Coal conveying control system and method for thermal power plant

Abstract

This application relates to a coal conveying control system and method for thermal power plants. The method includes the following steps: S1: Real-time acquisition of coal conveying system operation data through a pre-set multi-source sensor network; S2: Construction of a time series prediction model based on an attention mechanism to dynamically predict the operating status of the coal conveying system; S3: Real-time optimization calculations are performed at edge computing nodes to generate a control instruction set including belt speed, coal feeder opening, and transfer station coordination strategies; S4: Control instructions are sent to the actuators through a 5G-MEC network, while key data is uploaded to the cloud for model iteration and updates; S5: Safety protection thresholds are dynamically adjusted based on equipment vibration spectrum analysis results, and a graded protection mechanism is triggered when a coal blockage risk is detected. This application effectively improves the performance of the coal conveying system and achieves green energy saving.

Description

Technical Field

[0001] This application relates to the field of thermal power generation, and in particular to a coal conveying control system and method for thermal power plants. Background Technology

[0002] With the development of the power industry and the increasing demand for electricity, large-capacity, large-unit thermal power plants have gradually become dominant in my country. The coal conveying system of thermal power plants is a crucial link in ensuring the stable operation of the power plant. Traditional coal conveying control systems mostly use relay control and manual operation, which suffer from problems such as low automation, high failure rate, and low efficiency.

[0003] In actual production at thermal power plants, constant-speed belt conveyors and screening and crushing equipment with fixed coal particle size are usually used to transport coal to boilers to ensure the reliability of electricity supply for users. As a result, low load rate of coal conveyor belts and low boiler operating efficiency are common.

[0004] Chinese patent application CN118938795A discloses a coal conveying control system and method for thermal power plants. The system includes a model prediction module for establishing an operating load prediction model and deriving the optimal boiler operating load based on this model; a load calculation module for plotting an operating load difference curve and determining the boiler coal consumption change based on this curve; a speed control module for determining the coal conveying capacity of the coal conveying system and controlling the operating speed of the conveyor belt based on the coal conveying capacity; a quality calculation module for calculating boiler operating efficiency and determining coal quality parameters based on the boiler operating efficiency; and a fineness control module for determining coal particle parameters, setting a target coal particle fineness based on the coal particle parameters, and controlling the screening and crushing equipment based on the target coal particle fineness. This application establishes an operating load prediction model to predict the optimal boiler operating load, calculates the coal consumption change, adjusts the coal conveying capacity, and controls the belt speed. However, the load prediction model relies too heavily on historical data and does not consider dynamic factors such as equipment aging and coal type changes. It suffers from problems such as subjective threshold setting, failure to consider multiple factors, lack of theoretical support for the formula, and insufficient real-time performance and robustness. Prediction errors and adjustment lags are prone to occur, leading to increased coal consumption and higher operating costs.

[0005] In view of the aforementioned related technologies, this application proposes a coal conveying control system and method for thermal power plants. Summary of the Invention

[0006] To address the aforementioned problems, this application provides a coal conveying control system and method for thermal power plants.

[0007] Firstly, this application provides a coal conveying control method for thermal power plants, employing the following technical solution:

[0008] A method for controlling coal conveying in a thermal power plant includes the following steps:

[0009] S1: Real-time acquisition of coal conveying system operation data through a pre-set multi-source sensor network, the operation data including coal flow parameters, equipment status parameters and environmental parameters;

[0010] S2: Construct a time series prediction model based on the attention mechanism to dynamically predict the operating status of the coal conveying system;

[0011] S3: Performs real-time optimization calculations on edge computing nodes to generate a set of control instructions that includes belt speed, coal feeder opening, and transfer station coordination strategies;

[0012] S4: Control commands are sent to the actuators via the 5G-MEC network, while key data is uploaded to the cloud for model iteration and updates;

[0013] S5: Dynamically adjusts the safety protection threshold based on the equipment vibration spectrum analysis results, and triggers a graded protection mechanism when a risk of coal blockage is detected.

[0014] Preferably, the multi-source sensor network in step S1 includes:

[0015] A cross-detection array consisting of a microwave moisture meter and a laser particle size analyzer is installed in the coal chute;

[0016] A distributed vibration sensor network arranged along the coal conveyor belt;

[0017] Infrared thermal imager mounted on a bracket at the transfer station;

[0018] An acoustic emission sensor is installed in the bearing housing of the lower drum of the coal conveyor belt.

[0019] Preferably, the cross-detection array employs the following calibration method:

[0020] Real-time acquisition of the moisture content of coal passing through the coal chute, measured by a microwave moisture meter and a laser particle size analyzer;

[0021] Determine whether the deviation between the moisture content ω1 measured by the microwave moisture analyzer and the moisture content ω2 estimated by the laser particle size analyzer exceeds 5%;

[0022] If the value is not exceeded, the calibration is successful.

[0023] If the value exceeds the limit, a pre-set fusion algorithm based on a BP neural network will be activated to compensate for the data. The compensation formula is: ω=αω1+(1-α)ω2;

[0024] Where ω is the compensated water content; α=σ(ω1,ω2,ΔT), ΔT is the change in ambient temperature, and σ is the sigmoid activation function.

