Self-learning controller based on neural network and method thereof

Abstract

The invention provides a neural network-based self-learning controller and a method thereof. The neural network-based self-learning controller comprises a neural network self-learning module, a state prediction and evaluation module, an input leveling module, a dynamic expansion and adjustment module and a deployment and debugging module which are in communication connection in sequence. The beneficial effects of the invention are that the intelligent level of the system is improved: through the self-learning characteristic of the neural network, the controller can automatically adapt to a complex industrial environment, manual intervention is reduced, and the intelligent and automatic levels of the system are improved; manual intervention and debugging are reduced: the system can automatically learn and adjust according to the operation condition of the equipment, so that the complexity of system deployment and maintenance is greatly reduced, and the labor cost is reduced; dynamic expansibility: the system supports dynamic expansion and can adapt to different equipment and load changes, and control logic does not need to be redesigned or the system does not need to be greatly modified; and rapid fault early warning: through prediction and analysis of the equipment state, the system can give out early warning in advance, equipment operation interruption or faults are avoided, and the reliability and safety of the whole system are improved.

Description

Technical Field

[0001] The invention belongs to the technical field of industrial automation control for coal mines, and in particular relates to a neural network-based self-learning controller and a method thereof. Background Art

[0002] In the field of industrial automation, traditional control systems usually rely on expert systems and knowledge bases to perform load balancing and state prediction of equipment. Its characteristics are to make judgments based on pre-set empirical models or rules. This approach has the following limitations in scenarios with diverse load changes and complex environments:

[0003] 1. Slow response speed: The expert system is based on preset rules and it is difficult to adaptively adjust the control logic in real time, resulting in delayed equipment load distribution and adjustment.

[0004] 2. Poor scalability: When new equipment or system expansion occurs, the knowledge base needs to be updated or the rules need to be redesigned, which makes the operation complicated.

[0005] 3. High maintenance cost: Traditional systems require frequent debugging and manual maintenance, especially for scenarios with dynamic load changes.

[0006] Traditional solutions cannot meet the requirements of multi-device load balancing and real-time adjustment in complex environments, so it is necessary to introduce new technical means to overcome these problems. Summary of the invention

[0007] In view of this, the present invention aims to propose a self-learning controller based on neural network and its method to solve the load management problem of traditional systems in complex multi-drive control, realize load balancing control, real-time status evaluation and prediction of multiple drive devices, and improve the scalability and deployment efficiency of the system.

[0008] To achieve the above object, the technical solution of the present invention is achieved as follows:

[0009] In a first aspect, the present invention provides a neural network-based self-learning controller, comprising a neural network self-learning module, a state prediction and evaluation module, an input leveling module, a dynamic expansion and adjustment module, and a deployment and debugging module that are sequentially connected in communication, wherein the neural network self-learning module is used to input device operation data and feedback results, and realize adaptive learning after multiple trainings; the state prediction and evaluation module is used to analyze the historical data of the device, so that the controller can predict the future operation state of the device; the input leveling module is used to perform load balancing adjustment between multiple controllers or drive devices; the dynamic expansion and adjustment module can support dynamic addition of new devices or adjustment of the load of existing devices without stopping or redesigning the control strategy; the deployment and debugging module is used to automatically adjust the control parameters to reduce the need for manual intervention and debugging;

[0010] Furthermore, in the neural network self-learning module, the control logic is:

[0011] The neural network model can automatically adjust the control strategy based on historical operating data and real-time status, ensuring that load distribution and drive load system adjustment are adaptive and do not rely on artificially set rules or empirical models;

[0012] The self-learning process can dynamically adjust the working status of each device according to the load characteristics and operating status of different devices, ensuring that each device operates within the optimal load range.

[0013] Furthermore, in the state prediction and evaluation module, the control logic is:

[0014] By analyzing the historical data of the equipment, the controller can predict the future operation status of the equipment, including:

[0015] Trends in equipment performance degradation: By analyzing current equipment load and past operating data, the controller can determine in advance whether equipment performance degradation or failure is likely;

[0016] Fault risk warning: When a possible equipment failure is predicted, the drive load system can issue a warning in advance to remind operators to perform maintenance or adjustments.

