Remote intelligent operation and maintenance method and system of MBR (Membrane Biological Reactor) distributed equipment adaptive to intelligent water affair

Abstract

The invention discloses an MBR distributed equipment remote intelligent operation and maintenance method and system adaptive to intelligent water affairs, and relates to the technical field of MBR distributed equipment remote operation and maintenance, and the method comprises the steps: collecting multi-dimensional operation parameters of distributed MBR stations, and generating a standardized time sequence state vector; identifying the current operation mode based on the standardized time sequence state vector, dynamically adjusting the internal weight of the membrane pollution prediction model by using an identification result, and outputting a membrane pollution index prediction value; constructing reinforcement learning state input by combining the membrane pollution index prediction value, the real-time electricity price, the agent cost and the membrane module operation duration, and generating a cleaning action decision containing the cleaning type, the cleaning strength and the cleaning opportunity; cleaning operation is executed according to the cleaning action decision, the flux recovery rate, energy consumption and medicament consumption after cleaning are collected, and a cleaning feedback performance index is formed; and the cleaning feedback performance index and the standardized time sequence state vector before cleaning are used for generating a soft label and a local statistical feature, and the soft label and the local statistical feature are uploaded to a cloud after being encrypted.

Description

Technical Field

[0001] This invention relates to the field of remote operation and maintenance technology for distributed MBR equipment, and in particular to a remote intelligent operation and maintenance method and system for distributed MBR equipment adapted to smart water management. Background Technology

[0002] Remote operation and maintenance technology for distributed MBR equipment refers to a comprehensive technological system that enables remote monitoring, fault early warning, intelligent diagnosis, and autonomous optimization control of miniaturized or skid-mounted wastewater treatment equipment from membrane bioreactors widely distributed in towns, rural areas, and industrial parks. This is achieved through the collaboration of the Internet of Things (IoT), edge computing, artificial intelligence (AI), and cloud platforms. Its core objective is to ensure the long-term stable, efficient, and low-cost operation of distributed MBR systems under unattended or minimally staffed conditions. Therefore, improving the intelligence and security of remote operation and maintenance of distributed MBR equipment using advanced technologies is one of the most pressing issues to be addressed.

[0003] Current MBR operation and maintenance mostly rely on fixed thresholds or static models for pollution judgment, which cannot adapt to dynamic changes in operating conditions such as different water quality, seasons, and loads. This leads to high false alarm rates and frequent false alarms in decentralized scenarios, resulting in excessive cleaning that wastes resources or delayed cleaning that causes membrane blockage. Furthermore, each MBR site operates independently, with no data sharing, resulting in few local model training samples and difficulties in updating. Uploading raw data to the cloud also faces problems such as limited communication bandwidth, data privacy leaks, and high compliance risks, making it difficult to deploy and iterate intelligent algorithms on a large scale. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a remote intelligent operation and maintenance method for distributed MBR equipment adapted to smart water management. This addresses the shortcomings of existing MBR operation and maintenance methods, which often rely on fixed thresholds or static models for pollution assessment. These methods are unable to adapt to dynamic changes in water quality, season, and load, leading to high false alarm rates and frequent missed alarms in distributed scenarios. This results in excessive cleaning, wasting resources, or delayed cleaning causing membrane blockage. Furthermore, each MBR site operates independently, with no data sharing, resulting in limited local model training samples and difficulties in updating. Uploading raw data to the cloud also presents challenges such as limited communication bandwidth, data privacy leaks, and high compliance risks, hindering the large-scale deployment and iteration of intelligent algorithms.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a remote intelligent operation and maintenance method for distributed MBR equipment adapted to smart water management, comprising:

[0008] Collect multi-dimensional operating parameters of distributed MBR sites and generate standardized time-series state vectors;

[0009] The current operating mode is identified based on the standardized time-series state vector, and the internal weights of the membrane fouling prediction model are dynamically adjusted using the identification results to output the predicted value of the membrane fouling index.

