An intelligent multi-stage wind power coal separation parameter optimization method and system

Abstract

The application discloses an intelligent multi-stage wind force coal separation parameter optimization method and system, which comprises multi-dimensional data acquisition, data optimization processing, state prediction model construction, parameter optimization model construction and separation parameter optimization. The original data is obtained through multi-dimensional data acquisition; the data optimization processing method of signal synchronization and alignment, noise suppression, feature extraction and fusion, data standardization and data set segmentation is adopted; the improved deep network model is used as the state prediction model, and through the coal quality fluctuation mechanism and adaptive feature perception, the separation state evolution trend containing the process significance can be analyzed, and high-quality and interpretable input is provided for the subsequent parameter optimization strategy; the hierarchical reinforcement learning model is used as the parameter optimization model, the coal quality state can be predicted in real time, the parameters can be intelligently adjusted under the rigid constraint of guaranteeing the separation efficiency, the long-term cumulative energy consumption is optimized, and the balance between the separation efficiency and the energy consumption is achieved.

Description

Technical Field

[0001] This invention relates to the field of multi-stage wind-powered coal preparation parameter optimization technology, specifically to an intelligent multi-stage wind-powered coal preparation parameter optimization method and system. Background Technology

[0002] Optimization of wind-powered coal preparation parameters refers to the process of rationally adjusting and optimizing key parameters such as air volume, vibration frequency, and air pressure during wind-powered coal preparation to maximize separation efficiency, minimize energy consumption, and optimize product quality. This can significantly improve the economic benefits of coal preparation plants, reduce production costs, and simultaneously increase clean coal recovery rate and product quality stability.

[0003] However, traditional multi-stage wind-powered coal preparation parameter optimization methods rely on predefined fixed rules, empirical thresholds, or shallow models for state judgment, making it difficult to effectively handle coal quality fluctuations, equipment state changes, and nonlinear coupling relationships between multiple parameters. This leads to technical problems such as parameter adjustment lag and unstable separation effect. Furthermore, traditional multi-stage wind-powered coal preparation parameter optimization methods often employ simple controllers based on fixed rules, static scheduling, or single-objective optimization, lacking the ability to perceive and adapt to complex and dynamic operating environments. They also cannot handle the multi-objective trade-off between separation efficiency and energy consumption, making it difficult to achieve optimal control. Summary of the Invention

[0004] To address the above issues and overcome the shortcomings of existing technologies, this invention provides an intelligent multi-stage wind-powered coal preparation parameter optimization method and system. Traditional multi-stage wind-powered coal preparation parameter optimization methods rely on predefined fixed rules, empirical thresholds, or shallow models for state judgment, making it difficult to effectively handle coal quality fluctuations, equipment state changes, and nonlinear coupling relationships between multiple parameters, leading to lag in parameter adjustment and unstable separation results. This solution creatively employs an improved deep network model as the state prediction model. By embedding a coal quality fluctuation mechanism and adaptive feature perception, it can extract the process-meaning evolution trend of the separation state from multi-sensor signals, while simultaneously integrating... The quantity-sensitive constraints provide high-quality, interpretable input for subsequent parameter optimization strategies. Addressing the technical challenges of traditional multi-stage wind-powered coal preparation parameter optimization methods, which often rely on simple controllers based on fixed rules, static scheduling, or single-objective optimization, lacking the ability to perceive and adapt to complex and dynamic operating environments, and unable to handle the multi-objective trade-off between sorting efficiency and energy consumption, thus hindering optimal control, this solution creatively employs a hierarchical reinforcement learning model as the parameter optimization model. Based on real-time predicted coal quality conditions, it intelligently allocates parameters such as airflow and vibration frequency under the rigid constraint of ensuring sorting efficiency, ultimately achieving optimization of long-term cumulative energy consumption and a balance between sorting efficiency and energy consumption.

[0005] The technical solution adopted by this invention is as follows: This invention provides an intelligent multi-stage wind-powered coal preparation parameter optimization method, which includes the following steps:

[0006] Step S1: Multidimensional data acquisition;

[0007] Step S2: Data optimization processing;

[0008] Step S3: Construction of the state prediction model;

[0009] Step S4: Parameter optimization model construction;

[0010] Step S5: Optimize sorting parameters.

[0011] Further, in step S1, the multidimensional data acquisition is used to collect multidimensional raw data required for multi-stage wind-powered coal preparation and separation parameter optimization. Specifically, through data acquisition, a parameter optimization raw dataset is obtained. The parameter optimization raw dataset specifically includes a parameter optimization historical raw dataset and a parameter optimization current raw dataset. Both the parameter optimization historical raw dataset and the parameter optimization current raw dataset contain raw coal property data, equipment operation data, process parameter data, and environmental data.

