Calcium-based particle optimal circulation heat storage performance prediction method based on CPFD numerical simulation

Abstract

The invention discloses a calcium-based particle optimal circulation heat storage performance prediction method based on CPFD numerical simulation. The calcium-based particle optimal circulation heat storage performance prediction method comprises the following steps: (1) carrying out a pressurized fluidized calcium-based particle heat storage performance experiment in a small fluidized bed; (2) constructing a three-dimensional mathematical model of the small fluidized bed experiment table, and performing multi-working-condition numerical simulation by using the three-dimensional mathematical model; (3) constructing a space-time model for predicting the optimal circulating heat storage performance of the calcium-based particles in the small fluidized bed in the pressurized fluidized state on the basis of the numerical simulation data; the method provides reference for promoting the calcium-based material to be further suitable for an actual process environment.

Description

Technical Field

[0001] The present invention relates to the field of energy technologies, and particularly to a method for predicting the optimal cyclic heat storage performance of calcium-based particles based on CPFD numerical simulation. Background Art

[0002] The calcium looping-thermal power plant system based on the calcium looping heat storage / release reaction principle generally consists of a solar calcination reactor, a carbonation reactor, a CO 2 compression and storage tank, a CaCO 3 and CaO solid material storage tank, etc. Currently, fluidized beds have been considered suitable as solar calcination reactors because they have extremely high heat transfer coefficients, high thermal diffusivities, efficient solid mixing, easily adaptable residence times, and the possibility of continuous operation (in the circulating fluidized bed mode). In addition, regarding carbonation reactors, some studies have shown that fixed-bed reactors and thermogravimetric analyzers are suitable reactors commonly used in current laboratory research. However, in most studies of CaL-CSP systems, fluidized beds are selected as carbonation reactors to maximize the power cycle integration efficiency in the carbonation zone. This is because the use of fluidized bed reactors offers more significant theoretical advantages. On the one hand, efficient gas-solid mixing and heat and mass transfer ensure higher thermal efficiency. On the other hand, with a large heat transfer coefficient, the required heat transfer surface is smaller, and due to the rapid establishment of thermal equilibrium, the system can quickly respond to load changes. In fact, most current CO 2 capture CaL pilot plants are also fluidized bed reactors. In addition, the carbonation pressure of the carbonation reaction is also one of the important factors affecting the heat storage performance of calcium-based materials. Increasing the carbonation pressure can make the exothermic reaction proceed at a faster rate. However, too high a carbonation pressure will lead to the complication of the design of equipment components.

[0003] Currently, although the carbonation reactors for heat storage of calcium-based materials have the same basic requirements as the existing CaL CO 2 capture reactors and are relatively mature technically, there are still few studies using pressurized fluidized beds as carbonation reactors in current research on calcium-based material heat storage experiments. At the same time, due to the complex and variable operating conditions of pressurized fluidized beds, it is too costly in terms of time and operating costs to find the optimal cyclic heat storage performance of calcium-based particles based on experiments. How to quickly and accurately obtain the verification of the cyclic heat storage performance under corresponding operating conditions has become the biggest problem. Summary of the Invention

[0004] Object of the Invention: The object of the present invention is to provide a method for predicting the optimal cyclic heat storage performance of calcium-based particles based on CPFD numerical simulation to solve the problems of high cost, long time, and inability to accurately know the operating conditions under the optimal cyclic heat storage performance of calcium-based pellets in a pressurized fluidized bed in the existing technical methods.

[0005] Technical solution: A method for predicting the optimal cyclic heat storage performance of calcium-based particles based on CPFD numerical simulation according to the present invention includes the following steps:

[0006] (1) Conduct an experiment on the heat storage performance of calcium-based particles in a pressurized fluidized state in a small fluidized bed;

[0007] (2) Construct a three-dimensional mathematical model of the small fluidized bed experimental platform and perform numerical simulations under multiple working conditions using the three-dimensional mathematical model;

[0008] (3) Based on the numerical simulation data, construct a spatio-temporal model for predicting the optimal cyclic heat storage performance of calcium-based particles in a pressurized fluidized state in a small fluidized bed.

