An autonomous optimization control method and system for an intelligent sludge treatment device

Abstract

The present invention provides an autonomous optimization control method and system for an intelligent sludge treatment device. An intelligent agent system based on a neural network is deployed in the sludge treatment unit, which receives the sludge moisture content and concentration as inputs and outputs the stirring speed and pressure control values. A star network system supporting parallel learning of multiple intelligent agents is established, including a central control node and multiple sludge treatment unit nodes. A distributed algorithm is used to achieve collaborative learning and knowledge transfer between the neural networks of multiple treatment units. The present invention solves the technical problem that the control strategy of traditional sludge treatment devices is fixed and cannot be adaptively adjusted according to different working conditions. Through the multi-agent collaborative learning and knowledge transfer mechanism, the autonomous optimization of the control strategy of the sludge treatment unit is realized, the processing efficiency of the device is improved, the energy consumption is reduced, and at the same time, the collaborative operation effect between multiple treatment units is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of sludge treatment equipment control, and particularly to an autonomous optimization control method and system for intelligent sludge treatment equipment. Background Art

[0002] With the rapid development of industry and urbanization, sludge treatment has become an important topic in the field of environmental protection. The quantity of sludge generated during the treatment of industrial wastewater and urban sewage continues to increase, and the requirements for sludge treatment equipment are also constantly rising. Existing sludge treatment equipment mainly realizes the reduction treatment of sludge by adjusting parameters such as stirring speed and pressure. During the treatment process, the stirring speed directly affects the shear force of the sludge and the effect of floc breakup, while the pressure parameter determines the degree of sludge dewatering. The regulation of these two key parameters has a decisive impact on the sludge treatment effect. Currently, most sludge treatment equipment adopts fixed control strategies or simple PID control methods, and conducts control according to preset parameters, with relatively simple control logic.

[0003] However, the existing technology has the following significant deficiencies in practical applications:

[0004] The properties of sludge are complex and variable, and the sludge generated from different sources and different processes vary greatly in characteristics such as moisture content, organic matter content, and particle size distribution. Traditional fixed control strategies are difficult to adapt to these differences, often using unified control parameters and unable to precisely regulate according to different sludge characteristics, resulting in low treatment efficiency;

[0005] During the sludge treatment process, key parameters such as the moisture content and concentration of the sludge will change dynamically over time. A single control mode cannot adaptively adjust the control strategy according to the real-time changes of these parameters, easily causing problems such as insufficient treatment or over-treatment. Especially when the treatment working conditions change suddenly, the adaptability of the existing control methods is significantly insufficient;

[0006] Large sewage treatment plants usually have multiple sludge treatment units operating in parallel, but there is a lack of an effective coordination mechanism between these units. Each treatment unit operates independently, unable to achieve experience sharing and the migration of optimization strategies, resulting in repeated trial and error under similar working conditions, not only reducing the overall system efficiency but also causing resource waste;

[0007] Energy consumption has always been an important issue in the sludge treatment process. Existing control methods rely too much on empirical parameters and are difficult to achieve energy conservation and consumption reduction while ensuring the treatment effect. Especially when dealing with different types of sludge, due to the lack of an intelligent optimization mechanism, conservative control strategies are often adopted, resulting in unnecessary energy waste;

[0008] Traditional control systems lack the ability to learn and optimize, and are unable to summarize experiences from historical operation data and improve control strategies. This results in long-term low operating efficiency of equipment, difficulty in meeting new sludge treatment requirements, and poor adaptability and scalability of the equipment.

[0009] These technical problems restrict the treatment efficiency and operation effect of sludge treatment equipment, not only affecting the quality and stability of sludge treatment, but also causing waste of energy resources and increasing operating costs. Therefore, there is an urgent need for an intelligent control method to improve the performance of sludge treatment equipment and achieve precise control and continuous optimization of the sludge treatment process. Summary of the Invention

[0010] The purpose of the present invention is to propose an autonomous optimization control method and system for intelligent sludge treatment equipment to solve the technical problems of fixed control strategies and inability to adaptively adjust of existing sludge treatment equipment, and improve sludge treatment efficiency and reduce energy consumption. To achieve the above purpose, the technical solutions adopted by the present invention are as follows:

[0011] An autonomous optimization control method for intelligent sludge treatment equipment includes the following steps:

[0012] Deploy an intelligent agent system based on neural network in each sludge treatment unit. Each of the neural networks receives the sludge moisture content and sludge concentration of the unit as inputs and outputs the control values of the stirring speed and pressure of the unit.

