Middle-deep layer buried pipe heat exchanger seepage thermal response rapid prediction method based on artificial intelligence

Abstract

The invention relates to the technical field of geothermal energy utilization and underground heat exchange, in particular to a medium-deep layer buried pipe heat exchanger seepage thermal response rapid prediction method based on artificial intelligence, and the method comprises the steps: obtaining temperature field snapshot data of a medium-deep layer buried pipe heat exchanger system under different operation conditions; performing orthogonal decomposition on the temperature field snapshot data to obtain main modal characteristics and corresponding amplitudes; the parameters of all the operation conditions serve as input variables, all the main modal features and the corresponding amplitudes serve as output variables, a BP neural network is trained, and an artificial intelligence agent model is constructed; and obtaining parameters of a target operation condition, inputting the parameters into the artificial intelligence agent model, outputting the amplitude of each main mode under the target operation condition, reconstructing a temperature field, and completing rapid prediction of the seepage thermal response of the medium-deep layer buried pipe heat exchanger under the target operation condition. According to the method, the temperature field distribution of the DBHE system under the action of underground water seepage can be quickly predicted.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geothermal energy utilization and underground heat exchange, and particularly relates to a middle-deep buried pipe heat exchanger seepage heat response rapid prediction method based on artificial intelligence, which is suitable for heat transfer analysis and heat conduction parameter inversion modeling considering the influence of underground water seepage. BACKGROUND

[0002] The middle-deep buried pipe heat exchanger system has a wide application prospect in the field of geothermal energy development due to its high heat extraction capacity and underground space utilization efficiency. However, the underground water seepage significantly affects the heat extraction efficiency and long-term operation sustainability of the DBHE system. Existing numerical modeling is mostly based on the thermal resistance method model, which often ignores the heat storage effect of the pipe wall, grouting material and surrounding rock, and is difficult to accurately simulate the coupling influence of seepage on the heat field evolution, and has large calculation overhead and low efficiency.

[0003] At present, the traditional method has insufficient modeling flexibility and data driving capability when facing long-term operation and multi-variable parameter change scenes. Therefore, there is an urgent need for a modeling method that is efficient, accurate and has intelligent analysis capability to realize the rapid prediction and optimization design of the thermal performance of the middle-deep buried pipe system under complex working conditions. SUMMARY

[0004] The purpose of the present application is to provide a middle-deep buried pipe heat exchanger seepage heat response rapid prediction method based on artificial intelligence, which is used for rapidly predicting the temperature field distribution, heat extraction capacity and key parameter inversion of the DBHE system under the action of underground water seepage.

[0005] To achieve the above purpose, the present application provides the following scheme:

[0006] A middle-deep buried pipe heat exchanger seepage heat response rapid prediction method based on artificial intelligence, comprising:

[0007] Obtaining temperature field snapshot data of the middle-deep buried pipe heat exchanger system under different operating conditions;

[0008] Orthogonal decomposition is performed on the temperature field snapshot data to obtain main modal characteristics and corresponding amplitudes;

[0009] Taking the parameters of each operating condition as input variables and the main modal characteristics and corresponding amplitudes as output variables, a BP neural network is trained to construct an artificial intelligence agent model;

[0010] The parameters of the target operating condition are input into the artificial intelligence agent model, the main modal amplitudes under the target operating condition are output, and the temperature field is reconstructed to complete the middle-deep buried pipe heat exchanger seepage heat response rapid prediction under the target operating condition.

[0011] Optionally, after obtaining the temperature field snapshot data, the method further comprises matrix arrangement on the temperature field snapshot data to construct a temperature sample matrix, and the temperature sample matrix is:

[0012] ;

[0013] wherein, is a total fluctuation sample of the temperature field, is a radial coordinate, is a vertical coordinate, is a first sampling time.

[0014] Optionally, the orthogonal decomposition on the temperature field snapshot data to obtain a main modal feature and a corresponding amplitude comprises:

[0015] setting a truncated free element and constructing a feature matrix;

[0016] solving eigenvalues and eigenvectors of the feature matrix to obtain a main modal function and a corresponding amplitude;

[0017] calculating a projection of the temperature field sample on a basis function space to obtain a spectral coefficient, and coupling reconstruction on the temperature field to output a reduced dimension temperature field.

[0018] Optionally, the artificial intelligence agent model comprises a three-layer BP neural network, and the BP neural network is used to establish a mapping relationship between a temperature modal amplitude and a system parameter.

[0019] Optionally, an input layer of the BP neural network is a physical parameter and an operating parameter of the target middle-deep buried pipe heat exchanger system, a hidden layer adopts a ReLU and a tanh activation function, and an output layer is a main modal amplitude prediction value.

