Water purification process based on edge intelligent regulation

Abstract

The application discloses a water purification process based on edge intelligent regulation and control, and relates to the technical field of intelligent regulation and control. The water purification process comprises initial data collection, data analysis and the like. The initial data collection collects equipment data of each initial stage of the water purification process as first equipment data. The data analysis sends the first equipment data to an edge computing center for analysis of the equipment data through an intelligent algorithm model to obtain an analysis result. Through edge-cloud collaborative architecture design, the edge equipment processes real-time data, the cloud performs big data modeling and model iteration updating, and the continuous optimization of the overall efficiency of the system is realized. Meanwhile, through the distributed deployment of encrypted edge intelligent equipment at each stage of the water purification process, real-time analysis of water quality sensor data, dynamic regulation of process parameters, independent decision-making of the edge end, rapid response in the case of sudden and high-frequency rapid fluctuation of raw water quality and in the case of no network or weak network environment, and reduction of the risk of user data leakage are realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, specifically to a water purification process based on edge intelligent control. Background Technology

[0002] Traditional water purification processes rely heavily on preset parameters or manual experience for control. When faced with complex and ever-changing water source quality and dynamic treatment needs, they often suffer from slow response, inefficient regulation, and low energy efficiency. In recent years, the Internet of Things (IoT) and artificial intelligence (AI) technologies have provided new ideas for optimizing water purification processes. However, centralized cloud-based intelligent processing models face challenges in terms of real-time performance, privacy, and reliability. Therefore, this invention proposes a water purification process with edge intelligent control.

[0003] In actual water treatment plant operation, precise control of process parameters at each stage is crucial to the subsequent treatment effect and cost. For example, adjusting the coagulant dosage is essential when facing a sharp increase in raw water turbidity due to heavy rain, requiring rapid and accurate adjustments to ensure sedimentation. Existing technologies have attempted to address this issue using intelligent methods, such as a cloud-based intelligent dosing system (patent publication number CN119118318A). This system collects raw water parameters, such as turbidity and pH, through online water quality monitoring instruments, transmits the data to a remote cloud platform, calculates the optimal dosing command using an algorithm model deployed in the cloud, and then sends it to the on-site dosing equipment for execution.

[0004] However, when dealing with sudden, high-frequency, and rapid fluctuations in raw water quality, such as the initial stages of heavy rain or industrial drainage impacts, the aforementioned equipment suffers from significant response delays due to its reliance on a remote cloud platform for centralized data processing and decision-making. Furthermore, it exhibits low user data privacy and security, as well as limited ability to self-iterate and update its models. Summary of the Invention

[0005] The purpose of this invention is to provide a water purification process based on edge intelligent control to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a water purification process based on edge intelligent control, comprising:

[0007] Initial data collection: Collect equipment data for each initial stage of the water purification process as the first equipment data;

[0008] Data analysis: The data from the first device is sent to the edge computing center, where it is analyzed using intelligent algorithm models to obtain the analysis results;

[0009] Command transmission: The edge computing center sends control commands to the corresponding equipment in the water purification process based on the analysis results; the equipment data of each stage of the water purification process after the control commands are sent are collected as the second equipment data;

[0010] Edge-cloud collaboration: edge devices process local data and model training, while the cloud performs global model aggregation and distribution.

[0011] Furthermore, the method for generating the intelligent algorithm model includes:

[0012] S1: Acquire data, acquire demand data, first estimated parameters and first equipment data, the first equipment data including first real-time parameters;

[0013] S2: Data comparison, compare the magnitudes of the first estimated parameter and the first real-time parameter to obtain the first adjustment data;

[0014] S3: Data adjustment, adjust the first real-time parameter according to the first adjustment data, and obtain the second device data, the second device data including the second real-time parameter and the corresponding first effect data;

[0015] S4: Feedback adjustment. First, compare the effect data with the demand data. If the demand is met, the second real-time parameter is qualified and stored. If the demand is not met, the second real-time parameter is unqualified. Return to steps S2-S4 until it is qualified and save the final real-time parameter.

