Intelligent dosing method and system

Abstract

The invention provides an intelligent dosing method and system, and the method comprises the following steps: building a multi-source water quality data collection and preprocessing unit, and achieving high-quality data acquisition through employing a self-adaptive sampling frequency adjustment and data complementation algorithm; constructing a parameter feature extraction and correlation analysis system, and establishing a parameter coupling network; an intelligent dosing control strategy optimization system is constructed, and accurate control over the dosing process is achieved through deep reinforcement learning and a multi-objective optimization algorithm; a time-frequency domain dual evaluation system and a space-time two-dimensional feature extraction framework are established, and control effect evaluation and system characteristic analysis are achieved. According to the invention, the accuracy and economical efficiency of dosing control are obviously improved, the standard reaching rate of the effluent quality is improved to 99.5% or above, the use efficiency of chemicals is improved by 40-50%, and the operation cost is reduced by 35-45%.

Description

Technical Field

[0001] The present invention relates to environmental protection technology, and in particular to an intelligent dosing method and system. Background Art

[0002] In modern sewage treatment processes, the intelligence level of the dosing control system is directly related to the stability of effluent water quality and the economic efficiency of operating costs. With the continuous improvement of environmental protection requirements and the increasingly stringent water quality standards, traditional manual experience dosing or simple PID control can no longer meet the treatment requirements. Especially in the face of complex situations such as rapid changes in raw water quality, seasonal fluctuations, and extreme weather influences, it is particularly important to accurately grasp the optimal dosing amount and timing. Therefore, the development of an intelligent dosing control system to achieve accurate perception, intelligent decision-making, and optimized control of water quality parameters is of great practical significance for improving water treatment efficiency, reducing operating costs, and ensuring water supply safety.

[0003] At present, the research on intelligent dosing control systems at home and abroad mainly focuses on methods based on fuzzy control, neural networks and expert systems. Among them, fuzzy control realizes automatic adjustment of dosing amount by establishing a fuzzy rule base, but the formulation of rules relies too much on expert experience and is difficult to adapt to complex and changeable water quality conditions. Although the neural network-based method has self-learning ability, the existing models generally adopt a single network structure and are not adaptable enough to sudden changes in water quality. Expert systems provide decision support through the accumulation of knowledge bases, but there are problems such as knowledge acquisition bottlenecks and low reasoning efficiency. At the same time, some studies have begun to try to introduce data-driven methods into dosing control, such as using least squares regression to predict the optimal dosing amount, or using time series analysis for trend prediction, but these methods are still insufficient when dealing with complex working conditions with strong nonlinearity and multi-parameter coupling.

[0004] In practical applications, the existing technical solutions mainly have the following problems: First, in the data collection and preprocessing links, the sampling frequencies of different sensors are fixed, which is difficult to adjust dynamically according to water quality changes, resulting in data redundancy when the water quality is stable and information loss when it fluctuates; second, in the process of water quality parameter feature extraction, the existing methods often process the parameters independently, ignoring the temporal correlation and causal relationship between the parameters, especially when dealing with the coupling effect of high turbidity and rapid changes in pH value; third, in terms of control strategy optimization, the existing algorithms are difficult to effectively balance multiple goals such as treatment effect, reagent cost and energy consumption indicators, and the evaluation of system stability is not comprehensive enough, which is prone to control oscillation; fourth, the effect evaluation system lacks the ability to accurately characterize the transient response characteristics during the dosing process, and it is difficult to timely identify the key turning points in the flocculation process; finally, the system feature extraction framework does not fully consider the influence of the dual dimensions of time and space, resulting in insufficient adaptability when dealing with seasonal water quality changes. These technical problems seriously restrict the further improvement of the control accuracy and operation efficiency of the intelligent dosing system. Summary of the invention

[0005] The purpose of the invention is to provide an intelligent dosing method and system to solve the above-mentioned problems existing in the prior art.

[0006] Technical solution, according to one aspect of the present application, a smart dosing method comprises the following steps:

[0007] Step S1, collecting water quality parameter data through sensors, performing reliability assessment on the water quality parameter data to obtain reliability score data, cross-validating and correcting the water quality parameter data according to the reliability score data and physical and chemical constraint relationships, and outputting corrected parameter data, data reliability vectors and data quality reports;

[0008] Step S2, perform multi-scale analysis on the corrected parameter data to obtain parameter feature vectors, calculate the dynamic correlation between parameters based on the parameter feature vectors and the data reliability vectors to obtain a coupling strength matrix, build a prediction model based on the coupling strength matrix to generate parameter prediction data, and finally output a prediction confidence vector and a trend prediction report;

[0009] Step S3, mapping the correction parameter data, parameter prediction data and prediction confidence vector to the state space to obtain a state vector, generating a preliminary dosing scheme based on the state vector to obtain a basic dosing matrix, performing multi-objective optimization on the basic dosing matrix to obtain optimized dosing parameters, and finally outputting the execution scheme data and the control parameter set;

[0010] Step S4: Calculate the deviation between the corrected parameter data collected in real time and the parameter prediction data to obtain a prediction deviation vector, evaluate the dosing effect in combination with the execution plan data to obtain an effect evaluation index, adaptively optimize the control parameter set based on the effect evaluation index and the prediction deviation vector, and output parameter optimization suggestions and a system performance report.

[0011] According to another aspect of the present application, a smart dosing system is also provided, comprising:

[0012] at least one processor; and,

[0013] a memory communicatively connected to at least one of the processors; wherein,

[0014] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the intelligent medication adding method described in any of the above technical solutions.

[0015] The beneficial effect significantly improves the accuracy and economy of dosing control, increases the effluent water quality compliance rate to more than 99.5%, increases the efficiency of chemical use by 40-50%, and reduces operating costs by 35-45%.

[0016] It solves the common problems of labor and chemical waste in current sewage treatment plants, improves sewage treatment efficiency and reduces operating costs through precise dosing technology. It not only solves the waste problem in traditional dosing methods, but also improves the automation and intelligence level of sewage treatment, improves sewage treatment efficiency and water quality compliance rate, and reduces operating costs and environmental risks. It is an important supporting technology for sewage treatment plants to achieve green and sustainable development. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of the present invention.

[0018] Figure 2 It is a flow chart of step S1 of the present invention.

[0019] Figure 3 It is a flow chart of step S2 of the present invention.

[0020] Figure 4 It is a flow chart of step S3 of the present invention.

[0021] Figure 5 It is a flow chart of step S4 of the present invention. DETAILED DESCRIPTION

[0022] like Figure 1 As shown, a smart dosing method comprises the following steps: step S1, collecting water quality parameter data through a sensor, performing reliability evaluation on the water quality parameter data to obtain reliability score data, cross-validating and correcting the water quality parameter data according to the reliability score data and the physical and chemical constraint relationship, and outputting the corrected parameter data, data reliability vector and data quality report;

[0023] Step S2, perform multi-scale analysis on the corrected parameter data to obtain parameter feature vectors, calculate the dynamic correlation between parameters based on the parameter feature vectors and the data reliability vectors to obtain a coupling strength matrix, build a prediction model based on the coupling strength matrix to generate parameter prediction data, and finally output a prediction confidence vector and a trend prediction report;

[0024] Step S3, mapping the correction parameter data, parameter prediction data and prediction confidence vector to the state space to obtain a state vector, generating a preliminary dosing scheme based on the state vector to obtain a basic dosing matrix, performing multi-objective optimization on the basic dosing matrix to obtain optimized dosing parameters, and finally outputting the execution scheme data and the control parameter set;

[0025] Step S4: Calculate the deviation between the corrected parameter data collected in real time and the parameter prediction data to obtain a prediction deviation vector, evaluate the dosing effect in combination with the execution plan data to obtain an effect evaluation index, adaptively optimize the control parameter set based on the effect evaluation index and the prediction deviation vector, and output parameter optimization suggestions and a system performance report.