[0020] Optionally, the method further comprises: inverting the optimal thermal conductivity coefficient in the model training process comprises:

[0021] constructing a target function minimization strategy according to the artificial intelligence agent model, inputting a difference between a predicted temperature field and a measured temperature field, and inverting to obtain an optimal thermal conductivity coefficient interval.

[0022] Optionally, the target function is:

[0023] ;

[0024] wherein, is a predicted value of an average circulating fluid temperature in a buried pipe at a first time, is a rock-soil thermal conductivity coefficient, ) is a spatial number of a prediction model, is a total spatial number of the prediction model, The number of tests is The average circulating fluid temperature measured value in the pipe at the moment, The total number of test data, The number of initial tests.

[0025] The beneficial effects of the present application are:

[0026] (1) High efficiency: through dimension reduction and agent modeling, the calculation speed is improved by more than 180 times;

[0027] (2) High precision: under various seepage conditions, the prediction error is controlled within a reasonable range;

[0028] (3) Intelligent: introducing artificial intelligence technology, with nonlinear modeling and inversion capability;

[0029] (4) Adaptability: suitable for complex conditions such as multiple formations, multiple properties, multiple time periods, and multiple heat loads;

[0030] (5) Strong expandability: can be extended to thermal response modeling of other types of underground heat exchange systems. BRIEF DESCRIPTION OF DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0032] Figure 1 It is a deep buried pipe heat exchanger system structure schematic diagram of the embodiment of the present application;

[0033] Figure 2 It is a schematic diagram of the heat field evolution mechanism under the influence of groundwater seepage of the embodiment of the present application;

[0034] Figure 3 It is a flow chart of a kind of seepage heat response rapid prediction method of deep buried pipe heat exchanger based on artificial intelligence of the embodiment of the present application;

[0035] Figure 4 It is a schematic diagram of artificial intelligence agent model structure of the embodiment of the present application;

[0036] Figure 5 It is a comparison diagram of model prediction and finite difference (C-N method) results of the embodiment of the present application;

[0037] Figure 6 It is a temperature recovery rate change curve diagram under different seepage velocities of the embodiment of the present application;

[0038] Figure 7A schematic diagram of the thermal conductivity inversion result of the embodiment of the present application. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0040] In order to make the above objectives, characteristics and advantages of the present application more apparent, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0041] The embodiment provides a quick prediction method for heat response of a middle-deep buried pipe heat exchanger based on artificial intelligence, and the method comprises the following steps:

[0042] Obtaining temperature field snapshot data of the middle-deep buried pipe heat exchanger system under different operating conditions;

[0043] Orthogonal decomposition is performed on the temperature field snapshot data to obtain main modal characteristics and corresponding amplitudes;

[0044] Taking the parameters of each operating condition as input variables and the main modal characteristics and corresponding amplitudes as output variables, a BP neural network is trained to construct an artificial intelligence agent model;

[0045] The parameters of a target operating condition are input into the artificial intelligence agent model, the main modal amplitudes under the target operating condition are output, and the temperature field is reconstructed, so that the quick prediction of the heat response of the middle-deep buried pipe heat exchanger under the target operating condition is completed.

[0046] Specifically, as shown in the figure, the following steps and contents are included: Figure 3

[0047] S1: POD agent model construction process.

[0048] As shown in the figures, first, the temperature field snapshot data of a certain middle-deep buried pipe heat exchanger system under different groundwater seepage velocities (0-10 m / s) and operating time periods (0-5 years) is obtained, and three-dimensional temperature distribution at -6 Figure 1 Figure 2 a total of 100 time points is collected. The matrices of all snapshots are arranged to construct a temperature sample matrix:

[0049]

[0050]

[0051] ​​​​​wherein, is the total fluctuation sample of the temperature field, , is the radial coordinate, is the vertical coordinate, is the first sampling time.

[0052] Specifically, temperature field snapshot data of the ground heat exchanger system at different times and operating conditions is obtained; based on the snapshot idea: the POD algorithm involves using data obtained from a high-dimensional system to construct a snapshot matrix. Then, the snapshot matrix is projected into a low-dimensional parameter space, in which the dispersion of the snapshot matrix is required to be maximized. The snapshot matrix available at a certain time point constitutes a 65x50 matrix , and the samples of the 50 different time layers taken are required to be linearly independent, and the step length between adjacent samples needs to be sufficiently long. In order to provide a more compact sparse linear combination in a low-dimensional space, help to reduce the calculation storage cost and simulation complexity, the sample matrix obtained is as follows:

[0053] .

[0054] S2: Orthogonal decomposition is performed on the snapshot data to obtain the principal mode function and the corresponding amplitude.