[0010] By combining the predicted value of membrane fouling index, real-time electricity price, reagent cost and membrane module runtime, a reinforcement learning state input is constructed to generate cleaning action decisions that include cleaning type, intensity and timing;

[0011] Based on the cleaning action decision, the cleaning operation is executed, and the flux recovery rate, energy consumption and reagent consumption after cleaning are collected to form cleaning feedback performance indicators;

[0012] The cleaning feedback performance indicators and the standardized time-series state vector before cleaning are used to generate soft tags and local statistical features, which are then encrypted and uploaded to the cloud.

[0013] The teacher model is updated in the cloud based on the soft tags and local statistical features, and the updated model parameters are sent to the edge side through knowledge distillation to complete the model iteration.

[0014] As a preferred embodiment of the remote intelligent operation and maintenance method for distributed MBR equipment adapted to smart water management as described in this invention, the specific steps for collecting multi-dimensional operating parameters of distributed MBR sites and generating standardized time-series state vectors are as follows:

[0015] At each MBR site, six types of sensor data were simultaneously acquired, including influent flow rate, mixed liquor suspended solids concentration, transmembrane pressure difference, dissolved oxygen, water temperature, and membrane flux-related pressure, to form the original observation sequence.

[0016] The original observation sequence is timestamped and linearly interpolated to eliminate data loss caused by communication delays or sensor failures.

[0017] The completed data for each dimension are mapped to a unified numerical range to form a normalized observation vector.

[0018] Normalized observation vectors at several consecutive time points are extracted using a fixed-length time window, input into a one-dimensional convolutional encoder, spatiotemporal coupling features are extracted, and a standardized temporal state vector with fixed dimensions is output. This serves as a unified input representation for subsequent operating condition identification and pollution prediction.

[0019] As a preferred embodiment of the remote intelligent operation and maintenance method for distributed MBR equipment adapted to smart water management as described in this invention, the specific steps of identifying the current operating mode based on standardized time-series state vectors and dynamically adjusting the internal weights of the membrane fouling prediction model using the identification results to output the membrane fouling index prediction value are as follows:

[0020] Standardize the time-series state vector Input a pre-trained operation pattern clustering model, which is trained offline based on historical operation data of the entire network to obtain several typical operation condition centers, and determine the current operation condition category through nearest neighbor matching;

[0021] The identified work condition categories are converted into embedding vectors. And input it into the membrane fouling prediction model along with the historical state sequence;

[0022] The membrane fouling prediction model employs a gated time-series network structure, and its gating coefficients are... Calculated by the following formula:

[0023] ;

[0024] in, This represents the Sigmoid activation function. For learnable weight matrix, For bias terms, This represents a vector concatenation operation;

[0025] Gating coefficient This is used to dynamically scale the contribution of historical data to the current prediction, outputting a predicted value for the membrane fouling index. It reflects the relative trend of the membrane flux decay rate per unit time.

[0026] As a preferred embodiment of the remote intelligent operation and maintenance method for distributed MBR equipment adapted to smart water management as described in this invention, the step of constructing a reinforcement learning state input by combining the predicted membrane fouling index, real-time electricity price, reagent cost, and membrane module runtime to generate cleaning action decisions including cleaning type, intensity, and timing is as follows:

[0027] Obtain the predicted value of membrane fouling index Time-of-use electricity prices published by the power grid Unit drug cost Cumulative runtime of membrane modules These are combined to form a reinforcement learning state vector;

[0028] The state vector is input into a lightweight policy network deployed on an edge device, which outputs the probability distribution of each cleaning action.

[0029] The cleaning action space includes four types of operations: no cleaning, backwashing, low-concentration chemical cleaning, and high-concentration chemical cleaning. Each type of operation is associated with an intensity level and an execution delay time.

[0030] The policy network is trained by maximizing the long-term cumulative reward, and the reward function is... The definition is as follows:

[0031] ;

[0032] in, For flux recovery rate, Energy consumption for cleaning For the cost of the medicine, , , These are preset weighting coefficients used to balance process performance and economic objectives;

[0033] The final cleaning action decision is determined based on the output sampling of the strategy network and then sent to the execution agency.

[0034] As a preferred embodiment of the remote intelligent operation and maintenance method for distributed MBR equipment adapted to smart water management as described in this invention, the specific steps of executing the cleaning operation based on the cleaning action decision and collecting the flux recovery rate, energy consumption, and chemical consumption after cleaning to form cleaning feedback performance indicators are as follows:

[0035] The control system drives the pumps, valves, and dosing devices to perform the cleaning actions as specified by the decision-making process.