[0025] Preferably, the optimization calculation in step S3 employs a multi-objective particle swarm optimization algorithm, with the objective function being:

[0026] min(E) = k1·Pt + k2·Σ(1 / MTBF) i )+k3·ΔQ;

[0027] Constraints:

[0028] C1:Q min ≤Qt≤0.9Q max ;

[0029] C2:f vib ≤0.8f threshold ;

[0030] Where Pt is the real-time power, k1 is the real-time power weighting coefficient; MTBF i The mean time between failures (MTBF) of the equipment is Σ(1 / MTBF). i The equipment reliability score is represented by k2, where k2 is the equipment reliability score, k3 is the flow rate fluctuation coefficient, and Qt is the coal flow rate. vib f is the vibration frequency of the equipment. threshold The vibration frequency safety threshold for the equipment.

[0031] Preferably, step S1, data acquisition includes the following preprocessing steps:

[0032] Wavelet noise reduction processing is performed on the vibration sensor signal;

[0033] Perform spatial registration on infrared thermal imaging data;

[0034] The Kalman filter algorithm is used to fuse multi-source moisture content measurements;

[0035] Time-frequency domain feature extraction is performed on the acoustic emission signal.

[0036] Preferably, the model iterative update in step S4 includes the following steps:

[0037] A model performance evaluation report is automatically generated weekly, which includes model accuracy, recall, and F1 score.

[0038] Incremental learning is triggered when the prediction error exceeds 2% for three consecutive days. When the prediction error exceeds 2% for three consecutive days, the incremental learning mechanism is started to retain the top 10% of important neurons in the neural network and optimize the model to adapt to new data.

[0039] Cross-factory transfer learning is conducted every quarter, transferring data from other factories to this model for iterative learning.

[0040] Establish a version rollback mechanism and save the model weights for the most recent 5 versions.

[0041] Preferably, step 7, the construction of a time series prediction model based on an attention mechanism for dynamically predicting the operating status of the coal conveying system, specifically includes the following steps:

[0042] S21. Data processing and matrix construction: The preprocessed multi-source sensor data is sliced ​​according to time windows to construct an input matrix Xt∈RN×12, which contains 12-dimensional features, and the time step N is 60.

[0043] S22. Network Architecture Construction: Build an improved LSTM-ATT network architecture, which includes a bidirectional LSTM layer with 64 hidden units and a dropout rate of 0.2; a multi-head attention layer with 4 heads and a query vector dimension of 32; and a fully connected output layer using LeakyReLU as the activation function.

[0044] S23. Model parameter initialization: The model parameters are initialized using the transfer learning method. The source domain data used includes 200 sets of historical operating conditions.

[0045] S24. Loss function definition: Define the loss function as L = λ1MAE + λ2DTW, where λ1 takes the value of 0.7 and λ2 takes the value of 0.3.

[0046] S25, Rolling Forecast Execution: By executing the rolling forecast operation, output the predicted coal flow rate Q and equipment health trend H for the next 5 minutes.

[0047] Preferably, the step of dynamically adjusting the safety protection threshold based on the equipment vibration spectrum analysis results and triggering the graded protection mechanism when a coal blockage risk is detected specifically includes the following steps:

[0048] S51. Spectrum Transformation and Feature Extraction: Perform FFT transformation on the vibration signal to extract the energy distribution features in the 1-2000Hz frequency band;

[0049] S52. Risk Coefficient Calculation: Calculate the coal blockage risk coefficient P=Σ(w i ·E i / E total ), where w i The pre-training frequency band weights for the i-th frequency band within the 1-2000Hz frequency band are set by the administrator based on historical training records, E i E represents the vibrational energy of the i-th frequency band within the 1-2000Hz frequency range. total It is the total vibrational energy within the 1-2000Hz frequency band;

[0050] S53. Dynamic adjustment of protection threshold: The protection threshold is dynamically adjusted according to the coal blockage risk coefficient P: When P is less than 0.3, the protection threshold T is set to 1.2T0; when P is between 0.3 (inclusive) and 0.5, T is equal to T0; when P is greater than or equal to 0.5, T is 0.8T0, where T0 is a preset benchmark protection threshold.

[0051] S54. Implementation of graded response strategy: Implement the pre-set graded response strategy based on the coal blockage risk coefficient P;

[0052] S55. Fault Diagnosis Report Generation: Automatically generate fault diagnosis reports after the fact, marking abnormal frequency band characteristics and scoring the effectiveness of handling.

[0053] Preferably, the graded response strategy includes: Level 1 response, when P is in the range of [0.3, 0.5), controlling the upstream equipment to reduce its speed to 85% Vmax and activating a pre-set high-definition camera for inspection; Level 2 response, when P is in the range of [0.5, 0.7), starting a pre-set air cannon clearing device and simultaneously pushing AR maintenance guidance to the maintenance personnel's smart terminal; Level 3 response, when P is greater than or equal to 0.7, performing a reverse shutdown with a shutdown delay of Δt = 1.2L / vcurrent and triggering an audible and visual alarm.

