direction trend cycles, where α = basic smoothing coefficient + (number of consecutive same-direction cycles / K) × adaptive gain factor; The most recent M same-direction periodic adjustments are weighted, where the weight decays according to the exponential function of α, and the calculation result is dynamically amplified by the nonlinear gain coefficient β, where the β value is determined by the variance σ2 of the adjustment sequence through β=1+log(1+σ2); The calculated nonlinear trend compensation value is limited within a preset safety boundary to generate a final trend prediction compensation term.

7. The method for adding medicine to a water production process unit of a water plant based on intelligent sensing and adaptive regulation as described in any one of claims 1 to 6, characterized in that: The dynamic adjustment model is constructed and trained according to the dosage adjustment value, water quality parameters and environmental condition parameters. After applying the dosage optimization value output by the dynamic adjustment model, the deviation between the outlet water quality parameters and the target value is monitored in real time, and the parameters of the dynamic adjustment model are updated adaptively. The trigger threshold adjustment of the fuzzy rule base and the reconstruction of the membership function are performed, including: Taking the dosage adjustment value, real-time water quality parameters and environmental condition parameters as input, a dynamic adjustment model with adjustable regression coefficient is established through linear regression algorithm. During the training process, the regression coefficient is optimized by minimizing the error between the predicted value and the actual dosage, and the optimal dosage value at the current moment is output; The current optimal value of the dosage is sent to the dosing mechanism for the dosing process, and the deviation between the measured value of the water quality parameter at the outlet and the target value is monitored in real time. When the cumulative deviation of N consecutive sampling cycles exceeds the preset threshold, the adaptive adjustment mechanism is triggered: The sliding window mechanism is used to select the latest N groups of operating data, and the regression coefficient is recalculated to minimize the mean square error between the predicted dosage and the actual dosage; Recalculate the rule confidence weight based on the statistical distribution characteristics of the deviation, and automatically eliminate redundant rules whose triggering frequency is lower than the set value; Based on the distribution change characteristics of historical operating data, the membership function parameters of the input dimension are dynamically reconstructed, and the turning point of the trapezoidal function is updated using the sliding window statistical method to ensure that the fuzzy quantization results match the current outlet water quality parameters; The updated dynamic adjustment model parameters, the optimized fuzzy rule base and the reconstructed membership function are synchronously applied to the next control cycle to form a closed-loop optimization link.

8. A dosing system for water production process units in a water plant based on intelligent sensing and adaptive regulation, characterized in that: include: A data stream generation module is used to collect water quality parameters and environmental operating condition parameters in real time through a multidimensional data perception network pre-deployed in at least part of the water production process section to obtain a multidimensional data stream with time series characteristics; The dosing amount prediction module is used to build a dosing demand prediction model, conduct spatiotemporal correlation analysis on multidimensional data streams and historical dosing data, establish a nonlinear mapping relationship between dynamic changes in water quality and dosing amount, and output dosing amount prediction values ​​corresponding to real-time water quality dynamic characteristics; The dosage adjustment module is used to map the dosage prediction value and the environmental condition parameters to the fuzzy decision space of the constructed fuzzy logic controller, quantify the fluctuation fuzzy characteristics of the input data through the membership function, and perform coupled reasoning based on the preset fuzzy rule base to output the dosage adjustment value with trend prediction compensation items; The dosage optimization and parameter updating module is used to build and train a dynamic adjustment model based on the dosage adjustment value, water quality parameters and environmental operating parameters. After applying the dosage optimization value output by the dynamic adjustment model, the deviation between the outlet water quality parameters and the target value is monitored in real time, and the parameters of the dynamic adjustment model are updated adaptively, the trigger threshold of the fuzzy rule base is adjusted, and the membership function is reconstructed.

9. A dosing device for water production process unit in a water plant based on intelligent sensing and adaptive regulation, characterized in that: include: A multi-dimensional data perception network, pre-deployed in at least part of the water production process, for multi-source heterogeneous data including water quality parameters and environmental condition parameters; Cloud platform, used to store collected multi-source heterogeneous data, historical operation data sets and equipment maintenance logs; A SCADA control platform, connected to the multi-dimensional data perception network and the cloud platform, respectively, for executing the dosing method for the water production process unit of the water plant based on intelligent sensing and adaptive regulation as described in any one of claims 1 to 7; Edge computing nodes are deployed in at least part of the water production process, and achieve computing power coordination with the SCADA control platform through dynamic containerized deployment.

10. A computer-readable medium having computer-executable instructions stored thereon, characterized in that: When the executable instructions are executed by the processor, the dosing method for the water production process unit of a water plant based on intelligent sensing and adaptive regulation as described in any one of claims 1 to 7 is implemented.

Images