[0036] After cleaning, continuously monitor the membrane flux stability value and calculate the flux recovery rate. :

[0037] ;

[0038] in, The steady-state membrane flux before cleaning, This represents the steady-state membrane flux after cleaning.

[0039] The changes in the electricity meter readings during the cleaning process are recorded synchronously to obtain the actual energy consumption;

[0040] Record the operating volume of the dosing metering pump, calculate the amount of chemical consumed based on the chemical concentration, and multiply by the unit cost to obtain the chemical cost;

[0041] Flux recovery rate The actual energy consumption and reagent cost are combined to form a set of cleaning feedback performance indicators, which are used for subsequent model updates and strategy evaluation.

[0042] As a preferred embodiment of the remote intelligent operation and maintenance method for MBR distributed equipment adapted to smart water management described in this invention, the specific steps of using cleaning feedback performance indicators and standardized time-series state vectors before cleaning to generate soft tags and local statistical features, and then uploading them to the cloud after encryption, are as follows:

[0043] The standardized temporal state vector before cleaning Input the cloud-based teacher model and obtain the output probability distribution as a soft label. The calculation method is as follows:

[0044] ;

[0045] in, This represents the forward inference function of the cloud-based teacher model. >1 represents the distillation temperature, used to smooth the output distribution;

[0046] The gradient of the local student model loss with respect to the parameters is calculated on the edge side, and the first and second moments of the gradient are extracted as local statistical features.

[0047] soft label The first and second moments are asymmetrically encrypted and uploaded to the cloud aggregation server via a secure transmission protocol.

[0048] As a preferred embodiment of the remote intelligent operation and maintenance method for distributed MBR equipment adapted to smart water management as described in this invention, the steps of updating the teacher model in the cloud based on soft tags and local statistical features, and then distributing the updated model parameters to the edge side through knowledge distillation to complete the model iteration are as follows:

[0049] The cloud aggregation server receives encrypted soft tags and local statistical features uploaded from multiple sites, decrypts them, performs a weighted average, and updates the teacher model parameters.

[0050] Soft tags were regenerated based on the historical status of each site using the updated teacher model. This is used as a supervisory signal to train the edge-side student model through knowledge distillation.

[0051] Wherein, the total distillation loss function Defined as:

[0052] ;

[0053] in, For marginal student models, For mission-related losses, ∈(0,1) controls the distillation intensity;

[0054] After training is completed, the compressed student model parameters are sent to the corresponding MBR site after digital signature and integrity verification, replacing the original model and completing a closed-loop iteration of cloud-edge collaboration.

[0055] Secondly, this invention provides a remote intelligent operation and maintenance system for distributed MBR equipment adapted to smart water management, comprising:

[0056] Multi-source perception fusion module, working condition adaptive prediction module, intelligent cleaning decision module, cleaning execution feedback module, edge security upload module, and cloud-edge collaborative evolution module;

[0057] The multi-source sensing fusion module is used to collect multi-dimensional operating parameters of each MBR site and generate a standardized time-series state vector.

[0058] The adaptive prediction module is used to identify the operating mode based on the standardized time-series state vector, dynamically adjust the weights of the membrane fouling prediction model, and output the predicted value of the membrane fouling index.

[0059] The intelligent cleaning decision module is used to combine the predicted value of membrane fouling index, electricity price, chemical cost and membrane module running time to generate cleaning action decisions that include cleaning type, intensity and timing.

[0060] The cleaning execution feedback module is used to perform cleaning operations and collect the flux recovery rate, energy consumption and reagent consumption after cleaning to form cleaning feedback performance indicators.

[0061] The edge security upload module is used to convert the standardized temporal state vector before cleaning and the cleaning feedback performance index into soft tags and local statistical features, and then encrypt and upload them to the cloud.

[0062] The cloud-edge collaborative evolution module is used to update the teacher model in the cloud and distribute the optimized model parameters to the edge side through knowledge distillation to complete the closed-loop iteration.

[0063] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the remote intelligent operation and maintenance method for MBR distributed equipment adapted to smart water management as described in the first aspect of the present invention.