[0017] Furthermore, in the input leveling module, the control logic is:

[0018] The drive load system can balance the load between multiple controllers or drive devices, ensuring that the load input of the devices is reasonably distributed to avoid overloading of a single device or waste of resources.

[0019] Furthermore, in the dynamic expansion and adjustment module, the control logic is:

[0020] The Drive Load System supports dynamically adding new equipment or adjusting the load of existing equipment without downtime or redesigning control strategies; reducing commissioning and deployment time.

[0021] Furthermore, in the deployment and debugging module, the control logic is:

[0022] Through the adaptive capabilities of neural networks, the system debugging process is simplified and human intervention is greatly reduced.

[0023] In a second aspect, based on the same concept, the present invention also provides a method of a self-learning controller based on a neural network, comprising the following steps:

[0024] S1, obtain data through the existing controller;

[0025] S2, preprocessing the data of the data channel;

[0026] S3, preload the data of each channel to establish a model;

[0027] S4, updating the model for driving load optimization;

[0028] S5. Optimize the data dynamic upgrade model of each channel;

[0029] In step S1, obtaining data through the existing controller includes:

[0030] In the working face control and belt drive control, the existing controller is responsible for obtaining the operating status data of the equipment, including information:

[0031] Current, voltage, and temperature sensor data;

[0032] Equipment load, speed drive performance data;

[0033] Equipment start and stop data, operating time, shutdown times, fault points, and fault frequency historical operating data;

[0034] The above data will be provided as input to the neural network of the self-learning controller for analysis and learning.

[0035] Furthermore, in step S2, the data of the data channel is preprocessed, including:

[0036] Data preprocessing refers to the process of cleaning, standardizing and normalizing the raw data obtained from the traditional controller sensors and various feedback modules. The drive load system first uses the communication port connected to the traditional controller to obtain the operating data of the equipment;

[0037] Establish a time series database and obtain equipment data through the time series database. Define the equipment as a continuous time series, define the equipment load data L(t) and power consumption P(t), where t represents time; the equipment load data is obtained from the coal quantity sensor until the load data is obtained, and the power consumption P data is input at the same time, and piecewise fitting is performed, which is expressed as:

[0038] ;

[0039] L is the load;

[0040] L1 is the load dividing point;

[0041] a1, b1c1 are constants used for nonlinear relationships in the first interval (L≤I L1);

[0042] a2, b2c2 are constants used for nonlinear relationships in the first interval (L>L1);

[0043] abc is a constant fitted based on historical data. I is divided into 3-5 stages for fitting and the data is stored.

[0044] Furthermore, in step S3, the data of each channel is preloaded to establish a model, including:

[0045] Preloading refers to inputting preprocessed data into the neural network for preliminary training. The operation content is:

[0046] Load multi-channel data in batches, and input the data of each channel into the neural network model for preliminary learning and training;

[0047] Learning is done based on the channels and manual operations of the on-site monitoring equipment. Unsupervised learning is used first, and supervised learning is done through historical data. The model is adaptively adjusted based on the data pattern.

[0048] The input data records P and L and the corresponding fitting data, records the label jump data, stores it in the database, and gives prompts to the operator, stores it in the database, makes judgments later, and finally completes the labeled learning for self-optimization and upgrade.

[0049] Furthermore, in step S4, the model is updated for driving load optimization, including:

[0050] After the initial training is completed, the neural network model can output load optimization suggestions based on multi-channel data; by comparing the operating status of each device, the neural network can provide a load distribution plan for each device. The operation content is:

[0051] Real-time optimization: Based on the real-time data input, the model automatically updates the weights and adjusts the load output of each drive device to maximize the overall operating efficiency of the equipment;

[0052] Load balancing algorithm: distributes the load to multiple drives to balance system resources;

[0053] Design three modes: load balancing with maximum load carrying capacity, load balancing with minimum power consumption, and load balancing combined with power optimization:

[0054] The combination of intermediate load balancing and power optimization is a model-based load forecasting, which uses a time series forecasting model to predict future load and power requirements; load distribution is performed based on the forecast results, and scheduling is performed in advance to adapt to future needs;

[0055] The relationship between the power consumption Pi and the load Li of each device i is defined as a nonlinear function Pi=f(Li), where f is a specific formula fitted by each device; the load distribution objective function is performed using the following comprehensive optimization objective:

[0056] ;

[0057] Constraints:

[0058] Total load distribution constraints:

[0059] ;

[0060] The load of each device does not exceed its maximum load carrying capacity:

[0061] ;

[0062] Finally, initialize the initialization function to assign initial load to each device , and adjust Li by gradient descent to minimize the total power consumption while satisfying the load distribution constraints.

