Ventilation system valve adjustment strategy calculation method based on deep learning

Abstract

The invention discloses a ventilation system valve adjustment strategy calculation method based on deep learning, and relates to the technical field of ventilation systems. The invention provides a ventilation system valve adjustment strategy calculation method based on deep learning, and the method comprises the steps: constructing a three-dimensional model of a local pipe fitting according to a ventilation system pipe network structure; according to the three-dimensional model of the local pipe fitting, numerical values of different flow velocities are obtained through analogue simulation, and a resistance coefficient of the pipe fitting is obtained through fitting; constructing a one-dimensional model according to a ventilation system pipe network structure; latin hypercube sampling is utilized, adjustment strategy sampling is carried out on the opening degree of each valve in the one-dimensional model to serve as initial parameters, and a ventilation parameter combination data set is obtained; training a deep learning algorithm model based on the ventilation parameter combination data set to obtain a valve adjustment strategy calculation model; and obtaining a valve adjusting strategy according to the target ventilation parameter. Therefore, by adopting the method, the valve adjusting strategy can be efficiently and accurately obtained, and the reliability of the simulation model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of ventilation systems, and in particular to a method for calculating a ventilation system valve adjustment strategy based on deep learning. Background Art

[0002] Spent fuel reprocessing plants play a vital role in the nuclear energy industry. In particular, the ventilation system of spent fuel reprocessing plants has the function of reducing the concentration of harmful gases and ensuring air circulation. Efficient control and regulation of the ventilation system is an important measure to ensure the production safety and personal safety of the reprocessing plants.

[0003] At present, the ventilation system of spent fuel reprocessing plants has many branches and a complex structure. The traditional manual mathematical derivation calculation method relies on complex mathematical derivation and a lot of practical experience, and the valve adjustment strategy is difficult to propose and the efficiency is low. Therefore, it is necessary to calculate the efficient valve adjustment strategy for the ventilation system. Summary of the invention

[0004] The purpose of the present invention is to provide a deep learning-based calculation method for a ventilation system valve adjustment strategy, which can accurately and efficiently output a valve adjustment strategy according to target ventilation parameters.

[0005] To achieve the above object, the present invention provides a method for calculating a ventilation system valve adjustment strategy based on deep learning, comprising the following steps:

[0006] S1. Construct a three-dimensional model of local pipe fittings according to the pipe network structure of the ventilation system;

[0007] S2. According to the three-dimensional model of the local pipe fitting, the values ​​at different flow rates are obtained through simulation, and the resistance coefficient of the pipe fitting is obtained by fitting;

[0008] S3. According to the ventilation system pipe network structure, determine the position of each component and construct a one-dimensional model;

[0009] S4, using Latin hypercube sampling, sampling the adjustment strategy of each valve opening in the one-dimensional model as the initial parameter to obtain a ventilation parameter combination data set;

[0010] S5. Training the deep learning algorithm model based on the ventilation parameter combination data set to obtain a valve adjustment strategy calculation model;

[0011] S6. According to the target ventilation parameters, the valve regulation strategy calculation model is used to output the valve regulation strategy.

[0012] Preferably, in step S1, constructing the three-dimensional model of the local pipe fitting includes setting the outlet of the local pipe fitting to a distance of twice the diameter, setting the fluid domain and dividing the unstructured grid.

[0013] Preferably, in step S2, the simulation includes setting the boundary conditions as the velocity inlet and the pressure outlet, and obtaining the inlet and outlet pressures of the local pipe fittings according to different fluid velocities at the inlet of the local pipe fittings by setting the residual, fluid density, dynamic viscosity and wall roughness, and then fitting to obtain the local resistance coefficient, as follows:

[0014]

[0015] Where ξ is the local resistance coefficient, ρ is the fluid density, v is the inlet velocity, and P in is the inlet pressure, P out is the outlet pressure.

[0016] Preferably, in step S3, the constructed one-dimensional model includes local pipe components, pressure source components, three-way components, valve components and fan components.

[0017] Preferably, step S4 comprises:

[0018] S41, sampling the adjustment strategy of each valve in the one-dimensional model according to the Latin hypercube sampling method, and randomly generating valve opening data;

[0019] S42, shuffling and recombining the order of each valve to form a valve opening combination data set;

[0020] S43, setting target ventilation parameters, fluid parameters, local resistance coefficient and valve opening, and completing the boundary condition setting of the one-dimensional model;

[0021] S44. Calculate and obtain the ventilation parameter combination data set of each key node.