[0013] Establish a star network system that supports parallel learning of multiple intelligent agents, including a central control node and multiple sludge treatment unit nodes.

[0014] Based on the star network system, through a distributed algorithm, realize collaborative learning and knowledge transfer between the neural networks of multiple sludge treatment units, and optimize the neural networks of each unit.

[0015] Each sludge treatment unit uses its optimized neural network to generate and execute the control instructions of the unit based on the real-time state data of the unit.

[0016] A further technical solution lies in that the step of deploying an intelligent agent system based on neural network in each sludge treatment unit includes:

[0017] Construct a three-layer fully connected neural network in each sludge treatment unit as the policy network of the unit. Among them, the input layer of the policy network receives the state parameters of the sludge treatment unit. The policy network includes at least one hidden layer for feature extraction and nonlinear mapping, and the output layer of the policy network generates the control parameters of the sludge treatment unit.

[0018] The number of input layer nodes of the policy network is the same as the dimension of the unit state vector, the state vector includes the sludge moisture content and sludge concentration of the unit, the hidden layer of the policy network includes multiple neurons, the number of output layer nodes of the policy network is the same as the dimension of the unit action vector, and the action vector includes the stirring speed and pressure of the unit;

[0019] Construct a reward function based on the sludge treatment efficiency for each sludge treatment unit, and the reward function is calculated by the ratio of the sludge reduction value of the unit to the energy consumption value per unit treatment volume of the unit, where the sludge reduction value is the ratio of the difference in the mass of inlet and outlet sludge per unit time of the unit to the treatment time, and the energy consumption value per unit treatment volume is the ratio of the total energy consumption of the unit to the total amount of treated sludge.

[0020] A further technical solution lies in: the steps of establishing a star network system that supports parallel learning of multiple agents include:

[0021] Establish the star network system using a network communication protocol, taking the central control node as the central node and the sludge treatment units as branch nodes;

[0022] Set up a local experience pool in each sludge treatment unit. The local experience pool adopts a circular queue structure and stores experience data including the identifier of the unit, the current state vector of the unit, the action vector executed by the unit, the reward value obtained by the unit, and the next state vector of the unit;

[0023] Set up a global experience pool in the central control node for storing and distributing high-quality experience data, and the high-quality experience data is screened based on the relative magnitudes of the reward values calculated by the reward functions of each sludge treatment unit;

[0024] Set up a communication retransmission and heartbeat detection mechanism to ensure the reliability of the star network system.

[0025] A further technical solution lies in: the star network system includes:

[0026] A central control node for overall coordination of the overall optimization process;

[0027] Multiple sludge treatment unit nodes, respectively establishing direct two-way communication connections with the central control node;

[0028] Independent communication channels are established between the central control node and each sludge treatment unit node;

[0029] A further technical solution lies in: the steps of realizing collaborative learning and knowledge transfer between the neural networks of multiple sludge treatment units through a distributed algorithm include:

[0030] The respective policy networks of multiple sludge treatment units are trained simultaneously using a distributed gradient descent algorithm to achieve parallel optimization of the sludge treatment units;

[0031] Based on the cosine similarity, calculate the similarity between the current state of the target sludge treatment unit and the historical states in the global experience pool. When the cosine similarity is greater than the preset similarity threshold, select the corresponding experience data and input the experience data into the policy network of the target sludge treatment unit for training to optimize the control strategy of the unit for similar working conditions;

[0032] The efficiency score is the average value of the reward values obtained by the sludge treatment unit within a preset time period;

[0033] The central control node periodically performs policy parameter migration. When the difference in efficiency scores between two sludge treatment units is greater than the preset efficiency threshold, the central control node obtains the policy network parameters from the sludge treatment unit with a higher efficiency score and updates the policy network parameters to the policy network of the sludge treatment unit with a lower efficiency score;

[0034] A further technical solution lies in that: the step of generating and executing a control instruction based on the real-time state data of the sludge treatment unit includes:

[0035] Each sludge treatment unit collects the sludge moisture content and sludge concentration data of the unit in real time, performs outlier detection, missing value processing, and numerical normalization on the data, and constructs the state vector of the unit;

[0036] Calculate the stirring speed value and pressure value of the unit through the policy network of the unit and perform validity verification. The validity verification includes determining whether the stirring speed value and the pressure value are within the allowable operating range of the equipment and determining whether the changes in the stirring speed value and the pressure value conform to the dynamic response characteristics of the equipment;

[0037] When the change range of the stirring speed value and pressure value of the unit exceeds the control parameter smoothing threshold, a linear interpolation method is used to generate the transition control sequence of the unit;

[0038] Encode the verified control parameters of the unit into a standard control instruction and execute it.