[0063] Furthermore, in step S5, the data dynamic update model of each channel is optimized, including:

[0064] During the operation of the driving load system, the neural network model can continue to receive real-time data from each channel and dynamically adjust and upgrade according to the load changes and operating status of the equipment;

[0065] The model continuously learns and optimizes through real-time input data. As the operating status of the equipment changes, the model is constantly updated to ensure that each device is always in the optimal load state.

[0066] Through the online learning mechanism of continuous labeling, the model can be iteratively updated according to the latest data of the equipment, gradually improving the prediction accuracy and load distribution efficiency, and fully self-adjusting.

[0067] Compared with the prior art, the neural network-based self-learning controller and method described in the present invention have the following advantages:

[0068] (1) The self-learning controller based on neural network and the method thereof described in the present invention improve the intelligence level of the system: through the self-learning characteristics of the neural network, the controller of the present invention can autonomously adapt to complex industrial environments, reduce human intervention, and improve the intelligence and automation level of the system.

[0069] (2) The neural network-based self-learning controller and method described in the present invention reduce manual intervention and debugging: the system can automatically learn and adjust according to the equipment operation status, greatly reducing the complexity of system deployment and maintenance and reducing labor costs.

[0070] (3) A neural network-based self-learning controller and method thereof described in the present invention, dynamic scalability: The system of the present invention supports dynamic expansion and can adapt to different devices and load changes without redesigning the control logic or significantly modifying the system.

[0071] (4) The present invention discloses a neural network-based self-learning controller and method thereof, which provides rapid fault warning: by predicting and analyzing the equipment status, the system can issue a warning in advance to avoid equipment operation interruption or failure, thereby improving the reliability and safety of the overall system. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0073] Figure 1 This is a schematic diagram of the principle of the self-learning controller according to an embodiment of the present invention;

[0074] Figure 2 A schematic diagram of a method flow of a self-learning controller according to an embodiment of the present invention;

[0075] Figure 3 The figure is a schematic diagram of the convenient deployment and debugging process described in an embodiment of the present invention. DETAILED DESCRIPTION

[0076] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0077] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and the like are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first", "second", and the like may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0078] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood by specific circumstances.

[0079] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0080] like Figures 1 to 3 As shown, a neural network-based self-learning controller includes control logic, and the control logic includes automatic learning and load balancing, state evaluation and prediction, multi-system input leveling, dynamic expansion and adjustment, and simplified deployment and debugging;

[0081] 1. Automatic learning and load balancing: The controller conducts self-learning of the equipment through neural networks, and can automatically adjust the load distribution according to the actual working conditions, so that the equipment can operate at the optimal efficiency.

[0082] 2. Status assessment and prediction: The neural network model can evaluate the current status of the equipment in real time, predict future operating trends, and provide early warning of equipment failure or performance degradation by learning the equipment's historical operating data and analyzing real-time data.

[0083] 3. Multi-system input leveling: In scenarios where there are multiple drive devices or controllers, the system can balance the input load to ensure reasonable load distribution between different devices and reduce overload or resource waste.

[0084] 4. Dynamic expansion and adjustment: The system has strong scalability and supports dynamic addition of new devices or load adjustment without redesigning the control strategy, which significantly reduces debugging and deployment time.

[0085] 5. Simplified deployment and debugging: Through the adaptive capabilities of neural networks, the system debugging process is simplified, greatly reducing manual intervention.

[0086] The present invention can be applied to the following scenarios:

[0087] 1. Belt multi-drive controller: Controls the load adjustment of multiple belt transmission lines to ensure that the drives on different belts maintain balanced operation when the load changes.

[0088] 2. Work face controller: In mines or industrial scenarios, it controls the collaborative work of multiple devices or machines to ensure that they perform load balancing and dynamic adjustments in real time according to changes in working conditions.

[0089] 3. Other major multi-drive equipment linkage systems that require power balance.