[0055] Using the POD technology to perform eigenvalue orthogonal decomposition, the characteristic mode and the corresponding whole coefficient are obtained, and the temperature field expression is reconstructed as:

[0056] ;

[0057] wherein, is the number of principal modes selected, generally accounting for more than 90% of the total energy.

[0058] Specifically, since the number of rows in the snapshot matrix is greater than the number of columns, the dimension of the matrix will be much larger than that of the matrix . Therefore, in order to reduce the calculation complexity in the eigenvalue decomposition process and avoid the generation of singular matrices, the truncated free element defined can accurately and efficiently implement the best orthogonal decomposition on the matrix :

[0059] ;

[0060] wherein, is a two-dimensional calculation domain in the spatial direction. And the array is constructed with the truncated free element as an element:

[0061] ;

[0062] is a 50x50 symmetric matrix, the eigenvalues and eigenvectors of the matrix are solved and , the generation process of the eigenvalues and eigenvectors here can be calculated as follows:

[0063] ;

[0064] The modal function is constructed as , wherein is calculated.

[0065] The spectral coefficients are obtained by calculating the projection of the temperature field sample on the basis function space, that is, , wherein T refers to the total fluctuation sample of the temperature field, and the temperature field is coupled and reconstructed, and finally the drilling outlet temperature and the sampling temperature at different times are output.

[0066] S3: Artificial intelligence BP neural network modeling.

[0067] An artificial intelligence agent model based on BP neural network is constructed, the parameters of each operating condition are taken as input variables, and the main modal characteristics and corresponding amplitudes are taken as output variables, and the temperature modal amplitude is predicted under the input system parameters. The system input parameters are organized into a vector:

[0068] ;

[0069] , wherein is the groundwater seepage velocity; is the formation thermal conductivity; t is the time; is the initial temperature. Other operating parameters such as burial depth, grouting parameters, etc. are also included, and the BP neural network is a multi-output function:

[0070] ;

[0071] The output is the amplitude prediction value of each modal, which is used for temperature field reconstruction.

[0072] As shown in Figure 4 , a three-layer BP neural network structure is adopted:

[0073] Input layer: seepage velocity, formation thermal conductivity, operating time, initial ground temperature and other physical parameters;

[0074] Hidden layer: ReLU or tanh activation function is used;

[0075] Output layer: main modal amplitude , which is used for reconstructing the temperature field.

[0076] The neural network is supervised trained using the training set, and the loss function is the root mean square error (RMSE). The validation set error tends to be stable after the 20th round. The final model can be used to quickly predict the temperature field distribution under the target parameters.

[0077] The training set comes from the snapshot data generated by the numerical simulation of three-dimensional unsteady ground heat exchanger. Under different operating conditions (such as burial depth, soil layer distribution, groundwater seepage velocity, grouting thermal conductivity, and operating time), the temperature field is solved by a high-precision numerical model, and the global temperature distribution is extracted at a specific time node to form a snapshot matrix. Then, the snapshot matrix is subjected to proper orthogonal decomposition (POD) to obtain the principal modal function and the corresponding amplitude coefficient sequence. In the supervised training process, each operating parameter is taken as an input variable, and the amplitude of each principal mode obtained by POD decomposition is taken as an output variable to construct an input-output mapping relationship. The root mean square error (RMSE) is used as the loss function for iterative optimization. The trained BP neural network can quickly predict the amplitude of each principal mode under given new operating parameters, and reconstruct the temperature field through POD, thereby realizing efficient approximation of complex three-dimensional unsteady temperature distribution.

[0078] S4: Rapid prediction of thermal performance and inverse analysis of thermal conductivity parameters.

[0079] The efficiency of thermal performance prediction is compared, as shown in Figure 5 , under the same groundwater conditions, the temperature field is predicted using the proposed method and the traditional C-N finite difference method respectively. The results show that the average prediction error is controlled within ±0.35°C; the calculation time is reduced from 36 minutes (C-N) to 11.8 seconds, with a speed increase of about 183.57 times.

[0080] Thermal recovery analysis under seepage influence, as shown in Figure 6 , the temperature recovery rate after 5 years under different seepage velocities (0, 2.5×10 -7 , 5×10 -7 , 7.5×10 -7 , 10 -6 m / s) is simulated. The results show that:

[0081] The higher the seepage velocity, the faster the cold front moves, and the heat exchange capacity is enhanced in the early stage but the period is shortened; when the seepage velocity is 7.5×10 -7 m / s, the temperature recovery rate increases to 99.13%, higher than 91.23% without seepage.