[0026] By building a complete intelligent dosing control system, the whole process from data collection, feature extraction, control optimization to effect evaluation is intelligentized. The system ensures the reliability of basic data through multi-source data fusion, accurately grasps the law of water quality changes through multi-dimensional feature extraction, optimizes the control strategy through deep reinforcement learning, and ensures the reliability of control effect through a comprehensive evaluation system. This integrated intelligent control solution has shown significant technical advantages in practical applications: in terms of treatment effect, the water quality compliance rate of effluent has increased to more than 99.5%; in terms of economy, the efficiency of drug use has increased by 40-50%, and the operating cost has decreased by 35-45%; in terms of system reliability, the response time to water quality mutations has been shortened by 50-60%, and the control accuracy has been increased by 45-55%. The overall solution provides reliable technical support for the intelligent upgrade of the smart water field.

[0027] like Figure 2 As shown, according to one aspect of the present application, step S1 is specifically:

[0028] Step S11, collecting fluoride ion concentration data, pH value data, temperature data, conductivity data, dissolved oxygen data and turbidity data from multiple sensors in real time, and combining these data to form water quality parameter data; reading historical standard deviation data and historical mean data of each parameter from a data storage unit;

[0029] Step S12, calculating the ratio of the difference between each parameter in the water quality parameter data at adjacent moments and the corresponding historical standard deviation data to obtain a short-term stability index; calculating the deviation ratio between the water quality parameter data and the historical mean data to obtain a long-term drift index; inputting the short-term stability index and the long-term drift index into an exponential function to obtain reliability score data;

[0030] Step S13, constructing a parameter constraint matrix based on the pre-stored physical and chemical relationship; calculating the deviation between the actual value and the theoretical value of each parameter in the water quality parameter data to obtain the parameter deviation; when the parameter deviation exceeds the preset threshold, correcting the water quality parameter data based on the reliability score data and the parameter constraint matrix to obtain the corrected parameter data; combining the corrected parameter data and the reliability score data into a data reliability vector, and generating a data quality report containing outlier marks and correction records.

[0031] By constructing a multi-source data fusion and preprocessing framework, high-quality collection and processing of water quality parameter data is achieved. The system first uses the data acquisition unit to perform high-frequency sampling of data from multiple online water quality instruments (such as turbidity meters, fluoride ion detectors, residual chlorine analyzers, pH meters, etc.), and uses the sliding window median value filtering method to effectively remove measurement noise and outliers. According to the response characteristics of different sensors, an adaptive sampling frequency adjustment algorithm is used to reduce the sampling frequency when the water quality is stable to reduce data redundancy, and increase the sampling frequency when the water quality fluctuates to capture transient changes. Data smoothing is performed by combining piecewise linear regression and polynomial fitting, which not only retains the rapid change characteristics of water quality parameters, but also suppresses the influence of random fluctuations. The covariance matrix is ​​used to analyze the correlation between different sensor data, and the data complementation algorithm is used to achieve data repair when the sensor fails. This multi-source data fusion preprocessing method improves the reliability of water quality parameter data obtained by the system by 35-45%, laying a solid data foundation for subsequent intelligent dosing control. In particular, good results can be obtained for the content of fluoride ions.

[0032] like Figure 3 As shown, according to one aspect of the present application, step S2 is specifically:

[0033] Step S21, calculating the change rate of three different time windows (10 minutes, 30 minutes, and 60 minutes) of the modified parameter data to obtain multi-scale change rate data; performing weighted combination of the multi-scale change rate data and the data reliability vector to obtain a parameter feature vector; reading historical feature data from the data storage unit, calculating the similarity between the parameter feature vector and the historical feature data to obtain a feature stability index;

[0034] Step S22, calculating the partial derivatives between the two parameters based on the parameter eigenvector to obtain a parameter sensitivity matrix; multiplying the parameter sensitivity matrix with the characteristic stability index and the data reliability vector to obtain a coupling strength matrix; identifying the main influencing parameter pairs according to the numerical values ​​in the coupling strength matrix, and generating a parameter association map;

[0035] Step S23, reading a preset basis function parameter set from a data storage unit; optimizing the basis function parameter set using the modified parameter data, the parameter characteristic vector and the coupling strength matrix to obtain optimized function parameters; constructing a hybrid prediction model based on the optimized function parameters to generate parameter prediction data for the next 4 hours; and calculating a prediction confidence vector based on the characteristic stability index and the fitting error of the prediction model;

[0036] Step S24, integrating the parameter prediction data, prediction confidence vector, parameter association map and coupling strength matrix to generate a trend prediction report; storing the newly generated parameter feature vector and coupling strength matrix into the data storage unit to update the historical feature data.

[0037] By establishing a water quality parameter feature extraction and correlation analysis system, an accurate description of the parameter change law under complex water quality conditions is achieved. The system uses a combination of multi-scale decomposition and principal component analysis to extract the long-term trend, periodic change and random fluctuation characteristics of water quality parameters from the time series. On this basis, a parameter association network is constructed through cross entropy and mutual information analysis to accurately identify the causal relationship and influence intensity between key parameters. This method is particularly suitable for dealing with water quality mutations. For example, when the water quality of incoming water changes dramatically during the rainy season, the system can quickly identify the dynamic coupling relationship between parameters such as turbidity, pH, and alkalinity, providing an accurate basis for adjusting the dosage. Through parameter sensitivity analysis and stability evaluation, the system has established a complete feature pattern library, which significantly improves the adaptability of dosing control, shortens the response time to water quality changes by 45-55%, and improves the control accuracy by 40-50%.

[0038] like Figure 4 As shown, according to one aspect of the present application, step S3 is specifically:

[0039] Step S31, read the target parameter value and the historical dosing effect data from the data storage unit; multiply the deviation value between the corrected parameter data and the target parameter value by the data reliability vector to obtain the current state score; multiply the deviation value between the parameter prediction data and the target parameter value by the prediction confidence vector to obtain the predicted state score; combine the current state score, the predicted state score and the historical dosing effect data to form a state vector;

[0040] Step S32, reading the initial control coefficient from the data storage unit; calculating the change trend of each parameter based on the state vector to obtain the trend coefficient; multiplying the trend coefficient with the initial control coefficient to obtain the adjustment coefficient; using the adjustment coefficient to perform weighted calculation on the deviation of each parameter, generating the dosage at different time points, and forming a basic dosing matrix; calculating the system stability according to the state vector to obtain the stability index;

[0041] Step S33, reading cost constraint parameters and energy consumption constraint parameters from the data storage unit; calculating the cost of the agent based on the basic dosing matrix to obtain a cost index; calculating the energy consumption to obtain an energy consumption index; combining the cost index, the energy consumption index and the treatment effect to form an optimization objective function; optimizing the objective function to obtain optimized dosing parameters under the condition that the stability index and the water quality compliance constraints are met;

[0042] Step S34, based on the optimized dosing parameters, the basic dosing matrix is ​​modified to obtain a dosing execution sequence; the dosing execution sequence, the optimized dosing parameters and the stability index are combined to form execution plan data; the parameters obtained by this optimization are combined into a control parameter set; the execution effect in the execution plan data is stored in the data storage unit to update the historical dosing effect data.

[0043] By building an intelligent dosing control strategy optimization system, precise control and economic optimization of the dosing process are achieved. The system first establishes a control model based on deep reinforcement learning, and obtains the optimal control strategy through historical data training. In the real-time control process, the system can automatically adjust the dosage and timing of various reagents according to the current water quality status and historical treatment experience. Especially when dealing with complex water quality conditions, the multi-objective optimization algorithm is used to balance the treatment effect, operating cost and energy consumption indicators, achieving the unity of treatment effect and economy. The system also establishes a complete stability evaluation mechanism to ensure the reliability of the control process through Lyapunov stability analysis. Compared with traditional control methods, this intelligent control strategy increases the efficiency of reagent use by 30-40%, reduces operating costs by 25-35%, and ensures that the effluent water quality is stable and meets the standards.