[0009] Furthermore, step (1) is specifically as follows: Specifically, the rotary granulation method is used to prepare experimental calcium-based pellets with 5% TiO 2 as an inert doping carrier and 10% bagasse as a pore-forming agent; among them, the particle size is 0.6 - 1 mm, and a cyclic heat storage experiment is carried out in a small pressurized fluidized bed.

[0010] Furthermore, in step (1), the specific process of the cyclic heat storage experiment is as follows: First, after the temperature in the constant temperature section rises to the specified carbonation temperature, where the carbonation temperature is 800 - 900 °C; weigh about 40 g of the sample, quickly put it into the central heating tube at one time and seal it with a flange; during the carbonation reaction stage, pure CO 2 is introduced, the gas flow rate is adjusted to quickly increase the pressure to the specified pressure, and then a pressurized fluidized carbonation reaction is carried out. After the first carbonation ends in 30 min, the temperature is adjusted to the calcination temperature of 800 - 900 °C; during this process, the pure CO 2 atmosphere remains unchanged. When the specified calcination temperature is reached, the reaction gas is converted to pure N 2 , the gas flow rate is adjusted to keep the sample in a bubbling fluidized state; during the calcination stage, a gas analyzer is used to record the CO 2 concentration in the outlet tail gas in real time until the CO 2 concentration drops to 0, and the sample is completely calcined. Repeat the above steps, which is the pressurized carbonation cyclic heat storage experiment of calcium-based pellets; among them, the range of the analyzer is 0 - 100% CO2, and the accuracy is 0.01%.

[0011] Further, in step (1), before each carbonation / calcination experiment is started, an empty bed experiment is carried out once as a blank control without placing samples, and one cycle is completed according to the above steps; the fluidized bed used in the experiment is as follows: the inner diameter of the central heating tube is 40 mm, the height is 1300 mm, the material of the central heating tube is 0Cr25Ni20 high-temperature resistant alloy steel, with a thickness of 6-8 mm; the furnace body is heated by resistance wires, and the upper 300 mm of the reaction tube is not heated; the designed temperature: the highest temperature of the reactor is 950 °C, and the working conditions are set at 800-900 °C; among them, it is divided into 3 groups according to 50 °C, namely 800, 850, and 900 °C; the designed pressure meets the requirement of a maximum of 1.0 MPa, and the working conditions are set at 0.1-0.8 MPa, among which it is divided according to 0.1 MPa, specifically 0.1 MPa, 0.2 MPa, 0.3 MPa, 0.4 MPa, 0.5 MPa, 0.6 MPa, 0.7 MPa, 0.8 MPa.

[0012] Further, step (2) is specifically as follows: According to the designed dimensions of the experimental bench, an accurate three-dimensional reduction of the small fluidized bed is carried out, including accurate modeling of the pipe diameter, bed height, length, definition of the heating surface, etc. At the same time, the grid is divided in the Barracuda software and compared and calibrated with 5 groups of historical data under the experimental conditions to ensure that the error of the three-dimensional mathematical model is <10%; after the calibration of the three-dimensional mathematical model is completed, numerical simulation is carried out using the Barracuda professional software to expand and supplement the experimental condition parameters, including: the temperature is extended from 800 °C - 900 °C to 1200 °C, divided according to 50 °C, specifically 950, 1000, 1050, 1100, 1150, and 1200 °C; the pressure is extended to 2 MPa, divided according to 0.1 MPa, specifically 0.9, 1.0, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9, and 2 MPa; at the same time, multiple groups of monitoring nodes are set in the three-dimensional mathematical model of the small fluidized bed, mainly for CO 2 emissions during the absorption and calcination processes of CO 2 are detected, and the time-related data set is saved to form a database.

[0013] Further, step (3) is specifically as follows: The gas flow direction in the small fluidized bed is transformed into a directed adjacency matrix, and the matrix scale is a square matrix of the selected number of nodes, with the row as the starting point pointing to the column as the fluidization direction between nodes; the training set of the spatio-temporal prediction model comes from experimental data and numerical simulation data. By taking the CO2 content, temperature and pressure changes, and node coordinates as different-dimensional convolutions, spatial features are constructed; the relationship between different moments between nodes is extracted as time features, and a spatio-temporal prediction model is constructed based on the graph convolutional neural network and the gated convolutional unit.