[0012] Further, in step S2, the data optimization processing is used to optimize the collected raw data, specifically including the following steps:

[0013] Step S21: Signal synchronization and alignment, used to eliminate time deviation of multi-channel data. Specifically, the dynamic time warping algorithm is used to align multi-channel data with different sampling rates and transmission delays to a unified time base to obtain a time-consistent data sequence.

[0014] Step S22: Noise suppression, used to improve signal quality, specifically by using wavelet transform combined with an adaptive threshold denoising method, and using an adaptive filter based on accelerometer signals to process vibration noise, resulting in high-quality, low-noise process parameter data;

[0015] Step S23: Feature extraction and fusion, used to extract meaningful features, specifically time domain, frequency domain and nonlinear dynamic features, and retain key features through mutual information feature selection algorithm to obtain a feature set with low redundancy and high information content;

[0016] Step S24: Data standardization, used to handle missing data and standardize it. Specifically, a time series generative adversarial network is used to synthesize reasonable missing data segments, and the data is normalized to obtain complete and standardized feature data.

[0017] Step S25: Dataset segmentation, used to obtain training data and validation data, specifically involves segmenting the original dataset of the parameter optimization history.

[0018] The current parameter optimization dataset is processed by signal synchronization and alignment, noise suppression, feature extraction and fusion, and data standardization to obtain the current parameter optimization dataset. The historical parameter optimization dataset is processed by signal synchronization and alignment, noise suppression, feature extraction and fusion, data standardization, and dataset segmentation to obtain the parameter optimization training set and parameter optimization test set.

[0019] Further, in step S3, the state prediction model is constructed to build a model that can accurately predict the coal quality state during multi-stage wind-powered coal preparation. Specifically, an improved deep network model is constructed as the state prediction model. The improved deep network model performs state prediction by combining the coal quality oscillation mechanism, time-varying convolution field and energy-sensitive sparse coding to obtain the state prediction result.

[0020] The construction of the state prediction model specifically includes the following steps:

[0021] Step S31: Design of a coal quality oscillation module to simulate coal quality fluctuation characteristics and their coupling. The steps include:

[0022] Step S311: Initialize coal quality oscillation parameters to set the initial state of the oscillator. Specifically, by setting the natural frequency, initial phase and coupling strength parameters of each oscillator, an oscillator system that can simulate the real coal quality fluctuation law is constructed to obtain the initialized oscillator state.

[0023] Step S312: Oscillator coupling, used to define the interaction between oscillators. Specifically, it learns the interaction strength between each oscillator through a neural network, simulates the coupling relationship between different coal quality parameters, and obtains a dynamically changing coupling strength matrix.

[0024] Step S313: Achieve phase synchronization, which is used to evaluate and adjust the synchronization degree of the oscillator. Specifically, the synchronization degree of the oscillator system is evaluated by calculating the average phase difference, and an adjustment mechanism is designed to maintain an appropriate synchronization level to obtain a stable phase synchronization state.

[0025] Step S32: Time-varying convolutional field design, used to extract phase-sensitive features. Specifically, multiply the basic convolutional kernel element-wise with a sine function modulated by the oscillator phase to obtain a multi-scale time-varying convolutional kernel. Convolve the input data with time-varying convolutional kernels of different scales respectively. Calculate the weights of the output features of the time-varying convolutional kernels of each scale based on the attention mechanism, and sum them up to obtain multi-scale fused features.

[0026] Step S33: Energy-sensitive sparse coding, used for feature selection under energy cost constraints. Specifically, a regularization term for feature energy cost is added to the objective function of traditional sparse coding, so that the coding process considers both the information content and energy cost of the features, and obtains energy-sensitive sparse coefficients.

[0027] Step S34: Construct a coal quality state feature vector to predict coal quality state. Specifically, sparse coefficients are incorporated into the multi-scale fusion features and processed by an independent multilayer perceptron to obtain the coal quality state feature vector, which is used as the state prediction result output by the model.

[0028] Step S35: Construct and train the model. Specifically, the improved deep network model is constructed by integrating the coal quality oscillation module design, the time-varying convolutional field design, the energy-sensitive sparse coding, and the construction of coal quality state feature vectors. The model is trained and its performance is verified based on the parameter optimization training set and the parameter optimization test set to obtain the improved deep network model as the state prediction model.