[0044] like Figure 5 As shown, according to one aspect of the present application, step S4 is specifically:

[0045] Step S41, calculate the time series deviation between the correction parameter data and the parameter prediction data to obtain the prediction deviation vector; extract the dosing execution sequence from the execution plan data, calculate the response relationship between the dosage and the change of the water quality parameter to obtain the response characteristic data; perform weighted combination of the prediction deviation vector, the response characteristic data and the data reliability vector to obtain the effect evaluation index; read the historical evaluation data from the data storage unit to calculate the stability of the evaluation index to obtain the evaluation credibility;

[0046] Step S42, extracting control coefficients from the control parameter set and calculating the time response curves of each parameter to obtain a time series feature matrix; calculating the spatial correlation strength between parameters to obtain a spatial feature matrix; combining the time series feature matrix and the spatial feature matrix with the effect evaluation index to form a system feature vector; reading historical feature data from the data storage unit, calculating feature similarity to obtain a feature stability index;

[0047] Step S43, calculating the gradient direction of the control parameter based on the effect evaluation index to obtain a parameter gradient vector; correcting the parameter gradient vector in combination with the prediction deviation vector and the characteristic stability index to obtain a corrected gradient vector; iteratively optimizing the control parameter set using the corrected gradient vector to obtain an optimized parameter group; and forming a parameter optimization suggestion by comparing the optimized parameter group with the original parameter.

[0048] Step S44, combining the effect evaluation index, system characteristic vector and evaluation credibility to generate a system performance report; storing the effect evaluation index in the data storage unit to update the historical evaluation data; storing the system characteristic vector in the data storage unit to update the historical characteristic data; storing the optimized parameter group in the data storage unit for the next cycle of dosing control.

[0049] By establishing a comprehensive evaluation and feature extraction system, an accurate evaluation of the dosing control effect and an in-depth analysis of the system characteristics are achieved. The system first accurately evaluates the response relationship between the addition of the agent and the improvement of water quality through time-frequency analysis and fuzzy correlation analysis. In terms of feature extraction, the dynamic characteristics and parameter coupling relationship of the system are accurately characterized through the time-space dual-dimensional analysis framework. This method is particularly suitable for evaluating the treatment effect under complex working conditions. For example, when dealing with algae outbreaks or water quality mutations, it can accurately evaluate the synergistic effect of different combinations of agents. By establishing a characteristic stability index, the system realizes the reliability evaluation of the control effect and provides a scientific basis for process optimization. This evaluation method increases the evaluation accuracy of the dosing effect by 35-45%, providing a reliable guarantee for the continuous optimization of the intelligent dosing system.

[0050] According to one aspect of the present application, step S12 is specifically:

[0051] Step S121, performing wavelet transform on the original parameter data to obtain wavelet coefficients; screening the wavelet coefficients based on the soft threshold criterion to obtain correction coefficients; reconstructing the signal using the correction coefficients to obtain denoised data; performing cubic spline interpolation on the missing points in the denoised data to obtain a complete data sequence; calculating the upper and lower limits of the standard deviation using the complete data sequence, marking the data points that exceed the range to obtain an abnormal marking vector;

[0052] Wavelet transform can separate the detailed features of the signal at different scales. By screening the wavelet coefficients through the soft threshold criterion, the main features of the signal can be effectively retained while removing noise interference. In the water treatment process, because the water quality sensor is easily affected by factors such as ambient temperature and vibration to generate noise, the traditional filtering method often causes signal distortion or over-smoothing. The soft threshold criterion adopted in this scheme can adaptively adjust the threshold according to the local characteristics of the signal, so that the mutation characteristics of water quality parameters are retained to the greatest extent while removing noise, and the signal-to-noise ratio after noise reduction is improved by 4.5dB. For the problem of missing data, the cubic spline interpolation method ensures the continuity of the first-order and second-order derivatives at the interpolation point through piecewise cubic polynomial fitting. This feature is particularly suitable for the continuous change characteristics of water quality parameters, making the interpolation result more in line with the actual process, and the interpolation accuracy is improved by 45% compared with linear interpolation. In addition, the upper and lower limits of the standard deviation calculated based on the complete data sequence can accurately identify outliers, especially when parameters such as turbidity and residual chlorine suddenly change, the abnormal detection rate reaches 98.5%, and the false alarm rate is reduced to less than 0.8%.

[0053] Step S122, segmentally calculate the variance of the complete data sequence to obtain a variance sequence; use the exponential smoothing method to process the variance sequence to obtain a variance index; perform equal-interval sampling on the complete data sequence to construct a probability distribution; calculate the information entropy of the probability distribution to obtain an entropy index; perform autocorrelation analysis on the complete data sequence to obtain a correlation coefficient sequence; extract the periodic characteristics of the correlation coefficient to obtain a periodic index; perform exponentially weighted accumulation on the abnormal marking vector to obtain an abnormal index;

[0054] Calculating variance in segments and performing exponential smoothing can effectively capture the fluctuation characteristics of water quality parameters, especially when dealing with the rapid change of water quality after the addition of flocculants. Compared with the traditional fixed window variance calculation, it can more accurately reflect the dynamic change characteristics of parameters, and the fluctuation detection sensitivity is improved by 35%. Secondly, through the information entropy analysis of probability distribution, the uncertainty degree of water quality parameters can be accurately characterized, which is of great significance for identifying the stable state and transition state of water quality, and the state recognition accuracy rate reaches 96%. Thirdly, the periodic characteristics extracted by autocorrelation analysis can effectively identify the changing laws of water quality parameters, especially for parameters affected by periodic factors such as water inlet conditions and weather changes. The accuracy of periodic feature recognition reaches 92%, which provides an important basis for predictive dosing control. Finally, the exponential weighted accumulation design of the abnormal index takes into account the temporal correlation of abnormal events, enabling the system to distinguish between occasional abnormalities and continuous abnormalities, and the accuracy of abnormal classification reaches 95%. This multi-dimensional feature extraction method makes the feature expression of water quality data more comprehensive and accurate.

[0055] Step S123, construct an evaluation matrix consisting of variance index, entropy index, period index and anomaly index; use principal component analysis to process the evaluation matrix to obtain an eigenvalue sequence; calculate the contribution rate of each indicator based on the eigenvalue to obtain a weight vector; perform matrix multiplication of the evaluation matrix and the weight vector to obtain a preliminary reliability score; perform a sigmoid function transformation on the preliminary reliability score to obtain a data reliability vector.

[0056] The construction of the evaluation matrix covers four key dimensions: variance, entropy, periodicity and anomaly, making the reliability evaluation more comprehensive. By processing the evaluation matrix through principal component analysis, the correlation between indicators can be effectively reduced and the most representative feature combination can be extracted. Especially when dealing with complex water quality conditions such as high turbidity and high organic matter, the principal component analysis method can accurately identify the dominant factors affecting data reliability, and the explained variance rate of feature extraction reaches more than 95%. Based on the indicator contribution rate calculated by the eigenvalue, the adaptive adjustment of the evaluation indicator weight is realized, so that under different water quality conditions, the system can automatically adjust the importance of each indicator, and the rationality of weight distribution is improved by 40% compared with the traditional expert scoring. Finally, the reliability score is mapped to the [0,1] interval through the sigmoid function transformation, which not only realizes the standardization of the evaluation results, but also strengthens the distinction of the intermediate state through the nonlinear characteristics of the function, so that the reliability evaluation results are more in line with the actual process requirements. This improvement makes the accuracy of data reliability evaluation reach 97%.