[0039] The present invention also provides an autonomous optimization control system for an intelligent sludge treatment device, including: multiple sludge treatment units and a central control node, where:

[0040] The sludge treatment unit is a sludge reduction treatment functional unit with a stirring device and a pressurizing device, including:

[0041] A feeding system for transporting sludge to be treated;

[0042] A treatment chamber provided with the stirring device and the pressurizing device;

[0043] A discharging system for discharging the treated sludge;

[0044] A data acquisition module for acquiring sludge moisture content, sludge concentration data, inlet and outlet sludge mass data, treatment time data, and energy consumption data;

[0045] A policy network module including a three-layer fully connected neural network structure for outputting a stirring speed value and a pressure value according to the sludge moisture content and the sludge concentration data;

[0046] A control execution module for executing control instructions corresponding to the stirring speed value and the pressure value;

[0047] The central control node is connected to a plurality of the sludge treatment units through a network system for coordinating the collaborative learning and knowledge transfer of the plurality of sludge treatment units.

[0048] A further technical solution lies in that: the sludge treatment unit further includes a local experience pool for storing experience data including unit identification, current state vector, executed action vector, obtained reward value, and next state vector;

[0049] The central control node includes a global experience pool and a distributed training module. The global experience pool is used for storing and distributing high-quality experience data. The distributed training module is used for executing a distributed gradient descent algorithm to train the policy network and performing policy parameter transfer based on an efficiency score.

[0050] The beneficial effects of the present invention are as follows:

[0051] (1) The intelligent adaptive control effect is remarkable: The present invention adopts an intelligent agent system based on a neural network, which dynamically outputs optimal stirring speed and pressure control values by receiving sludge moisture content and concentration data in real time. Compared with traditional fixed control strategies, the present invention can automatically adjust treatment parameters according to different sludge characteristics, making the control strategy more accurate and flexible.

[0052] (2) The multi-unit collaborative learning advantage is prominent: The present invention innovatively establishes a star network system and a distributed algorithm framework to realize collaborative learning and knowledge transfer among multiple treatment units. When a certain unit obtains high-quality treatment experience, it can quickly share it with other units to avoid the process of repeated trial and error.

[0053] (3) The experience pool is designed scientifically and efficiently: In the present invention, a local experience pool is set up in each processing unit, and a global experience pool is set up at the central control node. The circular queue structure is adopted to efficiently store the operation experience. Through the calculation of cosine similarity, the system can quickly retrieve and utilize historical high-quality experience to provide optimization reference for new working conditions.

[0054] (4) The system has strong reliability and scalability: The present invention adopts the distributed gradient descent algorithm for parallel training, and sets up a communication retransmission and heartbeat detection mechanism to ensure the stable and reliable operation of the system. At the same time, the star network architecture facilitates the rapid access of new processing units and the migration of optimization strategies. The system has strong scalability and good adaptability. Description of the Drawings

[0055] Figure 1 It is a flow chart of the autonomous optimization control method for intelligent sludge treatment equipment.

[0056] Figure 2 It is a flow chart of the deployment of the intelligent agent system based on neural network.

[0057] Figure 3 It is a diagram of the star network system architecture.

[0058] Figure 4 It is a flow chart of the collaborative learning and knowledge transfer of the neural network in the sludge treatment unit.

[0059] Figure 5 It is a diagram of the architecture of the autonomous optimization control system for intelligent sludge treatment equipment. Detailed Embodiments

[0060] Next, in combination with the drawings in the embodiments of the present invention, the technical solutions in the embodiments of the invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0061] Many specific details are set forth in the following description in order to provide a thorough understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0062] Referring to Figure 1 , the main steps of the method provided by the present invention are as follows:

[0063] (1) Deploy an agent system based on neural network in each sludge treatment unit. Each neural network receives the sludge moisture content and sludge concentration of the unit as inputs and outputs the control values of the stirring speed and pressure of the unit.