[0054] Secondly, this application provides a coal conveying control system for a thermal power plant, which adopts the following technical solution:

[0055] A coal conveying control system for a thermal power plant, comprising:

[0056] The data acquisition module is used to collect real-time operating data of the coal conveying system through a pre-set multi-source sensor network. The operating data includes coal flow parameters, equipment status parameters, and environmental parameters.

[0057] The dynamic prediction module is used to build a time series prediction model based on the attention mechanism to dynamically predict the operating status of the coal conveying system.

[0058] The edge optimization module is used to perform real-time optimization calculations on edge computing nodes and generate a set of control instructions that include belt speed, coal feeder opening and transfer station coordination strategies.

[0059] The data transmission module is used to send control commands to the actuators via the 5G-MEC network, and at the same time upload key data to the cloud for model iteration and updates;

[0060] The risk control module is used to dynamically adjust the safety protection threshold based on the results of equipment vibration spectrum analysis, and triggers a graded protection mechanism when a risk of coal blockage is detected.

[0061] In summary, this application includes at least one of the following beneficial technical effects:

[0062] By deeply integrating multi-source sensing, dynamic optimization, and intelligent decision-making technologies, the system first collects operational data through a multi-source sensor network to build a predictive model that relies on comprehensive multi-source data for dynamic prediction, significantly improving prediction accuracy. Secondly, it performs real-time optimization calculations using a multi-objective particle swarm optimization algorithm at edge computing nodes, generating precise control command sets. Efficient communication and model iteration are achieved through a 5G-MEC network, enabling precise and efficient operation control and model optimization, thus improving the operational efficiency of the coal conveying system. Finally, based on equipment vibration spectrum analysis, the system dynamically adjusts safety protection thresholds. When a risk of coal blockage is detected, a graded protection mechanism is triggered, effectively reducing the probability of equipment failure, ensuring safe equipment operation, and achieving effective performance improvement and energy conservation in the coal conveying system.

[0063] Vibration spectrum analysis of equipment can accurately calculate the coal blockage risk coefficient, dynamically adjust the protection threshold based on the risk coefficient, and reasonably match different risk levels to improve the safety and adaptability of the coal conveying system under different operating conditions. At the same time, based on the risk coefficient, a graded response strategy is implemented to resolve the coal blockage risk before it occurs, ensuring the normal operation of the coal conveying system and effectively improving the operational stability of the coal conveying system.

[0064] Faced with complex and ever-changing environments and equipment operating conditions, the coal moisture content measured by microwave moisture meters and laser particle size analyzers is compared in real time. When the measurement deviation exceeds the allowable range, a compensation mechanism is automatically triggered, and a fusion algorithm based on a BP neural network is started. Combined with the change in ambient temperature, the compensation coefficient α is calculated using the sigmoid activation function to compensate for the moisture content and dynamically adjust the data. This eliminates the need for frequent manual intervention, ensuring long-term stable and accurate data and maintaining the efficient and stable operation of the coal conveying control system. Attached Figure Description

[0065] Figure 1 This is a flowchart of a coal conveying control method for a thermal power plant according to an embodiment of this application;

[0066] Figure 2 This is a flowchart of the calibration method for the cross-detection array in the embodiments of this application;

[0067] Figure 3 This is a flowchart of the method for preprocessing multi-source sensor network data acquisition in an embodiment of this application;

[0068] Figure 4 This is a flowchart of a method for dynamically predicting the operating status of a coal conveying system in an embodiment of this application;

[0069] Figure 5 This is a flowchart of the method for iterative updating of the time series prediction model in the embodiments of this application;

[0070] Figure 6This is a flowchart of the method for triggering a graded protection mechanism when a risk of coal blockage is detected in an embodiment of this application;

[0071] Figure 7 This is a system block diagram of a coal conveying control system for a thermal power plant, as described in an embodiment of this application.

[0072] Explanation of reference numerals in the attached diagram: 1. Data acquisition module; 2. Dynamic prediction module; 3. Edge optimization module; 4. Data transmission module; 5. Risk control module. Detailed Implementation

[0073] The following is in conjunction with the appendix Figure 1-7 This application will be described in further detail.

[0074] The existing coal conveying system includes coal yard equipment (such as bucket wheel stacker-reclaimers, coal pushers, etc.), coal feeders, coal conveyor belts, transfer stations, coal chutes and other main conveying equipment, as well as some auxiliary equipment such as crushers, vibrating screens, impurity removal equipment (such as iron removers, wood chip separators), electronic belt scales and automatic samplers.

[0075] This application discloses a coal conveying control method for a thermal power plant. (Refer to...) Figure 1 A method for controlling coal conveying in a thermal power plant, comprising the following steps:

[0076] S1: Real-time acquisition of coal conveying system operation data through a pre-set multi-source sensor network, the operation data including coal flow parameters, equipment status parameters and environmental parameters;

[0077] S2: Construct a time series prediction model based on the attention mechanism to dynamically predict the operating status of the coal conveying system;

[0078] S3: Perform real-time optimization calculations on the edge computing node to generate a control instruction set that includes belt speed, coal feeder opening, and transfer station coordination strategies. In this example, the edge computing node uses the NVIDIA Jetson AGX Xavier edge controller, which has the following built-in features: a real-time data cleaning module (sliding window of 60 seconds, step size of 5 seconds); a multi-objective particle swarm optimization algorithm library (population size of 50, number of iterations of 100); and a safety assessment engine (vibration spectrum resolution of 0.1Hz).

