An AI-driven intelligent medicine injection control system and method

Abstract

The application provides an AI-driven intelligent dosing control system and method. A plurality of key dosing parameters are extracted from a plurality of parameter monitoring data streams of a target sedimentation tank, and long-period steady-state trend characteristics and short-period transient fluctuation characteristics of water body changes are extracted based on all the key dosing parameters. The short-period transient fluctuation characteristics are matched with fluctuation characteristics corresponding to historical water body purification failure cases of the target sedimentation tank, and then an associated compensation strategy for the target sedimentation tank to cope with abnormal fluctuations under the current dosing state is retrieved. A dosing compensation coefficient of a dosing pipeline is determined by combining the associated compensation strategy with the current dosing state. A dosing amount prediction value is predicted according to the long-period steady-state trend characteristics and dosing deviation data of the last dosing control period. An optimized dosing instruction for the water body of the target sedimentation tank is generated based on the dosing compensation coefficient and the dosing amount prediction value. The technical solution provided by the application can accurately identify short-period transient fluctuations of water quality in the water dosing process.

Description

Technical Field

[0001] This application relates to the field of drug delivery control technology, and more specifically, to an AI-driven intelligent drug delivery control system and method. Background Technology

[0002] In the field of water treatment, water quality assurance and cost control are of paramount importance. Traditional dosing methods, such as manual dosing based on experience or control methods based on simple parameters (turbidity, pH value, etc.), have many drawbacks. Manual dosing is not only labor-intensive, but also difficult to accurately control the dosage due to real-time changes in water quality and quantity, which can easily lead to waste of chemicals or substandard water quality. Simple parameter control models cannot fully reflect the complex characteristics of water quality and have obvious lag. With industrialization and urbanization, water pollution has become more serious, and water quality fluctuations have become more frequent and severe, placing higher demands on the accuracy of chemical dosing.

[0003] In existing dosing control, firstly, sensors are used to collect water quality parameters and operating condition data in real time, which are then transmitted to the controller. The controller calculates the optimal dosage based on preset algorithms (such as PID control, fuzzy control, neural network models, etc.) and combined with water quality targets and historical data. Then, it drives actuators such as metering pumps to complete the dosing. However, traditional dosing control often relies on models built based on fixed parameters (such as simple turbidity and pH value) or on manual experience. These methods cannot capture short-term transient fluctuations in water quality during the dosing process in a timely and accurate manner. For example, in the event of a sudden pollution event in the source water (pollutants entering the river, causing turbidity to increase in a short period of time), the fixed model cannot quickly adjust the dosage, resulting in a dosing lag in the dosing system. This can easily lead to overdosing and secondary pollution of the water body. Therefore, how to accurately identify short-term transient fluctuations in water quality during the dosing process and thus avoid overdosing and secondary pollution of the water body has become a challenge for the industry. Summary of the Invention

[0004] This application provides an AI-driven intelligent dosing control system and method that can accurately identify short-cycle transient fluctuations in water quality during the dosing process.

[0005] In a first aspect, this application provides an AI-driven intelligent drug delivery control method, comprising the following steps:

[0006] In response to the intelligent dosing trigger command, it collects multi-parameter monitoring data streams in real time when dosing is performed in the target sedimentation tank;

[0007] Based on the multi-parameter monitoring data stream, multiple key dosing parameters are extracted for dosing the target sedimentation tank. All key dosing parameters are then input into a multi-scale time-series feature extraction model to extract long-term steady-state trend features and short-term transient fluctuation features of water body changes.

[0008] The short-cycle transient fluctuation characteristics are matched with the fluctuation characteristics corresponding to historical water purification failure cases in the target sedimentation tank. When the similarity between the short-cycle transient fluctuation characteristics and any water purification failure case exceeds the adaptive threshold, the associated compensation strategy of the target sedimentation tank to deal with abnormal fluctuations under the current dosing state is retrieved. Then, the dosing compensation coefficient of the dosing pipeline is determined by combining the associated compensation strategy with the current dosing state.

[0009] Obtain the dosing deviation data of the target sedimentation tank in the previous dosing control cycle, and then predict the dosing amount of water in the target sedimentation tank in the next dosing control cycle based on the long-term steady-state trend characteristics and the dosing deviation data.

[0010] Based on the dosing compensation coefficient and the predicted dosing amount, an optimized dosing instruction is generated for the next dosing control cycle of the target sedimentation tank.

[0011] In some embodiments, the multiple key dosing parameters extracted from the multi-parameter monitoring data stream for dosing the target sedimentation tank specifically include:

[0012] The multi-parameter monitoring data stream is normalized to obtain a standardized data stream;

[0013] The standardized data stream is divided into multiple sub-data streams using a sliding window algorithm;

[0014] Principal component analysis is performed on each sub-data stream to obtain the dimensionality-reduced feature vector set;

[0015] Based on the preset key parameter feature template, pattern matching is performed in the feature vector set to determine multiple key dosing parameters when dosing the target sedimentation tank.

[0016] In some embodiments, inputting all key dosing parameters into a multi-scale time-series feature extraction model to extract long-term steady-state trend features and short-term transient fluctuation features of water body changes specifically includes:

[0017] All key dosing parameters are constructed into a data matrix according to time series, and the data matrix is ​​subjected to discrete wavelet transform to obtain the coefficient matrix of different frequency sub-bands;

[0018] Time pooling is performed on the coefficient matrix of the low-frequency subband to extract the long-term steady-state trend characteristics of water body changes.

[0019] A sliding time window is used to extract local features from the coefficient matrix of the high-frequency subband, thereby obtaining the short-period transient fluctuation characteristics of water body changes.

[0020] In some embodiments, matching the short-cycle transient fluctuation characteristics with the fluctuation characteristics corresponding to historical water purification failure cases in the target sedimentation tank specifically includes:

[0021] The fluctuation characteristics corresponding to the historical water purification failure cases are normalized to construct a standardized historical feature library.