[0029] Furthermore, in step S4, the parameter optimization model construction is used to construct the model required for multi-stage wind-powered coal preparation and separation parameter optimization, specifically by constructing a hierarchical reinforcement learning model as the parameter optimization model;

[0030] The construction of the parameter optimization model specifically includes the following steps:

[0031] Step S41: State space design, used to define the state representation of reinforcement learning, specifically by combining the coal quality state feature vector, the oscillator feature vector, the equipment operation feature, the process parameter feature and the environmental feature into a state vector to obtain the state space of reinforcement learning;

[0032] Step S42: Action space design, used to define the actions of reinforcement learning. Specifically, the actions are divided into three levels: high-level macro policies, mid-level mode selection, and low-level parameter fine-tuning, to obtain a hierarchical action space.

[0033] The macro strategy at the higher level is used to design a time segmentation strategy, specifically by dividing a future time window into multiple sub-intervals of unequal length and assigning an operating mode to each sub-interval to obtain a time resource allocation scheme.

[0034] The middle layer mode selection is used to define the operating mode library. Specifically, it predefines multiple equipment operating modes with different energy consumption and sorting performance characteristics. Each equipment operating mode includes a specific air volume level, vibration frequency level, and algorithm complexity, resulting in a set of selectable equipment operating modes, providing options for time segmentation.

[0035] The lower-level parameter fine-tuning specifically involves continuously fine-tuning parameters, including but not limited to airflow and vibration frequency, based on the selected equipment operating mode to obtain refined parameter settings.

[0036] Step S43: Reward function design, used to design the reward function for reinforcement learning, specifically to construct a multi-objective reward function that includes three aspects: energy efficiency, product quality, and equipment health, to obtain a comprehensive reward function;

[0037] Step S44: Design a hierarchical policy optimization algorithm to optimize the hierarchical policy. Specifically, update the policy parameters of the high-level, middle-level and low-level layers through the policy gradient method to obtain the optimized hierarchical policy.

[0038] Step S45: Construct a parameter optimization model, specifically by integrating the state space design, action space design, reward function design, and hierarchical policy optimization algorithm to construct a hierarchical reinforcement learning model, which serves as the parameter optimization model.

[0039] Further, in step S5, the sorting parameter optimization specifically involves using the current parameter optimization dataset as input to the state prediction model to predict the coal quality state, obtaining the state prediction result, and the parameter optimization model performing multi-level wind-powered coal preparation sorting parameter optimization based on the state prediction result, with the optimized stratification strategy being used as the sorting parameter optimization decision result.

[0040] This invention provides an intelligent multi-stage wind-powered coal preparation and separation parameter optimization system, comprising a multi-dimensional data acquisition module, a data optimization processing module, a state prediction model construction module, a parameter optimization model construction module, and a separation parameter optimization module;

[0041] The multidimensional data acquisition module is used to acquire raw data, obtain parameter-optimized raw dataset by acquiring raw data, and send the parameter-optimized raw dataset to the data optimization processing module.

[0042] The data optimization processing module is used for data optimization processing. Through data optimization processing, it obtains the current parameter optimization dataset, parameter optimization training set, and parameter optimization test set, and sends the current parameter optimization dataset to the sorting parameter optimization module, and sends the parameter optimization training set and the parameter optimization test set to the state prediction model construction module.

[0043] The state prediction model construction module is used to construct a state prediction model. By constructing an improved deep network model, a state prediction model is obtained, and the state prediction model is sent to the sorting parameter optimization module.

[0044] The parameter optimization model construction module is used to construct the parameter optimization model. By constructing a hierarchical reinforcement learning model, the parameter optimization model is obtained, and the parameter optimization model is sent to the sorting parameter optimization module.

[0045] The sorting parameter optimization module is used to optimize sorting parameters. It processes the current data by using the state prediction model in combination with the parameter optimization model and outputs the sorting parameter optimization decision result.

[0046] The beneficial effects achieved by the present invention using the above solution are as follows:

[0047] (1) In view of the technical problems of traditional multi-stage wind-powered coal preparation parameter optimization methods, which rely on predefined fixed rules, empirical thresholds or shallow models for state judgment, it is difficult to effectively handle coal quality fluctuations, equipment state changes and nonlinear coupling relationships between multiple parameters, resulting in parameter adjustment lag and unstable sorting effect. This solution creatively adopts an improved deep network model as the state prediction model. By embedding the coal quality fluctuation mechanism and adaptive feature perception, it can analyze the sorting state evolution trend with process significance from the multi-sensor signals. At the same time, it integrates energy-sensitive constraints to provide high-quality and interpretable input for subsequent parameter optimization strategies.