[0079] S4: Control commands are sent to the actuators via the 5G-MEC network, while key data is uploaded to the cloud for model iteration and updates. The cloud platform can be a self-developed cloud platform or a commercially available industrial cloud platform. In addition, the cloud platform is equipped with a digital twin system (built based on Unity 3D, with a geometric error of <2mm), a model iteration module (supporting the TensorFlow 2.8 framework), and a cross-plant federated learning server (communication protocol MQTT+SSL).

[0080] S5: Based on equipment vibration spectrum analysis, the safety protection threshold is dynamically adjusted, triggering a tiered protection mechanism when a coal blockage risk is detected. First, operational data is collected through a multi-source sensor network to build a predictive model. Dynamic prediction based on comprehensive multi-source data significantly improves prediction accuracy. Second, edge computing nodes utilize a multi-objective particle swarm optimization algorithm in real-time to generate precise control command sets. Efficient communication and model iteration are achieved through a 5G-MEC network, enabling precise and efficient operation control and model optimization, thus improving the operating efficiency of the coal conveying system. Finally, based on equipment vibration spectrum analysis, the safety protection threshold is dynamically adjusted, triggering a tiered protection mechanism when a coal blockage risk is detected. This effectively reduces the probability of equipment failure, ensures safe equipment operation, and achieves the desired effects of improving coal conveying system performance and energy conservation.

[0081] The multi-source sensor network in step S1 above includes:

[0082] A cross-detection array consisting of a microwave moisture meter (model HYDRA Probe 2.0) and a laser particle size analyzer (Malvern Panalytical Insitec) is installed in the coal chute;

[0083] A distributed vibration sensor network is arranged along the coal conveyor belt; in this embodiment, a triaxial vibration sensor (PCB 356A32, sampling rate 10kHz) is arranged every 20m along the coal conveyor belt, and multiple sampling nodes form a vibration network;

[0084] An infrared thermal imager is installed on the support frame of the transfer station; in this example, an infrared thermal imager (FLIR A700, resolution 640×480) is installed on the support frame of the transfer station to monitor the temperature of the belt joint, with a sampling interval of 10 seconds; in addition, a temperature and humidity composite sensor (Sensirion SHT45, accuracy ±1.5%RH) is deployed on the top of the coal conveying corridor.

[0085] An acoustic emission sensor is installed in the bearing housing of the drum under the coal conveyor belt. In this embodiment, the acoustic emission sensor (Physical Acoustics Nano30, frequency response range 100kHz-1MHz) is installed in the bearing housing and fixed by a magnetic base. Through the setup of a multi-source sensor network, coal characteristic data can be accurately acquired, and the operating status, operating environment, and equipment health of each piece of equipment in the coal conveying system can be monitored comprehensively. This facilitates dynamic prediction based on comprehensive multi-source data, helping to improve prediction accuracy.

[0086] Reference Figure 2 The cross-detection array employs the following calibration method:

[0087] A1. Real-time acquisition of the moisture content of coal passing through the coal drop pipe, as measured by a microwave moisture meter and a laser particle size analyzer;

[0088] A2. Determine whether the deviation between the moisture content ω1 measured by the microwave moisture meter and the moisture content ω2 estimated by the laser particle size analyzer exceeds 5%;

[0089] A3. If the value is not exceeded, the calibration is passed;

[0090] A4. If the value exceeds the limit, the pre-set fusion algorithm based on the BP neural network will be activated to compensate for the data. The compensation formula is: ω=αω1+(1-α)ω2.

[0091] Where ω represents the compensated moisture content; α = σ(ω1, ω2, ΔT), ΔT is the change in ambient temperature, and σ is the sigmoid activation function. Facing complex and ever-changing environments and equipment operating conditions, the system compares the coal moisture content measured by a microwave moisture meter and a laser particle size analyzer in real time. If the measurement deviation exceeds the allowable range, an automatic compensation mechanism is triggered, initiating a fusion algorithm based on a BP neural network. Combining the change in ambient temperature, the system uses the sigmoid activation function to calculate the compensation coefficient α, compensating for the moisture content and dynamically adjusting the data. This eliminates the need for frequent manual intervention, ensuring long-term data stability and accuracy, and maintaining the efficient and stable operation of the coal conveying control system.

[0092] Reference Figure 3 The data acquisition in step S1 includes the following preprocessing steps:

[0093] B1. Wavelet Denoising: Wavelet denoising is performed on the vibration sensor signal; in this example, the db4 wavelet basis function is selected, and the number of layers is 5.

[0094] B2. Spatial Registration: Perform spatial registration on the infrared thermal imaging data; where the registration accuracy is ≤0.5 pixels.