[0064] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the remote intelligent operation and maintenance method for MBR distributed equipment adapted to smart water management as described in the first aspect of the present invention.

[0065] The beneficial effects of this invention are as follows: By introducing a condition-adaptive gated time-series prediction mechanism, the membrane fouling model can dynamically adjust the weights of historical states according to the real-time operating mode, effectively improving the prediction accuracy under complex scenarios such as water quality fluctuations and seasonal changes. Combined with multi-objective reinforcement learning to construct a cleaning decision framework, flux recovery, energy consumption, and reagent costs are uniformly incorporated into the optimization objectives, achieving a leap from experience-based cleaning to economic-process synergistic optimal cleaning. Through encrypted upload based on soft tags and gradient statistical features and a cloud-edge knowledge distillation mechanism, the model can continuously evolve without transmitting the original data, ensuring data security and supporting the efficient iteration of intelligent operation and maintenance systems for large-scale distributed equipment. This significantly reduces the total lifecycle operation and maintenance cost while improving the reliability of the MBR system. Attached Figure Description

[0066] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 A flowchart for a remote intelligent operation and maintenance method for distributed MBR equipment adapted to smart water management.

[0068] Figure 2 A schematic diagram of a remote intelligent operation and maintenance system for distributed MBR equipment adapted to smart water management. Detailed Implementation

[0069] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0070] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0071] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0072] Reference Figures 1-2As one embodiment of the present invention, this embodiment provides a remote intelligent operation and maintenance method for distributed MBR equipment adapted to smart water management, including the following steps:

[0073] S1. Collect multi-dimensional operating parameters of distributed MBR sites and generate standardized time-series state vectors.

[0074] Furthermore, at each MBR site, six types of sensor data are simultaneously acquired, including influent flow rate, mixed liquor suspended solids concentration, transmembrane pressure difference, dissolved oxygen, water temperature, and membrane flux-related pressure, to form the original observation sequence.

[0075] The original observation sequence is timestamped and linearly interpolated to eliminate data loss caused by communication delays or sensor failures.

[0076] The completed data for each dimension are mapped to a unified numerical range to form a normalized observation vector.

[0077] Normalized observation vectors at several consecutive time points are extracted using a fixed-length time window, input into a one-dimensional convolutional encoder, spatiotemporal coupling features are extracted, and a standardized temporal state vector with fixed dimensions is output. This serves as a unified input representation for subsequent operating condition identification and pollution prediction.

[0078] It should be noted that the synchronous acquisition and standardized processing of the multi-source sensor data mentioned above solves the data fragmentation problem caused by heterogeneous equipment and unstable communication in distributed MBR sites. By extracting spatiotemporal coupling features through a one-dimensional convolutional encoder, not only is key dynamic information preserved, but the input dimension is also compressed, providing a unified, compact and semantically rich state representation foundation for the efficient operation of the subsequent lightweight model at the edge.

[0079] S2. Based on the standardized time-series state vector, identify the current operating mode and use the identification results to dynamically adjust the internal weights of the membrane fouling prediction model, and output the predicted value of the membrane fouling index.

[0080] Furthermore, the standardized temporal state vector Input a pre-trained operation pattern clustering model. The operation pattern clustering model is trained offline based on historical operation data of the entire network to obtain several typical operation condition centers. The current operation condition category is determined by nearest neighbor matching.

[0081] The identified work condition categories are converted into embedding vectors. And input it into the membrane fouling prediction model along with the historical state sequence;

[0082] The membrane fouling prediction model employs a gated time-series network structure, and its gating coefficients are... Calculated by the following formula:

[0083] ;

[0084] in, This represents the Sigmoid activation function. For learnable weight matrix, For bias terms, This represents a vector concatenation operation;

[0085] Gating coefficient This is used to dynamically scale the contribution of historical data to the current prediction, outputting a predicted value for the membrane fouling index. It reflects the relative trend of the membrane flux decay rate per unit time.

[0086] It should be noted that the operation mode clustering model is based on offline construction of typical operating condition centers using historical data from the entire network, enabling individual sites to identify operating conditions without requiring a large amount of local labeled data; the gated time series network introduces operating condition embedding vectors to dynamically adjust the weights of historical states, breaking through the limitation of traditional static prediction models in terms of weak generalization ability under scenarios of sudden load changes or seasonal transitions, and effectively improving the adaptability and robustness of membrane fouling trend prediction.