[0022] Preferably, in step S5, the deep learning algorithm includes full connection between neurons, determining the neural network structure through a particle swarm optimization algorithm, and optimizing it using an adaptive matrix estimation algorithm.

[0023] Preferably, the adaptive matrix estimation algorithm is as follows:

[0024] m t =β1·m t-1 +(1-β1)·g t

[0025]

[0026]

[0027]

[0028]

[0029] In the formula, m t 、v t They represent the first-order moment and the second-order moment respectively, β1 and β2 are the decay rates of the first-order moment and the second-order moment respectively, and g t is the current gradient; are the deviation corrections for the first-order moment and the second-order moment, α is the learning rate, and ε is a constant.

[0030] Preferably, the particle swarm optimization algorithm is as follows:

[0031] v i '=w×v i +c1×rand×(p best_i -p resent_i )+c2×rand×(g best -p resent_i )

[0032] +c2×rand×(g best -p resent_i )

[0033] p' resent_i =p resent_i +v i '

[0034] In the formula, v i ' represents the updated velocity of particle i, v i represents the current speed of particle i, rand represents a random number, p' resent_i represents the updated position of particle i, p resent_i represents the current position of particle i, p best_i represents the individual optimal position found by particle i, g best represents the global optimal position found by all particles, w is the inertia weight, and c1 and c2 are learning factors.

[0035] Therefore, the present invention adopts the above-mentioned ventilation system valve adjustment strategy calculation method based on deep learning, which has the following technical effects:

[0036] (1) The three-dimensional model local simulation and one-dimensional model system simulation are combined for modeling, and the neural network structure optimization algorithm based on particle swarm optimization is used to increase the reliability of the simulation model and the accuracy of the feedforward neural network.

[0037] (2) The valve regulation strategy calculation model can calculate the nonlinear relationship between the opening of each valve and the ventilation parameters of each key node. The corresponding valve regulation strategy can be obtained by using the changing target ventilation parameters. The ventilation system can be adjusted according to the strategy to improve the ventilation system regulation efficiency.

[0038] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flow chart of the calculation method of the valve adjustment strategy of the ventilation system based on deep learning;

[0040] Figure 2 The three-dimensional model diagram of the ventilation system pipe network structure of the spent fuel reprocessing plant in the embodiment of the ventilation system valve adjustment strategy calculation method based on deep learning is shown in FIG. Figure 2 (a) is the overall pipe network structure diagram. Figure 2 (b) is the first local pipe structure, Figure 2 (c) is the second partial pipe structure;

[0041] Figure 3 The one-dimensional model diagram of the overall pipe network structure of the ventilation system of the spent fuel reprocessing plant in the embodiment of the calculation method of the ventilation system valve adjustment strategy based on deep learning is shown. Figure 3 (a) is a one-dimensional model diagram. Figure 3 (b) is the resistance coefficient curve of the local pipe;

[0042] Figure 4 The present invention is a working principle diagram of deep learning in an embodiment of a ventilation system valve adjustment strategy calculation method based on deep learning.

[0043] Reference numerals

[0044] 1. Main road; 2. Branch road 1; 3. Branch road 2; 4. Valve 1; 5. Valve 2; 6. Valve 3; 7. Fan element; 8. Pressure source atmosphere; 9. Blowing instrument gallery; 10. Corridor; 11. Three-way element. DETAILED DESCRIPTION

[0045] The present invention can be explained in more detail by the following examples. The purpose of disclosing the present invention is to protect all changes and improvements within the scope of the present invention. The present invention is not limited to the following examples.

[0046] like Figure 1 As shown, the present invention provides a method for calculating a ventilation system valve adjustment strategy based on deep learning, comprising the following steps:

[0047] S1. According to the overall pipe network structure of the ventilation system of the spent fuel reprocessing plant (including main line 1, branch line 1 2 and branch line 2 3), divide the local pipe fittings that need to be modeled; build a three-dimensional model according to the actual size of the local pipe fittings of the ventilation system of the spent fuel reprocessing plant, and extend the outlet of the local pipe fittings by a distance of twice the diameter, such as Figure 2 As shown. Figure 2As shown in (b), for the first partial pipe, the diameter of the pipe outlet is De=2ab / (a+b), where a and b are the length and width of the pipe outlet, respectively, and the length of the pipe outlet is 2De. Figure 2 As shown in (c), for the second partial pipe, the diameter of the pipe outlet is D, and the length of the pipe outlet is 2D.