[0095] B3. Moisture content calibration: The Kalman filter algorithm is used to fuse moisture content measurements from multiple sources;

[0096] B4. Frequency Band Feature Extraction: Time-frequency domain feature extraction is performed on the acoustic emission signal; in this example, the focus is on extracting the energy proportion of the 1-3MHz frequency band. Through the above preprocessing, wavelet denoising is used to improve signal quality, spatial registration with an accuracy of ≤0.5 pixels is performed on the infrared thermal imaging data to ensure image accuracy, and the Kalman filter algorithm is used to fuse multi-source moisture content measurements. By comprehensively utilizing information from multiple data sources, measurement errors are reduced, and the accuracy of moisture content data is significantly improved. Furthermore, the frequency band containing rich equipment fault information is extracted, which helps to accurately identify abnormal conditions inside the equipment, providing key evidence for early diagnosis and prevention of equipment faults and reducing the risk of sudden equipment failures.

[0097] Reference Figure 4The construction of a time series prediction model based on an attention mechanism for dynamic prediction of the operating status of a coal conveying system includes the following steps:

[0098] S21. Data Processing and Matrix Construction: The preprocessed multi-source sensor data is sliced ​​according to time windows to construct an input matrix Xt∈RN×12, which contains 12-dimensional features, with a time step N of 60. The 12-dimensional features include coal flow parameters: instantaneous flow rate, moisture content, and particle size distribution standard deviation; equipment parameters: vibration RMS value, bearing temperature, and motor current harmonic distortion rate; and environmental parameters: corridor temperature and humidity gradient.

[0099] S22. Network Architecture Construction: Build an improved LSTM-ATT network architecture, which includes a bidirectional LSTM layer with 64 hidden units and a dropout rate of 0.2; a multi-head attention layer with 4 heads and a query vector dimension of 32; and a fully connected output layer using LeakyReLU as the activation function.

[0100] S23. Model parameter initialization: The model parameters are initialized using the transfer learning method. The source domain data used includes 200 sets of historical operating conditions.

[0101] S24. Loss function definition: Define the loss function as L = λ1MAE + λ2DTW, where λ1 takes the value of 0.7 and λ2 takes the value of 0.3.

[0102] S25. Rolling Prediction Execution: By executing the rolling prediction operation, the predicted coal flow rate Q and equipment health trend H for the next 5 minutes are output. Through the above steps, a matrix is ​​constructed by slicing 12-dimensional feature data of coal flow, equipment, environment, etc., according to time windows, fully exploring the correlation between different types of data and providing rich information for accurate prediction. An improved LSTM-ATT network is used, with bidirectional LSTM capturing long-term dependencies, multi-head attention focusing on key features, and fully connected layers combined with the LeakyReLU activation function to improve the model's ability to predict complex operating states. On this basis, transfer learning is used to initialize parameters, and 200 sets of historical operating condition source domain data are used to accelerate the model training speed, reduce training time costs, and adapt to new data more quickly. Finally, the loss function combines MAE and DTW to more comprehensively and accurately measure the prediction error, improve the prediction accuracy, and realize the rolling prediction output of coal flow rate and equipment health for the next 5 minutes, providing timely and effective data support for operation control and achieving the effect of effectively improving the model's prediction accuracy.

[0103] The optimization calculation in step S3 above uses a multi-objective particle swarm optimization algorithm, and the objective function is:

[0104] min(E)=k1·Pt+k2·Σ(1 / MTBFi)+k3·ΔQ;

[0105] Constraints:

[0106] C1:Q min ≤Qt≤0.9Q max In this embodiment, Q min =200t / h, Q max =800t / h;

[0107] C2:f vib ≤0.8f threshold In this embodiment, f threshold =45Hz;

[0108] Where Pt is the real-time power, k1 is the real-time power weighting coefficient; MTBFi is the mean time between failures (MTBF), Σ(1 / MTBF_i) is the equipment reliability score (a higher score indicates a higher likelihood of equipment failure), k2 is the equipment reliability score; ΔQ is the flow rate fluctuation, k3 is the flow rate fluctuation coefficient; Qt is the coal flow rate, fvib is the equipment vibration frequency, and f threshold The vibration frequency safety threshold for the equipment is defined. In this embodiment, the weighting coefficients k1=0.6, k2=0.3, and k3=0.1 are determined using the Analytic Hierarchy Process (AHP). Employing a multi-objective particle swarm optimization algorithm, it can quickly respond to real-time data changes in the coal conveying system without transmitting data to the cloud for processing, significantly reducing data processing latency. It rapidly generates control command sets and promptly adjusts belt speed, feeder opening, and transfer station coordination strategies. This achieves a balance between reducing real-time power consumption, improving equipment reliability, and minimizing flow fluctuations, ensuring the real-time and efficient operation of the coal conveying system and effectively improving its performance while reducing energy consumption.

[0109] Reference Figure 5 The model iterative update in step S4 includes the following steps:

[0110] C1. Weekly Report Generation: A weekly model performance evaluation report is automatically generated, which includes model accuracy, recall, and F1 score.

[0111] C2. Daily error monitoring: Incremental learning is triggered when the prediction error exceeds 2% for 3 consecutive days. When the prediction error exceeds 2% for 3 consecutive days, the incremental learning mechanism is started to retain the top 10% of important neurons in the neural network and optimize the model to adapt to new data.

