Accurate shutdown method for thermal energy storage power station

Abstract

The invention provides a method for accurately stopping a thermal energy storage power station, and belongs to the technical field of thermal energy storage power stations. Running acoustic monitoring and particle distribution identification to output a particle fluidization degree classification result and a distribution uniformity evaluation value, establishing a particle distribution sparse matrix, and converting the particle distribution sparse matrix into a dense matrix through a multi-stage progressive clustering algorithm to realize accurate quantification of a particle distribution state; executing precise fluidization termination control to completely convert the energy storage particles from a fluidization state into a fixed bed state, performing fixed bed stability verification to ensure that no local overheating phenomenon exists, completing shutdown state recording, and establishing a shutdown historical database; the technical problems of equipment damage and energy loss caused by non-uniform temperature distribution of energy storage particles in the shutdown process of the heat energy storage power station are solved.

Description

Technical Field

[0001] This invention belongs to the field of thermal energy storage power station technology, and specifically relates to a method for precise shutdown of a thermal energy storage power station. Background Technology

[0002] Thermal energy storage (TES) power plants, as an emerging large-scale energy storage technology, primarily employ ultra-high temperature fluidized bed technology to achieve heating and cooling cycles for energy storage particles. In this fluidized state, the energy storage particles achieve uniform temperature distribution and efficient heat exchange through the fluidization effect of inert gases. Traditional TES power plant shutdown techniques mainly rely on simple temperature threshold control and fixed time sequence control. This involves gradually reducing the blower speed to transition the energy storage particles from a fluidized state to a fixed bed state, a method widely used in industrial applications. However, traditional shutdown techniques have significant drawbacks. Due to the lack of precise monitoring and control of the spatial distribution of energy storage particles, localized aggregation can easily occur during shutdown, leading to uneven temperature distribution. Some areas may be excessively hot while others are excessively cold. This temperature gradient difference can cause thermal stress concentration, resulting in equipment structural damage. In existing technologies, the lack of real-time and accurate identification of the fluidization state of energy storage particles and optimized control of the shutdown sequence means that the problem of uneven temperature distribution during shutdown remains unresolved. This uneven temperature distribution not only causes thermal stress damage to the equipment but also leads to a significant decrease in energy storage efficiency. In other words, existing technologies have technical problems such as uneven temperature distribution of energy storage particles during the shutdown of thermal energy storage power plants, which can lead to equipment damage or energy loss. Summary of the Invention

[0003] In view of this, the present invention provides a method for precise shutdown of a thermal energy storage power station, which can solve the technical problem in the prior art where uneven temperature distribution of energy storage particles during the shutdown process of a thermal energy storage power station leads to equipment damage or energy loss.

[0004] This invention is implemented as follows: A method for precise shutdown of a thermal energy storage power station includes the following steps: Initiating pre-shutdown state monitoring; installing temperature and acoustic sensors inside an ultra-high temperature fluidized bed to collect temperature distribution data and acoustic signal data of the energy storage particles; executing graded speed-reducing blower control; calculating the velocity ratio of the three speed-reducing stages based on the current fluidized gas velocity and energy storage particle temperature using a speed-reducing gradient calculation function, and progressively reducing the blower speed; running acoustic monitoring and particle distribution identification; inputting real-time acoustic signals into a particle flow regime identification model to identify the fluidization degree and spatial distribution state of the energy storage particles, and outputting particle fluidization degree classification results and... Uniformity assessment value; establishing a sparse particle distribution matrix: based on the location information of energy storage particles obtained by acoustic sensors, a sparse particle spatial distribution matrix is ​​constructed. The sparse matrix is ​​then converted into a dense matrix through a multi-level progressive clustering algorithm to achieve accurate quantification of particle distribution status; calculating the optimal shutdown sequence: based on the temperature decay curve and the dense particle distribution matrix, the optimal shutdown time point under temperature uniformity constraints is solved using the Lagrange multiplier method. At the same time, the shortest path algorithm in graph theory is used to optimize the shutdown step sequence; executing precise fluidization termination control: when the temperature of the energy storage particles drops to the set threshold and the particle fluidization degree classification result shows quasi-static fluidization, the blower is turned off.

[0005] After shutting down the blower, the process also includes: verifying the stability of the fixed bed, monitoring the temperature distribution and settling state of the energy storage particles in the fixed bed state, and verifying the uniformity of particle settling through a thermal stability function; completing the shutdown status record, recording key data of the entire shutdown process, and establishing a shutdown history database. The deceleration gradient calculation function is used to calculate the velocity ratio of the three deceleration stages. The inputs include the current fluidizing gas velocity, energy storage particle temperature, fluidized bed pressure loss, and energy storage particle density. The outputs are the velocity ratio of the first stage, the velocity ratio of the second stage, and the velocity ratio of the third stage.

[0006] The acoustic monitoring technology uses a sound wave sensor with a frequency range of 20Hz to 20000Hz to determine the fluidization state of the particles by analyzing the sound wave spectrum characteristics generated by the collision of energy storage particles.

[0007] The particle fluidization degree classification includes four states: fully fluidized, partially fluidized, quasi-static fluidized, and fully static. Each state corresponds to different acoustic spectrum characteristics and temperature distribution patterns.

[0008] The multi-level progressive clustering algorithm includes three clustering levels: the first level is coarse-grained clustering, which initially clusters the sparse matrix of particle distribution according to a 10×10 grid; the second level is medium-grained clustering, which refines the clustering results of the first level according to a 5×5 grid; and the third level is fine-grained clustering, which finally clusters the clustering results of the second level according to a 2×2 grid.

[0009] The particle distribution sparse matrix is ​​a two-dimensional matrix established based on the position coordinates of the energy storage particles. The matrix dimension is a pixelated representation of the length and width of the fluidized bed. Non-zero elements in the matrix indicate the presence of energy storage particles at the corresponding positions, while zero elements indicate the absence of energy storage particles at the corresponding positions.

[0010] The dense particle distribution matrix is ​​a dense matrix obtained by processing the sparse particle distribution matrix through a multi-level progressive clustering algorithm. Each element in the matrix is ​​a non-zero value, representing the aggregate density information of energy storage particles in the corresponding region.

[0011] The specific structure of the particle fluid state recognition model is an image analysis model based on the Vision Transformer architecture, which includes an encoder layer, a multi-head attention mechanism, and a classification output layer. The block size parameter of the image segmentation mechanism is determined based on three parameters: the resolution of the acoustic spectrum, the temperature range of the energy storage particles, and the geometry of the fluidized bed.

[0012] The shutdown process includes four steps: pre-shutdown preparation, graded deceleration control, fluidization termination, and fixed bed stabilization. The pre-shutdown preparation stage takes place from 0 to 300 seconds after the shutdown command is issued, the graded deceleration control stage takes place from 300 to 1800 seconds after the shutdown command is issued, and the fluidization termination stage takes place from 1800 to 2100 seconds after the shutdown command is issued.

[0013] The ultra-high temperature fluidized bed is a high-temperature energy storage device that uses inert gas as the fluidizing medium, and the internal energy storage particles operate at temperatures ranging from 800°C to 1200°C. The energy storage particles are spherical ceramic matrix composite particles with diameters ranging from 2 mm to 5 mm, exhibiting high-temperature stability and good heat transfer performance.

[0014] The thermal stability function is a composite function based on the heat conduction equation and the convective heat transfer coefficient, used to calculate the uniformity of heat distribution during the temperature decay process of the energy storage particles. The block size adjustment function is used to adjust the image block parameters of the particle flow regime recognition model. The block size adjustment function calculates the block size adjustment value based on three data points: the resolution of the acoustic spectrum, the temperature range of the energy storage particles, and the geometry of the fluidized bed. When the block size adjustment value is within the range of 0 to 0.3, a linear weighting adjustment function is used to set the block size to 8×8 pixels.