[0016] Furthermore, the method for generating the first estimated parameter includes the following steps:

[0017] M1: Database establishment, the data includes: demand data for each stage of the water purification process, corresponding estimated parameters, and parameter sensitivity matrix;

[0018] M2: The generation of the first predicted parameter includes the following steps:

[0019] Step 1: Demand matching retrieval. Analyze the water quality demand characteristics of the current process stage and perform multi-level matching in the database, including: exact matching, finding completely identical demand records; fuzzy matching, retrieving historical demands with 85-100% similarity in key indicators; and extended matching, associating with common parameter templates of the same process stage.

[0020] Step 2: Parameter fusion algorithm, which performs weighted fusion on the retrieved historical parameters:

[0021] First estimated parameter = Σ(parameter value × weight coefficient) / total weight;

[0022] Total weight = Timeliness factor × Historical success rate;

[0023] Time factor = a 当前时间-记录时间a is 0.7-0.9, the current time refers to the system real-time time when the time factor is calculated, and the recording time refers to the time when the water quality data was last updated. Both the current time and the recording time are in hours.

[0024] Step 3: New requirement processing. When the historical matching degree is 84-70%, construct a virtual process model, perform parameter space scanning, select the effect data that is closest to the requirement as the first predicted parameter, and store the new parameter pair in the database for verification.

[0025] Furthermore, the method for changing the first estimated parameter includes the following steps:

[0026] N1: Triggering condition: The deviation between the first effect data and the required data is 5-20%;

[0027] N2: The method for changing the first predicted parameter includes the following steps:

[0028] Step 1: Deviation Calculation, calculate the deviation of the current process section's required indicators;

[0029] Step 2: Parameter sensitivity calculation. Call the preset sensitivity matrix, perform real-time perturbation verification, and simulate the effect change rate by fluctuating the first estimated parameter by ±5%. Sensitivity coefficient = effect change rate / parameter change rate.

[0030] Step 3: Calculate the adjustment amount of the first estimated parameter. The adjustment amount of the first estimated parameter = -Σ(sensitivity coefficient × index deviation) × damping coefficient; the initial value of the damping coefficient is 0.7, which decreases to 0.3 as the first estimated parameter is changed; the parameter adjustment shall not exceed the safe range of the equipment.

[0031] Step 4: Generate new predicted parameters. The second predicted parameter = the first predicted parameter + the adjustment amount of the first predicted parameter.

[0032] Furthermore, the edge-cloud collaboration includes the following steps:

[0033] Q1: Initialize the global model. The initial global model is generated in the cloud and distributed to each edge device.

[0034] Q2: Local training on edge devices. Edge devices use the "local training module" of the federated learning algorithm to clean and extract features from local data, and use the local data to iteratively optimize the initial global model.

[0035] Q3: Cloud-based secure aggregation model: Edge devices encrypt and upload local model parameters to the cloud. The "aggregation module" using the federated learning algorithm performs a weighted average of the parameters from multiple edge devices to generate a new global model.

[0036] Q4: New Global Model Distribution and Iteration: The cloud distributes the aggregated new global model to each edge device, repeating steps Q2-Q4 until the model achieves the required effect.

[0037] Furthermore, the edge device encrypts and uploads local model parameters to the cloud, including: the edge device determines the sensitivity of the data; for highly sensitive data, it uploads the locally trained model parameters or parameter update amount back to the cloud; for low-sensitivity data, it uploads the local training raw data, model parameters or parameter update amount back to the cloud.

[0038] Furthermore, the initial data includes: raw water flow rate, turbidity, suspended solids particle size distribution, coagulant dosage, sedimentation tank water turbidity, decanting level, inlet and outlet pressure values ​​of each filter, activated carbon adsorption rate, precision filtration accuracy, heavy metal content, organic matter concentration, pH value, residual chlorine concentration, disinfectant retention time, and microbial content.

[0039] Compared with the prior art, the beneficial effects of the present invention are: a water purification process based on edge intelligent control, through edge-cloud collaborative architecture design, edge devices process real-time data, and the cloud performs big data modeling and model iterative updates, thereby achieving continuous optimization of the overall system performance.