[0022] The short-cycle transient fluctuation features are converted into feature vectors of the same dimension as the standardized historical feature library;

[0023] The cosine similarity algorithm is used to calculate the similarity between the transformed short-period transient fluctuation features and each feature vector in the standardized historical feature library.

[0024] In some embodiments, when the similarity between the short-cycle transient fluctuation characteristics and any water purification failure case exceeds an adaptive threshold, the specific compensation strategies for the target sedimentation tank to cope with abnormal fluctuations under the current dosing state include:

[0025] When the similarity between the short-cycle transient fluctuation characteristics and any water purification failure case exceeds the adaptive threshold, all water purification failure cases that exceed the adaptive threshold are filtered out.

[0026] Based on the current dosing parameters of the target sedimentation tank, a dosing parameter identifier is constructed, and target failure cases that match the current dosing parameter identifier are retrieved from water purification failure cases that exceed the adaptive threshold.

[0027] Extract the pre-associated and stored compensation strategies from the target failure cases as the corresponding associated compensation strategies for the target sedimentation tank under the current dosing conditions.

[0028] In some embodiments, determining the dosing compensation coefficient of the dosing pipeline by combining the associated compensation strategy with the current dosing status specifically includes:

[0029] The compensation parameters included in the associated compensation strategy are normalized with the drug administration status parameters in the current drug administration status, wherein the drug administration status parameters include the dosage and the drug administration speed.

[0030] Different weights are assigned to each drug administration state parameter based on its influence on the drug administration effect, in order to construct a parameter weight matrix;

[0031] The dosing status fusion index of the current target sedimentation tank is determined based on the normalized compensation parameters and the corresponding weights in the parameter weight matrix.

[0032] The drug dosing status fusion index is substituted into the preset compensation coefficient calculation model for calculation to obtain the drug dosing compensation coefficient of the drug dosing pipeline.

[0033] In some embodiments, multi-parameter monitoring data during the dosing of chemicals in the target sedimentation tank are collected by a multi-source sensor network deployed in the target sedimentation tank.

[0034] Secondly, this application provides an AI-driven intelligent drug delivery control system for executing an AI-driven intelligent drug delivery control method, including:

[0035] The data acquisition module is used to respond to intelligent dosing trigger commands and collect multi-parameter monitoring data streams in real time when dosing is performed in the target sedimentation tank;

[0036] The processing module is used to extract multiple key dosing parameters when dosing the target sedimentation tank based on the multi-parameter monitoring data stream, and input all key dosing parameters into a multi-scale time series feature extraction model to extract long-term steady-state trend features and short-term transient fluctuation features of water body changes.

[0037] The processing module is also used to perform similarity matching between the short-cycle transient fluctuation characteristics and the fluctuation characteristics corresponding to historical water purification failure cases of the target sedimentation tank. When the similarity between the short-cycle transient fluctuation characteristics and any water purification failure case exceeds the adaptive threshold, the module retrieves the associated compensation strategy of the target sedimentation tank to deal with abnormal fluctuations under the current dosing state, and then determines the dosing compensation coefficient of the dosing pipeline by combining the associated compensation strategy with the current dosing state.

[0038] The processing module is also used to obtain the dosing deviation data of the target sedimentation tank in the previous dosing control cycle, and then predict the dosing amount of water in the target sedimentation tank in the next dosing control cycle based on the long-term steady-state trend characteristics and the dosing deviation data.

[0039] The execution module is used to generate optimized dosing instructions for the next dosing control cycle of the target sedimentation tank based on the dosing compensation coefficient and the predicted dosing amount.

[0040] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described AI-driven intelligent drug delivery control method.

[0041] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned AI-driven intelligent drug delivery control method.

[0042] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0043] The AI-driven intelligent dosing control system and method provided in this application first responds to an intelligent dosing trigger command by real-time collecting multi-parameter monitoring data streams during dosing in the target sedimentation tank; secondly, it extracts multiple key dosing parameters based on the multi-parameter monitoring data streams during dosing in the target sedimentation tank, and inputs all key dosing parameters into a multi-scale time-series feature extraction model to extract long-term steady-state trend features and short-term transient fluctuation features of water body changes; further, it performs similarity matching between the short-term transient fluctuation features and the fluctuation features corresponding to historical water purification failure cases in the target sedimentation tank, and when the short-term transient fluctuation features are detected... When the similarity between the state fluctuation characteristics and any water purification failure case exceeds an adaptive threshold, the associated compensation strategy for the target sedimentation tank to cope with abnormal fluctuations under the current dosing state is retrieved. Then, the dosing compensation coefficient of the dosing pipeline is determined by combining the associated compensation strategy with the current dosing state. Next, the dosing deviation data of the target sedimentation tank in the previous dosing control cycle is obtained. Then, the predicted dosing amount of water in the target sedimentation tank in the next dosing control cycle is predicted based on the long-term steady-state trend characteristics and the dosing deviation data. Finally, the optimized dosing instruction for the target sedimentation tank in the next dosing control cycle is generated based on the dosing compensation coefficient and the predicted dosing amount.