[0048] (2) In view of the technical problems of traditional multi-stage wind-powered coal preparation parameter optimization methods, which are mostly simple controllers based on fixed rules, static scheduling or single-objective optimization, lack the ability to perceive and adapt to complex and dynamic operating environments, and cannot handle the multi-objective trade-off between sorting efficiency and energy consumption, making it difficult to achieve optimal control, this scheme creatively adopts a hierarchical reinforcement learning model as the parameter optimization model. Based on the real-time predicted coal quality status, under the rigid constraint of ensuring sorting efficiency, it can intelligently allocate parameters such as air volume and vibration frequency, and finally achieve the optimization of long-term cumulative energy consumption, thus achieving a balance between sorting efficiency and energy consumption. Attached Figure Description

[0049] Figure 1 A flowchart illustrating an intelligent multi-stage wind-powered coal preparation parameter optimization method provided by the present invention;

[0050] Figure 2 A schematic diagram of a module for an intelligent multi-stage wind-powered coal preparation and separation parameter optimization system provided by the present invention;

[0051] Figure 3 A flowchart illustrating the data optimization process in step S2;

[0052] Figure 4 A flowchart illustrating the process of building the state prediction model for step S3;

[0053] Figure 5A schematic diagram of the process for constructing the parameter optimization model in step S4.

[0054] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0055] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0056] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0057] Example 1, see Figure 1 This invention provides an intelligent multi-stage wind-powered coal preparation parameter optimization method, which includes the following steps:

[0058] Step S1: Multidimensional data acquisition;

[0059] Step S2: Data optimization processing;

[0060] Step S3: Construction of the state prediction model;

[0061] Step S4: Parameter optimization model construction;

[0062] Step S5: Optimize sorting parameters.

[0063] Example 2, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S1, the multidimensional data acquisition is used to collect multidimensional raw data required for multi-stage wind-powered coal preparation and sorting parameter optimization. Specifically, through data acquisition, the parameter optimization raw dataset is obtained. The parameter optimization raw dataset specifically includes the parameter optimization historical raw dataset and the parameter optimization current raw dataset. Both the parameter optimization historical raw dataset and the parameter optimization current raw dataset contain raw coal property data, equipment operation data, process parameter data, and environmental data.

[0064] The raw coal property data specifically includes particle size distribution data, density distribution data, ash content data, moisture content data, and gangue content data. The equipment operation data specifically includes fan speed data, vibrator frequency data, air pressure data, current data, and equipment temperature data. The process parameter data specifically includes feed rate data, air volume distribution data, discharge cycle data, and bed thickness data. The environmental data specifically includes ambient temperature and humidity data and air pressure data.

[0065] Example 3, see Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above embodiment. In step S2, the data optimization processing is used to optimize the collected raw data, specifically including the following steps:

[0066] Step S21: Signal synchronization and alignment, used to eliminate time deviation of multi-channel data. Specifically, the dynamic time warping algorithm is used to align multi-channel data with different sampling rates and transmission delays to a unified time base to obtain a time-consistent data sequence.

[0067] Step S22: Noise suppression, used to improve signal quality, specifically by using wavelet transform combined with an adaptive threshold denoising method, and using an adaptive filter based on accelerometer signals to process vibration noise, resulting in high-quality, low-noise process parameter data;

[0068] Step S23: Feature extraction and fusion, used to extract meaningful features, specifically time domain, frequency domain and nonlinear dynamic features, and retain key features through mutual information feature selection algorithm to obtain a feature set with low redundancy and high information content;

[0069] Step S24: Data standardization, used to handle missing data and standardize it. Specifically, a time series generative adversarial network is used to synthesize reasonable missing data segments, and the data is normalized to obtain complete and standardized feature data.

[0070] Step S25: Dataset segmentation, used to obtain training data and validation data, specifically involves segmenting the original dataset of the parameter optimization history.

[0071] The current parameter optimization dataset is processed by signal synchronization and alignment, noise suppression, feature extraction and fusion, and data standardization to obtain the current parameter optimization dataset. The historical parameter optimization dataset is processed by signal synchronization and alignment, noise suppression, feature extraction and fusion, data standardization, and dataset segmentation to obtain the parameter optimization training set and parameter optimization test set.

[0072] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the state prediction model is constructed to build a model that can accurately predict the coal quality state during multi-stage wind-powered coal preparation. Specifically, an improved deep network model is constructed as the state prediction model. The improved deep network model performs state prediction by combining the coal quality oscillation mechanism, time-varying convolution field and energy-sensitive sparse coding to obtain the state prediction result.