[0087] S3. Combine the predicted value of membrane fouling index, real-time electricity price, reagent cost and membrane module running time to construct reinforcement learning state input, and generate cleaning action decisions including cleaning type, intensity and timing.

[0088] Furthermore, obtain the predicted value of the membrane fouling index. Time-of-use electricity prices published by the power grid Unit drug cost Cumulative runtime of membrane modules These are combined to form a reinforcement learning state vector;

[0089] The state vector is input into a lightweight policy network deployed on an edge device, which outputs the probability distribution of each cleaning action.

[0090] The cleaning action space includes four types of operations: no cleaning, backwashing, low-concentration chemical cleaning, and high-concentration chemical cleaning. Each type of operation is associated with an intensity level and an execution delay time.

[0091] The policy network is trained by maximizing the long-term cumulative reward, and the reward function is... The definition is as follows:

[0092] ;

[0093] in, For flux recovery rate, Energy consumption for cleaning For the cost of the medicine, , , These are preset weighting coefficients used to balance process performance and economic objectives;

[0094] The final cleaning action decision is determined based on the output sampling of the strategy network and then sent to the execution agency.

[0095] It should be noted that by integrating membrane fouling predictions with external economic variables (such as time-of-use electricity prices and reagent costs) and equipment aging indicators (running time) to construct a reinforcement learning state space, cleaning decisions not only respond to process requirements but also proactively respond to operational cost constraints. The lightweight strategy network is deployed at the edge to ensure the generation of multi-objective optimization actions that take into account flux recovery, energy saving, and membrane life protection under low latency conditions, thereby realizing a paradigm shift from passive response to proactive collaborative optimization in operation and maintenance.

[0096] S4. Execute cleaning operations based on the cleaning action decision, and collect the flux recovery rate, energy consumption and reagent consumption after cleaning to form cleaning feedback performance indicators.

[0097] Furthermore, the control system drives the pumps, valves, and dosing devices to perform the cleaning actions as specified by the decision.

[0098] After cleaning, continuously monitor the membrane flux stability value and calculate the flux recovery rate. :

[0099] ;

[0100] in, The steady-state membrane flux before cleaning, This represents the steady-state membrane flux after cleaning.

[0101] The changes in the electricity meter readings during the cleaning process are recorded synchronously to obtain the actual energy consumption;

[0102] Record the operating volume of the dosing metering pump, calculate the amount of chemical consumed based on the chemical concentration, and multiply by the unit cost to obtain the chemical cost;

[0103] Flux recovery rate The actual energy consumption and reagent cost are combined to form a set of cleaning feedback performance indicators, which are used for subsequent model updates and strategy evaluation.

[0104] It should be noted that by quantifying the flux recovery rate after cleaning, actual energy consumption, and reagent costs, a measurable and comparable set of feedback performance indicators is constructed, providing a real closed-loop signal for strategy evaluation and model training. This feedback mechanism avoids relying on theoretical estimation or human experience judgment, enabling the system to have the foundation for continuously calibrating decision-making capabilities based on empirical data, and ensuring that the intelligent operation and maintenance effect is verifiable and traceable.

[0105] S5. Use the cleaning feedback performance indicators and the standardized time-series state vector before cleaning to generate soft tags and local statistical features, and then upload them to the cloud after encryption.

[0106] Furthermore, the standardized temporal state vector before cleaning... Input the cloud-based teacher model and obtain the output probability distribution as a soft label. The calculation method is as follows:

[0107] ;

[0108] in, This represents the forward inference function of the cloud-based teacher model. >1 represents the distillation temperature, used to smooth the output distribution;

[0109] The gradient of the local student model loss with respect to the parameters is calculated on the edge side, and the first and second moments of the gradient are extracted as local statistical features.

[0110] soft label The first and second moments are asymmetrically encrypted and uploaded to the cloud aggregation server via a secure transmission protocol.