[0040] Meanwhile, by distributing encrypted edge intelligent devices at each stage of the water purification process, real-time analysis of water quality sensor data and dynamic adjustment of process parameters are achieved. The edge device independently completes decision-making, enabling rapid response to sudden and high-frequency rapid fluctuations in raw water quality, as well as in environments without or with weak networks, and reducing the risk of user data leakage. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the edge intelligent control logic of the present invention;

[0042] Figure 2 This is a schematic diagram of the edge intelligent control logic of the water purification process of the present invention;

[0043] Figure 3 This is a schematic diagram of the intelligent algorithm model of the present invention;

[0044] Figure 4 This is a schematic diagram of the first estimated parameter generation method of the present invention;

[0045] Figure 5 This is a schematic diagram of the first estimated parameter changing method of the present invention;

[0046] Figure 6 This is a schematic diagram of the water purification process of the present invention;

[0047] Figure 7 This is a schematic diagram of edge-cloud collaboration according to the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Example 1: In order to solve the technical problems mentioned in the background art, such as Figure 1 and Figure 2 As shown, the present invention provides a technical solution: a water purification process based on edge intelligent control, comprising:

[0050] Step 1: Initial Data Collection. Collect equipment data from each initial stage of the water purification process as the first set of equipment data, including:

[0051] During the pretreatment stage, the raw water flow rate is 0.5 m³ / h. 3 / h, turbidity 10 NTU and suspended solids particle size distribution 0.1 mm;

[0052] During the coagulation and sedimentation stage, the coagulant dosage is 5 mg / L, the turbidity of the sedimentation tank water is 1 NTU, and the decanting level is 1 m.

[0053] During the filtration stage, the pressure difference between the inlet and outlet of each filter is 0.1 bar, the activated carbon adsorption efficiency is 60%, and the precision filtration accuracy is 1 μm.

[0054] During the adsorption stage, the heavy metal content was 0 mg / L, the organic matter concentration was 0 mg / L, and the pH value was 6.5.

[0055] During the sterilization and disinfection stage, the residual chlorine concentration was 0.1 mg / L, the disinfectant retention time was 10 min, and the microbial content was 100 CFU / ml.

[0056] Step 2: Data analysis and command transmission. The first device data is sent to the edge computing center. The intelligent algorithm model analyzes the device data to obtain the analysis results. The edge computing center sends control commands to the corresponding devices in the water purification process based on the analysis results. The device data of each stage of the water purification process after the control commands are sent is collected as the second device data.

[0057] like Figure 3 As shown, a method for generating intelligent algorithm models is proposed:

[0058] Specifically, the process begins by acquiring data, including demand data, a first estimated parameter, and first equipment data, where the first equipment data includes a first real-time parameter. Next, data comparison is performed, comparing the first estimated parameter and the first real-time parameter to obtain first adjustment data. Then, data adjustment is implemented, adjusting the first real-time parameter based on the first adjustment data to acquire second equipment data, which includes a second real-time parameter and corresponding first effect data. Finally, feedback adjustment is performed, comparing the first effect data with the demand data. If the demand is met, the second real-time parameter is considered qualified and stored; otherwise, if the demand is not met, the second real-time parameter is considered unqualified, and the process of repeating the "data comparison-data adjustment-feedback adjustment" steps is repeated until the desired result is achieved, and the final real-time parameter is saved.

[0059] like Figure 4 As shown, a method for generating the first predicted parameters is proposed:

[0060] Specifically, the database is established, including: demand data for each stage of the water purification process, corresponding estimated parameters, and a parameter sensitivity matrix; demand matching retrieval, analyzing the water quality demand characteristics of the current process segment, and performing multi-level matching in the database, including: exact matching, finding completely identical demand records; fuzzy matching, retrieving historical demands with a key indicator similarity of 85-100%; extended matching, associating with common parameter templates for the same process stage; and a parameter fusion algorithm, weighting and fusing the retrieved historical parameters, where the first estimated parameter = Σ(parameter value × weight coefficient) / total weight, the total weight = timeliness factor × historical success rate, and the timeliness factor = a 当前时间-记录时间 a is 0.7-0.9. The current time refers to the real-time system time when calculating the time factor, and the record time refers to the time when the water quality data was last updated. Both the current time and the record time are in hours. For new demand processing, when the historical matching degree is 84-70%, a virtual process model is constructed, a parameter space scan is performed, the effect data that is closest to the demand is selected as the first estimated parameter, and the new parameter pair is stored in the database for verification.