[0073] The construction of the state prediction model specifically includes the following steps:

[0074] Step S31: Design of a coal quality oscillation module to simulate coal quality fluctuation characteristics and their coupling. The steps include:

[0075] Step S311: Initialize coal quality oscillation parameters to set the initial state of the oscillator. Specifically, by setting the natural frequency, initial phase, and coupling strength parameters of each oscillator, an oscillator system that can simulate the real coal quality fluctuation law is constructed to obtain the initialized oscillator state. The formula used is as follows:

[0076] ;

[0077] In the formula, This represents the phase of the a-th oscillator. This represents the natural frequency of the a-th oscillator, initialized using the dominant fluctuation frequency from historical coal quality data. This represents the coupling strength between the a-th oscillator and the b-th oscillator, initialized by calculating the mutual information between the oscillator outputs. This represents the phase of the b-th oscillator. This represents the phase offset between the a-th oscillator and the b-th oscillator. Let represent the external disturbance term of the a-th oscillator, which is Gaussian white noise with a mean of 0;

[0078] Step S312: Oscillator Coupling, used to define the interaction between oscillators. Specifically, it involves learning the interaction strength between each oscillator through a neural network, simulating the coupling relationship between different coal quality parameters, and obtaining a dynamically changing coupling strength matrix. The formula used is as follows:

[0079] ;

[0080] In the formula, This represents the coupling strength between the a-th oscillator and the b-th oscillator, obtained through the neural network. This represents the sigmoid function. Represents the coupling weight matrix. Indicates the coupling bias term. This represents the eigenvector of the a-th oscillator. This represents the eigenvector of the b-th oscillator. Indicates a splicing operation;

[0081] Step S313: Achieve phase synchronization, which is used to evaluate and adjust the synchronization degree of the oscillator. Specifically, the synchronization degree of the oscillator system is evaluated by calculating the average phase difference, and an adjustment mechanism is designed to maintain an appropriate synchronization level to obtain a stable phase synchronization state.

[0082] Step S32: Time-varying convolutional field design, used to extract phase-sensitive features. Specifically, multiply the basic convolutional kernel element-wise with a sine function modulated by the oscillator phase to obtain a multi-scale time-varying convolutional kernel. Convolve the input data with time-varying convolutional kernels of different scales respectively. Calculate the weights of the output features of the time-varying convolutional kernels of each scale based on the attention mechanism, and sum them up to obtain multi-scale fused features.

[0083] Step S33: Energy-sensitive sparse coding, used for feature selection under energy cost constraints. Specifically, a regularization term for feature energy cost is added to the objective function of traditional sparse coding, so that the encoding process considers both the information content and energy cost of the features, resulting in energy-sensitive sparse coefficients. The formula used is as follows:

[0084] ;

[0085] In the formula, Let x represent the sparse coding objective function, and let x represent the input independent variable. This indicates the calculation of the L2 norm. This indicates the computation of the L1 norm, where Dm represents the pre-trained dictionary matrix, and Sc represents the sparse coefficient vector. and These represent distinct sparse coding regularization parameters. This represents the energy cost of the c-th dimension of the input feature vector. This represents the sparse coefficients of the c-th dimension of the sparse coefficient vector;

[0086] Step S34: Construct a coal quality state feature vector to predict coal quality state. Specifically, sparse coefficients are incorporated into the multi-scale fusion features and processed by an independent multilayer perceptron to obtain the coal quality state feature vector, which is used as the state prediction result output by the model.

[0087] Step S35: Construct and train the model. Specifically, the improved deep network model is constructed by integrating the coal quality oscillation module design, the time-varying convolutional field design, the energy-sensitive sparse coding, and the construction of coal quality state feature vectors. The model is trained and its performance is verified based on the parameter optimization training set and the parameter optimization test set to obtain the improved deep network model as the state prediction model.

[0088] By performing the above operations, this solution addresses the technical problems of traditional multi-stage wind-powered coal preparation parameter optimization methods, which rely on predefined fixed rules, empirical thresholds, or shallow models for state judgment. These methods struggle to effectively handle coal quality fluctuations, equipment state changes, and nonlinear coupling relationships between multiple parameters, leading to lag in parameter adjustment and unstable sorting results. This solution creatively adopts an improved deep network model as the state prediction model. By embedding a coal quality fluctuation mechanism and adaptive feature perception, it can extract the evolution trend of the sorting state with process significance from multi-sensor signals. At the same time, it integrates energy-sensitive constraints, providing high-quality and interpretable input for subsequent parameter optimization strategies.

[0089] Example 5, see Figure 1 , Figure 2 and Figure 5 This embodiment is based on the above embodiment. In step S4, the parameter optimization model is constructed to build the model required for multi-stage wind-powered coal preparation and separation parameter optimization. Specifically, it is constructed by building a hierarchical reinforcement learning model as the parameter optimization model.

[0090] The construction of the parameter optimization model specifically includes the following steps:

[0091] Step S41: State space design, used to define the state representation of reinforcement learning, specifically by combining the coal quality state feature vector, the oscillator feature vector, the equipment operation feature, the process parameter feature and the environmental feature into a state vector to obtain the state space of reinforcement learning;

[0092] Step S42: Action space design, used to define the actions of reinforcement learning. Specifically, the actions are divided into three levels: high-level macro policies, mid-level mode selection, and low-level parameter fine-tuning, to obtain a hierarchical action space.