[0014] Furthermore, the specific process of constructing a spatio-temporal prediction model based on a graph convolutional neural network and a gated convolutional unit is as follows:

[0015] (a) Construct the graph topology of the calcium-based particle fluidized bed using the data of the flow direction and monitoring points saved in the calcium-based particle fluidized bed obtained by the CPFD numerical simulation method: the directed flow matrix A between N monitoring nodes, the pressure / temperature feature matrix X 1 , CO 2 parameter feature matrix X 2 , air velocity feature matrix X 3 , the set of monitoring point coordinates P; Add time parameters to the different feature matrix labels of N monitoring nodes.

[0016] (b) Use G=(V, E) to describe the graph topology. Each monitoring node is a vertex. V represents the set of vertices, and E is the set of edges between nodes; A represents the adjacency matrix of the graph, that is, the directed flow matrix with flow direction between monitoring nodes, used to describe the edge E;

[0017] (c) Use the graph convolutional network to obtain spatial features;

[0018] (d) Use the gated recurrent unit to extract temporal features;

[0019] (e) Combine spatial and temporal features for modeling.

[0020] Furthermore, the specific method of using the graph convolutional network to obtain spatial features is as follows: Construct a filter in the Fourier domain, which acts on the monitoring nodes and their first-order neighborhoods to extract the spatial features between the monitoring nodes; The spatial feature extraction can be made more accurate through the directed flow adjacency matrix A. The mathematical expression of the graph convolutional network is:

[0021] X (m+1) = Sigmoid(D -1 / 2 (A + I N )D -1 / 2 X m W m )

[0022] Furthermore, the specific method of using the gated recurrent unit to extract temporal features is as follows: Represent the data at different times as different layers, and establish connections between the neurons between the layers of the recurrent neural network; Among them, the state at the current moment affects the state at the next moment, and a gated structure is designed on the basis of the recurrent neural network to enable selective transmission of information.

[0023] Furthermore, the combination of spatial and temporal feature modeling is as follows: First, use the graph convolutional network to extract spatial features, then use the gated recurrent unit to extract temporal features, and introduce CPFD numerical simulation data as the true value in the training parameters. Design a loss function to minimize the prediction error. The loss function error uses the mean squared error formula:

[0024] Loss = |Y - Y’| 2

[0025] Finally, the mean absolute error will be used as the evaluation index to form a spatio-temporal prediction model for the optimal cyclic heat storage performance of calcium-based particles in a fluidized bed.

[0026] Beneficial effects: Compared with the prior art, the present invention has the following remarkable advantages: The present invention mainly aims at granulated calcium-based pellets, and calibrates a three-dimensional mathematical model with an allowable error range using the heat storage performance data in the pressurized fluidized state experiment. Based on the CPFD numerical simulation technology, multi-condition numerical simulations are carried out in professional software to further expand the experimental data. Combining neural network technology, a spatio-temporal prediction model for the optimal cyclic heat storage performance of calcium-based particles in a small pressurized fluidized bed is constructed. The carbonation pressure, carbonation temperature, fluidization velocity, and calcination temperature on the cyclic heat storage performance of calcium-based pellets are predicted in advance. This model can accurately predict the optimal conditions, calculate quickly, and the results are accurate, providing a reference for promoting the further application of calcium-based materials in the actual process environment. Brief Description of the Drawings

[0027] Figure 1 is the preparation process of the calcium-based pellets of the present invention;

[0028] Figure 2 is a schematic diagram of the pressurized fluidized bed experimental device of the present invention;

[0029] Figure 3 is the logic diagram of the prediction model for the optimal cyclic heat storage performance of calcium-based particles in a pressurized fluidized bed based on spatio-temporal modeling of the present invention;

[0030] Figure 4 is a schematic diagram of the construction of the three-dimensional mathematical model of the present invention. Detailed Embodiments

[0031] The technical solutions of the present invention will be further described below with reference to the accompanying drawings.