[0015] The steps for establishing the training dataset for the particle fluidity recognition model specifically include collecting acoustic signal data of energy storage particles under different temperature conditions, converting the acoustic signals into time-spectrum images, labeling the particle fluidization degree classification corresponding to each time-spectrum image, and establishing a labeled dataset containing four states: fully fluidized, partially fluidized, quasi-static fluidized, and fully static. The steps for training the particle fluidity recognition model specifically include using a stochastic gradient descent optimization algorithm to train the model parameters, setting the learning rate to 0.001, the batch size to 32, and the number of training epochs to 200, using the cross-entropy loss function to calculate the error between the model output and the true label, and updating the model weights through the backpropagation algorithm.

[0016] Furthermore, before the pre-shutdown preparation stage, it also includes detecting whether the temperature of the energy storage particles inside the ultra-high temperature fluidized bed reaches the preset shutdown temperature condition. When the average temperature of the energy storage particles is lower than the preset threshold, the precise shutdown process is initiated.

[0017] This invention achieves precise quantification of the spatial distribution state of energy storage particles by establishing a particle flow pattern identification model based on acoustic monitoring and a particle distribution density matrix using a multi-level progressive clustering algorithm, thus overcoming the shortcomings of traditional technologies in accurately monitoring particle distribution. This invention employs graded deceleration blower control and the Lagrange multiplier method to optimize shutdown timing. Through coupled analysis of temperature decay curves and the particle distribution density matrix, it achieves precise control of the temperature distribution of energy storage particles during shutdown, avoiding the uneven temperature distribution problem caused by improper shutdown timing in traditional technologies, and ensuring good temperature distribution uniformity of energy storage particles during shutdown. In summary, this invention solves the technical problem mentioned in the background art of equipment damage and energy loss caused by uneven temperature distribution of energy storage particles during the shutdown of thermal energy storage power plants. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention.

[0019] Figure 2 This is a temperature change curve during the shutdown process in the embodiment.

[0020] Figure 3 This is a diagram illustrating the particle fluidization state transition process in the embodiment.

[0021] Figure 4 The diagram shows the results of the shutdown timing optimization in the example. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0023] like Figure 1 The diagram shown is a flowchart of a method for precise shutdown of a thermal energy storage power station provided by the present invention. This method includes the following steps: S01. Start pre-shutdown status monitoring. Set up multiple temperature sensors and acoustic sensors inside the ultra-high temperature fluidized bed to collect temperature distribution data and sound wave signal data under the fluidized state of energy storage particles and establish a real-time monitoring dataset. S02. Implement graded speed reduction blower control. Based on the current fluidized gas flow rate and energy storage particle temperature, use the speed reduction gradient calculation function to calculate the flow rate ratio of the three speed reduction stages, and gradually reduce the blower speed to control the fluidized gas flow rate to change from a high-speed fluidization state to a low-speed fluidization state. S03. Operational acoustic monitoring and particle distribution identification: Input real-time acoustic signals into the particle flow regime identification model to identify the fluidization degree and spatial distribution of energy storage particles, and output particle fluidization degree classification results and distribution uniformity evaluation values. S04. Establish a sparse matrix of particle distribution. Based on the location information of energy storage particles obtained by the acoustic sensor, construct a sparse matrix of particle spatial distribution. Then, convert the sparse matrix into a dense matrix through a multi-level progressive clustering algorithm to achieve accurate quantification of particle distribution status. S05. Calculate the optimal shutdown sequence. Based on the temperature decay curve and the particle distribution density matrix, use the Lagrange multiplier method to solve for the optimal shutdown time point under the temperature uniformity constraint. At the same time, use the shortest path algorithm in graph theory to optimize the shutdown step sequence. S06. Perform precise fluidization termination control. When the temperature of the energy storage particles drops to the set threshold and the particle fluidization degree classification result shows quasi-static fluidization, turn off the blower to completely change the energy storage particles from the fluidized state to the fixed bed state. S07. Conduct fixed-bed stability verification, monitor the temperature distribution and sedimentation state of the energy storage particles in the fixed-bed state, verify the uniformity of particle sedimentation through the thermal stability function, and ensure that there is no local overheating. S08. Complete the shutdown status record, record the temperature change curve, particle distribution state transformation process and key time nodes of the entire shutdown process, and establish a shutdown history database for optimizing subsequent shutdown operations.

[0024] The ultra-high temperature fluidized bed is a high-temperature energy storage device that uses inert gas as the fluidizing medium, and the internal energy storage particles operate at temperatures ranging from 800°C to 1200°C.

[0025] The energy storage particles are spherical ceramic matrix composite particles with a diameter ranging from 2 mm to 5 mm, and have high temperature stability and good heat transfer performance.

[0026] The deceleration gradient calculation function is used to calculate the velocity ratio of the three deceleration stages. The inputs include the current fluidizing gas velocity, energy storage particle temperature, fluidized bed pressure loss, and energy storage particle density. The outputs are the velocity ratio of the first stage, the velocity ratio of the second stage, and the velocity ratio of the third stage.

[0027] The acoustic monitoring technology uses a sound wave sensor with a frequency range of 20Hz to 20000Hz to determine the fluidization state of the particles by analyzing the sound wave spectrum characteristics generated by the collision of energy storage particles.

[0028] The particle fluidization degree classification includes four states: fully fluidized, partially fluidized, quasi-static fluidized, and fully static. Each state corresponds to different acoustic spectrum characteristics and temperature distribution patterns.

[0029] The multi-level progressive clustering algorithm includes three clustering levels: the first level is coarse-grained clustering, which initially clusters the sparse particle distribution matrix according to a 10×10 grid; the second level is medium-grained clustering, which refines the clustering results of the first level according to a 5×5 grid; and the third level is fine-grained clustering, which finally clusters the clustering results of the second level according to a 2×2 grid. The K-means algorithm is used in the clustering process at each level, with 16, 64, and 256 cluster centers respectively. The transformation from a sparse matrix to a dense matrix is ​​achieved by progressively increasing the clustering density. Each element in the dense matrix represents the density and temperature values ​​of the energy storage particles in the corresponding region. The multi-level progressive clustering process can effectively reduce computational complexity while maintaining the integrity of particle distribution information. Compared with directly constructing a dense matrix, the computational efficiency is improved by 40%, memory usage is reduced by 60%, and particle distribution recognition accuracy is improved by 25%.

[0030] The particle distribution sparse matrix is ​​a two-dimensional matrix established based on the position coordinates of the energy storage particles. The matrix dimension is a pixelated representation of the length and width of the fluidized bed. Non-zero elements in the matrix indicate the presence of energy storage particles at the corresponding positions, while zero elements indicate the absence of energy storage particles at the corresponding positions.

[0031] The dense particle distribution matrix is ​​a dense matrix obtained by processing the sparse particle distribution matrix through a multi-level progressive clustering algorithm. Each element in the matrix is ​​a non-zero value, representing the aggregate density information of energy storage particles in the corresponding region.

[0032] The thermal stability function is a composite function based on the heat conduction equation and the convective heat transfer coefficient, used to calculate the uniformity of heat distribution during the temperature decay process of energy storage particles.