[0093] The macro strategy at the higher level is used to design a time segmentation strategy, specifically by dividing a future time window into multiple sub-intervals of unequal length and assigning an operating mode to each sub-interval to obtain a time resource allocation scheme.

[0094] The middle layer mode selection is used to define the operating mode library. Specifically, it predefines multiple equipment operating modes with different energy consumption and sorting performance characteristics. Each equipment operating mode includes a specific air volume level, vibration frequency level, and algorithm complexity, resulting in a set of selectable equipment operating modes, providing options for time segmentation.

[0095] The lower-level parameter fine-tuning specifically involves continuously fine-tuning parameters, including but not limited to airflow and vibration frequency, based on the selected equipment operating mode to obtain refined parameter settings.

[0096] Step S43: Reward function design, used to design the reward function for reinforcement learning. Specifically, it involves constructing a multi-objective reward function that includes three aspects: energy efficiency, product quality, and equipment health, to obtain a comprehensive reward function. The formula used is as follows:

[0097] ;

[0098] In the formula, This indicates an energy efficiency bonus. This indicates a product quality reward. Indicates a health reward for the device. Indicates actual energy consumption. Indicates the baseline energy consumption. Indicates the energy efficiency coefficient. Represents the information gain at time t. This represents the multilayer perceptron function used to output the sorting urgency score. This represents the multilayer perceptron function used to output the coal quality risk score, where PS represents the coal quality state feature vector. This represents the wear penalty weighting coefficient. This represents the overload penalty weighting coefficient. Indicates equipment wear and tear penalty. This indicates a penalty for equipment overload. This represents the value of the comprehensive reward function. This represents the energy efficiency reward weighting coefficient. This represents the product quality reward weighting coefficient. This represents the weighting coefficient for equipment health rewards;

[0099] Step S44: Design a hierarchical policy optimization algorithm to optimize the hierarchical policy. Specifically, update the policy parameters of the high-level, middle-level and low-level layers through the policy gradient method to obtain the optimized hierarchical policy.

[0100] Step S45: Construct a parameter optimization model, specifically by integrating the state space design, action space design, reward function design, and hierarchical policy optimization algorithm to construct a hierarchical reinforcement learning model, which serves as the parameter optimization model.

[0101] By performing the above operations, this solution addresses the technical problems of traditional multi-stage wind-powered coal preparation parameter optimization methods, which often rely on simple controllers based on fixed rules, static scheduling, or single-objective optimization. These methods lack the ability to perceive and adapt to complex and dynamic operating environments and cannot handle the multi-objective trade-off between sorting efficiency and energy consumption, making it difficult to achieve optimal control. This solution creatively adopts a hierarchical reinforcement learning model as the parameter optimization model. Based on real-time predicted coal quality conditions, it can intelligently allocate parameters such as air volume and vibration frequency under the rigid constraint of ensuring sorting efficiency, ultimately achieving the optimization of long-term cumulative energy consumption and achieving a balance between sorting efficiency and energy consumption.

[0102] Example 6, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S5, the sorting parameter optimization specifically involves using the current parameter optimization dataset as the input of the state prediction model to predict the coal quality state and obtain the state prediction result. Based on the state prediction result, the parameter optimization model performs multi-level wind-powered coal preparation sorting parameter optimization, and the resulting optimized stratification strategy is used as the sorting parameter optimization decision result.

[0103] Example 7, see Figure 1 and Figure 2 Based on the above embodiments, this embodiment provides an intelligent multi-stage wind-powered coal preparation and sorting parameter optimization system, including a multi-dimensional data acquisition module, a data optimization processing module, a state prediction model construction module, a parameter optimization model construction module, and a sorting parameter optimization module;

[0104] The multidimensional data acquisition module is used to acquire raw data, obtain parameter-optimized raw dataset by acquiring raw data, and send the parameter-optimized raw dataset to the data optimization processing module.

[0105] The data optimization processing module is used for data optimization processing. Through data optimization processing, it obtains the current parameter optimization dataset, parameter optimization training set, and parameter optimization test set, and sends the current parameter optimization dataset to the sorting parameter optimization module, and sends the parameter optimization training set and the parameter optimization test set to the state prediction model construction module.

[0106] The state prediction model construction module is used to construct a state prediction model. By constructing an improved deep network model, a state prediction model is obtained, and the state prediction model is sent to the sorting parameter optimization module.