[0032] As Figure 1 -4 shows, an embodiment of the present invention provides a method for predicting the optimal cyclic heat storage performance of calcium-based particles based on CPFD numerical simulation, including the following steps:

[0033] (1) Using the rotary granulation method, experimental calcium-based pellets (particle size: 0.6 - 1 mm) were prepared with 5% (mass fraction) TiO2 as the inert doping carrier and 10% bagasse as the pore former, and the cyclic heat storage experiment was carried out in a small pressurized fluidized bed. First, after the temperature in the constant temperature section rose to the specified carbonation temperature (800 - 900 °C), about 40 g of the sample was weighed and quickly put into the central heating tube at one time and sealed with a flange. In the carbonation reaction stage, pure CO2 was introduced, the gas flow was adjusted to quickly increase the pressure to the specified pressure, and then the pressurized fluidized carbonation reaction was carried out. After the first carbonation (30 min) ended, the temperature was adjusted to the calcination temperature (800 - 900 °C). During this process, the pure CO2 atmosphere was kept unchanged to avoid calcination of the sample. When the specified calcination temperature was reached, the reaction gas was converted to pure N2, and the gas flow was adjusted to keep the sample in a bubbling fluidized state. During the calcination stage, a gas analyzer (range: 0 - 100% CO2, accuracy: 0.01%) was used to record the CO2 concentration in the outlet tail gas in real time until the CO2 concentration dropped to 0, and the sample was completely calcined. Repeating the above steps is the pressurized carbonation cyclic heat storage experiment of calcium-based pellets. Each time before starting the cyclic carbonation / calcination experiment, an empty bed experiment was carried out as a blank control without placing the sample, and one cycle was completed according to the above steps. The fluidized bed used in this experiment was: the inner diameter of the central heating tube was 40 mm, the height was 1300 mm, the material of the central heating tube was 0Cr25Ni20 high-temperature resistant alloy steel, with a thickness of 6 - 8 mm; the furnace body was heated by resistance wires, and the upper 300 mm of the reaction tube was not heated. Design temperature: the maximum reactor temperature was 950 °C, and the working conditions were set at 800 - 900 °C (divided into 3 groups according to 50 °C, 800, 850, 900 °C); the design pressure met the maximum requirement of 1.0 MPa, and the working conditions were set at 0.1 - 0.8 MPa, divided according to 0.1 MPa, specifically 0.1 MPa, 0.2 MPa, 0.3 MPa, 0.4 MPa, 0.5 MPa, 0.6 MPa, 0.7 MPa, 0.8 MPa.

[0034] About 40 g of the added material, and the static height of the material bed layer was 10 cm. The designed length of the preheater was 50 - 60 mm, the thickness of the insulation layer was 50 - 100 mm, the designed temperature was about 400 - 600 °C, and it was heated by resistance wires, and it needed to meet a pressure of 2.0 MPa. The present invention uses the effective heat storage conversion rate and the effective heat storage density to measure the heat storage performance of the material. The effective conversion rate represents the ratio of the mass of CaO actually reacting during each carbonation process to the total mass of the sample before the carbonation reaction. According to the CO 2 concentration, as shown in Equation (1):

[0035]

[0036] In the formula, N is the number of heat storage cycles; XN is the effective conversion rate of the Nth heat storage cycle of the calcium-based pellets; M CaO is the molar mass of CaO, g / mol; m 0 is the total mass of the sample before the carbonation reaction, g; t 0 is the time required for the complete calcination reaction, min; Q is the gas flow rate, L / min; α CO2 (N,T) is the CO in the tail gas at the t-th minute of the Nth heat storage cycle 2 concentration, vol%; α CO2，0 (0,T) is the CO in the tail gas at the t-th minute during the empty bed experiment 2 concentration, vol%.

[0037] The effective heat storage density represents the maximum heat that can be released per unit mass of the calcium-based material during each carbonation process, as shown in Equation (2).

[0038]

[0039] In the formula, Q g,N is the mass heat storage density of the calcium-based material in the Nth heat storage cycle, kJ / kg; △H 0 is the reaction heat of the carbonation reaction under standard conditions, which is calculated as 178 kJ / mol in this article.