[0044] Therefore, this application can accurately identify short-period transient fluctuations in water quality during chemical dosing. Firstly, it collects multi-parameter monitoring data streams from the target sedimentation tank in real time, providing rich and timely raw information for subsequent dosing control, ensuring data timeliness and comprehensiveness, and avoiding dosing decision errors due to missing or delayed data. Secondly, it extracts multiple key dosing parameters from the multi-parameter monitoring data streams during dosing in the target sedimentation tank, and inputs all key dosing parameters into a multi-scale time-series feature extraction model to extract long-period steady-state trend characteristics and short-period transient fluctuation characteristics of water quality changes. This is achieved by analyzing massive amounts of multi-parameter data. Key dosing parameters are screened from the monitoring data stream, focusing on core data that substantially affects dosing control, reducing redundant information interference. Key parameters are decomposed into long-term steady-state trends and short-term transient fluctuation characteristics to identify short-term transient fluctuations in water quality during dosing. This provides structured feature evidence for accurately controlling water quality change patterns and formulating dosing strategies, enhancing the system's ability to perceive changes in operating conditions at different time scales. The short-term transient fluctuation characteristics are then matched with the fluctuation characteristics corresponding to historical water purification failure cases in the target sedimentation tank. When a short-term transient fluctuation characteristic is detected to be similar to any water purification failure case... When the fluctuation exceeds the adaptive threshold, the system retrieves the associated compensation strategies for the target sedimentation tank under the current dosing state to address abnormal fluctuations. Then, by combining these strategies with the current dosing state, the system determines the dosing compensation coefficient for the dosing pipeline, enabling rapid response and precise compensation for sudden abnormal conditions and reducing the probability of water purification failure. Based on long-term steady-state trend characteristics and dosing deviation data from the previous dosing control cycle in the target sedimentation tank, the system predicts the predicted dosing amount for the next dosing control cycle, proactively estimating the dosing requirement for the next cycle and providing predictive guidance for dosing control, thus reducing the risk of adverse drug reactions due to indiscriminate dosing. This addresses the issues of agent waste or poor purification effects. Based on the dosing compensation coefficient and the predicted dosing amount, optimized dosing instructions for the target sedimentation tank water are generated. The dosing compensation coefficient dynamically corrects the predicted dosing amount, eliminating the deviation between real-time operating conditions and the predicted value. Then, optimized dosing instructions are generated according to the communication protocol encoding of the dosing equipment, precisely driving the dosing equipment to execute operations. This ensures the efficient and stable operation of the dosing system, improving the economic efficiency of agent use while guaranteeing water purification effects, and thus avoiding secondary pollution of the water body due to excessive dosing. In summary, the technical solution provided in this application can accurately identify short-cycle transient fluctuations in water quality during the dosing process. Attached Figure Description

[0045] Figure 1 This is an exemplary flowchart of an AI-driven intelligent drug delivery control method according to some embodiments of this application;

[0046] Figure 2 This is an exemplary flowchart illustrating the determination of key dosing parameters according to some embodiments of this application;

[0047] Figure 3 This is a schematic diagram of the structure of an AI-driven intelligent drug delivery control system according to some embodiments of this application;

[0048] Figure 4 This is a schematic diagram of the structure of a computer device implementing an AI-driven intelligent drug delivery control method according to some embodiments of this application. Detailed Implementation

[0049] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] refer to Figure 1 The figure is an exemplary flowchart of an AI-driven intelligent drug delivery control method according to some embodiments of this application. The figure mainly includes the following steps:

[0051] In step S101, in response to the intelligent dosing trigger command, multi-parameter monitoring data streams are collected in real time when dosing is performed in the target sedimentation tank.

[0052] In practice, when the dosing control system receives the intelligent dosing trigger command, it uses a multi-source sensor network (such as turbidity sensor, pH sensor, reagent concentration probe, temperature sensor, conductivity sensor and ultrasonic suspended solids concentration meter) deployed in the target sedimentation tank to monitor multi-parameter data during dosing. The multi-parameter monitoring data is then timestamped and standardized by edge computing nodes to form a multi-parameter monitoring data stream with time series labels. The multi-parameters include at least raw water turbidity, pH value, dosing concentration, tank temperature, conductivity and suspended solids concentration.

[0053] It should be noted that the multi-parameter monitoring data stream in this application represents the set of relevant water parameters when the water in the target sedimentation tank is purified by chemical dosing. The determination of the multi-parameter monitoring data stream can effectively provide real-time decision-making basis and closed-loop control foundation for the AI-driven intelligent dosing system.

[0054] In step S102, multiple key dosing parameters for dosing the target sedimentation tank are extracted based on the multi-parameter monitoring data stream, and all key dosing parameters are input into a multi-scale time-series feature extraction model to extract long-term steady-state trend features and short-term transient fluctuation features of water body changes.

[0055] In some embodiments, reference Figure 2 As shown in the figure, this is an exemplary flowchart for determining key dosing parameters according to some embodiments of this application. In this embodiment, the extraction of multiple key dosing parameters for dosing the target sedimentation tank based on the multi-parameter monitoring data stream can be achieved by the following steps:

[0056] In step S1021, the multi-parameter monitoring data stream is normalized to obtain a standardized data stream;

[0057] In step S1022, the standardized data stream is divided into multiple sub-data streams using a sliding window algorithm;

[0058] In step S1023, principal component analysis is performed on each sub-data stream to obtain the dimensionality-reduced feature vector set;

[0059] In step S1024, pattern matching is performed on the feature vector set based on the preset key parameter feature template to determine multiple key dosing parameters when dosing the target sedimentation tank.

[0060] In specific implementation, firstly, the Z-score normalization algorithm is used to process most multi-parameter monitoring data streams to eliminate the dimensional differences between different monitoring parameters, resulting in a standardized data stream. Secondly, a sliding window algorithm is used to divide the standardized data stream into multiple sub-data streams according to the window length (usually 15-30 minutes) and step size (usually set to half the window length) determined based on the hydraulic retention time of the sedimentation tank. Then, principal component analysis is performed on each sub-data stream. By calculating the covariance matrix, obtaining eigenvalues ​​and eigenvectors, and selecting the top few eigenvectors with a cumulative contribution rate exceeding 85%, a dimensionality-reduced eigenvector set is obtained. The cumulative contribution rate is the sum of the proportions of each principal component eigenvalue to the sum of the total eigenvalues ​​in the principal component analysis, used to measure... The principal components retained after dimensionality reduction have a comprehensive representation ability of the original data information. Finally, based on the preset key parameter feature template, pattern matching is performed in the feature vector set to determine multiple key dosing parameters when dosing the target sedimentation tank. That is, the preset key parameter feature template is obtained. This key parameter feature template refers to the feature pattern generated by cluster analysis of historical multi-parameter monitoring data streams (such as using the K-means algorithm). It will not be elaborated here. The cosine similarity between each feature vector in the feature vector set and the key parameter feature template is calculated. A similarity threshold is set (which can be set according to actual needs, usually 0.75). The original water body parameter combinations corresponding to feature vectors with cosine similarity exceeding the similarity threshold are determined as key dosing parameters.