[0112] C3. Quarterly Transfer Learning: Conduct cross-plant transfer learning every quarter, transferring data from other plants to this model for iterative learning; specifically, feature distillation techniques can be used to transfer valuable knowledge from other plants to this model, improving the model's generalization ability;

[0113] C4. Establish a version rollback mechanism: Establish a version rollback mechanism and save the model weights of the most recent 5 versions. Through the above steps, the time series prediction model is continuously monitored to promptly grasp the model status. Incremental learning is triggered when the prediction error exceeds 2% for 3 consecutive days, retaining key neurons to optimize the model, enabling it to quickly adapt to new data and improve prediction accuracy. Each quarter, feature distillation technology is used to transfer valuable knowledge from other plants to this model, improving the model's generalization ability and effectively improving model performance and prediction accuracy.

[0114] Reference Figure 6 The safety protection threshold is dynamically adjusted based on the equipment vibration spectrum analysis results. When a risk of coal blockage is detected, the graded protection mechanism is triggered, which specifically includes the following steps:

[0115] S51. Spectrum Transformation and Feature Extraction: Perform FFT transformation on the vibration signal to extract the energy distribution features within the 1-2000Hz frequency band; In this embodiment, the frequency bands are divided as follows: 1-100Hz, 100-500Hz, and 500-2000Hz. The pre-trained frequency band weights w1, w2, and w3 for the corresponding three frequency bands are obtained by training a logistic regression classifier using historical fault data, with w1=0.2, w2=0.5, and w3=0.3.

[0116] S52. Risk Coefficient Calculation: Calculate the coal blockage risk coefficient P=Σ(w i ·E i / E total ), where w i The pre-training frequency band weights for the i-th frequency band within the 1-2000Hz frequency band are set by the administrator based on historical training records, E i E represents the vibrational energy of the i-th frequency band within the 1-2000Hz frequency range. total It is the total vibrational energy within the 1-2000Hz frequency band;

[0117] S53. Dynamic adjustment of protection threshold: The protection threshold is dynamically adjusted according to the coal blockage risk coefficient P: When P is less than 0.3, the protection threshold T is set to 1.2T0; when P is between 0.3 (inclusive) and 0.5, T is equal to T0; when P is greater than or equal to 0.5, T is 0.8T0, where T0 is a preset benchmark protection threshold.

[0118] S54. Implementation of graded response strategy: Implement the pre-set graded response strategy based on the coal blockage risk coefficient P.

[0119] S55. Fault Diagnosis Report Generation: A fault diagnosis report is automatically generated post-incidentally, annotating abnormal frequency band characteristics and assigning an effectiveness score to the handling measures. Through the above steps, vibration spectrum analysis of the equipment can accurately calculate the coal blockage risk coefficient. Based on the risk coefficient, protection thresholds are dynamically adjusted to reasonably match different risk levels, improving the safety and adaptability of the coal conveying system under different operating conditions. Simultaneously, a graded response strategy is implemented based on the risk coefficient to resolve coal blockage risks before they occur, ensuring the normal operation of the coal conveying system and effectively improving its operational stability.

[0120] In addition, the automatic generation of fault diagnosis reports after the incident, which marks the abnormal frequency band and the handling score, helps to summarize experience and continuously optimize the protection mechanism and fault handling process.

[0121] The aforementioned tiered response strategy includes: Level 1 response: when P is in the range of [0.3, 0.5), control the upstream equipment to reduce its speed to 85% of Vmax and activate a pre-set AI high-definition camera (5 seconds / frame) for inspection; Level 2 response: when P is in the range of [0.5, 0.7), activate a pre-set air cannon to clear blockages and simultaneously push AR maintenance guidance to the maintenance personnel's smart terminal; Level 3 response: when P is greater than or equal to 0.7, execute a reverse-sequence shutdown with a shutdown delay Δt = 1.2L / vcurrent (L is the length of the coal conveyor belt, and vcurrent is the operating speed of the coal conveyor belt), and trigger an audible and visual alarm (≥90dB). By implementing the tiered response strategy, differentiated measures are implemented based on different ranges of the coal blockage risk coefficient P. When the risk is low, relatively mild measures such as speed reduction and inspection are prioritized to avoid excessive intervention; when the risk increases, more aggressive measures are taken to rationally allocate resources, reduce unnecessary equipment wear and costs, and improve the efficiency and effectiveness of fault response through coordination among all links.

[0122] This application also discloses a coal conveying control system for a thermal power plant. (Refer to...) Figure 7 A coal conveying control system for a thermal power plant includes:

[0123] Data acquisition module 1 is used to collect real-time operating data of the coal conveying system through a pre-set multi-source sensor network. The operating data includes coal flow parameters, equipment status parameters and environmental parameters.