[0040] (2) According to the designed dimensions of the experimental bench, accurately three-dimensionally restore the small fluidized bed, including accurately modeling the pipe diameter, bed height, length, definition of the heating surface, etc. At the same time, divide the grid in the Barracuda software and compare and calibrate it with 5 groups of historical data under the experimental conditions to ensure that the error of the three-dimensional mathematical model is <10%. After the calibration of the three-dimensional mathematical model is completed, use professional software for numerical simulation to expand and supplement the experimental condition parameters, including: expanding the temperature from 800°C - 900°C to 1200°C, dividing it by 50°C, specifically divided into 950, 1000, 1050, 1100, 1150 and 1200°C; to explore the thermal cycle performance of calcium-based particles under high-pressure environments, expand the pressure to 2 MPa, divide it by 0.1 MPa, specifically 0.9, 1.0, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9 and 2 MPa; at the same time, set multiple monitoring nodes in the three-dimensional mathematical model of the small fluidized bed to mainly detect the CO2 emissions during the CO2 absorption and calcination processes, and save the time-related data set to form a database.

[0041] (3) Convert the gas flow direction in the small fluidized bed into a directed adjacency matrix. The matrix size is a square matrix of the selected number of nodes, with the rows pointing to the columns as the fluidization direction between nodes. The training set of the spatio-temporal prediction model is derived from experimental data and numerical simulation data. By taking the CO2 content, temperature and pressure changes, and node coordinates as different-dimensional convolutions, spatial features are constructed. The relationship between different time points between nodes is extracted as time features. A spatio-temporal prediction model is constructed based on the graph convolutional neural network and the gated convolutional unit. The specific process is as follows:

[0042] a) Construct the graph topology of the calcium-based particle fluidized bed using the flow direction in the calcium-based particle fluidized bed obtained by the CPFD numerical simulation method and the data saved at the monitoring points: the directed flow matrix A between N monitoring nodes, the pressure / temperature feature matrix X 1 , the CO2 parameter feature matrix X 2 , the gas velocity feature matrix X 3 , and the set P of monitoring point coordinates; Add time parameters to the different feature matrix labels of N monitoring nodes.

[0043] b) Describe the graph topology with G=(V, E). Each monitoring node is a vertex. V represents the set of vertices, and E is the set of edges between nodes. A represents the adjacency matrix of the graph, that is, the directed flow matrix with the flow direction between monitoring nodes, which is used to describe the edge E. The process of graph structure prediction is to learn a mapping function f between the topological graph G and the feature matrix X, where n represents the length of the input time series, and T represents the length of the output prediction series:

[0044] [X t+1 ,…,X t+T =f(G,(X t-n+1 ,…,X t ))

[0045] Since the coordinates of the monitoring nodes remain unchanged, only the pressure / temperature, CO2 parameters, gas velocity, and time change. Use graph convolution to extract features. For the N nodes after the coordinates are determined, it is necessary to define the spectral domain graph convolution for the pressure / temperature, CO2 parameters, gas velocity, and time of each node:

[0046] F(f·g)=F(f)*F(g)

[0047] where f and g represent two original signals, F(f) represents the Fourier transform of f, · represents the product operator, and * represents the convolution operator. After performing the inverse Fourier transform and Hadamard multiplication-related processing on the graph convolution expression definition, the calculation is converted into matrix multiplication; When constructing the convolutional neural network, the Chebyshev network is introduced to parameterize the convolutional kernel, avoiding the eigenvalue decomposition of the Laplacian matrix and reducing the time complexity; To avoid overfitting, normalization is used to process the weight matrix, and finally:

[0048] X (m+1) = h(D -1 / 2 (A + I N )D -1 / 2 X m W m )

[0049] where h represents the activation function, A is the directed flow adjacency matrix extracted from the three - dimensional mathematical model, D is the degree matrix of (A + I N ) and X represents the feature matrix, while W is the trainable parameter in the neural network.

[0050] c) The graph convolutional network obtains spatial features: By constructing a filter in the Fourier domain, which acts on the monitoring nodes and their first - order neighborhoods to extract the spatial features between the monitoring nodes. The spatial feature extraction can be made more accurate through the directed flow adjacency matrix A. The mathematical expression of the graph convolutional network is:

[0051] X (m+1) = Sigmoid(D -1 / 2 (A + I N )D -1 / 2 X m W m )

[0052] d) The gated recurrent unit extracts temporal features: The data at different times represents different layers. Connections are established between the neurons of different layers in the recurrent neural network layer, and the state at the current moment can affect the state at the next moment. A gating structure is designed based on the recurrent neural network to allow selective information transmission.