[0061] like Figure 5 As shown, a method for changing the first predicted parameter is proposed:

[0062] Specifically, this includes: triggering conditions, where the deviation between the first effect data and the demand data is 5-20%; deviation calculation, calculating the deviation of the current process section's demand index; parameter sensitivity calculation, calling a preset sensitivity matrix, real-time perturbation verification, simulating the effect change rate for ±5% fluctuation of the first estimated parameter, sensitivity coefficient = effect change rate / parameter change rate; calculation of the first estimated parameter adjustment amount, the first estimated parameter adjustment amount = -Σ(sensitivity coefficient × index deviation) × damping coefficient, the initial value of the damping coefficient is 0.7, decreasing to 0.3 with the number of changes to the first estimated parameter, ensuring that the parameter adjustment does not exceed the equipment's safe range; generation of a new estimated parameter, the second estimated parameter = first estimated parameter + first estimated parameter adjustment amount;

[0063] like Figure 6 As shown, a water purification process flow is proposed:

[0064] Specifically, the process includes: Pretreatment stage: Raw water enters the tank through the inlet pipe, adjusting the flow rate and quality; Coarse filtration removes larger suspended solids, silt, leaves, and other impurities from the raw water; Washing removes organic colloids and sediments through chemical dosing, stirring, and sedimentation; Coagulation and sedimentation stage: Coagulants are added, such as iron and aluminum salts, to the water, reacting with suspended solids and colloidal substances to form larger aggregates; Sedimentation sends the coagulated water to a sedimentation tank, where suspended solids and colloidal substances settle to form coagulated sediment; Decanting further removes remaining suspended solids and colloidal substances from the supernatant in the sedimentation tank; Filtration stage: Sand filtration removes small amounts of suspended solids and colloidal substances from the supernatant through a sand filter; Activated carbon filtration further removes suspended solids and colloidal substances from the supernatant. Water filtered through a sand filter is further filtered, then through an activated carbon filter to remove organic matter, odors, and color. A final stage of fine filtration removes residual fine suspended solids and colloidal substances. The water then undergoes adsorption pretreatment, where the filtered water is fed into an adsorber pretreatment unit where chemicals are added and pH is adjusted to improve water quality. The pretreated water is then passed through the adsorber, which is filled with adsorbents such as activated carbon and resin to remove organic matter, heavy metals, and other harmful substances. The sterilization and disinfection stage involves adding disinfectants such as sodium hypochlorite and ozone to kill bacteria, viruses, and other microorganisms. The water is then kept in a disinfection tank for a specific time to ensure complete elimination of bacteria and other microorganisms. Finally, a deodorizing agent is added to the disinfected water to remove any residual disinfectant odor. The purified water is then discharged.

[0065] It should be noted that when the decanting level is 0-1.5m, the device sends an instruction to the sedimentation step to adjust the sedimentation tank flow rate to 0.8m / h; during the filtration stage, when the activated carbon adsorption rate is 50-70%, the peripheral equipment sends an instruction to the activated carbon filtration step to adjust and replace the activated carbon; during the sterilization and disinfection stage, when the residual chlorine concentration is 0-0.3mg / L or the microbial content is 100-50CFU / ml, the peripheral equipment sends an instruction to the disinfectant addition step to adjust the disinfectant dosage to increase to 0.1mg / L, and sends an instruction to the disinfectant holding step to adjust the disinfectant holding time to extend to 25min.

[0066] Step 3: Edge-cloud collaboration, where edge devices process local data and model training, while the cloud performs global model aggregation and distribution;

[0067] like Figure 7 As shown, an edge-cloud collaborative working method is proposed:

[0068] Specifically, an initial global model is generated in the cloud and distributed to each edge device. The edge devices use the "local training module" of the federated learning algorithm to clean and extract features from their local data, and use the local data to iteratively optimize the initial global model. The edge devices then encrypt and upload the local model parameters to the cloud. The cloud management platform uses the "aggregation module" of the federated learning algorithm to perform a weighted average of the parameters from multiple edge devices, generate a new global model, and distribute it to each edge device. The above steps are repeated until the model achieves the required effect.