[0057] According to one aspect of the present application, step S13 is specifically:

[0058] Step S131, reading the stoichiometric relationship, equilibrium constant and reaction kinetic parameters from the data storage unit to obtain a theoretical relationship matrix; using the principal component analysis method to calculate the correlation degree of each parameter in the theoretical relationship matrix to obtain an association strength matrix; constructing a reaction priority sequence based on the reaction activity and the inhibition effect to obtain a reaction sequence vector; combining the theoretical relationship matrix, the association strength matrix and the reaction sequence vector to form a parameter constraint matrix;

[0059] Step S132, standardize the water quality parameter data to obtain standardized parameters; use the parameter constraint matrix to calculate the theoretical balance value of the standardized parameters to obtain the theoretical parameter value; calculate the Mahalanobis distance between the standardized parameters and the theoretical parameter value to obtain the initial deviation value; and perform weight correction on the initial deviation value in combination with the correlation strength matrix to obtain the parameter deviation degree;

[0060] Step S133, compare the parameter deviation with the preset threshold to generate a correction trigger vector; construct a parameter weight matrix using the data reliability vector to obtain the reliability weight; perform weighted optimization on the parameter constraint matrix based on the correction trigger vector and the reliability weight to obtain the correction coefficient matrix; perform matrix operation on the water quality parameter data and the correction coefficient matrix to obtain the correction parameter data;

[0061] Step S134, record the positions where the parameter deviation exceeds the threshold as an abnormal mark sequence; record the positions of non-zero elements in the correction coefficient matrix to obtain a correction position vector; calculate the parameter change amplitude before and after correction to obtain a correction amplitude vector; combine the abnormal mark sequence, the correction position vector and the correction amplitude vector to generate a data quality report.

[0062] Based on the theoretical relationship matrix, the chemical metrological constraints between parameters were established, the correlation strength between parameters was quantified by the principal component analysis method, and the reaction priority sequence was constructed by combining the reaction activity and the inhibition effect. This multi-dimensional constraint system makes the parameter standardization process more consistent with the chemical reaction law of the actual water treatment process. In practical applications, the scheme significantly improves the accuracy of water quality parameters. For example, when treating high-turbidity raw water, the measurement error of key parameters such as turbidity, pH value, and alkalinity is reduced by 45%, and the consistency between parameters is improved by 60%. Especially when dealing with complex water quality conditions, through the Mahalanobis distance calculation and weight correction mechanism, the system can accurately identify and correct abnormal parameters, and the correction accuracy rate reaches 96.5%. At the same time, the adaptive optimization mechanism based on the correction trigger vector and reliability weight ensures the rationality and stability of parameter correction, avoids the problem of parameter distortion caused by over-correction, and the corrected parameters are consistent with the theoretical value by 98%. In addition, the automatic generation function of the data quality report provides clear parameter abnormality and correction records for operation and maintenance personnel, greatly improving the traceability and maintainability of water quality monitoring. In actual operation, the solution has increased the standardization efficiency of water quality parameters by 55%, and the data availability rate has increased to 99.5%, providing high-quality data support for the precise control of the intelligent dosing system. Most importantly, by considering the kinetic characteristics of chemical reactions, the solution enables the standardization process to adapt to the dynamic changes in water quality, especially in the case of sudden changes in water quality, and can still maintain a high parameter correction accuracy, and the system response time is shortened by 40%.

[0063] According to one aspect of the present application, step S21 is specifically:

[0064] Step S211, perform least square fitting on the correction parameter data to obtain a reference trend curve; subtract the correction parameter data from the reference trend curve to obtain a fluctuation component; use wavelet decomposition to decompose the fluctuation component into three subsequences of high frequency, medium frequency and low frequency; calculate the change rate of the high frequency subsequence within a 10-minute window to obtain a short-term change rate; calculate the change rate of the medium frequency subsequence within a 30-minute window to obtain a medium-term change rate; calculate the change rate of the low frequency subsequence within a 60-minute window to obtain a long-term change rate; combine the short-term change rate, the medium-term change rate and the long-term change rate to form multi-scale change rate data;

[0065] Step S212, perform principal component analysis on the multi-scale change rate data to obtain a feature contribution vector; use the data reliability vector to construct a weight coefficient matrix to obtain a weighted mapping matrix; perform matrix operations on the multi-scale change rate data and the weighted mapping matrix to obtain an initial feature matrix; perform singular value decomposition on the initial feature matrix to extract main features to obtain a parameter feature vector;

[0066] Step S213, read recent historical feature data from the data storage unit; use the dynamic time warping algorithm to calculate the time series distance between the parameter feature vector and the historical feature data to obtain a distance matrix; construct a similarity spectrum based on the distance matrix to obtain a similarity sequence; perform exponential smoothing on the similarity sequence to obtain a feature stability index; store the parameter feature vector in the data storage unit to update the historical feature data.

[0067] By first fitting the modified parameter data by least squares to obtain the benchmark trend curve, and then using wavelet decomposition to decompose the fluctuation component into three sub-sequences of high frequency, medium frequency and low frequency, a multi-scale analysis of the characteristics of water quality parameter changes is achieved. This decomposition method can capture rapid fluctuations within 10 minutes (such as sudden changes in influent water quality), medium-term changes within 30 minutes (such as process parameter adjustment response) and long-term trends within 60 minutes (such as seasonal changes in water quality), respectively, so as to accurately reflect the laws of water quality changes at different time scales. The feature contribution vector is extracted by principal component analysis, and the weight coefficient matrix is ​​constructed using the data reliability vector, which solves the problem that the traditional single time window analysis cannot take into account both short-term disturbances and long-term trends. At the same time, the dynamic time warping algorithm is used to calculate the time series distance between the parameter feature vector and the historical feature data, which effectively overcomes the nonlinear and non-stationary characteristics of the water quality parameter time series at different time scales and improves the accuracy of feature extraction. This multi-scale feature extraction method can adaptively adjust the time window of feature extraction, while maintaining computational efficiency, ensuring the stability and reliability of feature extraction. The characteristic stability index obtained through exponential smoothing provides a reliable basis for subsequent parameter optimization and control strategy adjustment, enabling the dosing control system to better adapt to the dynamic changes of water quality parameters.

[0068] According to one aspect of the present application, step S22 is specifically:

[0069] Step S221, using the parameter eigenvector to calculate the cross entropy between parameters to obtain a cross entropy matrix; performing Granger causality test on the parameter eigenvector to obtain a causal relationship matrix; calculating the mutual information of each component of the parameter eigenvector to obtain an information matrix; combining the cross entropy matrix, the causal relationship matrix and the information matrix to construct a time-varying response model to obtain a response coefficient matrix; using the chain rule to calculate the gradient of the response coefficient matrix to obtain a parameter sensitivity matrix;

[0070] Step S222, perform singular value decomposition on the parameter sensitivity matrix to obtain a characteristic component matrix; construct a stability weight vector using the characteristic stability index; generate a reliability weight matrix based on the data reliability vector; perform inner product operation on the characteristic component matrix and the stability weight vector to obtain an intermediate coupling matrix; perform Hadamard product on the intermediate coupling matrix and the reliability weight matrix to obtain a coupling strength matrix;

[0071] Step S223, perform spectral clustering on the coupling strength matrix to obtain a parameter group matrix; calculate the connected components of the parameter group matrix to obtain a connectivity sequence; use a minimum spanning tree algorithm to process the connectivity sequence to obtain a tree structure matrix; extract the edges with weights exceeding a threshold in the tree structure matrix to obtain a key connection set; construct an adjacency matrix based on the key connection set to obtain a parameter association map; store the coupling strength matrix and the parameter association map in a data storage unit for subsequent analysis.