[0048] Save the constructed 3D model as stl format, import it into CFD simulation software, set the fluid domain and divide the grid, use unstructured grid, set the parameter y + is 4.6, the thickness of the first boundary layer is 0.1 mm, the number of boundary layers is 10, and the layer thickness growth rate is 1.2.

[0049] S2. Select the calculation model SST-w in the CFD simulation software, set the boundary conditions to velocity inlet and pressure outlet, and calculate the residual to 10 -5 , the fluid density is 1.225kg / m 3 , the dynamic viscosity is 1.7894×10 -5 Pa·s, the fluid velocity was set to 4, 5, 6, 7, 8, and 9 m / s, and the wall roughness was 0.025 mm.

[0050] After grid independence verification, the inlet and outlet pressures of the local pipe fitting at different inlet velocities were calculated, and the local resistance coefficient was fitted.

[0051] The calculation formula of the local resistance coefficient is as follows:

[0052]

[0053] Where ξ is the local resistance coefficient, ρ is the fluid density, v is the inlet velocity, and P in is the inlet pressure, P out is the outlet pressure.

[0054] Based on the calculation formula of the local resistance coefficient, 1 / 2ρv 2 is the horizontal axis, ΔP=P in -P out The vertical axis is the slope of the fitted curve, which is the local resistance coefficient of the local pipe.

[0055] S3. According to the overall pipe network structure of the ventilation system of the spent fuel reprocessing plant, a one-dimensional model is constructed in the fluid system simulation software FloMaster to determine the positions of local pipe fittings, pressure source components (pressure source atmosphere 8, blowing instrument gallery 9, corridor 10), three-way components 11, valve components (valve 1 4, valve 2 5, valve 3 6) and fan components 7, and connect them into a one-dimensional model of the overall pipe network structure of the ventilation system of the spent fuel reprocessing plant, such as Figure 3Set the target ventilation parameters for the blowing instrument gallery 9 and gallery 10, set the fluid parameters, and set the local resistance coefficient for the local pipes.

[0056] S4. Latin hypercube sampling is performed on the valve openings in the overall pipe network structure of the spent fuel reprocessing plant ventilation system. First, the valve opening range is set to 0 to 1, divided into 300 equal parts, and a valve opening is randomly generated in each small interval; secondly, the order of each valve is shuffled and recombined to form a valve opening combination data set, and the data set is input into the valve element of FloMaster; then, the ventilation parameters of each key node of the one-dimensional model are numerically calculated to obtain the ventilation parameter combination data set of each key node under different valve opening combinations.

[0057] S5. Provide the ventilation parameter combination data set to the feedforward neural network algorithm model, use the ReLU activation function, take the mean square error as the loss function, and optimize it using the adaptive matrix estimation algorithm to obtain the valve regulation strategy calculation model. Among them, the neurons are fully connected, and the neural network structure is determined by the particle swarm optimization algorithm. The working principle is as follows Figure 4 shown.

[0058] The formula for mean square error is as follows:

[0059]

[0060] Among them, y i is the actual observed value, is the predicted value; n is the number of observations.

[0061] The formula of the adaptive matrix estimation (Adam) algorithm is as follows:

[0062] m t =β1·m t-1 +(1-β1)·g t

[0063]

[0064]

[0065]

[0066]

[0067] In the formula, m t and v t They represent the first-order moment and the second-order moment respectively; β1 and β2 are the attenuation rates of the first-order moment and the second-order moment respectively, which are usually set to β1 = 0.9 and β2 = 0.999; g t is the current gradient; are the deviation corrections for the first-order moment and the second-order moment respectively; α is the learning rate; ε is a constant, usually 10 -8 .

[0068] Particle swarm optimization algorithm, as follows:

[0069] v i '=w×v i +c1×rand×(p best_i -p resent_i )+c2×rand×(g best -p resent_i )

[0070] +c2×rand×(g best -p resent_i )

[0071] p' resent_i =p resent_i +v i '

[0072] In the formula, v i ' represents the updated velocity of particle i, v i represents the current speed of particle i, rand represents a random number, p' resent_i represents the updated position of particle i, p resent_i represents the current position of particle i, p best_i represents the individual optimal position found by particle i, g best represents the global optimal position found by all particles, w is the inertia weight; c1 and c2 are learning factors, usually set to c1=c2=2.