[0069] Example 2: The difference from Example 1 is that the adjustments made in this example include:

[0070] Step 1: Initial data collection, adjusting all data to: raw water flow rate 3.5m³ / h 3 / h, turbidity 510 NTU and suspended solids particle size distribution 3.5 mm, coagulant dosage 30 mg / L, sedimentation tank water turbidity 25 NTU and decanting liquid level 2 m, pressure difference between inlet and outlet of each filter 0.25 bar, activated carbon adsorption efficiency 80% and precision filtration accuracy 4.5 μm, heavy metal content 0.25 mg / L, organic matter concentration 2.5 mg / L, pH value 7.5, residual chlorine concentration 0.5 mg / L, disinfectant retention time 20 min and microbial content 50 CFU / ml;

[0071] Step 2: Adjust data analysis and command transmission when the inflow rate is 10-3m³. 3When the turbidity is 1000-500 NTU, the edge device sends a command to the washing step to adjust the dosage to 80 ml / min and the stirring speed to 250 rpm; when the suspended solids particle size is 5-3 mm, the edge device sends a command to the primary screening and filtration step to adjust the screen mesh size to 2 mm; when the turbidity of the sedimentation tank water is 50-20 NTU, the edge device sends a command to the coagulant addition step to adjust the coagulant dosage to 5 mg / L and a command to the decanting step to adjust the decanting time to 20 min.

[0072] Example 3: The difference between this example and Examples 1 and 2 is that the adjustments made in this example include:

[0073] Step 1: Initial data collection, adjusting all data to: raw water flow rate 5m³ / h 3 / h, turbidity 1000 NTU and suspended solids particle size distribution 5 mm, coagulant dosage 50 mg / L, sedimentation tank water turbidity 50 NTU and decanting liquid level 3 m, pressure difference between inlet and outlet of each filter 0.5 bar, activated carbon adsorption efficiency 99% and precision filtration accuracy 10 μm, heavy metal content 0.5 mg / L, organic matter concentration 5 mg / L, pH value 8.5, residual chlorine concentration 1 mg / L, disinfectant retention time 30 min and microbial content 10 FU / ml;

[0074] Step 2: Adjust data analysis and command transmission when the inflow rate is 10-3m³. 3 When the turbidity is 1000-500 NTU, the edge device sends a command to the washing step to adjust the dosage to 100 ml / min and the stirring speed to 300 rpm; when the suspended solids particle size is 5-3 mm, the edge device sends a command to the primary screening and filtration step to adjust the screen mesh size to 3 mm; when the sedimentation tank water turbidity is 50-20 NTU, the edge device sends a command to the coagulant addition step to adjust the coagulant dosage to 10 mg / L and to the decanting step to adjust the decanting time to 30 min; when the pressure difference between the sand filter inlet and outlet is 0.5-0.3 bar, the edge device sends a command to the backwashing device to shorten the backwashing cycle to 8 hours and increase the backwashing intensity to 12 L / (m³). 2 The edge device sends instructions to the filtration equipment when the precision filtration accuracy is 10-5μm, and sends instructions to the filtration equipment to replace the filter element and adjust the filtration accuracy to 3μm; when the heavy metal content in the adsorption stage is 0.5-0.3mg / L, the edge device sends instructions to the adsorption equipment to adjust the pH value to 7.0 and to start the adsorbent regeneration; when the organic matter concentration is 5-3mg / L, the edge device sends instructions to the adsorption step to increase the activated carbon dosage by 20%.

[0075] In the event of sudden heavy rain, this invention compares with traditional intelligent control systems for water purification processes:

[0076] Table 1: Comparison of the Invention with Traditional Intelligent Control Systems for Water Purification Processes

[0077]

[0078]

[0079] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended embodiments and their equivalents.