[0064] (2) Establish a star network system to support parallel learning of multiple agents, including a central control node and multiple sludge treatment unit nodes. Each sludge treatment unit establishes a two-way communication connection with the central control node to achieve storage and sharing of empirical data.

[0065] (3) Based on the star network system, through a distributed algorithm, achieve collaborative learning and knowledge transfer between the neural networks of multiple sludge treatment units, including distributed gradient descent training, reuse of similar experiences, and transfer of policy parameters, to optimize the neural networks of each unit.

[0066] (4) Each sludge treatment unit uses the optimized neural network to generate and execute control instructions based on real-time status data, including steps such as data preprocessing, calculation of control parameters, validity verification, and instruction execution.

[0067] As a preferred embodiment of the present invention, as Figure 2 shown, in step (1), the deployment process of the agent system based on neural network is as follows:

[0068] (1-1) For each sludge treatment unit, construct a three-layer fully connected neural network as the policy network of the unit. This policy network includes an input layer, a hidden layer, and an output layer. Among them, the input layer is used to receive the state parameters of the sludge treatment unit; the hidden layer contains multiple neuron nodes for feature extraction and non-linear mapping of the input data; the output layer is used to generate the control parameters of the sludge treatment unit. Specifically, the number of nodes in the input layer of the policy network is the same as the dimension of the unit state vector, and the state vector contains the sludge moisture content and sludge concentration data collected in real time by the unit; preferably, the ReLU activation function is used in the hidden layer of the policy network, and preferably, it contains 30-50 neuron nodes; the number of nodes in the output layer of the policy network is the same as the dimension of the unit action vector, and the action vector contains the stirring speed and pressure control values of the unit.

[0069] (1-2) Construct a reward function based on sludge treatment efficiency for each sludge treatment unit. This reward function is calculated by the ratio of the sludge reduction value of the unit to the energy consumption value per unit treatment volume of the unit, and is used to evaluate the quality of the control strategy. Among them, the calculation formula for the sludge reduction value is: the ratio of the difference in the mass of inlet and outlet sludge of the unit within a unit time to the treatment time; the calculation formula for the energy consumption value per unit treatment volume is: the total energy consumption of the unit divided by the total amount of treated sludge. Preferably, using this calculation method can simultaneously consider the reduction effect of sludge treatment and energy consumption, and achieve a balance between treatment efficiency and energy consumption.

[0070] (1 - 3) Preferably, data pre - processing is performed in each sludge treatment unit, and the collected sensor data is standardized. Preferably, it specifically includes: detecting outliers in the sludge moisture content and sludge concentration data and removing data that significantly deviates from the normal range; repairing missing data using interpolation methods; normalizing the data to the interval [-1, 1] to improve the training effect and prediction accuracy of the neural network. Preferably, a data caching mechanism is set up to regularly save the pre - processed historical data for the offline training and online optimization of the neural network.

[0071] As a preferred embodiment of the present invention, referring to Figure 3 , in step (2), the star - shaped network system includes a central control node and multiple sludge treatment unit nodes. Among them, the central control node is used as the central node, and the sludge treatment units are used as branch nodes. Each sludge treatment unit node establishes a direct two - way communication connection with the central control node. This network structure can support multiple sludge treatment units to perform parallel learning and achieve collaborative optimization of control strategies.

[0072] The design process of the communication module of the star - shaped network system in step (2 - 1) includes: establishing a star - shaped network system using a network communication protocol. Preferably, the TCP / IP protocol is used to achieve reliable communication between nodes to ensure the integrity and real - time nature of data transmission. Preferably, the design of the communication protocol includes message type definition, data format specification, and communication timing requirements, supporting various service requirements such as empirical data transmission and parameter synchronization.

[0073] The design process of the experience pool module of the star - shaped network system in step (2 - 2) includes: setting up a local experience pool in each sludge treatment unit. The local experience pool adopts a circular queue structure and is used to store experience data including the identifier of the unit, the current state vector of the unit, the action vector executed by the unit, the reward value obtained by the unit, and the next state vector of the unit. Preferably, the capacity of the local experience pool is 1000 - 2000 pieces of experience data. When the storage space is insufficient, new experience data will replace the earliest stored data.