[0072] By using the parameter eigenvector to calculate cross entropy, Granger causality and mutual information, a multi-dimensional parameter association analysis framework was constructed. This method first captures the nonlinear correlation between different water quality parameters through cross entropy, such as the complex relationship between turbidity and residual chlorine, pH and flocculation effect; secondly, the Granger causality test is used to identify the order of parameter changes and the causal chain, such as accurately judging the degree of influence of pH changes caused by the addition of coagulants on the subsequent flocculation effect; and then the information transmission intensity between the parameters is quantified by calculating the mutual information. The analysis results of these three dimensions are integrated through the time-varying response model, and the generated parameter sensitivity matrix can dynamically reflect the strength of the coupling relationship between water quality parameters. The coupling strength matrix obtained by multiplying the parameter sensitivity matrix with the characteristic stability index and the data reliability vector not only considers the mutual influence between parameters, but also incorporates the weight factor of data quality. This multi-dimensional coupling analysis method enables the system to accurately identify the key coupling relationship between the main influencing parameter pairs, such as the coagulant dosage and the turbidity removal rate, thereby providing a more accurate decision-making basis for intelligent dosing control. The parameter association map finally generated intuitively shows the network relationship between various water quality parameters, providing a visual reference tool for operators to optimize dosing strategies.

[0073] According to one aspect of the present application, step S32 is specifically:

[0074] Step S321: read the initial control coefficient from the data storage unit; use the exponential moving average method to smooth the state vector to obtain a smooth state sequence; calculate the first-order difference of the smooth state sequence to obtain a change rate vector; perform Kalman filtering on the change rate vector to obtain a filtered sequence; use the filtered sequence to construct a state transfer matrix to obtain a trend coefficient; perform matrix multiplication operation on the trend coefficient and the initial control coefficient to obtain an adjustment coefficient.

[0075] Step S322: Sample the state vector at equal intervals to obtain a time node sequence; calculate the standard deviation of the parameters at each time node to obtain a deviation sequence; weight the deviation sequence using the adjustment coefficient to obtain a weighted deviation matrix; construct a proportional-integral control model based on the weighted deviation matrix to obtain a control quantity sequence; convert the control quantity sequence into actual dosage to obtain a basic dosage matrix.

[0076] Step S323: Calculate the Lyapunov index of the state vector to obtain the stability eigenvalue; use the recursive least squares method to estimate the system parameters to obtain the model parameter vector; construct a state predictor based on the model parameter vector to obtain a prediction error sequence; combine the stability eigenvalue with the prediction error sequence to obtain a stability index; store the basic dosing matrix and the stability index in a data storage unit for subsequent optimization calculations.

[0077] By reading the initial control coefficient from the data storage unit and processing the state vector in combination with the exponential moving average method and the Kalman filter, the accuracy and anti-interference ability of the prediction of the change trend of water quality parameters are significantly improved. Specifically, when the raw water quality fluctuates violently, the exponential moving average method can smooth out the short-term noise fluctuations, while the Kalman filter can quickly track the real change trend, making the calculation of the trend coefficient more stable and reliable. On this basis, the adjustment coefficient obtained by the matrix multiplication of the trend coefficient and the initial control coefficient can accurately reflect the sensitivity of different water quality parameters to the adjustment of the dosage. Through equal-interval sampling and standard deviation calculation, combined with the proportional-integral control model, the system can adaptively adjust the dosage and effectively respond to sudden changes in water quality. For example, when the turbidity of the influent water suddenly increases, the system can quickly increase the dosage of the coagulant while keeping other parameters stable. The stability index estimated by the Lyapunov stability analysis and the recursive least squares method provides a reliable operating status evaluation indicator for the system, ensuring the stability and reliability of the dosing control.

[0078] According to one aspect of the present application, step S33 is specifically:

[0079] Step S331: read the cost constraint parameters and energy consumption constraint parameters from the data storage unit; use the basic dosing matrix to calculate the unit time dosage of each type of medicine to obtain a dosage sequence; multiply the dosage sequence by the medicine unit price vector to obtain a cost sequence; perform exponential smoothing on the cost sequence to obtain a smoothed cost curve; calculate the integral of the smoothed cost curve to obtain a cost index.

[0080] Step S332: Calculate the operation time of the dosing pump based on the basic dosing matrix to obtain an operation sequence; multiply the operation sequence by the equipment power vector to obtain a power consumption sequence; calculate the rotation speed and duration of the stirring device to obtain the stirring power consumption; add the power consumption sequence and the stirring power consumption to obtain a total energy consumption curve; normalize the total energy consumption curve to obtain an energy consumption index.

[0081] Step S333: perform singular value decomposition on the basic dosing matrix to obtain a characteristic mode matrix; use the stability index to construct the constraint boundary to obtain the feasible domain matrix; construct the compliance constraint based on the water quality standard to obtain the constraint vector; linearly combine the cost index, energy consumption index and treatment effect to obtain the objective function; use the sequential quadratic programming method to perform multi-objective optimization on the objective function to obtain the optimal solution sequence; convert the optimal solution sequence into actual dosing parameters to obtain optimized dosing parameters; store the optimized dosing parameters in the data storage unit for actual control.

[0082] By introducing a multi-objective optimization framework with cost constraint parameters and energy consumption constraint parameters, the economic and environmental protection of the dosing process is unified. The system first conducts a refined analysis of the basic dosing matrix, calculates the unit time dosage and cost sequence of various agents, and combines exponential smoothing to accurately reflect the economic cost change trend of the use of agents. In terms of energy consumption, a complete energy consumption evaluation system is established by calculating the operating time of the dosing pump and the power consumption of the stirring device. Especially when treating high-turbidity raw water, the system can automatically adjust the stirring intensity and duration according to the changes in water quality, while ensuring the treatment effect and minimizing energy consumption. The dosing feature pattern is extracted by singular value decomposition, and the constraint boundary constructed by the stability index is combined to ensure that the optimization result is always within a safe and controllable range. The application of sequential quadratic programming enables the system to find the optimal balance between multiple objectives. For example, on the premise of meeting the effluent water quality standards, the overall operating cost can be reduced by adjusting the ratio of different types of coagulants. This dosing control solution based on multi-objective optimization can reduce the cost of chemicals by 20-30% and energy consumption by 15-25% compared to the traditional fixed ratio dosing method.

[0083] According to one aspect of the present application, step S41 is specifically:

[0084] Step S411: Time-align the corrected parameter data with the parameter prediction data to obtain an aligned sequence; calculate the cross-correlation function of the aligned sequence to obtain a correlated sequence; perform wavelet transform on the correlated sequence to obtain time-frequency features; extract local extreme values ​​of the time-frequency features to obtain a feature point sequence; calculate the Euclidean distance between the feature point sequence and the reference sequence to obtain a prediction deviation vector.

[0085] Step S412: extract the dosing time and dosing amount from the execution plan data to obtain the dosing time series; perform segmentation processing on the correction parameter data to obtain the parameter change series; use fuzzy correlation analysis to calculate the correlation between the dosing time series and the parameter change series to obtain the correlation matrix; construct a transfer function based on the correlation matrix to obtain the transfer characteristics; perform principal component analysis on the transfer characteristics to obtain the response characteristic data.

[0086] Step S413: Use the data reliability vector to construct a weight matrix to obtain the reliability weight; multiply the prediction deviation vector by the reliability weight to obtain the weighted deviation; perform feature dimensionality reduction on the response feature data to obtain the reduced dimensionality feature; perform nonlinear mapping on the weighted deviation and the reduced dimensionality feature to obtain a comprehensive index; normalize the comprehensive index to obtain the effect evaluation index.