[0090] The main technical features of the self-learning controller include: neural network self-learning module, state prediction and evaluation module, input leveling module, dynamic expansion and adjustment module, and deployment and debugging module;

[0091] 1. Neural network self-learning:

[0092] The controller uses a multi-layer neural network model, which realizes adaptive learning through input of equipment operation data and feedback results after multiple trainings. The specific features are as follows:

[0093] The neural network model automatically adjusts the control strategy based on historical operating data and real-time status, ensuring that load distribution and system adjustments are adaptive and do not rely on artificially set rules or empirical models.

[0094] The self-learning process can dynamically adjust (partition adjustment based on the relationship between load and power consumption) the working status of each device according to the load characteristics and operating status of different devices, ensuring that each device operates within the optimal load range.

[0095] 2. Equipment status prediction and evaluation:

[0096] By analyzing the historical data of the equipment, the controller is able to predict the future operating status of the equipment. This includes predictions of the following aspects:

[0097] Trend of equipment performance degradation: By analyzing the current equipment load and past operating data, the controller can determine in advance whether the equipment may experience performance degradation or failure (balancing will not cause overload or overload of a single device, and the updated model is used to drive the balancing strategy mode in load optimization).

[0098] Fault risk warning: When a possible equipment failure is predicted, the system can issue a warning in advance to remind operators to perform maintenance or adjustments.

[0099] 3. Multi-system input leveling:

[0100] The system can balance the load between multiple controllers or drive devices. This mechanism ensures that the load input of the devices is reasonably distributed to avoid overloading of a single device or waste of resources.

[0101] In multiple drive load systems, the system dynamically levels the input to optimize the operating efficiency of each controller or drive.

[0102] 4. Dynamic expansion and adjustment:

[0103] The system supports dynamically adding new devices or adjusting the load of existing devices without downtime or redesigning control strategies. Through the self-learning characteristics of neural networks, the system can automatically adapt to the operating environment and load requirements of new devices, simplifying the scalability and deployment process of the system.

[0104] 5. Convenient deployment and debugging:

[0105] Compared with traditional systems, the controller of the present invention can reduce a lot of debugging work during initial deployment:

[0106] The system can automatically adjust control parameters through self-learning function, reducing the need for manual intervention and debugging. The system debugging time is greatly shortened, suitable for rapid deployment in different industrial environments, convenient deployment and debugging, such as Figure 3 shown.

[0107] This application uses neural networks to provide automatic learning of multiple drive loads, automatically learns to complete load balancing according to the load of the controlled target device, and gives evaluations and predictions of the operating status of each device. (Difference: the traditional system uses a knowledge base with an expert system to complete prediction and learning. The system currently uses neural networks for self-learning and prediction to complete multi-system input leveling). Compared with traditional expert systems, the system dynamically increases the expansion of the adjustment system, making deployment and debugging more convenient. It can be enabled to be installed on a belt multi-drive controller or a working surface controller to achieve control and autonomous adjustment

[0108] The present application particularly relates to a controller that uses a neural network for self-learning and load balancing. This controller can be widely used in scenarios that require intelligent regulation and load management, such as belt transmission systems and working surface controllers. The system is mainly used in industrial automation, especially in complex multi-drive control systems, including but not limited to coal mines, factory assembly lines, mining transportation equipment and other fields.

[0109] Advantages of the present invention:

[0110] Improve the intelligence level of the system: Through the self-learning characteristics of the neural network, the controller of the present invention can autonomously adapt to complex industrial environments, reduce human intervention, and improve the intelligence and automation level of the system.

[0111] Reduce manual intervention and debugging: The system can automatically learn and adjust according to the operation status of the equipment, which greatly reduces the complexity of system deployment and maintenance and reduces labor costs.

[0112] Dynamic scalability: The system of the present invention supports dynamic expansion and can adapt to different devices and load changes without redesigning the control logic or significantly modifying the system.

[0113] Rapid fault warning: By predicting and analyzing the status of equipment, the system can issue early warning to avoid equipment interruption or failure, thereby improving the reliability and safety of the overall system.

[0114] A method for a self-learning controller based on a neural network:

[0115] System Description: Based on the traditional controller, the controller is upgraded and a self-learning system is built. Through the traditional communication interface, control data is obtained and data collection is completed. Through data acquisition and self-learning in about a week, the overall control data collection is completed and automatically calculated. According to the prefabricated model optimization, the new on-site model is completed, and the calculation superposition update optimization is performed.