[0061] It should be noted that the key dosing parameters in this application refer to the core parameters extracted from the multi-parameter monitoring data stream that play a decisive role in the water purification effect and dosing control. These parameters usually include key indicators that reflect the water quality characteristics, the operating status of the dosing system, and the working condition of the sedimentation tank. Determining the key dosing parameters can reveal the long-term steady-state trend and short-term transient fluctuations of water changes, and can effectively improve the system's ability to predict the patterns of water quality changes.

[0062] In some embodiments, inputting all key dosing parameters into a multi-scale time-series feature extraction model to extract long-term steady-state trend features and short-term transient fluctuation features of water body changes can be achieved through the following steps:

[0063] All key dosing parameters are constructed into a data matrix according to time series, and the data matrix is ​​subjected to discrete wavelet transform to obtain the coefficient matrix of different frequency sub-bands;

[0064] Time pooling is performed on the coefficient matrix of the low-frequency subband to extract the long-term steady-state trend characteristics of water body changes.

[0065] A sliding time window is used to extract local features from the coefficient matrix of the high-frequency subband, thereby obtaining the short-period transient fluctuation characteristics of water body changes.

[0066] In practice, firstly, a data matrix is ​​constructed by arranging all key dosing parameters in time series form. This involves arranging the time-series observations of each key dosing parameter into a matrix, where rows represent time points and columns represent different key dosing parameters. Next, a discrete wavelet transform is performed on the data matrix. This is achieved by selecting appropriate wavelet basis functions (such as Daubechies wavelets) and decomposition levels to decompose the data matrix into coefficient matrices of different frequency sub-bands. These sub-bands include low-frequency and high-frequency sub-bands. The low-frequency sub-bands contain long-term trend information, while the high-frequency sub-bands contain short-term fluctuation information. Then, a time pooling operation (e.g., average pooling) is performed on the coefficient matrix of the low-frequency sub-bands, and the resulting low-frequency components are used as long-term steady-state trend characteristics of water body changes. Finally, a sliding time window technique is applied to the coefficient matrix of the high-frequency sub-bands. Appropriate window length and step size are set (the specific values ​​can be determined according to actual needs), and statistical characteristics (such as standard deviation) are calculated within each local window as short-term transient fluctuation characteristics of water body changes.

[0067] It should be noted that, in this embodiment, the coefficient matrix refers to the coefficient matrix generated in different frequency sub-bands after the data matrix of key dosing parameters constructed according to the time series is subjected to discrete wavelet transform. Its element values ​​reflect the decomposition coefficients of the key dosing parameters in the corresponding frequency sub-bands. In this application, the long-period steady-state trend characteristics characterize the slow change trend of water parameters over a longer time scale, reflecting the overall evolution trend of water quality indicators or operating parameters. In this application, the short-period transient fluctuation characteristics capture the rapid fluctuation characteristics of water parameters over a shorter time scale, used to reflect the instantaneous change characteristics of parameters caused by sudden changes in influent load or instantaneous dosing of chemicals. In the control process, determining the long-term steady-state trend characteristics and short-term transient fluctuation characteristics is crucial. By understanding the long-term steady-state trend characteristics, we can grasp the changing patterns of water parameters over a longer period of time, and formulate a basic dosing strategy accordingly. This ensures that the dosage of chemicals matches the long-term purification needs of the water body, avoiding overdosing or underdosing due to insufficient prediction of long-term trends. On the other hand, short-term transient fluctuation characteristics can capture instantaneous changes in water parameters. When the fluctuation characteristics are detected to match historical abnormal conditions, a dynamic compensation mechanism can be triggered to adjust the dosing parameters in real time. This allows the dosing system to respond quickly to sudden changes in water quality and prevents a decrease in the treatment efficiency of the sedimentation tank due to short-term fluctuations.

[0068] In step S103, the short-cycle transient fluctuation characteristics are matched with the fluctuation characteristics corresponding to historical water purification failure cases in the target sedimentation tank. When the similarity between the short-cycle transient fluctuation characteristics and any water purification failure case exceeds the adaptive threshold, the associated compensation strategy for the target sedimentation tank to deal with abnormal fluctuations under the current dosing state is retrieved. Then, the dosing compensation coefficient of the dosing pipeline is determined by combining the associated compensation strategy with the current dosing state.

[0069] In some embodiments, the similarity matching between the short-cycle transient fluctuation characteristics and the fluctuation characteristics corresponding to historical water purification failure cases in the target sedimentation tank can be achieved by the following steps:

[0070] The fluctuation characteristics corresponding to the historical water purification failure cases are normalized to construct a standardized historical feature library.

[0071] The short-cycle transient fluctuation features are converted into feature vectors of the same dimension as the standardized historical feature library;

[0072] The cosine similarity algorithm is used to calculate the similarity between the transformed short-period transient fluctuation features and each feature vector in the standardized historical feature library.