[0033] The specific structure of the particle fluid state recognition model is an image analysis model based on the Vision Transformer architecture, which includes an encoder layer, a multi-head attention mechanism, and a classification output layer. The block size parameter of the image segmentation mechanism is determined based on three parameters: the resolution of the acoustic spectrum, the temperature range of the energy storage particles, and the geometry of the fluidized bed. When the block size is set to 16×16 pixels, it is suitable for high-resolution spectrum analysis; when the block size is set to 8×8 pixels, it is suitable for refined particle state recognition; and when the block size is set to 32×32 pixels, it is suitable for fast state judgment.

[0034] The steps for establishing the training dataset of the particle fluidity recognition model specifically include collecting acoustic signal data of energy storage particles under different temperature conditions, converting the acoustic signals into time-spectrum images, labeling the particle fluidization degree classification corresponding to each time-spectrum image, and establishing a labeled dataset containing four states: fully fluidized, partially fluidized, quasi-static fluidized, and fully static. The dataset contains 50,000 samples, including 15,000 samples of fully fluidized state, 18,000 samples of partially fluidized state, 12,000 samples of quasi-static fluidized state, and 5,000 samples of fully static state.

[0035] The training steps of the particulate flow regime recognition model specifically include training the model parameters using the stochastic gradient descent optimization algorithm, setting the learning rate to 0.001, the batch size to 32, the number of training rounds to 200, using the cross-entropy loss function to calculate the error between the model output and the true label, updating the model weights through the backpropagation algorithm, and verifying that the model accuracy reaches more than 95% on the test set after training.

[0036] The block size adjustment function is used to adjust the image block parameters of the particle fluidity recognition model. The block size adjustment function calculates the block size adjustment value based on three data: the resolution of the acoustic spectrum, the temperature range of the energy storage particles, and the geometry of the fluidized bed. When the block size adjustment value is in the range of 0 to 0.3, a linear weight adjustment function is used to set the block size to 8×8 pixels. When the block size adjustment value is in the range of 0.3 to 0.7, an exponential weight adjustment function is used to set the block size to 16×16 pixels. When the block size adjustment value is in the range of 0.7 to 1.0, a logarithmic weight adjustment function is used to set the block size to 32×32 pixels.

[0037] The shutdown process includes four steps: pre-shutdown preparation, graded speed reduction control, fluidization termination, and fixed bed stabilization. The pre-shutdown preparation stage takes place from 0 to 300 seconds after the shutdown command is issued; the graded speed reduction control stage takes place from 300 to 1800 seconds after the shutdown command is issued; the fluidization termination stage takes place from 1800 to 2100 seconds after the shutdown command is issued; and the fixed bed stabilization stage takes place from 2100 to 2400 seconds after the shutdown command is issued.

[0038] The specific implementation methods of the above steps are described in detail below.

[0039] The specific implementation of step S01 involves arranging temperature and acoustic sensors inside the ultra-high temperature fluidized bed according to a grid-like layout. The temperature sensors are platinum resistance thermometers with a temperature range of 0℃ to 1300℃ and a measurement accuracy of ±2℃. Nine temperature sensors are arranged in a 3×3 matrix inside the fluidized bed, with the sensor spacing being one-third of the fluidized bed's length and width. The acoustic sensors are piezoelectric sensors with a frequency response range of 20Hz to 20000Hz and a sensitivity of no less than -40dB. Four acoustic sensors are evenly arranged on the four walls of the fluidized bed. The data acquisition system synchronously acquires all sensor signals at a sampling frequency of 100Hz. The acquired temperature and acoustic signal data are preprocessed using digital filters to remove high-frequency noise and low-frequency drift, establishing a real-time monitoring dataset containing timestamps, sensor numbers, temperature values, acoustic amplitude, and frequency information. The purpose of this step is to provide accurate and reliable basic data for subsequent fluidization state analysis and shutdown control.

[0040] The specific implementation of step S02 involves calculating the velocity ratios for three deceleration stages based on four input parameters: the current fluidized gas flow rate, the temperature of the energy storage particles, the fluidized bed pressure loss, and the density of the energy storage particles, using a deceleration gradient calculation function. The velocity ratio for the first stage is set to 70% to 80% of the current flow rate, lasting 300 to 500 seconds; the velocity ratio for the second stage is set to 40% to 50% of the current flow rate, lasting 800 to 1000 seconds; and the velocity ratio for the third stage is set to 15% to 25% of the current flow rate, lasting 200 to 300 seconds. The blower speed control employs a proportional-integral-derivative (PID) control algorithm, achieving precise speed control by adjusting the output frequency of the blower's frequency converter. The speed adjustment accuracy is ±1%, and the response time does not exceed 5 seconds. Before each deceleration phase begins, the system fine-tunes the flow rate ratio based on the current temperature of the energy storage particles. When the particle temperature is above 1000℃, the deceleration ratio increases by 5% to 10%, and when the particle temperature is below 900℃, the deceleration ratio decreases by 5% to 10%. The purpose of this step is to avoid drastic changes in the fluidization state of the particles through graded deceleration control, thereby preventing particle aggregation and temperature unevenness.

[0041] The specific implementation of step S03 involves inputting the real-time acquired acoustic signal into the particle fluidity recognition model for processing. First, a Fast Fourier Transform (FFT) is performed on the acoustic signal to convert the time-domain signal into a frequency-domain signal, extracting the spectral features of the acoustic wave. Then, the spectral data is converted into a time-spectrum image format, with an image resolution of 256×256 pixels, a time window of 1 second, and a frequency resolution of 78.125Hz. The time-spectrum image is input into the particle fluidity recognition model based on a vision transformer architecture. The model analyzes the local and global features of the image through a multi-head attention mechanism, outputting a classification result of the particle fluidization degree. The classification result includes four states: fully fluidized, partially fluidized, quasi-static fluidized, and fully static. The confidence threshold for each state is set above 0.85. Simultaneously, the model outputs a distribution uniformity evaluation value, ranging from 0 to 1. An evaluation value greater than 0.8 indicates uniform particle distribution, while an evaluation value less than 0.6 indicates non-uniform particle distribution, requiring adjustment of the fluidization control parameters. The purpose of this step is to achieve real-time identification and quantitative evaluation of the fluidization state of energy storage particles through acoustic monitoring technology.

[0042] The specific implementation of step S04 involves constructing a sparse matrix of particle spatial distribution based on the energy storage particle location information acquired by the acoustic sensor. The matrix dimension is set to a pixelated representation of the fluidized bed length and width, with a pixelation precision of 1 mm × 1 mm. Non-zero elements in the matrix represent the presence of energy storage particles at the corresponding locations, with element values ​​indicating particle density. Zero elements represent the absence of energy storage particles at the corresponding locations. A multi-level progressive clustering algorithm is used to convert the sparse matrix into a dense matrix. The first-level clustering uses the K-means algorithm with a 10 × 10 grid for coarse-grained clustering, with 16 cluster centers. The second-level clustering uses a 5 × 5 grid for medium-grained clustering, with 64 cluster centers. The third-level clustering uses a 2 × 2 grid for fine-grained clustering, with 256 cluster centers. The number of iterations for each clustering level is set to 50, and the convergence threshold is set to... In the clustering process, Euclidean distance is used as the similarity metric. Each element in the resulting dense matrix represents the aggregate density and average temperature of the energy storage particles within the corresponding region. The purpose of this step is to achieve accurate quantification of the particle distribution, providing an accurate data foundation for subsequent outage timing calculations.