[0082] Inverse analysis of thermal conductivity parameters, as shown in Figure 7 , based on the trained model, a minimum target function strategy is constructed, the difference between the predicted temperature field and the measured temperature field is input, and the optimal thermal conductivity coefficient interval is obtained by inversion, as shown in the following process:

[0083] Objective function construction:

[0084] ;

[0085] In the formula, For the first The average circulating fluid temperature in the buried pipe calculated by the heat transfer model at any time. o C), The thermal conductivity of the soil and rock is... ) represents the spatial layer number of the prediction model. For the first The average temperature of the fluid in the buried pipe, measured at all times, is approximately the average of the inlet and outlet fluid temperatures. o C), The total number of tests in the experiment. This represents the initial number of tests.

[0086] To make the objective function The minimum value can be obtained through optimal control methods. Here, we define... ,use Evaluation by value The optimal problem. If The closer the value is to 1, the more likely it is that the simplified low-order model of the POD algorithm can be used for field testing, and the initial estimate of the thermal conductivity of the soil and rock is reliable.

[0087] The average fluid temperature is expressed as a linear combination of the modal vector and the spectral coefficients, as shown in the following equation:

[0088] ;

[0089] Substituting the above equation into the objective function, we get:

[0090] ;

[0091] Consider the following optimization control problem:

[0092] ;

[0093] The final predicted optimal thermal conductivity of the soil and rock is:

[0094] W / (m·K);

[0095] The inversion results can be used for formation thermal conductivity estimation, heat exchanger design optimization, and construction guidance.

[0096] The method improves the calculation speed by more than 180 times through dimension reduction and proxy modeling; the prediction error is controlled within a reasonable range under various seepage conditions; the introduction of artificial intelligence technology has nonlinear modeling and inversion capabilities; it is suitable for complex conditions such as multiple formations, multiple properties, multiple time periods, and multiple heat loads; and it can be extended to thermal response modeling of other types of underground heat exchange systems.

[0097] The above-described embodiments are merely descriptions of the preferred modes of the present application and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements to the technical solutions of the present application made by those skilled in the art shall fall within the protection scope determined by the claims of the present application.

Claims

1. An artificial intelligence-based rapid prediction method for the heat response of a medium-deep buried pipe heat exchanger, characterized in that, The method comprises the following steps: obtaining temperature field snapshot data of a middle-deep buried pipe heat exchanger system under different operating conditions; performing orthogonal decomposition on the temperature field snapshot data to obtain main modal characteristics and corresponding amplitudes; training a BP neural network by taking parameters of each operating condition as input variables and taking the main modal characteristics and corresponding amplitudes as output variables, and constructing an artificial intelligence agent model; obtaining parameters of a target operating condition, inputting the parameters into the artificial intelligence agent model, outputting main modal amplitudes under the target operating condition, and reconstructing a temperature field to complete rapid prediction of seepage heat response of the middle-deep buried pipe heat exchanger under the target operating condition.

2. The method according to claim 1, wherein, After obtaining the temperature field snapshot data, the method further comprises matrix arrangement of the temperature field snapshot data to construct a temperature sample matrix, which is: ； wherein is the total fluctuation sample of the temperature field, is the radial coordinate, is the vertical coordinate, is the first sampling time.

3. The method according to claim 1, wherein, performing orthogonal decomposition on the temperature field snapshot data to obtain main modal characteristics and corresponding amplitudes, which comprises: setting a truncated free element and constructing a characteristic matrix; solving eigenvalues and eigenvectors of the characteristic matrix to obtain main modal functions and corresponding amplitudes; calculating projections of temperature field samples on a basis function space to obtain spectral coefficients, and coupling and reconstructing the temperature field to output a reduced temperature field.

4. The method according to claim 1, wherein, The artificial intelligence agent model comprises a three-layer BP neural network, and the BP neural network is used to establish a mapping relationship between temperature modal amplitudes and system parameters.

5. The method according to claim 4, wherein, The input layer of the BP neural network is physical parameters and operating parameters of the middle-deep buried pipe heat exchanger, the hidden layer adopts ReLU and tanh activation functions, and the output layer is a main modal amplitude prediction value.

6. The method according to any one of claims 1-5, wherein, The method further comprises: inverting the optimal thermal conductivity coefficient during model training, which comprises: constructing a target function minimization strategy according to the artificial intelligence agent model, inputting differences between predicted temperature fields and measured temperature fields, and inverting to obtain an optimal thermal conductivity coefficient interval.

7. The method according to claim 6, wherein, The target function is: ； wherein, is the average circulating fluid temperature prediction value in the ground heat exchanger at the is the average circulating fluid temperature prediction value in the ground heat exchanger at the is the geotechnical thermal conductivity, is the number of spatial layers of the prediction model, is the total number of spatial layers of the prediction model, is the average circulating fluid temperature prediction value in the ground heat exchanger at the is the average circulating fluid temperature prediction value in the ground heat exchanger at the is the total number of tests of experimental data, is the number of initial tests.

Images