[0107] The parameter optimization model construction module is used to construct the parameter optimization model. By constructing a hierarchical reinforcement learning model, the parameter optimization model is obtained, and the parameter optimization model is sent to the sorting parameter optimization module.

[0108] The sorting parameter optimization module is used to optimize sorting parameters. It processes the current data by using the state prediction model in combination with the parameter optimization model and outputs the sorting parameter optimization decision result.

[0109] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0110] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0111] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. An intelligent method for optimizing multi-stage wind-powered coal preparation and separation parameters, characterized in that: The method includes the following steps: S1: Multidimensional data acquisition. Through data acquisition, the original dataset for parameter optimization is obtained. Specifically, the original dataset for parameter optimization includes the historical original dataset for parameter optimization and the current original dataset for parameter optimization. S2: Data optimization processing, which optimizes the collected raw data to obtain the current parameter optimization dataset, parameter optimization training set, and parameter optimization test set; S3: State prediction model construction, used to construct a model that can accurately predict the coal quality state during multi-stage wind-powered coal preparation. Specifically, an improved deep network model is constructed as the state prediction model. The improved deep network model performs state prediction by combining the coal quality oscillation mechanism, time-varying convolution field and energy-sensitive sparse coding to obtain the state prediction result. S4: Parameter optimization model construction, used to build the model required for multi-stage wind-powered coal preparation and separation parameter optimization, specifically by constructing a hierarchical reinforcement learning model as the parameter optimization model; S5: Sorting parameter optimization, specifically, the current parameter optimization dataset is used as the input of the state prediction model to predict the coal quality state and obtain the state prediction result. Based on the state prediction result, the parameter optimization model performs multi-level wind-powered coal preparation sorting parameter optimization, and the optimized hierarchical strategy is used as the sorting parameter optimization decision result.

2. The intelligent multi-stage wind-powered coal preparation parameter optimization method according to claim 1, characterized in that: The construction of the state prediction model specifically includes the following steps: Step S31: Design of a coal quality oscillation module to simulate coal quality fluctuation characteristics and their coupling. The steps include: Step S311: Initialize coal quality oscillation parameters to set the initial state of the oscillator. Specifically, by setting the natural frequency, initial phase and coupling strength parameters of each oscillator, an oscillator system that can simulate the real coal quality fluctuation law is constructed to obtain the initialized oscillator state. Step S312: Oscillator coupling, used to define the interaction between oscillators. Specifically, it learns the interaction strength between each oscillator through a neural network, simulates the coupling relationship between different coal quality parameters, and obtains a dynamically changing coupling strength matrix. Step S313: Achieve phase synchronization, which is used to evaluate and adjust the synchronization degree of the oscillator. Specifically, the synchronization degree of the oscillator system is evaluated by calculating the average phase difference, and an adjustment mechanism is designed to maintain an appropriate synchronization level to obtain a stable phase synchronization state. Step S32: Time-varying convolutional field design, used to extract phase-sensitive features. Specifically, multiply the basic convolutional kernel element-wise with a sine function modulated by the oscillator phase to obtain a multi-scale time-varying convolutional kernel. Convolve the input data with time-varying convolutional kernels of different scales respectively. Calculate the weights of the output features of the time-varying convolutional kernels of each scale based on the attention mechanism, and sum them up to obtain multi-scale fused features. Step S33: Energy-sensitive sparse coding, used for feature selection under energy cost constraints. Specifically, a regularization term for feature energy cost is added to the objective function of traditional sparse coding, so that the coding process considers both the information content and energy cost of the features, and obtains energy-sensitive sparse coefficients. Step S34: Construct a coal quality state feature vector to predict coal quality state. Specifically, sparse coefficients are incorporated into the multi-scale fusion features and processed by an independent multilayer perceptron to obtain the coal quality state feature vector, which is used as the state prediction result output by the model. Step S35: Construct and train the model. Specifically, the improved deep network model is constructed by integrating the coal quality oscillation module design, the time-varying convolutional field design, the energy-sensitive sparse coding, and the construction of coal quality state feature vectors. The model is trained and its performance is verified based on the parameter optimization training set and the parameter optimization test set to obtain the improved deep network model as the state prediction model.