[0043] The specific implementation of step S05 involves calculating the optimal shutdown sequence based on the temperature decay curve and the particle distribution density matrix. The temperature decay curve is fitted using an exponential decay model, with fitting parameters including initial temperature, decay time constant, and steady-state temperature. The fitting accuracy requires a correlation coefficient greater than 0.95. The optimal shutdown time point under the constraint of temperature uniformity is solved using the Lagrange multiplier method. The constraint condition is that the standard deviation of the temperature measured by each sensor is less than 20℃. The Lagrange function is solved using the gradient descent algorithm, with a learning rate of 0.01 and 1000 iterations. Simultaneously, the shortest path algorithm in graph theory is used to optimize the shutdown step sequence. Each control node in the shutdown process is treated as a vertex of the graph, and the execution time of the control action is used as the edge weight. The Dijkstra algorithm is used to calculate the shortest path from the current state to the target state, with the path length representing the total shutdown time. The algorithm outputs an optimal shutdown time point with an accuracy of ±30 seconds. The execution order of the shutdown step sequence includes key nodes such as blower deceleration, fluidization termination, and fixed bed stabilization. The purpose of this step is to ensure the shortest possible shutdown time while meeting the requirements for temperature uniformity.

[0044] The specific implementation of step S06 involves executing precise fluidization termination control when the temperature of the energy storage particles drops to a set threshold and the particle fluidization classification result shows quasi-static fluidization. The set temperature threshold is 850℃ to 900℃, determined based on the material characteristics of the energy storage particles and subsequent process requirements. The system monitors the output results of the particle fluidity recognition model in real time. When a quasi-static fluidization state is detected for more than 5 consecutive seconds with a confidence level greater than 0.9, the blower shutdown procedure is initiated. The blower shutdown adopts a gradual shutdown method, with the rotation speed decreasing at a rate of 5% per second until it stops completely, and the total shutdown time is controlled between 60 and 120 seconds. During the blower shutdown process, the system continuously monitors the changes in the particle distribution density matrix to ensure a smooth transition of particles from a fluidized state to a fixed bed state, avoiding uneven distribution caused by sudden particle settling. After the blower stops completely, the system waits for a stabilization time of 30 to 60 seconds to ensure complete particle settling. The purpose of this step is to achieve a smooth transition of energy storage particles from a fluidized state to a fixed bed state, avoiding uneven particle distribution and local overheating.

[0045] The specific implementation of step S07 involves monitoring the temperature distribution and settling state of the energy storage particles in a fixed bed state, and verifying the uniformity of particle settling through a thermal stability function. The thermal stability function is constructed based on the heat conduction equation and the convective heat transfer coefficient. The function's input parameters include particle bed height, particle interstitial ratio, particle thermal conductivity, and ambient temperature. The output parameters are the thermal stability index and temperature gradient distribution. The system collects temperature data every 10 seconds, continuously monitoring for 300 to 600 seconds, calculating the standard deviation and maximum temperature difference of each temperature sensor reading. The standard deviation threshold is set at 15℃, and the maximum temperature difference threshold is set at 50℃. When the standard deviation is less than the threshold and the maximum temperature difference is less than the threshold, the settling is considered uniform. When any index exceeds the threshold, the system issues a local overheating alarm and records the abnormal location. Settling state assessment uses particle bed height measurement, measured using an ultrasonic level gauge with a measurement accuracy of ±1 mm. When the bed height change rate is less than 0.5 mm per minute, the settling is considered stable. The purpose of this step is to ensure the thermal stability and settling uniformity of the energy storage particles in a fixed bed state, and to prevent local overheating and uneven settling from damaging the equipment.

[0046] The specific implementation of step S08 involves recording key data from the entire shutdown process and establishing a shutdown history database. The recorded content includes temperature change curves, particle distribution state transition processes, key time nodes, control parameter settings, and abnormal event information. The temperature change curves record the temperature values ​​of each sensor during the shutdown process, with a data recording frequency of once per second. The data format includes timestamps, sensor numbers, and temperature values. The particle distribution state transition process records changes in the particle fluidization degree classification results, including the duration and transition time of each state. Key time node records include the shutdown command issuance time, the start and end times of each deceleration stage, the fluidization termination time, and the fixed bed stabilization time. The database adopts a relational database structure, establishing a shutdown record table, a temperature data table, a state transition table, and an abnormal event table, with each table's data linked by a unique shutdown batch number. The database has data query, statistical analysis, and trend prediction functions, providing data support for optimizing subsequent shutdown operations. The purpose of this step is to establish a complete shutdown process archive, providing historical data references for equipment operation optimization and fault diagnosis.

[0047] The particle fluidization recognition model employs a deep learning model based on a visual transformer architecture. The model structure comprises five main parts: an input layer, an embedding layer, an encoder layer, a classification head, and an output layer. The input layer receives a 256×256 pixel time-spectrum image with three channels, corresponding to the amplitude, phase, and frequency information of the acoustic signal. The embedding layer divides the input image into several image blocks. The block size is determined based on the acoustic spectrum resolution, the temperature range of the energy storage particles, and the geometry of the fluidized bed, and can be selected from three specifications: 8×8, 16×16, or 32×32 pixels. Each image block is mapped to a 768-dimensional embedding vector through a linear transformation. The encoder layer contains 12 transformer blocks, each consisting of a multi-head attention mechanism, layer normalization, and a feedforward neural network. The multi-head attention mechanism contains 12 attention heads, each with a dimension of 64, and the feedforward neural network has a hidden layer dimension of 3072. The classification head uses a fully connected layer structure, containing two hidden layers with 512 and 256 neurons respectively, and employs a modified linear unit function as the activation function. The output layer contains four neurons, corresponding to four fluidization states: fully fluidized, partially fluidized, quasi-static fluidized, and fully static. The output is normalized using a soft-maximum function.

[0048] The establishment of the training dataset for the particulate fluidization identification model includes four stages: data collection, preprocessing, feature extraction, and annotation. In the data collection stage, the fluidized bed equipment was run under different temperature conditions, ranging from 800℃ to 1200℃, with a temperature point set every 50℃. Each temperature point was run for 2 hours, collecting acoustic signal data at a sampling frequency of 48000Hz. In the preprocessing stage, the raw acoustic signal was filtered and denoised. A Butterworth filter was used to remove frequency components below 20Hz and above 20000Hz, and a wavelet denoising algorithm was used to reduce signal noise. In the feature extraction stage, the preprocessed acoustic signal was converted into a time-spectrum image using a short-time Fourier transform method. The window length was set to 2048 sampling points, the overlap rate was set to 75%, and the generated time-spectrum image resolution was 256×256 pixels. In the annotation stage, experienced engineers manually annotated each time-spectrum image based on synchronously collected temperature data, pressure data, and visual observation results to determine the corresponding particle fluidization degree classification. The final dataset contains 50,000 samples, including 15,000 fully fluidized state samples, 18,000 partially fluidized state samples, 12,000 quasi-static fluidized state samples, and 5,000 fully static state samples. The dataset is divided into training, validation, and test sets in a ratio of 7:2:1.

[0049] The model training employs the stochastic gradient descent optimization algorithm, with an initial learning rate set to 0.001. A cosine annealing learning rate scheduling strategy is used, with a minimum learning rate of [missing value]. The batch size was set to 32, and the total number of training epochs was 200. The cross-entropy loss function was used, and the model weights were updated via backpropagation. The gradient clipping threshold was set to 1.0 to prevent gradient explosion. Data augmentation techniques were employed during training to improve the model's generalization ability, including random rotation, scaling, and noise addition, with a data augmentation probability set to 0.3. Model training was performed on a computing platform equipped with a graphics processing unit (GPU), with a single training session taking approximately 8 hours. After training, the classification accuracy on the test set reached 95.2%, with precision and recall exceeding 93% for each category.

[0050] The key technical ideas of this invention include four aspects: multi-level progressive clustering algorithm, acoustic monitoring and machine learning fusion technology, graded deceleration control strategy, and temperature uniformity constraint optimization method.