[0073] S6. According to the provided data set, the nonlinear relationship between the input value and the output is calculated by the model trained by the feedforward neural network algorithm. When the target ventilation parameter of the ventilation system of the actual spent fuel reprocessing plant changes, the target ventilation parameter combination is provided to the valve adjustment strategy calculation model, so as to output the corresponding valve adjustment strategy.

[0074] Therefore, the present invention adopts the above-mentioned deep learning-based ventilation system valve adjustment strategy calculation method, which can efficiently and accurately provide a ventilation system valve adjustment strategy, realize rapid adjustment of the ventilation system, and increase the reliability of the simulation model.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for calculating the valve adjustment strategy of a ventilation system based on deep learning, characterized in that: The following steps are involved: S1. Construct a three-dimensional model of local pipe fittings according to the pipe network structure of the ventilation system; S2. According to the three-dimensional model of the local pipe fitting, the values ​​at different flow rates are obtained through simulation, and the resistance coefficient of the pipe fitting is obtained by fitting; S3. According to the ventilation system pipe network structure, determine the position of each component and construct a one-dimensional model; S4, using Latin hypercube sampling, sampling the adjustment strategy of each valve opening in the one-dimensional model as the initial parameter to obtain a ventilation parameter combination data set; S5. Training the deep learning algorithm model based on the ventilation parameter combination data set to obtain a valve adjustment strategy calculation model; S6. According to the target ventilation parameters, the valve regulation strategy calculation model is used to output the valve regulation strategy.

2. A method for calculating a ventilation system valve adjustment strategy based on deep learning according to claim 1, characterized in that: In step S1, constructing a three-dimensional model of a local pipe fitting includes setting the outlet of the local pipe fitting to a distance of twice the diameter, setting a fluid domain, and dividing an unstructured grid.

3. The method for calculating the ventilation system valve adjustment strategy based on deep learning according to claim 1, characterized in that: In step S2, the simulation includes setting the boundary conditions as the velocity inlet and the pressure outlet, and obtaining the inlet and outlet pressures of the local pipe fittings according to the different fluid velocities at the inlet of the local pipe fittings by setting the residual, fluid density, dynamic viscosity and wall roughness, and then fitting to obtain the local resistance coefficient, as follows: Where ξ is the local resistance coefficient, ρ is the fluid density, v is the inlet velocity, and P in is the inlet pressure, P out is the outlet pressure.

4. A method for calculating a ventilation system valve adjustment strategy based on deep learning according to claim 1, characterized in that: In step S3, the constructed one-dimensional model includes local pipe fitting components, pressure source components, three-way components, valve components and fan components.

5. The method for calculating the valve adjustment strategy of a ventilation system based on deep learning according to claim 1, characterized in that: Step S4 includes: S41, sampling the adjustment strategy of each valve in the one-dimensional model according to the Latin hypercube sampling method, and randomly generating valve opening data; S42, shuffling and recombining the order of each valve to form a valve opening combination data set; S43, setting target ventilation parameters, fluid parameters, local resistance coefficient and valve opening, and completing the boundary condition setting of the one-dimensional model; S44. Calculate and obtain the ventilation parameter combination data set of each key node.

6. A method for calculating a ventilation system valve adjustment strategy based on deep learning according to claim 1, characterized in that: In step S5, the deep learning algorithm includes full connection between neurons, determining the neural network structure through a particle swarm optimization algorithm, and optimizing it using an adaptive matrix estimation algorithm.

7. A method for calculating a ventilation system valve adjustment strategy based on deep learning according to claim 6, characterized in that: The adaptive matrix estimation algorithm is as follows: m t =β1·m t-1 +(1-β1)·g t In the formula, m t 、v t They represent the first-order moment and the second-order moment respectively, β1 and β2 are the decay rates of the first-order moment and the second-order moment respectively, and g t is the current gradient, are the deviation corrections for the first-order moment and the second-order moment, α is the learning rate, and ε is a constant.

8. A method for calculating a ventilation system valve adjustment strategy based on deep learning according to claim 6, characterized in that: Particle swarm optimization algorithm, as follows: v i '=w×v i +c1×rand×(p best_i -p resent_i )+c2×rand×(g best -p resent_i ) +c2×rand×(g best -p resent_i ) p' resent_i =p resent_i +v i ' In the formula, v i ' represents the updated velocity of particle i, v i represents the current speed of particle i, rand represents a random number, p' resent_i represents the updated position of particle i, p resent_i represents the current position of particle i, p best_i represents the individual optimal position found by particle i, g best represents the global optimal position found by all particles, w is the inertia weight, and c1 and c2 are learning factors.

Images