[0074] The design process of the central management module of the star - shaped network system in step (2 - 3) includes: setting up a global experience pool in the central control node, which is used to store and distribute high - quality experience data. The high - quality experience data is screened based on the relative magnitude of the reward values calculated by the reward functions of each sludge treatment unit. Preferably, a reward value screening threshold is set, and only the experience data with reward values in the top 30% is saved. Preferably, the global experience pool is updated regularly to maintain the timeliness of the data.

[0075] The reliability guarantee process of the star network system in step (2-4) includes: setting up a communication retransmission and heartbeat detection mechanism to ensure the reliability of the star network system. Specifically, when communication failure is detected, the system automatically retransmits data packets; the online status of each node is detected through regular heartbeat messages, and when a node is found to be offline, fault handling is performed in a timely manner. Preferably, the upper limit of the retransmission times is set to 3 times, and the heartbeat detection period is set to 5 seconds. Preferably, the system also includes an encryption and authentication mechanism for communication data to ensure the security of data transmission.

[0076] As a preferred embodiment of the present invention, referring to Figure 4 , in step (3), the collaborative learning and knowledge transfer between the neural networks of multiple sludge treatment units are realized through a distributed algorithm, including the following steps:

[0077] Step (3-1) Use the distributed gradient descent algorithm to train the respective policy networks of multiple sludge treatment units simultaneously to achieve parallel optimization of the sludge treatment units. Preferably, the training process is as follows: each sludge treatment unit calculates the parameter gradient of the policy network based on local experience data; sends the locally calculated gradient information to the central control node; the central control node aggregates the collected gradient information to generate a global gradient update value; distributes the global gradient update value to each sludge treatment unit for updating the respective policy network parameters. Preferably, the Adam optimizer is used for parameter update, the initial value of the learning rate is set to 0.001, and it is dynamically adjusted during the training process.

[0078] Step (3-2) Calculate the similarity between the current state of the target sludge treatment unit and the historical states in the global experience pool based on the cosine similarity. When the cosine similarity is greater than the preset similarity threshold, select the corresponding experience data and input the experience data into the policy network of the target sludge treatment unit for training to optimize the control strategy of the unit for similar working conditions. Preferably, the specific implementation process includes: calculating the cosine similarity between the current state vector of the target sludge treatment unit and the historical state vectors in the global experience pool; preferably, setting the similarity threshold to 0.85; screening the experience data with similarity greater than the threshold from the global experience pool and adding this experience data to the local experience pool of the target unit. Preferably, the amount of experience data added each time does not exceed 20% of the local experience pool capacity. When performing supplementary training, the mini-batch stochastic gradient descent method is used. Preferably, the batch size is set to 32, the number of training epochs is set to 10, and the learning rate is set to 0.0001.

[0079] Step (3-3): Set up an efficiency evaluation mechanism and perform policy parameter migration. Among them, the efficiency score is the average value of the reward values obtained by the sludge treatment unit within a preset time period. The central control node periodically performs policy parameter migration. When the difference in efficiency scores between two sludge treatment units is greater than a preset efficiency threshold, the central control node obtains the policy network parameters of the sludge treatment unit with a higher efficiency score and updates the policy network parameters to the policy network of the sludge treatment unit with a lower efficiency score. Preferably, the specific implementation process is as follows: Calculate the efficiency scores of each sludge treatment unit within a preset time period (preferably set to 24 hours); preferably, set the efficiency threshold to 20%; Sort the sludge treatment units according to the efficiency scores; When detecting a unit pair with an efficiency score difference exceeding the threshold, trigger policy parameter migration; Preferably, during the policy parameter migration process, directly copy and replace the network parameters of the high-efficiency unit with those of the low-efficiency unit to achieve rapid knowledge migration.

[0080] Through the above collaborative learning and knowledge migration mechanism, the operating experience of each sludge treatment unit can be fully utilized to accelerate the optimization process of the policy network and improve the control effect of the entire system. Among them, distributed training ensures learning efficiency, similarity calculation ensures the effectiveness of experience reuse, and policy migration promotes knowledge sharing between different units.