[0124] Dynamic prediction module 2 is used to build a time series prediction model based on the attention mechanism to dynamically predict the operating status of the coal conveying system;

[0125] Edge optimization module 3 is used to perform real-time optimization calculations on edge computing nodes to generate a set of control instructions that includes belt speed, coal feeder opening and transfer station coordination strategy;

[0126] Data transmission module 4 is used to send control commands to the actuators via the 5G-MEC network, and at the same time upload key data to the cloud for model iteration and updates;

[0127] Risk control module 5 dynamically adjusts safety protection thresholds based on equipment vibration spectrum analysis results, triggering a tiered protection mechanism when a coal blockage risk is detected. First, operational data is collected through a multi-source sensor network to build a predictive model. Dynamic prediction based on comprehensive multi-source data significantly improves prediction accuracy. Second, edge computing nodes utilize a multi-objective particle swarm optimization algorithm in real-time to generate precise control command sets. Efficient communication and model iteration are achieved through a 5G-MEC network, enabling precise and efficient operation control and model optimization, thus improving the operational efficiency of the coal conveying system. Finally, the dynamic adjustment of safety protection thresholds based on equipment vibration spectrum analysis triggers a tiered protection mechanism when a coal blockage risk is detected, effectively reducing the probability of equipment failure, ensuring safe equipment operation, and achieving effective performance improvement and energy conservation in the coal conveying system.

[0128] This application also discloses a computer-readable storage medium that stores a computer program that can be loaded by a processor and executed as described above. The computer-readable storage medium includes, for example, various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0129] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on these embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can still combine, add, delete, or otherwise adjust the features of the various embodiments of the present invention according to the circumstances without conflict or creative effort, thereby obtaining different technical solutions that do not fundamentally depart from the concept of the present invention. These technical solutions also fall within the scope of protection of the present invention.

Claims

1. A method for controlling coal conveying in a thermal power plant, characterized in that, Includes the following steps: S1: Real-time acquisition of coal conveying system operation data through a pre-set multi-source sensor network, the operation data including coal flow parameters, equipment status parameters and environmental parameters; S2: Construct a time series prediction model based on the attention mechanism to dynamically predict the operating status of the coal conveying system; S3: Performs real-time optimization calculations on edge computing nodes to generate a set of control instructions that includes belt speed, coal feeder opening, and transfer station coordination strategies; S4: Control commands are sent to the actuators via the 5G-MEC network, while key data is uploaded to the cloud for model iteration and updates; S5: Dynamically adjusts the safety protection threshold based on the equipment vibration spectrum analysis results, and triggers a graded protection mechanism when a risk of coal blockage is detected; The multi-source sensor network in step S1 includes: A cross-detection array consisting of a microwave moisture meter and a laser particle size analyzer is installed in the coal chute. A distributed vibration sensor network arranged along the coal conveyor belt; Infrared thermal imager mounted on a bracket at the transfer station; An acoustic emission sensor is installed in the bearing housing of the lower drum of the coal conveyor belt; The cross-detection array employs the following calibration method: Real-time acquisition of the moisture content of coal passing through the coal chute, measured by a microwave moisture meter and a laser particle size analyzer; Determine whether the deviation between the moisture content ω1 measured by the microwave moisture analyzer and the moisture content ω2 estimated by the laser particle size analyzer exceeds 5%; If the value is not exceeded, the calibration is successful. If the value exceeds the limit, a pre-set fusion algorithm based on a BP neural network will be activated to compensate for the data. The compensation formula is: ω=αω1+(1-α)ω2; Where ω is the compensated water content; α=σ(ω1,ω2,ΔT), ΔT is the change in ambient temperature, and σ is the sigmoid activation function.

2. The coal conveying control method for a thermal power plant according to claim 1, characterized in that, The optimization calculation in step S3 employs a multi-objective particle swarm optimization algorithm, with the objective function being: min(E) = k1-Pt+k2-∑ i( 1 / MTBF i )+k3-ΔQ; Constraints: C1:Q min ≤Qt≤0.9Q max ; C2:f vib ≤0.8f threshold ; Where Pt is the real-time power, k1 is the real-time power weighting coefficient; MTBF i Σ is the mean time between failures (MTBF) of the equipment. i (1 / MTBF i The score represents the equipment reliability rating. The higher the score, the more likely the equipment is to fail. k2 is the equipment reliability rating coefficient. ΔQ represents the flow rate fluctuation, k3 is the flow rate fluctuation coefficient; Qt is the coal flow rate, f vib f is the vibration frequency of the equipment. threshold The vibration frequency safety threshold for the equipment.

3. The coal conveying control method for a thermal power plant according to claim 1, characterized in that, The data acquisition in step S1 includes the following preprocessing steps: Wavelet noise reduction processing is performed on the vibration sensor signal; Perform spatial registration on infrared thermal imaging data; The Kalman filter algorithm is used to fuse multi-source moisture content measurements; Time-frequency domain feature extraction is performed on the acoustic emission signal.

4. The coal conveying control method for a thermal power plant according to claim 1, characterized in that: The model iterative update in step S4 includes the following steps: A model performance evaluation report is automatically generated weekly, which includes model accuracy, recall, and F1 score. When the prediction error exceeds 2% for three consecutive days, incremental learning is triggered. The incremental learning mechanism is activated to retain the top 10% of important neurons in the neural network and optimize the model to adapt to new data. Cross-factory transfer learning is conducted every quarter, transferring data from other factories to this model for iterative learning. Establish a version rollback mechanism and save the model weights for the most recent 5 versions.