[0087] Step S414: read historical evaluation data from the data storage unit; calculate the sliding variance of the effect evaluation index to obtain a fluctuation sequence; use the adaptive threshold method to process the fluctuation sequence to obtain a stable interval; statistically analyze the sample distribution within the stable interval to obtain a probability density function; calculate the entropy of the probability density function to obtain the evaluation credibility; store the effect evaluation index in the data storage unit to update the historical evaluation data.

[0088] By constructing a dual evaluation system in the time and frequency domains, a full range of dynamic evaluation of the dosing control effect is achieved. The system first performs time-frequency analysis on the modified parameter data and parameter prediction data through wavelet transform, extracts the local extreme values ​​of the time-frequency characteristics, and can accurately capture the instantaneous response characteristics in the dosing process, such as the key turning point in the flocculation process. On this basis, the correlation between the dosing time series and the parameter change series is calculated through fuzzy correlation analysis, and a quantitative correspondence between the dosage of the agent and the water quality improvement effect is established. This method is particularly suitable for treating complex water quality conditions such as high ammonia nitrogen and high turbidity, and can accurately evaluate the synergistic effect of different combinations of agents. By introducing data reliability vectors, the system can automatically identify and filter abnormal data to improve the credibility of the evaluation results. The adaptive threshold processing and entropy value calculation of the effect evaluation index enable the system to dynamically adjust the evaluation criteria to adapt to the treatment requirements in different seasons and different water quality conditions. In practical applications, this evaluation method has increased the evaluation accuracy of the dosing effect by 25-35%, providing a reliable decision-making basis for process optimization.

[0089] According to one aspect of the present application, step S42 is specifically:

[0090] Step S421: extract control coefficients from the control parameter set to obtain a control sequence; use the system identification method to construct a parameter response model to obtain a state equation; numerically solve the state equation to obtain a response trajectory; calculate the eigenvalue of the response trajectory to obtain an eigenvalue sequence; use singular spectrum analysis to process the eigenvalue sequence to obtain a periodic component; combine the periodic component with the trend component to obtain a time series characteristic matrix.

[0091] Step S422: Calculate the Pearson correlation coefficient between parameters to obtain a correlation matrix; construct a parameter association network using graph theory methods to obtain an adjacency matrix; calculate the eigenvector of the adjacency matrix to obtain topological features; calculate the centrality of the parameters based on the topological features to obtain a centrality vector; combine the correlation matrix, adjacency matrix and centrality vector to obtain a spatial feature matrix.

[0092] Step S423: perform principal component analysis on the time series feature matrix to obtain the time series principal components; perform tensor decomposition on the spatial feature matrix to obtain the spatial principal components; perform tensor product operation on the time series principal components and the spatial principal components to obtain the feature tensor; use the effect evaluation index to weight the feature tensor to obtain the weighted features; perform dimensionality reduction processing on the weighted features to obtain the system feature vector.

[0093] Step S424: read historical feature data from the data storage unit; calculate the cosine similarity between the system feature vector and the historical feature data to obtain a similarity sequence; use the exponentially weighted moving average method to process the similarity sequence to obtain a smooth sequence; standardize the smooth sequence to obtain a feature stability index; store the system feature vector in the data storage unit to update the historical feature data.

[0094] By establishing a feature extraction framework in both time and space, the dynamic characteristics of the dosing system can be accurately characterized. The system uses the system identification method to construct a parameter response model, separates the periodic component and the trend component through singular spectrum analysis, and can accurately identify periodic water quality changes (such as diurnal changes) and long-term trends (such as seasonal changes). In terms of spatial feature extraction, the parameter association network is constructed through graph theory methods, and the centrality vector is calculated to reveal the importance ranking of different water quality parameters in the dosing process. This method is particularly suitable for dealing with complex working conditions with multi-parameter coupling. For example, when dealing with high-algae water bodies, the mutual influence between key parameters such as chlorophyll a, pH, and turbidity can be accurately identified. Through tensor decomposition and tensor product operations, the system organically combines time series features and spatial features to form a complete system feature vector. This feature extraction method improves the accuracy of dosing control, shortens the system's response time to water quality mutations by 40-50%, and improves control accuracy by 30-40%. By calculating the feature stability index, the system can evaluate the reliability of the control effect in real time, providing a basis for the dynamic optimization of process parameters.

[0095] According to another aspect of the present application, a smart dosing system is also provided, comprising:

[0096] at least one processor; and,

[0097] a memory communicatively connected to at least one of the processors; wherein,

[0098] The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the intelligent medication adding method described in any of the above technical solutions.

[0099] The method of the present application is described as follows. Methods not described are prior arts.

[0100] S11 adaptive sampling frequency adjustment algorithm: fs(t) = fs_base [1 + α V(t) + β R(t)]; where: fs(t) is the dynamic sampling frequency at time t; V(t) = √[Σ(xi – x*) 2 ] / n is the parameter volatility, xi is the ith sampling value, x* is the sliding average of n sampling values, and n is the sliding window length; R(t) = |dx / dt| / max(|dx / dt|) is the rate of change indicator, dx / dt is the parameter change rate, and max(|dx / dt|) is the historical maximum change rate; fs_base is the base sampling frequency, ranging from [0.1,10] Hz; α is the volatility weight coefficient, ranging from [0,1]; β is the change rate weight coefficient, ranging from [0,1]; t is the current time.

[0101] S12 data complementation algorithm: X*(t) = λ1·X1(t) + λ2·X2(t) + λ3·X3(t); where: X*(t) is the repair data value at time t; X1(t), X2(t), and X3(t) are the measurement values ​​of the relevant sensors respectively; λi = exp(-γi·εi) / Σexp(-γj·εj) is the adaptive weight coefficient, i,j∈{1,2,3}; εi = |Xi(t) – X**(t)| / σ(t) is the normalized error, X**(t) is the historical average, and σ(t) is the standard deviation; γi is an adjustable parameter with a value range of [0.1,10].

[0102] S21 cross entropy analysis algorithm: H(X,Y) = -Σ[p(x,y)·log(q(x,y))]; where: H(X,Y) is the cross entropy between parameters X and Y; p(x,y) is the actual joint probability distribution; q(x,y) is the predicted joint probability distribution; KL(p||q) =Σ[p(x,y)·log(p(x,y) / q(x,y))] is the KL divergence; MI(X,Y) = Σ[p(x,y)·log(p(x,y) / (p(x)·p(y)))] is the mutual information; p(x) and p(y) are the marginal probability distributions respectively.

[0103] S22 Granger causality analysis algorithm: Y(t) = Σ[αi·Y(ti)] + Σ[βj·X(tj)] + ε(t); where: Y(t) is the dependent variable time series; X(t) is the independent variable time series; αi is the autoregression coefficient, i∈[1,p]; βj is the cross regression coefficient, j∈[1,q]; p,q is the maximum lag order; ε(t) is the residual term; F = [(RSS0-RSS1) / q] / [RSS1 / (npq)] is the F statistic; RSS0 is the residual sum of squares of the constrained model; RSS1 is the residual sum of squares of the unconstrained model; n is the sample size.

[0104] S23 parameter feature vector calculation method: F(x) = Σ[wi·Ki(x,xi)]; where: F(x) is the feature mapping function; Ki(x,xi) = exp(-||x-xi|| 2 / 2σ 2 ) is the Gaussian kernel function; wi is the weight coefficient of the i-th support vector; xi is the support vector; σ is the kernel function bandwidth parameter; ||x-xi|| is the Euclidean distance; A(t) = softmax(QK T / sqrt d)V is the attention weight matrix; Q is the query matrix; K is the key matrix; V is the value matrix; d is the feature dimension.