[0053] e) Combining spatial and temporal features for modeling: First, use the graph convolutional network to extract spatial features, then use the gated recurrent unit to extract temporal features, and introduce CPFD numerical simulation data as the true value in the training parameters. Design a loss function to minimize the prediction error. The loss function error uses the mean - squared - error formula:

[0054] Loss = |Y - Y'| 2

[0055] Finally, the mean absolute error is used as the evaluation index to form a spatio - temporal prediction model for the optimal cyclic heat storage performance in the calcium - based particle fluidized bed.

Claims

1. A method for predicting the optimal cycle heat storage performance of calcium-based particles based on CPFD numerical simulation, characterized in that: The following steps are involved: (1) Conducting heat storage performance experiments on pressurized fluidized calcium-based particles in a small fluidized bed; (2) Construct a three-dimensional mathematical model of a small fluidized bed test bench and use the three-dimensional mathematical model to perform numerical simulations of multiple working conditions; (3) Based on numerical simulation data, a spatiotemporal model for predicting the optimal cyclic heat storage performance of calcium-based particles in a pressurized fluidized bed in a small fluidized bed was constructed.

2. The method for predicting the optimal cycle heat storage performance of calcium-based particles based on CPFD numerical simulation according to claim 1 is characterized in that: Step (1) is specifically as follows: Specifically as follows: Use the rotary granulation method to prepare experimental calcium-based spherical particles with 5% TiO2 as an inert doping carrier and 10% sugarcane bagasse as a pore-forming agent; wherein the particle size is 0.6-1mm, and a cyclic heat storage experiment is carried out in a small pressurized fluidized bed.

3. The method for predicting the optimal cycle heat storage performance of calcium-based particles based on CPFD numerical simulation according to claim 2 is characterized in that: In step (1), the specific process of the cyclic heat storage experiment is as follows: first, wait until the temperature of the constant temperature section rises to the specified carbonation temperature, wherein the carbonation temperature is 800-900°C; weigh about 40g of sample, quickly put it into the central heating tube at one time and seal the flange; introduce pure CO2 in the carbonation reaction stage, adjust the gas flow rate to quickly increase the pressure to the specified pressure and then carry out pressurized fluidized carbonation reaction. After the first carbonation is completed for 30 minutes, adjust the temperature to the calcination temperature of 800-900°C; during this process, keep the pure CO2 atmosphere unchanged. When the specified calcination temperature is reached, the reaction gas is converted to pure N2, and the gas flow rate is adjusted to keep the sample in a bubbling fluidized state; in the calcination stage, use a gas analyzer to record the CO2 concentration in the outlet exhaust gas in real time until the CO2 concentration drops to 0, the sample is completely calcined, and repeat the above steps, which is the calcium-based pellet pressurized carbonation cyclic heat storage experiment; wherein the analyzer range is 0-100% CO2, and the accuracy is 0.01%.

4. The method for predicting optimal cycle heat storage performance of calcium-based particles based on CPFD numerical simulation according to claim 3 is characterized in that: In step (1), each time before the cyclic carbonation / calcination experiment is started, an empty bed experiment is carried out as a blank control, without placing a sample, and completing one cycle according to the above steps; the fluidized bed used in the experiment is: the inner diameter of the central heating tube is 40mm, the height is 1300mm, the material of the central heating tube is 0Cr25Ni20 high temperature resistant alloy steel, and the thickness is 6-8mm; the furnace body is heated by a resistance wire, and the upper 300mm of the reaction tube is not heated; design temperature: the reactor temperature is up to 950°C, and the operating condition is set to 800-900°C; wherein, it is divided into 3 groups according to 50°C, 800, 850, and 900°C; the design pressure meets the maximum demand of 1.0MPa, and the operating condition is set to 0.1-0.8MPa, wherein it is divided according to 0.1MPa, specifically 0.1MPa, 0.2MPa, 0.3MPa, 0.4MPa, 0.5MPa, 0.6MPa, 0.7MPa, and 0.8MPa.