[0073] In specific implementation, firstly, the fluctuation characteristics corresponding to the historical water purification failure cases are normalized using the Z-score normalization algorithm. This unifies historical fluctuation characteristics with different dimensions and numerical ranges to the same scale, eliminating interference caused by differences in data scale, and thus constructing a standardized historical feature library. This standardized historical feature library stores a set of fluctuation characteristic data under historical abnormal dosing conditions, and the fluctuation characteristics are historical short-cycle transient fluctuation characteristics under historical abnormal dosing conditions. Secondly, the short-cycle transient fluctuation characteristics are converted into feature vectors with the same dimension as the feature vectors in the standardized historical feature library through existing feature extraction and transformation operations (such as principal component analysis dimensionality reduction), ensuring that the two are comparable. Finally, the cosine similarity algorithm is used to calculate the cosine value of the angle between the converted short-cycle transient fluctuation feature vector and each feature vector in the standardized historical feature library as the similarity between the converted short-cycle transient fluctuation feature and the feature vector.

[0074] It should be noted that, in this embodiment, the standardized historical feature library refers to a feature set database formed by normalizing the fluctuation features extracted from historical water purification failure cases and storing them in a unified data structure. Each feature vector in this library corresponds to a fluctuation pattern under a historical abnormal operating condition, providing a benchmark for comparing real-time fluctuation features. In this application, similarity represents an index that measures the degree of similarity between short-cycle transient fluctuation feature vectors and feature vectors in the standardized historical feature library. It is used to measure the directional similarity between vectors. In dosing control, the core role of determining similarity is to provide a precise decision-making basis for the dynamic adjustment of dosing strategies by quantifying the degree of matching between the current short-cycle transient fluctuation features and the fluctuation features of historical water purification failure cases.

[0075] In some embodiments, when the similarity between the short-cycle transient fluctuation characteristics and any water purification failure case exceeds an adaptive threshold, the following steps can be used to retrieve the associated compensation strategy for the target sedimentation tank to cope with the abnormal fluctuations under the current dosing state:

[0076] When the similarity between the short-cycle transient fluctuation characteristics and any water purification failure case exceeds the adaptive threshold, all water purification failure cases that exceed the adaptive threshold are filtered out.

[0077] Based on the current dosing parameters of the target sedimentation tank, a dosing parameter identifier is constructed, and target failure cases that match the current dosing parameter identifier are retrieved from water purification failure cases that exceed the adaptive threshold.

[0078] Extract the pre-associated and stored compensation strategies from the target failure cases as the corresponding associated compensation strategies for the target sedimentation tank under the current dosing conditions.

[0079] In specific implementation, firstly, when the similarity between the short-cycle transient fluctuation characteristics and any water purification failure case exceeds an adaptive threshold, all water purification failure cases exceeding the adaptive threshold are filtered out. The adaptive threshold can be dynamically adjusted based on the statistical characteristics of historical similarity data and actual operating conditions to balance the risks of misjudgment and omission; this is not limited here. Secondly, a unique dosing parameter identifier is constructed according to preset rules based on the current dosing parameters of the target sedimentation tank. For example, different parameters are encoded and combined into a string, which serves as the dosing parameter identifier. Among the water purification failure cases exceeding the adaptive threshold, a string matching method is used to retrieve water purification failure cases that match the current dosing parameter identifier as target time-sensitive cases. Finally, the pre-associated and stored compensation strategies in the target failure cases are extracted as the corresponding associated compensation strategies for the target sedimentation tank under the current dosing. The compensation strategies are solutions formulated based on historical experience and expert knowledge for specific failure modes and dosing parameter combinations, such as adjusting the reagent ratio and changing the dosing frequency, thereby achieving dynamic adjustment and optimization of dosing control.

[0080] It should be noted that, in this embodiment, the water purification failure case refers to a specific working condition instance where the purification effect fails to meet the expected standard due to improper dosing control; in this embodiment, the target failure case refers to the water purification failure case that is closest to the current dosing state; in this embodiment, the dosing parameter identifier is a unique identifier representing the current dosing state, used to quickly locate and match cases in the case library; in this application, the association compensation strategy represents a set of operation schemes used to compensate for the dosing working condition. Through the association mapping with the target failure case, the association compensation strategy can effectively provide accurate operation guidance for real-time dosing control, so as to eliminate or mitigate the risk of water purification failure.

[0081] In some embodiments, determining the dosing compensation coefficient of the dosing pipeline by combining the associated compensation strategy with the current dosing status can be achieved through the following steps:

[0082] The compensation parameters included in the associated compensation strategy are normalized with the drug administration status parameters in the current drug administration status, wherein the drug administration status parameters include the dosage and the drug administration speed.

[0083] Different weights are assigned to each drug administration state parameter based on its influence on the drug administration effect, in order to construct a parameter weight matrix;

[0084] The dosing status fusion index of the current target sedimentation tank is determined based on the normalized compensation parameters and the corresponding weights in the parameter weight matrix.

[0085] The drug dosing status fusion index is substituted into the preset compensation coefficient calculation model for calculation to obtain the drug dosing compensation coefficient of the drug dosing pipeline.

[0086] In specific implementation, firstly, the Z-score normalization algorithm is used to normalize the compensation parameters (i.e., adjusted dosage and dosage rate) included in the associated compensation strategy and the dosage state parameters (i.e., actual dosage and dosage rate) in the current dosage state. This unifies parameters with different dimensions and numerical ranges to the same scale, eliminating the impact of data dimension differences on subsequent calculations. Secondly, based on long-term practical experience and process principles, the influence of each dosage state parameter (i.e., dosage and dosage rate) on the final dosage effect (i.e., water purification efficiency) is analyzed. Each dosage state parameter is assigned a corresponding weight to construct a parameter weight matrix. This parameter weight matrix clarifies the role of each dosage state parameter in the factors affecting the dosage effect. The relative importance is determined; then, the normalized compensation parameters and their corresponding weights in the parameter weight matrix are weighted and summed to obtain the dosing status fusion index of the current target sedimentation tank. The dosing status fusion index can comprehensively reflect the correlation between the compensation strategy and the current dosing status. Finally, the dosing status fusion index is substituted into the compensation coefficient calculation model (such as a linear regression model) that has been pre-fitted with historical data and verified and adjusted multiple times. The dosing compensation coefficient of the dosing pipeline is calculated through the compensation coefficient calculation model. This dosing compensation coefficient is used to quantitatively adjust the operating parameters of the dosing pipeline (such as increasing or decreasing the dosage by a certain percentage) to cope with the current abnormal dosing conditions. The compensation coefficient calculation model is as follows: (Where, K is the drug administration compensation coefficient;) , , , These are the regression coefficients of the model, which can be set according to actual needs; , The weights for drug dosage and drug delivery speed are respectively; C is the drug dosage parameter; V is the drug delivery speed parameter; and h is the drug delivery status fusion index.