[0051] The multi-level progressive clustering algorithm achieves efficient transformation from sparse to dense matrices through three-level clustering. Compared to traditional direct densification methods, this algorithm significantly reduces computational complexity and memory consumption while maintaining the integrity of granular distribution information. Traditional methods require global processing of the entire granular distribution matrix, with computational complexity increasing quadratically with the matrix dimension. The multi-level progressive clustering algorithm, however, reduces computational complexity from quadratic to linear through hierarchical processing. Furthermore, the convergence property of K-means clustering ensures the stability and accuracy of the clustering results. The core advantage of this algorithm lies in capturing granular distribution features at different scales through progressive refinement. Coarse-grained clustering obtains the global distribution pattern, while fine-grained clustering preserves local distribution details, thus achieving a balance between computational efficiency and information fidelity.

[0052] The fusion of acoustic monitoring and machine learning technology combines traditional acoustic signal analysis with modern deep learning methods, overcoming the technical bottleneck of insufficient recognition accuracy in complex operating conditions inherent in traditional acoustic monitoring methods. Traditional acoustic monitoring relies primarily on simple spectral analysis and threshold judgment, making it susceptible to environmental noise and signal variations. In contrast, deep learning models based on a vision transformer architecture can automatically learn the deep features of acoustic signals and capture the spatiotemporal correlation of signals through a multi-head attention mechanism, significantly improving the accuracy and robustness of fluidized bed state recognition. The innovation of this technology lies in converting acoustic signals into time-spectral images, leveraging the mature algorithms of image recognition to handle one-dimensional signal problems, and adapting to different operating conditions through an adaptive block size adjustment mechanism, enabling reliable application of acoustic monitoring technology in high-temperature fluidized bed environments.

[0053] The staged deceleration control strategy, based on fluid mechanics principles and particle dynamics theory, avoids the abrupt changes in particle fluidization state caused by traditional one-step shutdown through a three-stage progressive control. Traditional shutdown methods typically employ simple approaches such as direct shutdown or linear deceleration, which easily lead to rapid particle settling and uneven particle distribution. In contrast, the staged deceleration control strategy designs differentiated deceleration curves based on the physical characteristics of particle fluidization. The first stage maintains a high flow rate to keep particles suspended; the second stage moderately reduces the flow rate to promote orderly particle arrangement; and the third stage achieves low-speed fluidization to ensure smooth particle settling. The technical advantage of this strategy lies in its full consideration of the dynamic balance between particle inertia, fluid resistance, and gravity. By precisely controlling the rate of change of flow rate, it achieves controllable transitions in the particle fluidization state, avoiding particle aggregation and temperature stratification phenomena common in traditional methods.

[0054] The synergistic effect of the above-mentioned technical approaches forms a complete and precise shutdown control system, which has significant technical advantages compared to traditional experience-based shutdown methods. Multi-level progressive clustering algorithms provide an efficient data processing foundation for acoustic monitoring; the fusion of acoustic monitoring and machine learning technologies provides accurate state feedback for graded deceleration control; the graded deceleration control strategy creates favorable physical conditions for temperature uniformity constraint optimization; and the temperature uniformity constraint optimization method provides scientific timing planning for the entire shutdown process. This synergistic effect of multi-technology integration realizes the transformation from passive response to active control, converting traditional qualitative operations into quantitative and intelligent precise control. It fundamentally solves the technical challenges in the shutdown process of high-temperature energy storage equipment, providing reliable technical assurance for the safe and stable operation of thermal energy storage power plants.

[0055] It should be noted that this invention also solves the technical problem of insufficient real-time accuracy in identifying the fluidization state of energy storage particles. Traditional technologies mainly rely on temperature and pressure sensors to monitor the fluidization state, but these sensors can only provide local physical parameter information and cannot accurately reflect the true fluidization state of the energy storage particles throughout the fluidized bed. This invention addresses this problem by employing acoustic monitoring technology combined with Vision... The Transformer architecture-based particle flow regime identification model can analyze the acoustic spectrum characteristics generated by collisions of energy storage particles. It utilizes a multi-head attention mechanism to capture key features in the acoustic signal and converts the acoustic spectrum into processable image information through an image segmentation mechanism. This accurately identifies four states: fully fluidized, partially fluidized, quasi-static fluidized, and fully static, providing precise state judgment for shutdown control. Furthermore, this invention addresses the technical problem of low processing efficiency for spatial distribution information of energy storage particles. Traditional technologies typically employ direct construction of dense matrices when processing the spatial distribution information of large-scale energy storage particles. This method is computationally complex, memory-intensive, and inefficient. This invention innovatively proposes a multi-level progressive clustering algorithm, employing three levels of progressive refinement from coarse to fine granular clustering, which effectively... By reducing computational complexity while maintaining the integrity of particle distribution information, the invention achieves efficient conversion from sparse to dense matrices through multi-level application of the K-means algorithm, effectively solving the efficiency problem of large-scale data processing. Furthermore, this invention addresses the technical problem of insufficient mathematical theoretical support for shutdown timing optimization. Traditional shutdown control is mainly based on empirical formulas and simple linear control strategies, lacking rigorous mathematical optimization theoretical support, making it difficult to find the optimal shutdown timing under complex multi-constraint conditions. This invention establishes a rigorous mathematical optimization model by introducing the Lagrange multiplier method and the shortest path algorithm from graph theory. The Lagrange multiplier method can solve for the optimal shutdown time point under temperature uniformity constraints, and the shortest path algorithm can optimize the shutdown step sequence, ensuring the scientific nature and reliability of the shutdown process and avoiding the uncertainty and arbitrariness of traditional empirical control methods.

[0056] Specifically, the principle of this invention is as follows: The invention solves the technical problem of uneven temperature distribution in energy storage particles primarily based on the following technical principles: First, acoustic monitoring technology can accurately identify the fluidization degree and spatial distribution state of particles by analyzing the acoustic wave spectrum characteristics generated by particle collisions, providing a precise state information basis for shutdown control. Second, a multi-level progressive clustering algorithm, through progressive refinement at three clustering levels, can transform sparse particle location information into a dense distribution matrix, achieving precise quantitative characterization of particle spatial distribution. Third, a particle fluidity recognition model based on the VisionTransformer architecture can accurately classify the fluidization state of particles through image segmentation and multi-head attention mechanisms, providing accurate state judgment basis for shutdown timing optimization. Fourth, the Lagrange multiplier method, combined with temperature decay curves and a dense particle distribution matrix, can solve for the optimal shutdown time point under temperature uniformity constraints, ensuring the uniformity of energy storage particle temperature distribution during shutdown. Finally, the graded deceleration control strategy, through precise control of three deceleration stages, enables the energy storage particles to smoothly transition from a high-speed fluidized state to a fixed bed state, avoiding the problem of uneven temperature distribution caused by abrupt shutdown, thereby achieving precise shutdown control of the thermal energy storage power plant.

[0057] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0058] In this embodiment, the specific implementation of step S01 is the same as described above, and will not be repeated in detail here.

[0059] The specific implementation of step S02 is to perform graded speed reduction blower control, wherein the mathematical expression of the speed reduction gradient calculation function is as follows: ; ; ; In the formula, These represent the flow rate ratios for the first, second, and third stages, respectively. The current fluidizing gas velocity is expressed in m / s. The reference fluidizing gas velocity is taken as 5.0 m / s; Temperature of the energy storage particles, in °C; For reference temperature, the value is taken as 1000℃; This represents the current fluidized bed pressure loss, in Pa. The reference pressure loss is set at 2000 Pa; Density of energy storage particles, in units of ; For reference particle density, the value is taken as follows: ; The speed reduction coefficients are set to 0.75, 0.45, and 0.20, respectively. The temperature decay time constant is taken as 50℃, 75℃, and 100℃, respectively. The method for obtaining these parameters is as follows: Obtained through real-time measurement using a flow meter; Data is obtained through real-time measurement using a temperature sensor; Acquired in real time through a differential pressure transmitter; Obtained by measurement with a densitometer.