[0116] 1. Obtain data through traditional controller:

[0117] In the work surface control and belt drive control, the traditional controller is responsible for obtaining the operating status data of the equipment, including but not limited to the following information:

[0118] Sensor data such as current, voltage, temperature, etc.

[0119] Drive performance data such as device load and speed;

[0120] Equipment start and stop data, operating time, number of shutdowns, fault points, fault frequency and other historical operating data;

[0121] These data will be provided as input to the neural network of the self-learning controller for further analysis and learning.

[0122] 2. Preprocess the data of the data channel:

[0123] Data preprocessing refers to the process of cleaning, standardizing and normalizing the raw data obtained from the traditional controller sensors and various feedback modules. The purpose of this step is to ensure that the data input to the neural network is consistent and can reflect the actual operation of the equipment. Corresponding to each other, such as one start corresponds to one stop. In this step, the system first uses the communication port of the existing traditional controller to obtain the operation data of the equipment. These data come from the sensors or system monitoring modules of the equipment, usually including load data, power consumption data, operation status, fault information, etc.

[0124] At this time, it is necessary to establish a time series database and obtain equipment data through the time series database. The data can be a continuous time series, such as the equipment's load data L(t) and power consumption P(t), where t represents time. These data are the basic inputs for the system's neural network self-learning and will be further processed and used to train the model. The equipment's load data is obtained from the coal volume sensor, and the power consumption P data is input at the same time. In theory, there is a linear relationship between the two, but in practice, there may be many nonlinear force factors. Segmented fitting is performed under different circumstances, which can be expressed as:

[0125] ;

[0126] L is the load (unit: ampere, A);

[0127] L1 is the load cut-off point (unit: kg / s kilogram per second);

[0128] a1, b1c1 are constants used for nonlinear relationships in the first interval (L≤I L1);

[0129] a2, b2c2 are constants used for nonlinear relationships in the first interval (L>L1);

[0130] abc is a constant fitted based on historical data. I is divided into 3-5 stages for fitting and the data is stored.

[0131] 3. Preload the data of each channel to build the model:

[0132] Preloading refers to inputting preprocessed data into the neural network for preliminary training. This step uses existing data channels (each type of data is a channel, such as current, voltage, operating load, etc.) to initially upgrade the neural network model in preparation for the next step of load optimization.

[0133] Operation content:

[0134] The multi-channel data is loaded in batches (batching means dividing the data into stages, such as high power consumption stage, low power consumption stage, etc.), and the data of each channel (such as different drive devices, different sensor data) is input into the neural network model for preliminary learning and training.

[0135] Learning is performed based on the channels of on-site monitoring equipment and manual operations. Unsupervised learning is initially used, and supervised learning is performed through historical data. The model is adaptively adjusted based on data patterns (such as high-power consumption model, low-power consumption model, load classification, etc.).

[0136] The input data is recorded for P and L and the corresponding fitting data, and the record labels with larger jumps (based on the reasonable range predicted by the model. If it exceeds the range, it may be an equipment abnormality or a starter abnormality. Learning is to ensure liability balance) are stored in the database, and prompts are given to personnel operators. They can choose to store them in the database and make judgments later (such as: popping up a message on the server, and then improving the label based on the message). Finally, complete the labeled learning and perform self-optimization and upgrade.

[0137] 4. Update the model for drive load optimization:

[0138] After the initial training is completed, the neural network model will output load optimization suggestions based on the data from multiple channels. By comparing the operating status of each device, the neural network can provide an appropriate load distribution plan for each device to achieve load balancing.

[0139] Operation content:

[0140] Real-time optimization: Based on the real-time data input, the model automatically updates the weights and adjusts the load output of each drive device to maximize the overall operating efficiency of the equipment.

[0141] Load balancing algorithm: distributes the load to multiple drives to balance system resources.

[0142] Design 3 modes:

[0143] Maximum load carrying capacity Load balancing;

[0144] Minimum power consumption load balancing;

[0145] Combination of load balancing and power optimization;

[0146] The combination of intermediate load balancing and power optimization is a model-based load forecasting, which uses a time series forecasting model to predict future load and power requirements (based on the output results of each channel input model). Load distribution is performed based on the forecast results, and scheduling is performed in advance to adapt to future needs.