[0081] As a preferred implementation of the present invention, in step (4), each sludge treatment unit generates and executes control instructions based on real-time status data, including the following steps:

[0082] Step (4-1): Each sludge treatment unit collects the sludge moisture content and sludge concentration data of the unit in real time, performs outlier detection, missing value processing, and numerical normalization on the data, and constructs the status vector of the unit. Preferably, sensors are used to collect real-time data, and the sampling period is set to 1 second. First, outlier detection is performed on the collected data. Preferably, the 3σ criterion is used to identify abnormal data points. Repair processing is performed on the detected outliers and missing values. Preferably, the nearest neighbor interpolation method is used for data repair. Subsequently, the processed data is normalized and mapped to the [-1, 1] interval, and finally a standardized status vector is constructed.

[0083] Step (4-2) calculates the stirring speed value and pressure value of the unit through the unit's policy network and conducts validity verification. The validity verification includes determining whether the stirring speed value and pressure value are within the allowable operating range of the equipment, and determining whether the changes in the stirring speed value and pressure value conform to the dynamic response characteristics of the equipment. Specifically, first, the standardized state vector is input into the policy network to obtain the initial control value; then, it is checked whether the control value is within the equipment operating range. Preferably, the stirring speed range is 0-1000 rpm, and the pressure range is 0-2 MPa. At the same time, the change rate of the control value is calculated. Preferably, the maximum change rate of the stirring speed is set to 100 rpm / s, and the maximum change rate of the pressure is set to 0.2 MPa / s. When the control value exceeds the limit range, truncation processing is performed to ensure the effectiveness of the control parameters.

[0084] Step (4-3) When the change amplitude of the stirring speed value and pressure value of the unit exceeds the control parameter smoothing threshold, a linear interpolation method is used to generate a transition control sequence for the unit, and the verified control parameters are encoded into standard control instructions and executed. Specifically, first, the control parameter smoothing threshold is set. Preferably, the stirring speed change threshold is 50 rpm, and the pressure change threshold is 0.1 MPa. When the parameter change exceeds the threshold, preferably, a linear interpolation is used to generate a smooth transition sequence. Subsequently, the control parameters are encoded according to the equipment communication protocol to generate control instructions in a standard format, and the instructions are sent to the equipment execution layer through the control execution module. Preferably, an instruction execution feedback mechanism is set to monitor the execution status of the control instructions to ensure the reliability of the control process.

[0085] Through the above control instruction generation and execution mechanism, the stable operation of the sludge treatment equipment can be ensured, and the impact on the equipment caused by the drastic change of the control parameters can be avoided. At the same time, the perfect validity verification and smoothing processing mechanism can improve the reliability and stability of the control system.

[0086] Referring to Figure 5 , the autonomous optimization control system of the intelligent sludge treatment equipment provided by the present invention includes multiple sludge treatment units and a central control node. The system adopts a distributed architecture to realize the intelligent control and optimization of sludge treatment.

[0087] As a specific embodiment of the present invention, the sludge treatment unit is a sludge reduction treatment functional unit with a stirring device and a pressurizing device, and its specific structure includes:

[0088] A feeding system for transporting the sludge to be treated. Specifically, it includes a feeding pump, a pipeline system, and a flow metering device. Preferably, the feeding pump adopts a variable frequency speed regulation screw pump, which can adjust the feeding rate according to the process requirements; preferably, the flow metering device adopts an electromagnetic flowmeter to monitor the feeding amount in real time.

[0089] Treatment cavity, equipped with a stirring device and a pressurizing device. Specifically, the stirring device includes a variable-frequency motor and a stirring paddle, which are used to achieve sufficient mixing of the sludge; the pressurizing device includes a pressure pump and a pressure sensor, which are used to adjust and monitor the cavity pressure. Preferably, the stirring paddle adopts a double-propeller structure to improve the mixing effect; preferably, the measuring range of the pressure sensor is 0-5 MPa, and the accuracy class is 0.5 level.

[0090] Discharge system, used to discharge the treated sludge. Specifically, it includes a discharge pump, a discharge pipeline and a sludge metering device. Preferably, the discharge system is equipped with an automatic regulating valve, which can control the discharge rate according to the process requirements.

[0091] Data acquisition module, used to collect data such as sludge moisture content, sludge concentration, inlet and outlet sludge mass, treatment time and energy consumption. Specifically, it includes various sensors, a data acquisition card and a signal conditioning circuit. Preferably, the moisture content is measured by a microwave moisture meter, the concentration is monitored by an online concentration meter, and the energy consumption data is collected through an electric energy meter.