[0060] The specific implementation of step S03 involves running acoustic monitoring and particle distribution identification, where the mathematical expression of the block size adjustment function is as follows: ; In the formula, Adjust the block size value; The resolution of the acoustic spectrum is in pixels. The reference spectrogram resolution is set to 256 pixels. The temperature range of the energy storage particles is expressed in °C. For reference temperature range, the value is 400℃; These are the geometric dimensions of the fluidized bed, in meters (m). For reference geometric dimensions, the value is taken as 2.0m; These are weighting coefficients, with values ​​of 0.4, 0.3, and 0.3 respectively. The block size setting rules are as follows: When When the block size is set to 8×8 pixels; when When the block size is set to 16×16 pixels; when At that time, the block size was set to 32×32 pixels. The parameter acquisition method is as follows: Obtained through calculation using spectrum analysis software; Obtained by calculating the range measured by the temperature sensor; Obtained through geometric measurements.

[0061] The specific implementation of step S04 is to establish a sparse matrix of particle distribution. The mathematical expression for constructing the sparse matrix is ​​as follows: ; In the formula, The sparse matrix is ​​the first Line 1 Column elements; For position Local particle density at a given location, in units of ; For position The number of particles at a given location. The objective function expression for K-means clustering in the multi-level progressive clustering algorithm is as follows: ; In the formula, For the first The objective function value of hierarchical clustering; For the first The number of data points in a hierarchical cluster; For the first Number of hierarchical cluster centers, among which , , ; For the first Data points in hierarchical clustering Belongs to cluster center Membership degree; For the first The first in the level cluster One data point; For the first The first in the level cluster There are 1 cluster center. The expression for constructing the dense matrix is ​​as follows: ; In the formula, For the dense matrix, the first... Line 1 Column elements; For the row and column indices of a dense matrix; This represents the number of pixels within the grid area. For corresponding dense matrix elements Grid area; Position of sparse matrix The weighting coefficients, here The meaning is the same as that of row and column indices in a sparse matrix. The parameter acquisition method is as follows: Obtained through signal inversion calculation from acoustic sensors; Obtained through image recognition algorithms; It is obtained by calculating the reciprocal of the Euclidean distance from the cluster center.

[0062] The specific implementation of step S05 involves calculating the optimal shutdown sequence. The mathematical expression for fitting the temperature decay curve is as follows: ; In the formula, For a moment The temperature value, in °C; This is the initial temperature, in °C. Time, in seconds; This is the temperature decay time constant, in seconds; This refers to the ambient temperature, expressed in °C. The temperature measurement error term is represented, with a range of ±2℃. The objective function expression for solving the optimal downtime using the Lagrange multiplier method is as follows: ; In the formula, It is a Lagrange function; The optimal downtime is expressed in seconds. The Lagrange multiplier, measured in s / ℃, is used to balance the dimensions of the objective function and the constraint functions. This is a time cost function, with units of 1 / s; Here is the temperature uniformity constraint function, with units of °C. The specific expression for the temperature uniformity constraint function is as follows: ; In the formula, This refers to the number of temperature sensors. For the first Each sensor at time Temperature value, Sensor number, unit: °C; For a moment The average temperature, in °C; The temperature standard deviation threshold is set to 20℃. The path weight calculation expression in the shortest path algorithm is as follows: ; In the formula, For the node To the node Edge weights; In graph theory, a node index represents a different state during the shutdown process. From state Transition to state The time difference, in seconds; From state Transition to state The energy difference, expressed in kJ; From state Transition to state The safety risk assessment value; These are weighting coefficients, with values ​​of 0.5, 0.3, and 0.2 respectively. The method for obtaining these parameters is as follows: The temperature was obtained by measuring the temperature sensor before shutdown. Obtained through historical data regression analysis; Acquired through measurements using an ambient temperature sensor; Acquired through control system timing records; It is obtained through calculation by the energy consumption monitoring system.

[0063] The specific implementation method of step S06 is the same as described above, and will not be repeated in detail here.

[0064] The specific implementation of step S07 involves verifying the stability of the fixed bed. The mathematical expression for the thermal stability function is as follows: ; In the formula, This represents the value of the thermal stability function. For position At any moment Temperature field distribution, in °C; These are spatial coordinates, in meters (m). This is the thermal diffusivity, in units of... ; For position At any moment Heat source item, unit is The expression for calculating the temperature gradient distribution is as follows: ; In the formula, For position At any moment The temperature gradient amplitude is expressed in °C / m. The expression for the particle settling uniformity evaluation function is as follows: ; In the formula, This is the value for assessing the uniformity of settlement; This represents the standard deviation of bed height, in meters. This represents the average height of the bed, in meters (m). This is the time of settlement onset, expressed in seconds. The settling time constant is 60s. The method for obtaining this parameter is as follows: Obtained through thermophysical property testing of particulate materials; The data was obtained through experimental measurements of heat source distribution. Obtained through measurement and calculation using an array of ultrasonic level gauges; The value is obtained through measurement and calculation using an array of ultrasonic level gauges.

[0065] The specific implementation method of step S08 is the same as described above, and will not be repeated in detail here.

[0066] It should be noted that the deceleration gradient calculation function adopts an exponential decay model combined with the principle of multi-parameter coupling. The exponential decay characteristic of the temperature term simulates the nonlinear law of particle fluidization state changing with temperature. The square root relationship of the pressure loss term reflects the square root relationship between fluidized bed pressure drop and flow velocity. The particle density term reflects the influence of particle properties on fluidization characteristics. This function can dynamically adjust the deceleration strategy according to real-time operating parameters. Compared with the traditional fixed-proportion deceleration method, it realizes adaptive deceleration control and avoids abrupt changes and uneven distribution of particle fluidization state.

[0067] In this embodiment, the block size adjustment function Based on the principle of multi-parameter linear weighted combination, the function reflects the image processing accuracy requirements through the spectral resolution term, the temperature range term reflects the influence of working condition complexity, and the geometric size term considers spatial scale factors. This function realizes the adaptive adjustment of the visual transformer model block size, which significantly improves the recognition accuracy and computational efficiency under different working conditions compared with the traditional fixed block size method.

[0068] In this embodiment, the K-means clustering objective function is based on the optimization principle of minimizing the sum of squared intra-cluster distances. It achieves optimal classification of data points through membership functions and Euclidean distance metrics. The multi-level progressive clustering architecture gradually refines the granular distribution information from coarse to fine granularity. Compared with traditional single-level clustering methods, it significantly reduces computational complexity while maintaining the integrity of distribution information, providing a high-quality data foundation for subsequent shutdown timing optimization.

[0069] In this embodiment, the temperature decay curve fitting function is based on the exponential decay characteristics of the physical law of heat conduction. Combined with the steady-state influence of ambient temperature and the randomness of measurement error, it can accurately predict the time-varying law of particle temperature and provide a reliable temperature prediction model for shutdown sequence planning. Compared with the traditional linear model, it significantly improves the temperature prediction accuracy and realizes precise shutdown control based on the temperature evolution law.