[0147] The overall design is a comprehensive scheduling strategy that aims to optimize both load and power consumption, and improve the overall efficiency of the system by balancing the load and power consumption of the equipment. Since the relationship between power consumption and load is usually nonlinear, it is necessary to establish a model through a historical time series database to ensure that the load of each device (face application or belt application) is as uniform as possible to prevent some equipment from being overloaded.

[0148] The relationship between the power consumption Pi and the load Li of each device i can be expressed as a nonlinear function Pi=f(Li), where f is a specific formula fitted by each device. In order to consider both power consumption and load balancing, the following comprehensive optimization objective is used for the load distribution objective function:

[0149] ;

[0150] Constraints:

[0151] Total load distribution constraints:

[0152] ;

[0153] The load of each device does not exceed its maximum load carrying capacity:

[0154] ;

[0155] Finally, initialize the initialization function and assign the initial load to each device. , and then the gradient descent is used to adjust Li (control the coal falling, or speed up the belt, etc., which needs to be learned through labels) to minimize the total power consumption while satisfying the load distribution constraints.

[0156] 5. Optimize the data dynamic upgrade model of each channel:

[0157] During system operation, the neural network model will continue to receive real-time data from each channel and dynamically adjust and upgrade according to the load changes and operating status of the equipment. The model continuously learns and optimizes through real-time input data. As the operating status of the equipment changes, the model is constantly updated to ensure that each device is always in the optimal load state. Through the online learning mechanism of continuous labeling (online connection to the equipment, online monitoring of equipment status, and equipment operation is online monitoring), the model can iteratively update according to the latest data of the equipment, gradually improving the prediction accuracy and load distribution efficiency. Complete self-adjustment.

[0158] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A self-learning controller based on a neural network, characterized in that: It includes a neural network self-learning module, a state prediction and evaluation module, an input leveling module, a dynamic expansion and adjustment module, and a deployment and debugging module that are connected in communication in sequence. The neural network self-learning module is used to realize adaptive learning through inputting device operation data and feedback results after multiple trainings; the state prediction and evaluation module is used to analyze the historical data of the device so that the controller can predict the future operation status of the device; the input leveling module is used to perform load balancing adjustment among multiple controllers or drive devices; the dynamic expansion and adjustment module can support dynamic addition of new devices or adjustment of the load of existing devices without stopping or redesigning the control strategy; the deployment and debugging module is used to automatically adjust the control parameters to reduce the need for manual intervention and debugging.

2. A neural network-based self-learning controller according to claim 1, characterized in that: In the neural network self-learning module, the control logic is: The neural network model can automatically adjust the control strategy based on historical operating data and real-time status, ensuring that load distribution and drive load system adjustment are adaptive and do not rely on artificially set rules or empirical models; The self-learning process can dynamically adjust the working status of each device according to the load characteristics and operating status of different devices, ensuring that each device operates within the optimal load range.

3. A neural network-based self-learning controller according to claim 1, characterized in that: In the state prediction and evaluation module, the control logic is: By analyzing the historical data of the equipment, the controller can predict the future operation status of the equipment, including: Trends in equipment performance degradation: By analyzing current equipment load and past operating data, the controller can determine in advance whether equipment performance degradation or failure is likely; Fault risk warning: When a possible equipment failure is predicted, the drive load system can issue a warning in advance to remind operators to perform maintenance or adjustments.

4. A neural network-based self-learning controller according to claim 1, characterized in that: In the input leveling module, the control logic is: The drive load system can balance the load between multiple controllers or drive devices, ensuring that the load input of the devices is reasonably distributed to avoid overloading of a single device or waste of resources.

5. The neural network-based self-learning controller according to claim 1, characterized in that: In the dynamic expansion and adjustment module, the control logic is: The drive load system supports dynamically adding new equipment or adjusting the load of existing equipment without stopping or redesigning the control strategy; reducing commissioning and deployment time; In the deployment and debugging module, the control logic is: Through the adaptive capabilities of neural networks, the system debugging process is simplified and human intervention can be reduced.