Claims

1. A water purification process based on edge intelligent control, characterized in that: include: Initial data collection: Collect equipment data for each initial stage of the water purification process as the first equipment data; Data analysis: The data from the first device is sent to the edge computing center, where it is analyzed using intelligent algorithm models to obtain the analysis results; Command transmission: The edge computing center sends control commands to the corresponding equipment in the water purification process based on the analysis results; the equipment data of each stage of the water purification process after the control commands are sent are collected as the second equipment data; Edge-cloud collaboration: edge devices process local data and model training, while the cloud performs global model aggregation and distribution; The method for generating the intelligent algorithm model includes: S1: Acquire data, acquire demand data, first estimated parameters and first equipment data, the first equipment data including first real-time parameters; S2: Data comparison, compare the magnitudes of the first estimated parameter and the first real-time parameter to obtain the first adjustment data; S3: Data adjustment, adjust the first real-time parameter according to the first adjustment data, and obtain the second device data, the second device data including the second real-time parameter and the corresponding first effect data; S4: Feedback adjustment. First, compare the effect data with the demand data. If the demand is met, the second real-time parameter is qualified and stored. If the demand is not met, the second real-time parameter is unqualified. Return to steps S2-S4 until it is qualified and save the final real-time parameter. The method for generating the first estimated parameter includes the following steps: M1: Database establishment, the data includes: demand data for each stage of the water purification process, corresponding estimated parameters, and parameter sensitivity matrix; M2: The generation of the first predicted parameter includes the following steps: Step 1: Demand matching retrieval. Analyze the water quality demand characteristics of the current process stage and perform multi-level matching in the database, including: exact matching, finding completely identical demand records; fuzzy matching, retrieving historical demands with 85-100% similarity in key indicators; and extended matching, associating with common parameter templates of the same process stage. Step 2: Parameter fusion algorithm, which performs weighted fusion on the retrieved historical parameters: First estimated parameter = Σ(parameter value × weight coefficient) / total weight; Total weight = Timeliness factor × Historical success rate; Time factor = a 当前时间-记录时间 a is 0.7-0.9, the current time refers to the system real-time time when the time factor is calculated, and the recording time refers to the time when the water quality data was last updated. Both the current time and the recording time are in hours. Step 3: New requirement processing. When the historical matching degree is 84-70%, construct a virtual process model, perform parameter space scanning, select the effect data that is closest to the requirement as the first estimated parameter, and store the new parameter pair in the database for verification.

2. The water purification process based on edge intelligent control according to claim 1, characterized in that: The method for changing the first estimated parameter includes the following steps: N1: Triggering condition: The deviation between the initial effect data and the required data is 5-20%; N2: The method for changing the first predicted parameter includes the following steps: Step 1: Deviation Calculation, calculate the deviation of the current process section's required indicators; Step 2: Parameter sensitivity calculation. Call the preset sensitivity matrix, perform real-time perturbation verification, and simulate the effect change rate by fluctuating the first estimated parameter by ±5%. Sensitivity coefficient = effect change rate / parameter change rate. Step 3: Calculate the adjustment amount of the first estimated parameter. The adjustment amount of the first estimated parameter = -Σ(sensitivity coefficient × index deviation) × damping coefficient; the initial value of the damping coefficient is 0.7, which decreases to 0.3 as the first estimated parameter is changed; the parameter adjustment shall not exceed the safe range of the equipment. Step 4: Generate new predicted parameters. The second predicted parameter = the first predicted parameter + the adjustment amount of the first predicted parameter.

3. The water purification process based on edge intelligent control according to claim 1, characterized in that: The edge-cloud collaboration includes the following steps: Q1: Initialize the global model. The initial global model is generated in the cloud and distributed to each edge device. Q2: Local training on edge devices. Edge devices use the "local training module" of the federated learning algorithm to clean and extract features from local data, and use the local data to iteratively optimize the initial global model. Q3: Cloud-based secure aggregation model: Edge devices encrypt and upload local model parameters to the cloud. The "aggregation module" using the federated learning algorithm performs a weighted average of the parameters from multiple edge devices to generate a new global model. Q4: New Global Model Distribution and Iteration: The cloud distributes the aggregated new global model to each edge device, repeating steps Q2-Q4 until the model achieves the required effect.

4. The water purification process based on edge intelligent control according to claim 3, characterized in that: The edge device encrypts and uploads local model parameters to the cloud, including: the edge device determines the sensitivity of the data; for highly sensitive data, it uploads the locally trained model parameters or parameter updates back to the cloud; for low-sensitivity data, it uploads the original local training data, model parameters, or parameter updates back to the cloud.

5. The water purification process based on edge intelligent control according to claim 1, characterized in that: The initial data includes: raw water flow rate, turbidity, suspended solids particle size distribution, coagulant dosage, sedimentation tank water turbidity, decanting level, inlet and outlet pressure values ​​of each filter, activated carbon adsorption rate, precision filtration accuracy, heavy metal content, organic matter concentration, pH value, residual chlorine concentration, disinfectant retention time, and microbial content.

Images