[0087] It should be noted that, in this embodiment, the parameter weight matrix represents a set of weights constructed based on the degree of influence of the dosing status parameters on the dosing effect, used to measure the importance of each dosing status parameter in the calculation; in this embodiment, the dosing status fusion index is a comprehensive dosing status index characterizing the dosing pipeline; in this application, the dosing compensation coefficient characterizes the adjustment range of the actual dosing amount in the dosing pipeline. In dosing control, the determination of the dosing compensation coefficient plays a key role. By integrating the real-time dosing status and compensation strategy parameters, the dosing compensation coefficient transforms the theoretical compensation requirement into an executable quantitative adjustment instruction, enabling the dosing system to dynamically correct the dosing amount according to changes in water quality, avoiding waste of reagents or insufficient purification effect caused by a fixed dosing mode.

[0088] In step S104, the dosing deviation data of the previous dosing control cycle of the target sedimentation tank is obtained, and then the predicted dosing amount of water in the target sedimentation tank for the next dosing control cycle is predicted based on the long-term steady-state trend characteristics and the dosing deviation data.

[0089] In some embodiments, obtaining the dosing deviation data of the previous dosing control cycle in the target sedimentation tank can be achieved by the following steps:

[0090] Retrieve the actual dosage data and the preset dosage data for the corresponding period recorded in the previous dosing control cycle of the target sedimentation tank;

[0091] The actual dosage data is compared with the preset dosage data time by time, and the dosage difference at each time is calculated.

[0092] Statistical analysis was performed on the differences in drug dosage at all times to obtain the drug dosage deviation data for the previous drug control cycle.

[0093] In practice, firstly, the actual dosage data and the preset dosage data for the corresponding period are retrieved from the target sedimentation tank's operational database during the previous dosing control cycle. This data is typically stored in time series format, covering dosage information at each time point throughout the cycle. Secondly, a time-by-time comparison method is used to match the actual dosage data with the preset dosage data one-to-one. The dosage difference at each time point is calculated using subtraction operations. This difference directly reflects the difference between the actual dosage and the expected dosage at each time point. Finally, statistical analysis methods are applied to the dosage differences at all times to obtain the dosing deviation data for the previous dosing control cycle. The dosing deviation data includes multiple dosing deviation influencing factors, such as calculating the average, standard deviation, maximum, and minimum values ​​of the differences, thereby comprehensively assessing the degree and fluctuation of the dosage deviation from the preset value throughout the entire dosing control cycle.

[0094] It should be noted that the actual dosage data in this embodiment refers to the actual dosage record output by the dosing equipment during operation; the preset dosage data in this embodiment is the expected dosage set in advance based on water quality indicators and treated water volume; the dosing deviation data in this application is a comprehensive data indicator reflecting the dosage control during the dosing process.

[0095] In some embodiments, predicting the predicted dosage of the water body in the target sedimentation tank for the next dosing control cycle based on the long-term steady-state trend characteristics and the dosing deviation data can be achieved by the following steps:

[0096] The long-period steady-state trend characteristics are quantified into numerical feature vectors;

[0097] Construct a deviation correction matrix containing historical dosing deviation influencing factors based on the dosing deviation data;

[0098] The numerical feature vectors are weighted and fused with the deviation correction matrix to obtain the fused feature vectors;

[0099] The fused feature vector is input into a pre-trained time-series prediction model to calculate the predicted dosage of chemicals in the target sedimentation tank for the next dosing control cycle.

[0100] In specific implementation, firstly, the long-term steady-state trend features are transformed into numerical feature vectors using existing feature extraction methods. For example, the trend direction, amplitude, and period in the long-term steady-state trend features are encoded as element values ​​of a multi-dimensional vector, giving it a mathematical basis. Secondly, a deviation correction matrix containing historical dosing deviation influencing factors is constructed based on the dosing deviation data. That is, the deviation correction matrix analyzes the influencing factors of historical dosing deviations, organizes different types of deviation influencing factors into a matrix form according to preset weights, and then obtains the deviation correction matrix, which is used to quantify and correct systematic errors in dosing prediction. Then, matrix multiplication is performed... The numerical feature vector is weighted and fused with the deviation correction matrix to couple the long-term steady-state trend characteristics with the deviation correction factor, forming a fused feature vector that comprehensively reflects the long-term trend and historical deviation, thereby overcoming the limitations of single feature prediction. Finally, the fused feature vector is input into a time series prediction model (such as the autoregressive moving average model ARIMA) that has been trained using historical dosing data. This model uses the autocorrelation of time series and historical patterns to perform time series extrapolation on the fused feature vector, and finally outputs the predicted dosing amount of the target sedimentation tank in the future control cycle, providing a forward-looking quantitative basis for dosing control.

[0101] It should be noted that, in this embodiment, the numerical feature vector is a trend feature converted into a calculable numerical vector; the deviation correction matrix in this embodiment is a matrix structure that integrates historical dosing deviation influence factors and their weights; the fusion feature vector in this embodiment is a composite feature vector that integrates trend features and deviation correction factors; the time series prediction model in this embodiment is a mathematical model that predicts future values ​​based on the regularity of time series data, and is often used for parameter prediction in industrial processes; in this application, the predicted dosing value represents the expected dosing value of the target sedimentation tank within the future control cycle. As a key input for feedforward control, the predicted dosing value can adjust the execution parameters such as the dosing pump frequency and valve opening in advance based on the prediction of future operating conditions, avoiding excessive or insufficient dosing due to lagging control. For example, when it is known that the concentration of pollutants in the influent will increase, the predicted value will drive the system to increase the dosing in advance to ensure the purification effect.