6. A method of a self-learning controller based on a neural network, characterized in that: The method is implemented by a self-learning controller based on a neural network, and the method comprises the following steps: S1, obtain data through the existing controller; S2, preprocessing the data of the data channel; S3, preload the data of each channel to establish a model; S4, updating the model for driving load optimization; S5. Optimize the data dynamic upgrade model of each channel; In step S1, obtaining data through the existing controller includes: In the working face control and belt drive control, the existing controller is responsible for obtaining the operating status data of the equipment, including information: Current, voltage, and temperature sensor data; Equipment load, speed drive performance data; Equipment start and stop data, operating time, shutdown times, fault points, and fault frequency historical operating data; The above data will be provided as input to the neural network of the self-learning controller for analysis and learning.

7. The method of a neural network-based self-learning controller according to claim 6, characterized in that: In step S2, the data of the data channel is preprocessed, including: Data preprocessing refers to the process of cleaning, standardizing and normalizing the raw data obtained from the traditional controller sensors and various feedback modules. The drive load system first uses the communication port connected to the traditional controller to obtain the operating data of the equipment; Establish a time series database and obtain equipment data through the time series database. Define the equipment as a continuous time series, define the equipment load data L(t) and power consumption P(t), where t represents time; the equipment load data is obtained from the coal quantity sensor until the load data is obtained, and the power consumption P data is input at the same time, and piecewise fitting is performed, which is expressed as: ； L is the load; L1 is the load dividing point; a1, b1c1 are constants used for nonlinear relationships in the first interval (L≤I L1); a2, b2c2 are constants used for nonlinear relationships in the first interval (L>L1); abc is a constant fitted based on historical data. I is divided into 3-5 stages for fitting and the data is stored.

8. The method of a neural network-based self-learning controller according to claim 6, characterized in that: In step S3, the data of each channel is preloaded to establish a model, including: Preloading refers to inputting preprocessed data into the neural network for preliminary training. The operation content is: Load multi-channel data in batches, and input the data of each channel into the neural network model for preliminary learning and training; Learning is done based on the channels and manual operations of the on-site monitoring equipment. Unsupervised learning is used first, and supervised learning is done through historical data. The model is adaptively adjusted based on the data pattern. The input data records P and L and the corresponding fitting data, records the label jump data, stores it in the database, and gives prompts to the operator, stores it in the database, makes judgments later, and finally completes the labeled learning for self-optimization and upgrade.

9. The method of a neural network-based self-learning controller according to claim 6, characterized in that: In step S4, the model is updated for driving load optimization, including: After the initial training is completed, the neural network model can output load optimization suggestions based on multi-channel data; by comparing the operating status of each device, the neural network can provide a load distribution plan for each device. The operation content is: Real-time optimization: Based on the real-time data input, the model automatically updates the weights and adjusts the load output of each drive device to maximize the overall operating efficiency of the equipment; Load balancing algorithm: distributes the load to multiple drives to balance system resources; Design three modes: load balancing with maximum load carrying capacity, load balancing with minimum power consumption, and load balancing combined with power optimization: The combination of intermediate load balancing and power optimization is a model-based load forecasting, which uses a time series forecasting model to predict future load and power requirements; load distribution is performed based on the forecast results, and scheduling is performed in advance to adapt to future needs; The relationship between the power consumption Pi and the load Li of each device i is defined as a nonlinear function Pi=f(Li), where f is a specific formula fitted by each device; the load distribution objective function is performed using the following comprehensive optimization objective: ； Constraints: Total load distribution constraints: ； The load of each device does not exceed its maximum load carrying capacity: ； Finally, initialize the initialization function to assign initial load to each device , and adjust Li by gradient descent to minimize the total power consumption while satisfying the load distribution constraints.

10. The method of a neural network-based self-learning controller according to claim 6, characterized in that: In step S5, the data dynamic update model of each channel is optimized, including: During the operation of the driving load system, the neural network model can continue to receive real-time data from each channel and dynamically adjust and upgrade according to the load changes and operating status of the equipment; The model continuously learns and optimizes through real-time input data. As the operating status of the equipment changes, the model is constantly updated to ensure that each device is always in the optimal load state. Through the online learning mechanism of continuous labeling, the model can be iteratively updated according to the latest data of the equipment, gradually improving the prediction accuracy and load distribution efficiency, and fully self-adjusting.

Images