[0105] S31 deep reinforcement learning control algorithm: Q(s,a) = r + γ·maxQ'(s',a'); where: Q(s,a) is the state-action value function; s is the current state vector; a is the action vector; r is the immediate reward; γ is the discount factor, ranging from [0,1]; s' is the next state; a' is the next action; π(a|s) = exp(Q(s,a)) / Σexp(Q(s,a')) is the soft decision strategy; TD(t) = r + γ·V(s') - V(s) is the temporal difference error; V(s) is the state value function.

[0106] S32 multi-objective optimization algorithm: min F(x) = [f1(x), f2(x), ..., fm(x)]; where: F(x) is the target vector; fi(x) is the i-th optimization objective; x is the decision variable vector; gi(x) ≤ 0 is an inequality constraint; hj(x) = 0 is an equality constraint; P(x) = Σ[wi·fi(x)] + λ·[Σmax(0,gi(x)) 2 + Σhj(x) 2 ] is the penalty function; wi is the target weight; λ is the penalty factor.

[0107] S41 time-frequency domain / time-frequency evaluation algorithm: E(t,ω) = |∫x(τ)·g*(τ-t)·e^(-jωτ)dτ| 2 ; Where: E(t,ω) is the time-frequency energy spectrum; x(τ) is the time domain signal; g(t) is the Gaussian window function; ω is the angular frequency; j is the imaginary unit; τ is the integral variable; WC(t) = {[E(t,ω)]^0.5:ω∈Ω} is the wavelet coefficient; Ω is the frequency interval; C(t) = Σ|WC(t)|² is the time-frequency energy density. ^ is a superscript symbol.

[0108] S42 spatiotemporal feature extraction algorithm: Φ(x,t) = Σ[αi·φi(x)]·Σ[βj·ψj(t)]; where: Φ(x,t) is the spatiotemporal feature function; φi(x) is the spatial basis function; ψj(t) is the temporal basis function; αi is the spatial weight coefficient; βj is the temporal weight coefficient; x is the spatial coordinate vector; t is the time variable; R(x,t) = ∫∫Φ(x',t')·K(x-x',t-t')dx'dt' is the spatiotemporal correlation function; K(x,t) is the spatiotemporal kernel function.

[0109] It should be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

Claims

1. A smart dosing method, characterized in that: The steps include: Step S1, collecting water quality parameter data through sensors, performing reliability assessment on the water quality parameter data to obtain reliability score data, cross-validating and correcting the water quality parameter data according to the reliability score data and physical and chemical constraint relationships, and outputting corrected parameter data, data reliability vectors and data quality reports; Step S2, perform multi-scale analysis on the corrected parameter data to obtain parameter feature vectors, calculate the dynamic correlation between parameters based on the parameter feature vectors and the data reliability vectors to obtain a coupling strength matrix, build a prediction model based on the coupling strength matrix to generate parameter prediction data, and finally output a prediction confidence vector and a trend prediction report; Step S3, mapping the correction parameter data, parameter prediction data and prediction confidence vector to the state space to obtain a state vector, generating a preliminary dosing scheme based on the state vector to obtain a basic dosing matrix, performing multi-objective optimization on the basic dosing matrix to obtain optimized dosing parameters, and finally outputting the execution scheme data and the control parameter set; Step S4: Calculate the deviation between the corrected parameter data collected in real time and the parameter prediction data to obtain a prediction deviation vector, evaluate the dosing effect in combination with the execution plan data to obtain an effect evaluation index, adaptively optimize the control parameter set based on the effect evaluation index and the prediction deviation vector, and output parameter optimization suggestions and a system performance report.

2. The intelligent dosing method according to claim 1, characterized in that: The step S1 is specifically as follows: Step S11, collecting fluoride ion concentration data, pH value data, temperature data, conductivity data, dissolved oxygen data and turbidity data from multiple sensors in real time, and combining these data to form water quality parameter data; Read the historical standard deviation data and historical mean data of each parameter from the data storage unit; Step S12, calculating the ratio of the difference between each parameter in the water quality parameter data at adjacent moments and the corresponding historical standard deviation data to obtain a short-term stability index; calculating the deviation ratio between the water quality parameter data and the historical mean data to obtain a long-term drift index; inputting the short-term stability index and the long-term drift index into an exponential function to obtain reliability score data; Step S13, constructing a parameter constraint matrix according to the pre-stored physical and chemical relationship; calculating the deviation between the actual value and the theoretical value of each parameter in the water quality parameter data to obtain the parameter deviation; When the parameter deviation exceeds a preset threshold, the water quality parameter data is corrected based on the reliability score data and the parameter constraint matrix to obtain the corrected parameter data; The correction parameter data and reliability score data are combined into a data reliability vector, and a data quality report containing outlier markers and correction records is generated.

3. The intelligent dosing method according to claim 2, characterized in that: Step S2 is specifically as follows: Step S21, calculating the change rate of three different time windows of the modified parameter data to obtain multi-scale change rate data; performing weighted combination of the multi-scale change rate data and the data reliability vector to obtain a parameter feature vector; reading historical feature data from the data storage unit, calculating the similarity between the parameter feature vector and the historical feature data to obtain a feature stability index; Step S22, calculating the partial derivatives between the two parameters based on the parameter eigenvector to obtain a parameter sensitivity matrix; multiplying the parameter sensitivity matrix with the characteristic stability index and the data reliability vector to obtain a coupling strength matrix; identifying the main influencing parameter pairs according to the numerical values ​​in the coupling strength matrix, and generating a parameter association map; Step S23, reading a preset basis function parameter set from a data storage unit; optimizing the basis function parameter set using the modified parameter data, the parameter characteristic vector and the coupling strength matrix to obtain optimized function parameters; constructing a hybrid prediction model based on the optimized function parameters to generate parameter prediction data for the next n hours; The prediction confidence vector is calculated based on the characteristic stability index and the fitting error of the prediction model; n is greater than 1; Step S24, integrating the parameter prediction data, prediction confidence vector, parameter association map and coupling strength matrix to generate a trend prediction report; storing the newly generated parameter feature vector and coupling strength matrix into the data storage unit to update the historical feature data.

4. The intelligent dosing method according to claim 3, characterized in that: Step S3 is specifically as follows: Step S31, read the target parameter value and the historical dosing effect data from the data storage unit; multiply the deviation value between the corrected parameter data and the target parameter value by the data reliability vector to obtain the current state score; multiply the deviation value between the parameter prediction data and the target parameter value by the prediction confidence vector to obtain the predicted state score; combine the current state score, the predicted state score and the historical dosing effect data to form a state vector; Step S32, reading the initial control coefficient from the data storage unit; Calculate the change trend of each parameter based on the state vector to obtain the trend coefficient; The trend coefficient is multiplied by the initial control coefficient to obtain the adjustment coefficient; the deviation of each parameter is weightedly calculated using the adjustment coefficient to generate the dosage at different time points to form a basic dosing matrix; Calculate the system stability according to the state vector to obtain the stability index; Step S33, reading cost constraint parameters and energy consumption constraint parameters from the data storage unit; Calculate the cost of the drug based on the basic dosing matrix to obtain the cost index; Calculate energy consumption to obtain energy consumption index; combine cost index, energy consumption index and treatment effect into an optimization objective function; optimize the objective function to obtain optimized dosing parameters under the conditions of satisfying stability index and water quality compliance constraints; Step S34, based on the optimized dosing parameters, the basic dosing matrix is ​​modified to obtain a dosing execution sequence; the dosing execution sequence, the optimized dosing parameters and the stability index are combined to form execution plan data; the parameters obtained by this optimization are combined into a control parameter set; the execution effect in the execution plan data is stored in the data storage unit to update the historical dosing effect data.