[0111] It should be noted that using soft labels and gradient statistical features to replace the original operation data upload not only preserves the supervision information required for model training, but also effectively avoids the risk of leakage of sensitive operating condition data. The combination of asymmetric encryption and secure transmission protocol meets the strict requirements of smart water management platforms for data privacy and compliance, and provides a feasible technical path for large-scale cross-regional MBR equipment access.

[0112] S6. Update the teacher model in the cloud based on soft labels and local statistical features, and distribute the updated model parameters to the edge side through knowledge distillation to complete the model iteration.

[0113] Furthermore, the cloud aggregation server receives encrypted soft tags and local statistical features uploaded from multiple sites, decrypts them, performs a weighted average, and updates the teacher model parameters.

[0114] Soft tags were regenerated based on the historical status of each site using the updated teacher model. This is used as a supervisory signal to train the edge-side student model through knowledge distillation.

[0115] Wherein, the total distillation loss function Defined as:

[0116] ;

[0117] in, For marginal student models, For mission-related losses, ∈(0,1) controls the distillation intensity;

[0118] After training is completed, the compressed student model parameters are sent to the corresponding MBR site after digital signature and integrity verification, replacing the original model and completing a closed-loop iteration of cloud-edge collaboration.

[0119] It should be noted that the cloud-based teacher model achieves global experience accumulation by aggregating knowledge from multiple sites, while the edge student model approximates teacher performance under limited parameter conditions through knowledge distillation. This cloud-edge collaboration mechanism supports the model to complete incremental updates without interrupting on-site operation, solving the defect of traditional edge AI system deployment being fixed and ensuring that the intelligent operation and maintenance capabilities of distributed MBR equipment continue to evolve over time.

[0120] This embodiment also provides a remote intelligent operation and maintenance system for distributed MBR equipment adapted to smart water management, including:

[0121] Multi-source perception fusion module, working condition adaptive prediction module, intelligent cleaning decision module, cleaning execution feedback module, edge security upload module, and cloud-edge collaborative evolution module;

[0122] The multi-source sensing fusion module is used to collect multi-dimensional operating parameters of each MBR site and generate standardized time-series state vectors;

[0123] The adaptive prediction module is used to identify the operating mode based on the standardized time-series state vector, dynamically adjust the weights of the membrane fouling prediction model, and output the predicted value of the membrane fouling index.

[0124] The intelligent cleaning decision module is used to combine the membrane fouling index prediction, electricity price, chemical cost and membrane module running time to generate cleaning action decisions that include cleaning type, intensity and timing.

[0125] The cleaning execution feedback module is used to perform cleaning operations and collect data on flux recovery rate, energy consumption, and reagent consumption after cleaning, forming cleaning feedback performance indicators.

[0126] The edge security upload module is used to convert the standardized time-series state vector and cleaning feedback performance indicators before cleaning into soft tags and local statistical features, and then upload them to the cloud after encryption.

[0127] The cloud-edge co-evolution module is used to update the teacher model in the cloud and distribute the optimized model parameters to the edge side through knowledge distillation to complete the closed-loop iteration.

[0128] This embodiment also provides a computer device suitable for the remote intelligent operation and maintenance method of MBR distributed equipment adapted to smart water management, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the remote intelligent operation and maintenance method of MBR distributed equipment adapted to smart water management as proposed in the above embodiment.

[0129] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0130] This embodiment also provides a storage medium on which a computer program is stored. When executed by a processor, the program implements the remote intelligent operation and maintenance method for distributed MBR equipment adapted to smart water management, as proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0131] In summary, this invention introduces a condition-adaptive gating time-series prediction mechanism, enabling the membrane fouling model to dynamically adjust historical state weights based on real-time operating modes. This effectively improves prediction accuracy in complex scenarios such as water quality fluctuations and seasonal changes. By combining multi-objective reinforcement learning to construct a cleaning decision framework, flux recovery, energy consumption, and reagent costs are uniformly incorporated into the optimization objectives, achieving a leap from experience-based cleaning to economically and process-optimally coordinated cleaning. Through encrypted upload based on soft tags and gradient statistical features, as well as a cloud-edge knowledge distillation mechanism, the model can continuously evolve without transmitting original data. This ensures data security and supports the efficient iteration of intelligent operation and maintenance systems for large-scale distributed equipment, thereby significantly reducing the total lifecycle operation and maintenance costs while improving the reliability of the MBR system.