[0092] Policy network module, including a three-layer fully connected neural network structure, which is used to output the stirring speed value and the pressure value according to the sludge moisture content and sludge concentration data. Specifically, this module is deployed on an industrial control computer and runs a neural network algorithm using a real-time operating system. Preferably, the industrial control computer is configured with an industrial-grade CPU and a memory to ensure computing performance and stability.

[0093] Control execution module, used to execute the control instructions corresponding to the stirring speed value and the pressure value. Specifically, it includes a PLC controller, a frequency converter and an actuator. Preferably, a PLC system with redundancy function is adopted to improve control reliability.

[0094] Local experience pool, used to store experience data including unit identification, current state vector, executed action vector, obtained reward value and next state vector. Specifically, an industrial-grade database is used to implement data storage and management. Preferably, a data regular backup mechanism is set up to prevent data loss.

[0095] As a specific embodiment of the present invention, the central control node includes:

[0096] Global experience pool, used to store and distribute high-quality experience data. Specifically, a distributed database system is used to implement data storage and management, supporting multi-node concurrent access. Preferably, a data sharding storage mechanism is set up to improve data access efficiency.

[0097] The distributed training module is used to execute the distributed gradient descent algorithm to train the policy network and perform policy parameter migration based on the efficiency score. Specifically, it includes a training scheduler, a parameter synchronizer, and a performance evaluator. Preferably, a GPU acceleration card is used to improve the training speed, and a function of resuming training from a breakpoint is configured to avoid training interruption.

[0098] As a specific implementation manner of the present invention, each component of the system is connected through an industrial Ethernet. Specifically, a ring network topology structure is adopted to improve communication reliability, and a network redundancy protection function is configured. Preferably, network security protection measures are set, including mechanisms such as firewalls and access controls.

[0099] Through the above system implementation scheme, the intelligent control and optimization of the sludge treatment equipment can be realized. Among them, each functional module of the sludge treatment unit works together to ensure the stable operation of a single treatment unit; the central control node realizes the collaborative optimization of multiple units through the experience pool and distributed training; a reliable system integration scheme guarantees the stability and scalability of the entire system.

[0100] The above are only the preferred specific implementation manners of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the technical field, within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be within the protection scope of the present invention.

Claims

1. An autonomous optimization control method for an intelligent sludge treatment device, characterized in that, It includes the following steps: Deploy an agent system based on neural network in each sludge treatment unit. Each of the neural networks receives the moisture content and sludge concentration of the sludge in the unit as inputs and outputs the control values of the stirring speed and pressure of the unit; Establish a star network system that supports parallel learning of multiple agents, including a central control node and multiple sludge treatment unit nodes. A local experience pool is set in each of the sludge treatment units. The local experience pool adopts a circular queue structure and stores experience data including the identifier of the unit, the current state vector of the unit, the action vector executed by the unit, the reward value obtained by the unit, and the next state vector of the unit. A global experience pool is set in the central control node for storing and distributing high-quality experience data. The high-quality experience data is screened based on the relative magnitudes of the reward values calculated by the reward functions of the sludge treatment units; Based on the star network system, the collaborative learning and knowledge transfer between the neural networks of multiple sludge treatment units are realized through a distributed algorithm to optimize the neural networks of each unit. The central control node periodically performs policy parameter migration. When the difference in efficiency scores between two sludge treatment units is greater than a preset efficiency threshold, the central control node obtains the policy network parameters of the sludge treatment unit with a higher efficiency score and updates the policy network parameters to the policy network of the sludge treatment unit with a lower efficiency score; Each of the sludge treatment units uses its optimized neural network to generate and execute the control instructions of the unit based on the real-time state data of the unit; Build a three-layer fully connected neural network in each of the sludge treatment units as the policy network of the unit; The number of input layer nodes of the policy network is the same as the dimension of the unit state vector. The state vector includes the moisture content and sludge concentration of the sludge in the unit. The number of output layer nodes of the policy network is the same as the dimension of the unit action vector. The action vector includes the stirring speed and pressure of the unit; Use the distributed gradient descent algorithm to train the respective policy networks of multiple sludge treatment units simultaneously to achieve parallel optimization of the sludge treatment units.