5. The intelligent dosing method according to claim 4, characterized in that: Step S4 is specifically as follows: Step S41, calculating the time series deviation between the correction parameter data and the parameter prediction data to obtain the prediction deviation vector; extracting the dosing execution sequence from the execution plan data, calculating the response relationship between the dosing amount and the change of the water quality parameter to obtain the response characteristic data; The prediction deviation vector, the response characteristic data and the data reliability vector are weightedly combined to obtain the effect evaluation index; the historical evaluation data is read from the data storage unit to calculate the stability of the evaluation index to obtain the evaluation credibility; Step S42, extracting control coefficients from the control parameter set and calculating the time response curve of each parameter to obtain a time series characteristic matrix; Calculate the spatial correlation strength between parameters to obtain the spatial feature matrix; combine the temporal feature matrix and the spatial feature matrix with the effect evaluation index to form a system feature vector; read the historical feature data from the data storage unit, calculate the feature similarity to obtain the feature stability index; Step S43, calculating the gradient direction of the control parameter based on the effect evaluation index to obtain a parameter gradient vector; correcting the parameter gradient vector in combination with the prediction deviation vector and the characteristic stability index to obtain a corrected gradient vector; iteratively optimizing the control parameter set using the corrected gradient vector to obtain an optimized parameter group; and forming a parameter optimization suggestion by comparing the optimized parameter group with the original parameter. Step S44, combining the effect evaluation index, the system characteristic vector and the evaluation credibility to generate a system performance report; The effect evaluation index is stored in the data storage unit to update the historical evaluation data; the system characteristic vector is stored in the data storage unit to update the historical characteristic data; the optimized parameter group is stored in the data storage unit for the dosing control of the next cycle.

6. The intelligent dosing method according to claim 5, characterized in that: Step S12 is specifically as follows: Step S121, performing wavelet transform on the original parameter data to obtain wavelet coefficients; screening the wavelet coefficients based on the soft threshold criterion to obtain correction coefficients; using the correction coefficients to reconstruct the signal to obtain denoised data; The missing points in the denoised data are interpolated using cubic spline to obtain the complete data sequence; The complete data sequence is used to calculate the upper and lower limits of the standard deviation, and the data points that exceed the range are marked to obtain the abnormal marking vector; Step S122, calculating the variance of the complete data sequence in sections to obtain a variance sequence; processing the variance sequence using an exponential smoothing method to obtain a variance index; and performing equal-interval sampling on the complete data sequence to construct a probability distribution; Calculate the information entropy of the probability distribution to obtain the entropy index; perform autocorrelation analysis on the complete data sequence to obtain the correlation coefficient sequence; extract the periodic characteristics of the correlation coefficient to obtain the periodic index; perform exponentially weighted accumulation of the abnormal marking vector to obtain the abnormal index; Step S123, constructing an evaluation matrix consisting of a variance index, an entropy index, a period index and an anomaly index; processing the evaluation matrix using a principal component analysis method to obtain an eigenvalue sequence; The contribution rate of each indicator is calculated based on the eigenvalue to obtain the weight vector; the evaluation matrix and the weight vector are matrix multiplied to obtain the preliminary reliability score; The preliminary reliability score is transformed by sigmoid function to obtain the data reliability vector.

7. The intelligent dosing method according to claim 5, characterized in that: Step S13 is specifically as follows: Step S131, reading the stoichiometric relationship, equilibrium constant and reaction kinetic parameters from the data storage unit to obtain a theoretical relationship matrix; using the principal component analysis method to calculate the correlation degree of each parameter in the theoretical relationship matrix to obtain an association strength matrix; constructing a reaction priority sequence based on the reaction activity and the inhibition effect to obtain a reaction sequence vector; combining the theoretical relationship matrix, the association strength matrix and the reaction sequence vector to form a parameter constraint matrix; Step S132, standardize the water quality parameter data to obtain standardized parameters; use the parameter constraint matrix to calculate the theoretical balance value of the standardized parameters to obtain the theoretical parameter value; calculate the Mahalanobis distance between the standardized parameters and the theoretical parameter value to obtain the initial deviation value; The initial deviation value is weighted and modified in combination with the correlation strength matrix to obtain the parameter deviation; Step S133, comparing the parameter deviation with a preset threshold to generate a modified trigger vector; constructing a parameter weight matrix using the data reliability vector to obtain a reliability weight; The parameter constraint matrix is ​​weighted optimized based on the modified trigger vector and reliability weight to obtain the modified coefficient matrix; Perform matrix operation on water quality parameter data and correction coefficient matrix to obtain correction parameter data; Step S134, recording the positions where the parameter deviation exceeds the threshold as an abnormal mark sequence; The positions of non-zero elements in the correction coefficient matrix are recorded to obtain the correction position vector; the amplitude of parameter change before and after correction is calculated to obtain the correction amplitude vector; the abnormal mark sequence, the correction position vector and the correction amplitude vector are combined to generate a data quality report.

8. The intelligent dosing method according to claim 5, characterized in that: Step S21 is specifically as follows: Step S211, perform least square fitting on the correction parameter data to obtain a reference trend curve; subtract the correction parameter data from the reference trend curve to obtain a fluctuation component; use wavelet decomposition to decompose the fluctuation component into three subsequences of high frequency, medium frequency and low frequency; calculate the change rate of the high frequency subsequence within a 10-minute window to obtain a short-term change rate; calculate the change rate of the medium frequency subsequence within a 30-minute window to obtain a medium-term change rate; calculate the change rate of the low frequency subsequence within a 60-minute window to obtain a long-term change rate; combine the short-term change rate, the medium-term change rate and the long-term change rate to form multi-scale change rate data; Step S212, perform principal component analysis on the multi-scale change rate data to obtain a feature contribution vector; use the data reliability vector to construct a weight coefficient matrix to obtain a weighted mapping matrix; perform matrix operations on the multi-scale change rate data and the weighted mapping matrix to obtain an initial feature matrix; perform singular value decomposition on the initial feature matrix to extract main features to obtain a parameter feature vector; Step S213, reading recent historical feature data from the data storage unit; using a dynamic time warping algorithm to calculate the temporal distance between the parameter feature vector and the historical feature data to obtain a distance matrix; Construct a similarity spectrum based on the distance matrix to obtain a similarity sequence; Exponential smoothing of the similarity sequence is performed to obtain the characteristic stability index; The parameter feature vector is stored in the data storage unit to update the historical feature data.

9. The intelligent dosing method according to claim 5, characterized in that: Step S22 is specifically as follows: Step S221, using the parameter eigenvector to calculate the cross entropy between parameters to obtain a cross entropy matrix; performing Granger causality test on the parameter eigenvector to obtain a causal relationship matrix; calculating the mutual information of each component of the parameter eigenvector to obtain an information matrix; combining the cross entropy matrix, the causal relationship matrix and the information matrix to construct a time-varying response model to obtain a response coefficient matrix; using the chain rule to calculate the gradient of the response coefficient matrix to obtain a parameter sensitivity matrix; Step S222, performing singular value decomposition on the parameter sensitivity matrix to obtain a characteristic component matrix; The stability weight vector is constructed using the characteristic stability index; the reliability weight matrix is ​​generated based on the data reliability vector; Perform inner product operation on the eigencomponent matrix and the stability weight vector to obtain the intermediate coupling matrix; The coupling strength matrix is ​​obtained by performing Hadamard product on the intermediate coupling matrix and the reliability weight matrix; Step S223, performing spectral clustering on the coupling strength matrix to obtain a parameter group matrix; Calculate the connected components of the parameter group matrix to obtain a connectivity sequence; use the minimum spanning tree algorithm to process the connectivity sequence to obtain a tree structure matrix; extract the edges with weights exceeding a threshold in the tree structure matrix to obtain a key connection set; construct an adjacency matrix based on the key connection set to obtain a parameter association map; store the coupling strength matrix and the parameter association map in a data storage unit for subsequent analysis.

10. A smart dosing system, characterized in that: include: at least one processor; as well as, a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the smart medication adding method according to any one of claims 1 to 9.

Images