[0132] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for remote intelligent operation and maintenance of MBR decentralized equipment adapted to smart water affairs, characterized in that: The method comprises the following steps: Collecting multi-dimensional operating parameters of the dispersedly deployed MBR sites and generating a standardized time-series state vector; Identifying a current operating mode based on the standardized time-series state vector, dynamically adjusting internal weights of a membrane pollution prediction model using the identification result, and outputting a membrane pollution index prediction value; Combining the membrane pollution index prediction value, real-time electricity price, chemical cost, and membrane module operating time to construct a reinforcement learning state input, and generating a cleaning action decision including cleaning type, intensity, and timing; Executing the cleaning operation according to the cleaning action decision, collecting the flux recovery rate, energy consumption, and chemical consumption after cleaning, and forming a cleaning feedback performance index; Using the cleaning feedback performance index and the standardized time-series state vector before cleaning to generate soft labels and local statistical features, and uploading the encrypted soft labels and local statistical features to the cloud; Updating a teacher model based on the soft labels and local statistical features in the cloud, and distributing the updated model parameters to the edge side through knowledge distillation to complete model iteration.

2. The method of claim 1, wherein the method is characterized by: The method of collecting multi-dimensional operating parameters of the dispersedly deployed MBR sites and generating a standardized time-series state vector comprises the following steps: Synchronously acquiring six types of sensing data, including influent flow, mixed liquor suspended solids concentration, transmembrane pressure difference, dissolved oxygen, water temperature, and membrane flux related pressure, at each MBR site to form an original observation sequence; Performing time stamp alignment and linear interpolation completion on the original observation sequence to eliminate data missing caused by communication delay or sensor failure; Mapping each dimension of the completed data to a unified numerical interval to form a normalized observation vector; The normalized observation vectors of continuous time points are intercepted with a fixed length time window, input into a one-dimensional convolutional encoder, space-time coupling features are extracted, and a standardized time sequence state vector with fixed output dimension is output as a unified input representation for subsequent working condition recognition and pollution prediction.

3. The method of claim 2, wherein the method is characterized by: The method of identifying a current operating mode based on the standardized time-series state vector, dynamically adjusting internal weights of a membrane pollution prediction model using the identification result, and outputting a membrane pollution index prediction value comprises the following steps: standardized time-series state vector input a pre-trained operation mode clustering model, which is trained offline based on the historical operation data of the entire network to obtain a plurality of typical working condition centers, and determines the current working condition category through nearest neighbor matching; Converting the recognized operating condition class to an embedding vector and inputting the embedding vector into the membrane pollution prediction model together with the historical state sequence. The membrane fouling prediction model adopts a gated time series network structure, and a gating coefficient thereof is calculated by the following formula: is calculated by the following formula: ； wherein, represents a Sigmoid activation function, is a learnable weight matrix, is a bias term, represents a vector concatenation operation; Gating coefficients For dynamically scaling the contribution of the historical state to the current prediction, outputting a membrane fouling index prediction value Reflecting the relative trend of the membrane flux decay rate per unit time.

4. The method of claim 3, wherein the method is characterized by: The method of combining the membrane pollution index prediction value, real-time electricity price, chemical cost, and membrane module operating time to construct a reinforcement learning state input, and generating a cleaning action decision including cleaning type, intensity, and timing comprises the following steps: Obtaining a membrane fouling index prediction value , a time-of-use electricity price published by a power grid , a unit medicament cost , a cumulative operation time length of the membrane assembly , the combination forms a reinforcement learning state vector Inputting the state vector into a lightweight policy network deployed on an edge device, which outputs a probability distribution of each cleaning action; The cleaning action space includes four types of operations: no cleaning, backwashing, low-concentration chemical cleaning, and high-concentration chemical cleaning, each type of operation being associated with an intensity level and an execution delay time; The policy network is trained by maximizing the long-term cumulative reward, the reward function is defined as follows: ； wherein, is the flux recovery rate, is the cleaning energy consumption, is the chemical cost, , , is a preset weight coefficient for balancing the process performance and the economic target; Determining a final cleaning action decision according to the policy network output sampling and distributing it to the execution mechanism.