5. A coal conveying control method for a thermal power plant according to claim 3, characterized in that, The construction of a time series prediction model based on an attention mechanism for dynamically predicting the operating status of the coal conveying system specifically includes the following steps: S21. Data Processing and Matrix Construction: The preprocessed multi-source sensor data is sliced ​​according to time windows to construct the input matrix X. t ∈R N×12 It contains 12-dimensional features, and the time step N takes a value of 60; S22. Network Architecture Construction: Build an improved LSTM-ATT network architecture, which includes a bidirectional LSTM layer with 64 hidden units and a dropout rate of 0.2; a multi-head attention layer with 4 heads and a query vector dimension of 32; and a fully connected output layer using LeakyReLU as the activation function. S23. Model parameter initialization: The model parameters are initialized using the transfer learning method. The source domain data used includes 200 sets of historical operating conditions. S24. Loss function definition: Define the loss function as L = λ1MAE + λ2DTW, where λ1 takes the value of 0.7 and λ2 takes the value of 0.

3. S25, Rolling Forecast Execution: By executing the rolling forecast operation, output the predicted coal flow rate Q and equipment health trend H for the next 5 minutes.

6. The coal conveying control method for a thermal power plant according to claim 1, characterized in that: The dynamic adjustment of safety protection thresholds based on equipment vibration spectrum analysis results, and the triggering of a graded protection mechanism when coal blockage risk is detected, specifically includes the following steps: S51. Spectrum Transformation and Feature Extraction: Perform FFT transformation on the vibration signal to extract the energy distribution features in the 1-2000Hz frequency band; S52. Risk Coefficient Calculation: Calculate the coal blockage risk coefficient P=Σ i (w i ·E i / E total ), where w i The pre-training frequency band weights for the i-th frequency band within the 1-2000Hz frequency band are set by the administrator based on historical training records, E i E represents the vibrational energy of the i-th frequency band within the 1-2000Hz frequency range. total It is the total vibrational energy within the 1-2000Hz frequency band; S53. Dynamic adjustment of protection threshold: The protection threshold is dynamically adjusted according to the coal blockage risk coefficient P: when P is less than 0.3, the protection threshold T is set to 1.2T0; when 0.3≤P<0.5, T equals T0; when P is greater than or equal to 0.5, T is 0.8T0, where T0 is a preset benchmark protection threshold. S54. Implementation of graded response strategy: Implement the pre-set graded response strategy based on the coal blockage risk coefficient P; S55. Fault Diagnosis Report Generation: Automatically generate fault diagnosis reports after the fact, marking abnormal frequency band characteristics and scoring the effectiveness of handling.

7. A coal conveying control method for a thermal power plant according to claim 6, characterized in that: The graded response strategy includes: Level 1 response: when P is in the range of [0.3, 0.5), control the upstream equipment to reduce its speed to 85% of Vmax and activate the pre-set high-definition camera for inspection; Level 2 response: when P is in the range of [0.5, 0.7), start the pre-set air cannon clearing device and push AR maintenance guidance to the maintenance personnel's smart terminal; Level 3 response: when P is greater than or equal to 0.7, execute reverse shutdown with a shutdown delay of Δt = 1.2L / vcurrent and trigger an audible and visual alarm.

8. A coal conveying control system for a thermal power plant, characterized in that, include: The data acquisition module is used to collect real-time operating data of the coal conveying system through a pre-set multi-source sensor network. The operating data includes coal flow parameters, equipment status parameters, and environmental parameters. The dynamic prediction module is used to build a time series prediction model based on the attention mechanism to dynamically predict the operating status of the coal conveying system. The edge optimization module is used to perform real-time optimization calculations on edge computing nodes and generate a set of control instructions that include belt speed, coal feeder opening and transfer station coordination strategies. The data transmission module is used to send control commands to the actuators via the 5G-MEC network, and at the same time upload key data to the cloud for model iteration and updates; The risk control module is used to dynamically adjust the safety protection threshold based on the equipment vibration spectrum analysis results, and triggers a graded protection mechanism when a coal blockage risk is detected. The multi-source sensor network includes: A cross-detection array consisting of a microwave moisture meter and a laser particle size analyzer is installed in the coal chute; A distributed vibration sensor network arranged along the coal conveyor belt; Infrared thermal imager mounted on a bracket at the transfer station; An acoustic emission sensor is installed in the bearing housing of the lower drum of the coal conveyor belt; The cross-detection array employs the following calibration method: Real-time acquisition of the moisture content of coal passing through the coal chute, measured by a microwave moisture meter and a laser particle size analyzer; Determine whether the deviation between the moisture content ω1 measured by the microwave moisture analyzer and the moisture content ω2 estimated by the laser particle size analyzer exceeds 5%; If the value is not exceeded, the calibration is successful. If the value exceeds the limit, a pre-set fusion algorithm based on a BP neural network will be activated to compensate for the data. The compensation formula is: ω=αω1+(1-α)ω2; Where ω is the compensated water content; α=σ(ω1,ω2,ΔT), ΔT is the change in ambient temperature, and σ is the sigmoid activation function.

Images