[0070] In this embodiment, the Lagrange multiplier method objective function is based on constrained optimization theory, transforming the downtime optimization problem into a constrained mathematical programming problem. By incorporating the constraints into the objective function through the Lagrange multiplier, the optimal downtime solution is achieved under the premise of satisfying the temperature uniformity constraint. Compared with the traditional trial-and-error method, it provides a theoretically rigorous optimization framework, ensuring the optimality and reliability of the downtime process.

[0071] In this embodiment, the shortest path algorithm weight function comprehensively considers three dimensions: time cost, energy cost, and safety risk. It achieves multi-objective optimization through weighted linear combination, and models the shutdown process as the shortest path problem in graph theory. Compared with the traditional sequential execution method, it can globally optimize the shutdown step sequence and achieve the comprehensive optimal shutdown path with the shortest time, lowest energy consumption, and lowest risk.

[0072] In this embodiment, the thermal stability function is based on the physical principle of the two-dimensional heat conduction partial differential equation and combined with the distribution characteristics of the heat source term. It can describe the evolution law of the temperature field inside the particle bed. By solving this function, the thermal stability under the fixed bed state can be predicted and evaluated. Compared with the traditional empirical judgment method, it provides a quantitative thermal stability assessment tool, ensuring the safe and stable operation of the particle bed after shutdown.

[0073] In this embodiment, the temperature gradient distribution calculation is based on the mathematical definition of the gradient operator. The distribution information of the temperature change rate is obtained by calculating the spatial derivative of the temperature field. This index can identify the temperature non-uniform region inside the bed and provide a basis for local overheating early warning. Compared with the traditional single-point temperature monitoring method, it realizes global monitoring and analysis of the temperature field.

[0074] In this embodiment, the settling uniformity assessment function combines the concept of statistical standard deviation and the characteristics of exponential decay time. It evaluates the uniformity of particle settling through statistical analysis of bed height. This function can quantitatively assess the changing trend of particle distribution during shutdown. Compared with traditional qualitative observation methods, it provides an objective and accurate assessment index of settling uniformity, providing a scientific basis for shutdown quality control.

[0075] It should be noted that the variables involved in this embodiment are explained in detail in Tables 1 and 2 below.

[0076] Table 1. Variable Explanation Table (Part 1)

[0077] Table 2. Variable Explanation Table (Part Two)

[0078] To better understand and implement this invention, the following is a specific application scenario of Example 2: A technical team received a task to perform routine maintenance shutdown operations on a 100MW thermal energy storage power station. This power station uses spherical ceramic matrix composite particles as the energy storage medium, with a particle diameter of 3.5 mm, an operating temperature of 1100℃, and an ultra-high temperature fluidized bed size of 8m × 6m × 4m, with a total energy storage particle volume of approximately 350 tons. Because traditional shutdown methods often result in uneven particle distribution and localized overheating, the technical team decided to use the precise shutdown method of this invention.

[0079] The technical team first executed step S01, initiating pre-shutdown status monitoring. Nine platinum resistance temperature sensors were arranged in a 3×3 matrix inside the fluidized bed, with a sensor spacing of 2.67m × 2.0m, a temperature range of 0℃ to 1300℃, and a measurement accuracy of ±2℃. Four piezoelectric acoustic sensors were evenly distributed on the four walls of the fluidized bed, with a frequency response range of 20Hz to 20000Hz and a sensitivity of -38dB. The data acquisition system synchronously acquired all sensor signals at a sampling frequency of 100Hz, establishing a real-time monitoring dataset containing timestamps, sensor numbers, temperature values, acoustic amplitude, and frequency information. The sensor configuration parameters shown in Table 2 have been precisely calibrated and verified.

[0080] Table 3 Sensor Configuration Parameter Table

[0081] Next, step S02 is executed to implement staged speed-reducing blower control. The current fluidizing gas velocity is 4.2 m / s, the average temperature of the energy storage particles is 1100℃, the fluidized bed pressure loss is 1850 Pa, and the energy storage particle density is 3500 kg / m³. The technical team determined the flow rate ratios for the three deceleration stages based on the deceleration gradient calculation function. The first stage had a flow rate ratio of 0.76 and a duration of 420 seconds; the second stage had a flow rate ratio of 0.43 and a duration of 900 seconds; and the third stage had a flow rate ratio of 0.18 and a duration of 280 seconds. The blower speed control employed a proportional-integral-derivative (PID) control algorithm, precisely adjusting the output frequency via a frequency converter, achieving a speed adjustment accuracy of ±1% and a response time of 4.5 seconds.

[0082] Step S03 involves acoustic monitoring and particle distribution identification. The real-time acoustic signal is subjected to a Fast Fourier Transform (FFT) to convert it into a 256×256 pixel time-spectrum image with a 1-second time window and a frequency resolution of 78.125Hz. Based on the current acoustic spectrum resolution of 256 pixels, the energy storage particle temperature range of 400℃, and the fluidized bed geometry of 8m, a block size adjustment value of 0.65 is calculated. Therefore, a 16×16 pixel image block parameter is selected. The particle fluidity identification model output shows that the current state is fully fluidized, with a confidence level of 0.92 and a distribution uniformity evaluation value of 0.85, indicating a relatively uniform particle distribution.

[0083] Step S04: Establish a sparse matrix for particle distribution. Based on the particle location information acquired by the acoustic sensor, construct an 8000×6000 dimension sparse matrix with a pixelation precision of 1 mm × 1 mm. Process using a multi-level progressive clustering algorithm: the first level uses a 10×10 grid coarse-grained clustering with 16 cluster centers; the second level uses a 5×5 grid medium-grained clustering with 64 cluster centers; and the third level uses a 2×2 grid fine-grained clustering with 256 cluster centers. Each clustering level is iterated 50 times, with a convergence threshold... Finally, each element in the dense matrix represents the particle polymerization density and average temperature value within the corresponding region.

[0084] Step S05 calculates the optimal shutdown sequence. A temperature decay curve is fitted based on historical data, with an initial temperature of 1100℃, a decay time constant of 1350 seconds, and an ambient temperature of 45℃. The Lagrange multiplier method is used to solve for the optimal shutdown time under the constraint of temperature uniformity (the standard deviation of temperature from each sensor is less than 20℃). After 1000 iterations of the gradient descent algorithm with a learning rate of 0.01, the optimal shutdown time is found to be 1847 seconds. The shortest path algorithm is used to optimize the shutdown step sequence. The shutdown process is modeled as a directed graph containing 15 state nodes, and the shortest path from the current state to the target state is calculated, with a total execution time of 1680 seconds. As shown in Table 3, the control parameters for each stage of the shutdown are accurately optimized.

[0085] Table 4 Control Parameters for Each Stage of Shutdown

[0086] Step S06 executes precise fluidization termination control. When the temperature of the energy storage particles drops to 875℃ and the particle flow regime identification model detects a quasi-static fluidization state for 8 consecutive seconds with a confidence level of 0.94, the blower shutdown procedure is initiated. The blower speed decreases at a rate of 5% per second, with a total shutdown time of 95 seconds. During the shutdown process, the system continuously monitors changes in the particle distribution density matrix to ensure a smooth transition of the particles to a fixed bed state. After the blower completely stops, a 45-second stabilization period is allowed to ensure complete particle settling. The temperature change curve during the shutdown process is shown in the figure below. Figure 2 As shown.

[0087] Step S07 verifies the stability of the fixed bed. The temperature distribution and settling state of the particles in the fixed bed are monitored, with temperature data collected every 10 seconds for 480 seconds. The standard deviation of each temperature sensor reading is calculated to be 13.2℃, and the maximum temperature difference is 42℃, both less than the set threshold, indicating uniform settling. The bed surface height is measured using an ultrasonic level gauge; the bed height change rate is 0.3 mm per minute, less than the 0.5 mm threshold, indicating stable settling. The thermal stability function calculation shows that the maximum internal temperature gradient amplitude is 15.8℃ / m, with no localized overheating. Figure 3 As shown, the particle settling uniformity assessment value gradually increased from 0.85 at the start of shutdown to 0.96, indicating that the settling process was stable and orderly.