[0102] In step S105, an optimized dosing instruction for the next dosing control cycle of the target sedimentation tank is generated based on the dosing compensation coefficient and the predicted dosing amount.

[0103] In some embodiments, generating optimized dosing instructions for the next dosing control cycle of the target sedimentation tank based on the dosing compensation coefficient and the predicted dosing amount can be achieved by the following steps:

[0104] The predicted dosage value is corrected by the dosage compensation coefficient to obtain the corrected dosage value;

[0105] The corrected dosage values ​​are encoded according to the instruction format of the dosing equipment to generate optimized dosing instructions for the next dosing cycle of the target sedimentation tank.

[0106] In practice, firstly, the predicted dosage is corrected using a dosing compensation coefficient. This is achieved by multiplying the dosing compensation coefficient by the predicted dosage, resulting in a corrected dosage value. The dosing compensation coefficient is an adjustment coefficient derived from a specified calculation model, taking into account the differences between the current dosing status and the compensation strategy. Its function is to dynamically correct the predicted dosage based on the actual operating conditions. Secondly, the corrected dosage value is encoded according to the instruction format of the dosing equipment (such as a dosing valve) to generate an optimized dosing instruction for the next dosing cycle of the target sedimentation tank. This optimized dosing instruction can be accurately identified and executed by the dosing equipment, achieving precise control of the dosage.

[0107] It should be noted that the corrected dosage value in this embodiment is the dosage data that is more in line with the current working conditions after adjustment by the dosage compensation coefficient; the optimized dosage instruction in this application is generated according to the communication protocol encoding of the dosage equipment and is a set of instructions used to control the operation of the dosage equipment to ensure the accuracy and effectiveness of the dosage operation.

[0108] Furthermore, in another aspect of this application, in some embodiments, this application provides an AI-driven intelligent drug delivery control system, referencing... Figure 3 The figure is a schematic diagram of the structure of an AI-driven intelligent drug delivery control system according to some embodiments of this application. The AI-driven intelligent drug delivery control system includes: a data acquisition module 201, a processing module 202, and an execution module 203, which are described below:

[0109] The data acquisition module 201 in this application is mainly used to respond to the intelligent dosing trigger command and collect multi-parameter monitoring data streams in real time when dosing is performed in the target sedimentation tank;

[0110] Processing module 202, in this application, is mainly used to extract multiple key dosing parameters when dosing the target sedimentation tank based on the multi-parameter monitoring data stream, and input all key dosing parameters into a multi-scale time series feature extraction model to extract long-term steady-state trend features and short-term transient fluctuation features of water body changes;

[0111] The processing module 202 is further configured to perform similarity matching between the short-cycle transient fluctuation characteristics and the fluctuation characteristics corresponding to historical water purification failure cases of the target sedimentation tank. When the similarity between the short-cycle transient fluctuation characteristics and any water purification failure case exceeds the adaptive threshold, the module retrieves the associated compensation strategy of the target sedimentation tank to deal with abnormal fluctuations under the current dosing state, and then determines the dosing compensation coefficient of the dosing pipeline by combining the associated compensation strategy with the current dosing state.

[0112] In addition, the processing module 202 is also used to obtain the dosing deviation data of the previous dosing control cycle of the target sedimentation tank, and then predict the dosing amount of the target sedimentation tank in the next cycle based on the long-term steady-state trend characteristics and the dosing deviation data.

[0113] The execution module 203 in this application is mainly used to generate optimized dosing instructions for the next dosing control cycle of the target sedimentation tank based on the dosing compensation coefficient and the predicted dosing amount.

[0114] In addition, this application also provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described AI-driven intelligent drug delivery control method.

[0115] In some embodiments, reference Figure 4 The figure is a schematic diagram of the structure of a computer device implementing an AI-driven intelligent drug delivery control method according to some embodiments of this application. The AI-driven intelligent drug delivery control method in the above embodiments can... Figure 4 The computer device shown is used to implement this, and the computer device includes at least one processor 301, a communication bus 302, a memory 303, and at least one communication interface 304.

[0116] The processor 301 can be a general-purpose central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more devices used to control the execution of the AI-driven intelligent drug delivery control method described in this application.

[0117] The communication bus 302 can be used to transmit information between the aforementioned components.

[0118] The memory 303 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 303 may exist independently and be connected to the processor 301 via the communication bus 302. The memory 303 may also be integrated with the processor 301.

[0119] The memory 303 stores program code for executing the scheme of this application, and its execution is controlled by the processor 301. The processor 301 executes the program code stored in the memory 303. The program code may include one or more software modules. In the above embodiments, the determination of the AI-driven intelligent drug delivery control method can be achieved through the processor 301 and one or more software modules in the program code in the memory 303.

[0120] Communication interface 304 uses any transceiver-like device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0121] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0122] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0123] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described AI-driven intelligent drug delivery control method.