5. The method for predicting optimal cycle heat storage performance of calcium-based particles based on CPFD numerical simulation according to claim 1 is characterized in that: Step (2) is as follows: according to the design dimensions of the experimental platform, the small fluidized bed is accurately restored in three dimensions, including accurate modeling of the pipe diameter, bed height, length, heating surface definition, etc., and the grid is divided in the Barracuda software, and compared and calibrated with 5 sets of historical data under experimental conditions to ensure that the error of the three-dimensional mathematical model is less than 10%; After the calibration of the three-dimensional mathematical model, Barracuda software was used for numerical simulation to expand and supplement the experimental operating parameters, including: the temperature was expanded from 800℃-900℃ to 1200℃, divided by 50℃, specifically 950, 1000, 1050, 1100, 1150 and 1200℃; the pressure was expanded to 2MPa, divided by 0.1MPa, specifically 0.9, 1.0, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9 and 2MPa; at the same time, multiple groups of monitoring nodes were set in the three-dimensional mathematical model of the small fluidized bed, mainly to detect the CO2 emissions during CO2 absorption and calcination, and save time-related data sets to form a database.

6. The method for predicting the optimal cycle heat storage performance of calcium-based particles based on CPFD numerical simulation according to claim 1 is characterized in that: Step (3) is as follows: the gas flow direction in the small fluidized bed is converted into a directed adjacency matrix, the matrix size is a square matrix of the selected number of nodes, and the fluidization direction between nodes is taken as the starting point of the row and the column; the training set of the spatiotemporal prediction model is derived from experimental data and numerical simulation data, and the spatial features are constructed by convolving the CO2 content, temperature and pressure changes, and node coordinates as different dimensions; the relationship between nodes at different times is extracted as time features, and a spatiotemporal prediction model is constructed based on a graph convolutional neural network and a gated convolution unit.

7. The method for predicting optimal cycle heat storage performance of calcium-based particles based on CPFD numerical simulation according to claim 6 is characterized in that: The specific process of building a spatiotemporal prediction model based on graph convolutional neural network and gated convolutional unit is as follows: (a) The flow direction in the calcium-based granular fluidized bed and the data saved at the monitoring points obtained by the CPFD numerical simulation method are used to construct the topological structure of the calcium-based granular fluidized bed diagram: the directed flow matrix A between N monitoring nodes, the pressure / temperature characteristic matrix X1, the CO2 parameter characteristic matrix X2, the gas velocity characteristic matrix X3, and the monitoring point coordinate set P; the different characteristic matrices of the N monitoring nodes are labeled with time parameters. (b) Use G = (V, E) to describe the graph topology. Each monitoring node is a vertex. V represents the set of vertices, and E represents the set of edges between nodes. A represents the adjacency matrix of the graph, that is, the directed flow matrix with flow directions between monitoring nodes, which is used to describe the edge E. (c) Using graph convolutional networks to obtain spatial features; (d) Extracting temporal features using gated recurrent units; (e) Combining spatial and temporal features for modeling.

8. The method for predicting optimal cycle heat storage performance of calcium-based particles based on CPFD numerical simulation according to claim 7 is characterized in that: The specific method of using graph convolutional networks to obtain spatial features is as follows: a filter is constructed in the Fourier domain, acting on the monitoring nodes and their first-order domains to extract the spatial features between the monitoring nodes; the directed flow adjacency matrix A can make the spatial feature extraction more accurate. The mathematical expression of the graph convolutional network is: X (m+1) =Sigmoid(D -1 / 2 (A+I N )D -1 / 2 X m W m )。 9. The method for predicting optimal cycle heat storage performance of calcium-based particles based on CPFD numerical simulation according to claim 7 is characterized in that: The specific method of extracting time features using gated recurrent units is as follows: data at different times are represented in different layers, and connections are established between neurons in the recurrent neural network layers; the state at the current moment affects the state at the next moment, and a gating structure is designed based on the recurrent neural network to enable selective transmission of information.

10. The method for predicting optimal cycle heat storage performance of calcium-based particles based on CPFD numerical simulation according to claim 7, characterized in that: The modeling of combining spatial and temporal features is as follows: first use the graph convolutional network to extract spatial features, then use the gated recurrent unit to extract temporal features, and introduce CPFD numerical simulation data as the true value in the training parameters, design the loss function to minimize the prediction error, and the loss function error uses the square difference formula: Loss=|Y-Y’| 2 Finally, the mean absolute error will be used as an evaluation index to form a spatiotemporal prediction model for the optimal cycle heat storage performance in a calcium-based granular fluidized bed.

Images