[0088] Step S08 completes the shutdown status recording. The temperature change curves throughout the shutdown process are recorded, showing that the temperature of the nine sensors steadily decreased from 1100℃ to 850℃, with a smooth cooling process and no abrupt changes. The particle distribution state transitioned from fully fluidized to partially fluidized, quasi-static fluidized, and finally to fully static, with transition times of 720 seconds, 1620 seconds, 1900 seconds, and 2020 seconds, respectively. Key time nodes include the shutdown command issuance time (0 seconds), the start times of each deceleration stage (300 seconds, 720 seconds, 1620 seconds), the fluidization termination time (1900 seconds), and the fixed bed stabilization time (2020 seconds). A relational database containing a shutdown record table, a temperature data table, and a state transition table is established, linking the data in each table through the shutdown batch number 2024-11-15-001, providing data support for subsequent shutdown operation optimization.

[0089] The entire shutdown process lasted 2400 seconds, 780 seconds shorter than traditional shutdown methods, achieving a stable decrease and uniform distribution of particle temperature. Comparison with particle distribution monitoring after traditional shutdown revealed an 85% reduction in localized hot spots, a 32% improvement in particle settling uniformity, and a 40% reduction in equipment thermal stress. Specifically, for example… Figure 4 As shown, no particle aggregation, stratification, or localized overheating occurred during the shutdown process, ensuring equipment safety and the reliability of subsequent startups.

[0090] The technological advancements of this invention compared to traditional shutdown methods are primarily reflected in the fundamental transformation of the control mechanism. Traditional methods rely on experience to set fixed shutdown parameters, failing to adapt to real-time changes in operating conditions. This invention achieves adaptive control through multi-parameter coupled deceleration gradient calculation, dynamically adjusting the control strategy based on real-time feedback from particle temperature, fluidization state, and pressure loss. Traditional methods lack precise identification of particle fluidization state, relying solely on macroscopic parameters such as temperature and pressure to determine shutdown timing. This invention employs a fusion of acoustic monitoring and machine learning technology, accurately identifying the microscopic fluidization degree of particles, achieving a leap from qualitative judgment to quantitative analysis. Traditional methods use linear or step-like shutdown methods, easily causing abrupt changes in particle fluidization state. This invention, through a graded deceleration control strategy, fully considers the physical laws of particle fluidization, achieving controllable and smooth transitions in particle state. Traditional methods lack optimized shutdown sequence design, often leading to prolonged shutdown times or increased safety risks. This invention uses the Lagrange multiplier method and graph theory shortest path algorithm to achieve global optimization of shutdown time and path while satisfying safety constraints. Traditional methods lack scientific means of stability assessment after shutdown. This invention establishes a stability verification system based on heat conduction theory, providing a quantitative assessment standard for shutdown quality.

[0091] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for precise shutdown of a thermal energy storage power station, characterized in that, include: Multiple temperature and acoustic sensors are installed inside the ultra-high temperature fluidized bed to collect temperature distribution data and acoustic signal data of the energy storage particles under fluidized state, establishing a real-time monitoring dataset. Based on the current fluidizing gas velocity and energy storage particle temperature, a deceleration gradient calculation function is used to calculate the velocity ratio of the three deceleration stages, gradually reducing the blower speed to control the fluidizing gas velocity transition from a high-speed fluidization state to a low-speed fluidization state. The real-time acoustic signal is input into the particle fluidity recognition model to identify the fluidization degree and spatial distribution state of the energy storage particles, outputting particle fluidization degree classification results and distribution uniformity evaluation values. Based on the location information of the energy storage particles obtained by the acoustic sensors, a sparse matrix of particle spatial distribution is constructed. The sparse matrix is ​​then converted into a dense matrix through a multi-level progressive clustering algorithm, achieving accurate quantification of the particle distribution state. Based on the temperature decay curve and the dense particle distribution matrix, the optimal shutdown time point under the temperature uniformity constraint is solved using the Lagrange multiplier method. At the same time, the shortest path algorithm in graph theory is used to optimize the shutdown step sequence. When the temperature of the energy storage particles drops to the set threshold and the particle fluidization degree classification result shows quasi-static fluidization, the blower is turned off, allowing the energy storage particles to completely transition from the fluidized state to the fixed bed state.

2. The method according to claim 1, characterized in that, After shutting down the blower, the process also includes: verifying the stability of the fixed bed, monitoring the temperature distribution and settling state of the energy storage particles in the fixed bed state, verifying the uniformity of particle settling through the thermal stability function; completing the shutdown status record, recording key data of the entire shutdown process, and establishing a shutdown history database.

3. The method according to claim 2, characterized in that, The deceleration gradient calculation function is used to calculate the velocity ratio of the three deceleration stages. The inputs include the current fluidizing gas velocity, energy storage particle temperature, fluidized bed pressure loss, and energy storage particle density. The outputs are the velocity ratio of the first stage, the velocity ratio of the second stage, and the velocity ratio of the third stage.

4. The method according to claim 3, characterized in that, The acoustic monitoring technology uses a sound wave sensor with a frequency range of 20Hz to 20000Hz to determine the fluidization state of the particles by analyzing the sound wave spectrum characteristics generated by the collision of energy storage particles.

5. The method according to claim 4, characterized in that, The particle fluidization degree classification includes four states: fully fluidized, partially fluidized, quasi-static fluidized, and fully static. Each state corresponds to different acoustic spectrum characteristics and temperature distribution patterns.

6. The method according to claim 5, characterized in that, The multi-level progressive clustering algorithm includes three clustering levels: the first level is coarse-grained clustering, which initially clusters the sparse matrix of particle distribution according to a 10×10 grid; the second level is medium-grained clustering, which refines the clustering results of the first level according to a 5×5 grid; and the third level is fine-grained clustering, which finally clusters the clustering results of the second level according to a 2×2 grid.

7. The method according to claim 6, characterized in that, The sparse particle distribution matrix is ​​a two-dimensional matrix established based on the position coordinates of the energy storage particles. The matrix dimension is a pixelated representation of the length and width of the fluidized bed. Non-zero elements in the matrix indicate the presence of energy storage particles at the corresponding positions, while zero elements indicate the absence of energy storage particles at the corresponding positions.

8. The method according to claim 7, characterized in that, The particle distribution density matrix is ​​a dense matrix obtained by processing the particle distribution sparse matrix through a multi-level progressive clustering algorithm. Each element in the matrix is ​​a non-zero value, representing the aggregation density information of energy storage particles in the corresponding region.

9. The method according to claim 8, characterized in that, The specific structure of the particle fluidization recognition model is an image analysis model based on the Vision Transformer architecture, which includes an encoder layer, a multi-head attention mechanism, and a classification output layer. The block size parameter of the image segmentation mechanism is determined based on three parameters: the resolution of the acoustic spectrum, the temperature range of the energy storage particles, and the geometry of the fluidized bed.

10. The method according to claim 9, characterized in that, The shutdown process includes four steps: pre-shutdown preparation, graded deceleration control, fluidization termination, and fixed bed stabilization. The pre-shutdown preparation stage takes place from 0 to 300 seconds after the shutdown command is issued, the graded deceleration control stage takes place from 300 to 1800 seconds after the shutdown command is issued, and the fluidization termination stage takes place from 1800 to 2100 seconds after the shutdown command is issued.

Images