[0124] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0125] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. An AI-driven intelligent drug delivery control method, characterized in that, Includes the following steps: In response to the intelligent dosing trigger command, it collects multi-parameter monitoring data streams in real time when dosing is performed in the target sedimentation tank; Based on the multi-parameter monitoring data stream, multiple key dosing parameters are extracted for dosing the target sedimentation tank. All key dosing parameters are then input into a multi-scale time-series feature extraction model to extract long-term steady-state trend features and short-term transient fluctuation features of water body changes. The short-cycle transient fluctuation characteristics are matched with the fluctuation characteristics corresponding to historical water purification failure cases in the target sedimentation tank. When the similarity between the short-cycle transient fluctuation characteristics and any water purification failure case exceeds the adaptive threshold, the associated compensation strategy of the target sedimentation tank to deal with abnormal fluctuations under the current dosing state is retrieved. Then, the dosing compensation coefficient of the dosing pipeline is determined by combining the associated compensation strategy with the current dosing state. Obtain the dosing deviation data of the target sedimentation tank in the previous dosing control cycle, and then predict the dosing amount of water in the target sedimentation tank in the next dosing control cycle based on the long-term steady-state trend characteristics and the dosing deviation data. Based on the dosing compensation coefficient and the predicted dosing amount, an optimized dosing instruction is generated for the next dosing control cycle of the target sedimentation tank.

2. The method as described in claim 1, characterized in that, Based on the multi-parameter monitoring data stream, several key dosing parameters for dosing chemicals in the target sedimentation tank are extracted, including: The multi-parameter monitoring data stream is normalized to obtain a standardized data stream; The standardized data stream is divided into multiple sub-data streams using a sliding window algorithm; Principal component analysis is performed on each sub-data stream to obtain the dimensionality-reduced feature vector set; Based on the preset key parameter feature template, pattern matching is performed in the feature vector set to determine multiple key dosing parameters when dosing the target sedimentation tank.

3. The method as described in claim 1, characterized in that, All key dosing parameters are input into a multi-scale time-series feature extraction model to extract long-term steady-state trend features and short-term transient fluctuation features of water body changes, specifically including: All key dosing parameters are constructed into a data matrix according to time series, and the data matrix is ​​subjected to discrete wavelet transform to obtain the coefficient matrix of different frequency sub-bands; Time pooling is performed on the coefficient matrix of the low-frequency subband to extract the long-term steady-state trend characteristics of water body changes. A sliding time window is used to extract local features from the coefficient matrix of the high-frequency subband, thereby obtaining the short-period transient fluctuation characteristics of water body changes.

4. The method as described in claim 1, characterized in that, The similarity matching between the short-period transient fluctuation characteristics and the fluctuation characteristics corresponding to historical water purification failure cases in the target sedimentation tank specifically includes: The fluctuation characteristics corresponding to the historical water purification failure cases are normalized to construct a standardized historical feature library. The short-cycle transient fluctuation features are converted into feature vectors of the same dimension as the standardized historical feature library; The cosine similarity algorithm is used to calculate the similarity between the transformed short-period transient fluctuation features and each feature vector in the standardized historical feature library.

5. The method as described in claim 1, characterized in that, When the similarity between the short-period transient fluctuation characteristics and any water purification failure case exceeds an adaptive threshold, the specific compensation strategies for abnormal fluctuations in the target sedimentation tank under the current dosing state include: When the similarity between the short-cycle transient fluctuation characteristics and any water purification failure case exceeds the adaptive threshold, all water purification failure cases that exceed the adaptive threshold are filtered out. Based on the current dosing parameters of the target sedimentation tank, a dosing parameter identifier is constructed, and target failure cases that match the current dosing parameter identifier are retrieved from water purification failure cases that exceed the adaptive threshold. Extract the pre-associated and stored compensation strategies from the target failure cases as the corresponding associated compensation strategies for the target sedimentation tank under the current dosing conditions.

6. The method as described in claim 1, characterized in that, The dosing compensation coefficient for the dosing pipeline is determined by combining the associated compensation strategy with the current dosing status. Specifically, this includes: The compensation parameters included in the associated compensation strategy are normalized with the drug administration status parameters in the current drug administration status, wherein the drug administration status parameters include the dosage and the drug administration speed. Different weights are assigned to each drug administration state parameter based on its influence on the drug administration effect, in order to construct a parameter weight matrix; The dosing status fusion index of the current target sedimentation tank is determined based on the normalized compensation parameters and the corresponding weights in the parameter weight matrix. The drug dosing status fusion index is substituted into the preset compensation coefficient calculation model for calculation to obtain the drug dosing compensation coefficient of the drug dosing pipeline.

7. The method as described in claim 1, characterized in that, Multi-parameter monitoring data during chemical dosing in the target sedimentation tank are collected by a multi-source sensor network deployed in the target sedimentation tank.

8. An AI-driven intelligent drug delivery control system, used to execute the AI-driven intelligent drug delivery control method as described in any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to respond to intelligent dosing trigger commands and collect multi-parameter monitoring data streams in real time when dosing is performed in the target sedimentation tank; The processing module is used to extract multiple key dosing parameters when dosing the target sedimentation tank based on the multi-parameter monitoring data stream, and input all key dosing parameters into a multi-scale time series feature extraction model to extract long-term steady-state trend features and short-term transient fluctuation features of water body changes. The processing module is also used to perform similarity matching between the short-cycle transient fluctuation characteristics and the fluctuation characteristics corresponding to historical water purification failure cases of the target sedimentation tank. When the similarity between the short-cycle transient fluctuation characteristics and any water purification failure case exceeds the adaptive threshold, the module retrieves the associated compensation strategy of the target sedimentation tank to deal with abnormal fluctuations under the current dosing state, and then determines the dosing compensation coefficient of the dosing pipeline by combining the associated compensation strategy with the current dosing state. The processing module is also used to obtain the dosing deviation data of the target sedimentation tank in the previous dosing control cycle, and then predict the dosing amount of water in the target sedimentation tank in the next dosing control cycle based on the long-term steady-state trend characteristics and the dosing deviation data. The execution module is used to generate optimized dosing instructions for the next dosing control cycle of the target sedimentation tank based on the dosing compensation coefficient and the predicted dosing amount.

9. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing code, and the processor being configured to retrieve the code and execute the AI-driven intelligent drug delivery control method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the AI-driven intelligent drug delivery control method as described in any one of claims 1 to 7.

Images