2. The autonomous optimization control method of an intelligent sludge treatment device according to claim 1, characterized in that, The step of deploying an agent system based on neural network in each sludge treatment unit includes: The input layer of the policy network receives the state parameters of the sludge treatment unit. The policy network includes at least one hidden layer for feature extraction and non-linear mapping. The output layer of the policy network generates the control parameters of the sludge treatment unit. The hidden layer of the policy network includes multiple neurons; Construct a reward function based on sludge treatment efficiency for each of the sludge treatment units. The reward function is calculated by the ratio of the sludge reduction value of the unit to the energy consumption value per unit treatment volume of the unit. The sludge reduction value is the ratio of the difference in the mass of inlet and outlet sludge of the unit per unit time to the treatment time. The energy consumption value per unit treatment volume is the ratio of the total energy consumption of the unit to the total amount of sludge treated; 3. The autonomous optimization control method of an intelligent sludge treatment device according to claim 1, characterized in that, The step of establishing a star network system that supports parallel learning of multiple agents further includes: The star network system is established using a network communication protocol, with the central control node as the central node and the sludge treatment units as branch nodes; A communication retransmission and heartbeat detection mechanism is set up to ensure the reliability of the star network system.

4. An autonomous optimization control method for an intelligent sludge treatment device according to claim 3, characterized in that, The star network system includes: One central control node, which is used to overall coordinate and optimize the whole process; Multiple sludge treatment unit nodes, which respectively establish direct two-way communication connections with the central control node; Independent communication channels are established between the central control node and each sludge treatment unit node.

5. The autonomous optimization control method of an intelligent sludge treatment device according to claim 1, characterized in that, The steps of realizing collaborative learning and knowledge transfer between the neural networks of multiple sludge treatment units through a distributed algorithm further include: Based on the cosine similarity, calculate the similarity between the current state of the target sludge treatment unit and the historical states in the global experience pool. When the cosine similarity is greater than the preset similarity threshold, select the corresponding experience data, and input the experience data into the policy network of the target sludge treatment unit for training to optimize the control strategy of the unit for similar working conditions; The efficiency score is the average value of the reward values obtained by the sludge treatment unit within a preset time period.

6. The autonomous optimization control method of an intelligent sludge treatment device according to claim 1, characterized in that, The steps of generating and executing control instructions based on the real-time state data of the sludge treatment unit include: Each sludge treatment unit collects the sludge moisture content and sludge concentration data of the unit in real time, detects outliers, processes missing values and normalizes the data to construct the state vector of the unit; Calculate the stirring speed value and pressure value of the unit through the policy network of the unit and conduct validity verification. The validity verification includes judging whether the stirring speed value and the pressure value are within the allowable operating range of the equipment, and judging whether the changes of the stirring speed value and the pressure value conform to the dynamic response characteristics of the equipment; When the change range of the stirring speed value and pressure value of the unit exceeds the control parameter smoothing threshold, a linear interpolation method is used to generate the transition control sequence of the unit; Encode the verified control parameters of the unit into standard control instructions and execute them.

7. An autonomous optimization control system for an intelligent sludge treatment device, characterized in that, It includes multiple sludge treatment units and a central control node, where: The sludge treatment unit is a sludge reduction treatment functional unit with a stirring device and a pressurizing device, including: A feeding system, which is used to transport the sludge to be treated; A treatment cavity, which is provided with the stirring device and the pressurizing device; A discharging system, which is used to discharge the treated sludge; A data acquisition module, which is used to collect sludge moisture content, sludge concentration data, inlet and outlet sludge mass data, treatment time data and energy consumption data; A policy network module, which includes a three-layer fully connected neural network structure and is used to output a stirring speed value and a pressure value according to the sludge moisture content and the sludge concentration data; A control execution module, which is used to execute the control instructions corresponding to the stirring speed value and the pressure value; A local experience pool, which is used to store experience data including unit identification, current state vector, executed action vector, obtained reward value and next state vector; The central control node is connected to a plurality of the sludge treatment units through a network system and is used to coordinate the collaborative learning and knowledge transfer of the plurality of the sludge treatment units. The central control node includes: A global experience pool, which is used to store and distribute high-quality experience data, and the high-quality experience data is screened based on the relative magnitudes of the reward values calculated by the reward functions of the respective sludge treatment units; A distributed training module, which is used to execute a distributed gradient descent algorithm to train the policy network and perform policy parameter transfer based on an efficiency score.

Images