3. The intelligent multi-stage wind-powered coal preparation parameter optimization method according to claim 1, characterized in that: The construction of the parameter optimization model specifically includes the following steps: Step S41: State space design, used to define the state representation of reinforcement learning, specifically by combining the coal quality state feature vector, the oscillator feature vector, the equipment operation feature, the process parameter feature, and the environmental feature into a state vector to obtain the state space of reinforcement learning; Step S42: Action space design, used to define the actions of reinforcement learning. Specifically, the actions are divided into three levels: high-level macro policies, mid-level mode selection, and low-level parameter fine-tuning, to obtain a hierarchical action space. The macro strategy at the higher level is used to design a time segmentation strategy, specifically by dividing a future time window into multiple sub-intervals of unequal length and assigning an operating mode to each sub-interval to obtain a time resource allocation scheme. The middle layer mode selection is used to define the operating mode library. Specifically, it predefines multiple equipment operating modes with different energy consumption and sorting performance characteristics. Each equipment operating mode includes a specific air volume level, vibration frequency level, and algorithm complexity, resulting in a set of selectable equipment operating modes, providing options for time segmentation. The lower-level parameter fine-tuning specifically involves continuously fine-tuning parameters, including but not limited to airflow and vibration frequency, based on the selected equipment operating mode to obtain refined parameter settings. Step S43: Reward function design, used to design the reward function for reinforcement learning, specifically to construct a multi-objective reward function that includes three aspects: energy efficiency, product quality, and equipment health, to obtain a comprehensive reward function; Step S44: Design a hierarchical policy optimization algorithm to optimize the hierarchical policy. Specifically, update the policy parameters of the high-level, middle-level and low-level layers through the policy gradient method to obtain the optimized hierarchical policy. Step S45: Construct a management optimization model, specifically by integrating the state space design, action space design, reward function design, and design hierarchical policy optimization algorithm to construct a hierarchical reinforcement learning model, which serves as the parameter optimization model.

4. The intelligent multi-stage wind-powered coal preparation parameter optimization method according to claim 1, characterized in that: Both the historical raw dataset for parameter optimization and the current raw dataset for parameter optimization contain raw coal property data, equipment operation data, process parameter data, and environmental data.

5. The intelligent multi-stage wind-powered coal preparation parameter optimization method according to claim 1, characterized in that: The data optimization process specifically includes the following steps: Step S21: Signal synchronization and alignment, used to eliminate time deviation of multi-channel data. Specifically, the dynamic time warping algorithm is used to align multi-channel data with different sampling rates and transmission delays to a unified time base to obtain a time-consistent data sequence. Step S22: Noise suppression, used to improve signal quality, specifically by using wavelet transform combined with an adaptive threshold denoising method, and using an adaptive filter based on accelerometer signals to process vibration noise, resulting in high-quality, low-noise process parameter data; Step S23: Feature extraction and fusion, used to extract meaningful features, specifically time domain, frequency domain and nonlinear dynamic features, and retain key features through mutual information feature selection algorithm to obtain a feature set with low redundancy and high information content; Step S24: Data standardization, used to handle missing data and standardize it. Specifically, a time series generative adversarial network is used to synthesize reasonable missing data segments, and the data is normalized to obtain complete and standardized feature data. Step S25: Dataset segmentation, used to obtain training data and validation data, specifically involves segmenting the original dataset of the parameter optimization history. The current parameter optimization dataset is processed by signal synchronization and alignment, noise suppression, feature extraction and fusion, and data standardization to obtain the current parameter optimization dataset. The historical parameter optimization dataset is processed by signal synchronization and alignment, noise suppression, feature extraction and fusion, data standardization, and dataset segmentation to obtain the parameter optimization training set and parameter optimization test set.

6. An intelligent multi-stage wind-powered coal preparation parameter optimization system, used to implement the intelligent multi-stage wind-powered coal preparation parameter optimization method as described in any one of claims 1-5, characterized in that: It includes a multi-dimensional data acquisition module, a data optimization and processing module, a state prediction model construction module, a parameter optimization model construction module, and a sorting parameter optimization module.

7. The intelligent multi-stage wind-powered coal preparation and separation parameter optimization system according to claim 6, characterized in that: The multidimensional data acquisition module is used to acquire raw data, obtain parameter-optimized raw dataset by acquiring raw data, and send the parameter-optimized raw dataset to the data optimization processing module. The data optimization processing module is used for data optimization processing. Through data optimization processing, it obtains the current parameter optimization dataset, parameter optimization training set, and parameter optimization test set, and sends the current parameter optimization dataset to the sorting parameter optimization module, and sends the parameter optimization training set and the parameter optimization test set to the state prediction model construction module. The state prediction model construction module is used to construct a state prediction model. By constructing an improved deep network model, a state prediction model is obtained, and the state prediction model is sent to the sorting parameter optimization module. The parameter optimization model construction module is used to construct the parameter optimization model. By constructing a hierarchical reinforcement learning model, the parameter optimization model is obtained, and the parameter optimization model is sent to the sorting parameter optimization module. The sorting parameter optimization module is used to optimize sorting parameters. It processes the current data by using the state prediction model in combination with the parameter optimization model and outputs the sorting parameter optimization decision result.

Images