5. The method of claim 4, wherein the method is characterized by: The method of executing the cleaning operation according to the cleaning action decision, collecting the flux recovery rate, energy consumption, and chemical consumption after cleaning, and forming a cleaning feedback performance index comprises the following steps: The control system drives the pump, valve, and chemical feeding device to execute the operation specified in the cleaning action decision; After cleaning, the steady value of membrane flux is continuously monitored, and the flux recovery rate is calculated : ； wherein, J0is the steady state membrane flux before cleaning, J0is the steady state membrane flux after cleaning; Synchronously recording the change of the electric energy meter reading during the cleaning process to obtain the actual energy consumption; Recording the operating volume of the chemical dosing pump, combining the chemical concentration to calculate the chemical consumption, and multiplying the unit cost to obtain the chemical cost; The flux recovery rate The actual energy consumption, the medicine cost and the cleaning feedback performance index set are combined to be a cleaning feedback performance index set, which is used for subsequent model updating and strategy evaluation.

6. The method of claim 5, wherein the method is characterized by: The method of using the cleaning feedback performance index and the standardized time-series state vector before cleaning to generate soft labels and local statistical features, and uploading the encrypted soft labels and local statistical features to the cloud comprises the following steps: The normalized timing state vector before cleaning Input the cloud teacher model to obtain an output probability distribution as a soft label The calculation method is: ； wherein, represents a forward inference function of the cloud teacher model, T1 is a distillation temperature for smoothing the output distribution; Calculating the gradient of the local student model loss with respect to the parameters on the edge side, and extracting the first and second moments of the gradient as local statistical features; Soft labels The first and second moments are asymmetrically encrypted and uploaded to a cloud aggregation server via a secure transmission protocol.

7. The method of claim 6, wherein the method is characterized by: The teacher model is updated based on the soft label and local statistical characteristics in the cloud, and the updated model parameters are distributed to the edge side through knowledge distillation to complete model iteration. The specific steps are as follows: The cloud aggregation server receives the encrypted soft label and local statistical characteristics uploaded by multiple sites, performs weighted averaging after decryption, and updates the teacher model parameters; The historical states of each site are re-generated with the updated teacher model to obtain soft labels The soft labels are used as supervision signals to train the edge-side student model through knowledge distillation. wherein the distillation total loss function is defined as: ; wherein, is the edge student model, is the task-related loss, ∈(0,1) is a control distillation strength; After training, the compressed student model parameters are signed and verified for integrity, and then distributed to the corresponding MBR site to replace the original model, completing a closed-loop iteration of cloud-edge collaboration.

8. The MBR decentralized equipment remote intelligent operation and maintenance system adapted to smart water, based on the MBR decentralized equipment remote intelligent operation and maintenance method adapted to smart water according to any one of claims 1-7, characterized in that: It includes: Multi-source perception fusion module, working condition adaptive prediction module, intelligent cleaning decision module, cleaning execution feedback module, edge security upload module and cloud-edge collaborative evolution module; The multi-source perception fusion module is used for collecting multi-dimensional running parameters of each MBR site to generate a standardized time series state vector; The working condition adaptive prediction module is used for identifying the running mode based on the standardized time series state vector and dynamically adjusting the membrane pollution prediction model weight to output the membrane pollution index prediction value; The intelligent cleaning decision module is used to generate a cleaning action decision including cleaning type, intensity and time sequence by combining the membrane pollution index prediction value, electricity price, chemical cost and membrane module running time; The cleaning execution feedback module is used to execute cleaning operation and collect flux recovery rate, energy consumption and chemical consumption after cleaning to form cleaning feedback performance index; The edge security upload module is used to convert the standardized time series state vector before cleaning and the cleaning feedback performance index into soft label and local statistical characteristics, and upload them to the cloud after encryption; The cloud-edge collaborative evolution module is used for updating the teacher model in the cloud, and distributing the optimized model parameters to the edge side through knowledge distillation to complete the closed-loop iteration.

9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that: The processor executes the computer program to realize the steps of the MBR distributed equipment remote intelligent operation and maintenance method adapted to smart water in claim 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the MBR distributed equipment remote intelligent operation and maintenance method adapted to smart water in claim 1-7.

Images