Lateral Non-uniform Temperature Field Zoning and Active Control Methods for Rectangular Immersed Tunnels

Abstract

This invention discloses a method for zoning and actively controlling the transversely non-uniform temperature field in rectangular immersed tunnels. Specifically, it includes three steps: temperature field division within the tunnel area, active temperature control in each area, and temperature prediction error compensation. Temperature field division within the tunnel area includes initially determining the burner control area and quantitatively identifying the control area. Active temperature control in each area employs machine learning technology to train a neural network model to control the burner. Temperature prediction error compensation involves statistically analyzing the average temperature error of the five temperature control areas every 10 seconds, reducing the prediction error of burner power and improving the accuracy of restoring the temperature rise curves of the five temperature control areas. This method can solve the problem of zoning and actively controlling the transversely non-uniform temperature field in rectangular tunnel sections when using burners, providing technical convenience for research related to fire resistance in immersed tunnels.

Description

Technical Field

[0001] This invention relates to the field of fire resistance in immersed tunnel structures, specifically to a method for zoning and actively controlling the transverse non-uniform temperature field in rectangular immersed tunnels. Background Technology

[0002] Immersed tunnels, as an important form of underwater tunnel construction, are widely used in practical engineering. When a fire occurs in a tunnel, due to its semi-enclosed spatial structure and poor ventilation, a large amount of heat accumulates in the tunnel ceiling area and is transferred to the tunnel lining structure, creating a non-uniform temperature field. Immersed tunnels, located underwater, suffer even worse smoke extraction and heat dissipation conditions in the event of a fire, making personnel evacuation and rescue more difficult and causing more severe damage to the tunnel structure. To reduce tunnel damage caused by fire and minimize post-disaster repair costs, it is essential to study the fire resistance performance of immersed tunnels. However, current conditions make it difficult to conduct destructive tunnel fire resistance tests, thus limiting the scope of research on tunnel fire resistance performance.

[0003] Currently, tunnel fire tests are typically conducted in real in-situ tunnels and scaled-down laboratory tunnels, primarily focusing on the dynamics of tunnel fires, such as the temperature field distribution and smoke propagation. Considering safety and cost reduction, few researchers conduct actual tunnel fire resistance tests to study damage mechanisms and failure characteristics. Current fire resistance tests on tunnel lining structures in fire-resistant testing furnaces are modifications of existing building fire-resistant testing furnaces, altering the loading method. However, these modifications fail to replicate the non-uniform temperature field of a tunnel fire, making it difficult to accurately reflect the performance of tunnel lining structures in real fire conditions. The test results...

[0004] It lacks universality. At the same time, a few scholars have conducted in-situ test studies on immersed tunnels, mainly focusing on the study of temperature field and deformation, and rarely involving the study of fire resistance of immersed tunnel lining structures at high temperatures; most scholars use numerical simulation or laboratory scaled-down tests to study the stress state of immersed tunnel lining structures at high temperatures, but lack universality.

[0005] Due to the high cost and inherent risks of experimental research on immersed tunnel fires, numerical simulation methods are more suitable for studying the transverse temperature field of immersed tunnel fires. By comparing the results of experiments and numerical simulations with the same parameters, numerical simulation can more accurately reproduce the transverse temperature field of a tunnel under a real fire. When a tunnel fire occurs, the temperature distribution of the transverse temperature field is related to the distance from the fire source, the duration of combustion, and the cross-sectional shape. The factors influencing the temperature distribution of the tunnel cross-section are complex. Current research mainly focuses on the qualitative study of the temperature field distribution law in tunnel fires, with limited research on the quantitative division of the cross-sectional temperature field. Therefore, there is an urgent need for a method for the division and active control of the non-uniform temperature field in the transverse section. Summary of the Invention

[0006] To address the aforementioned technical shortcomings, the purpose of this invention is to provide a method for zoning and actively controlling the transverse non-uniform temperature field in rectangular immersed tunnels. This method can solve the problem of zoning and actively controlling the transverse non-uniform temperature field in the cross-section of a rectangular tunnel when using burners, thus providing technical convenience for research related to fire resistance of immersed tunnels.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0008] This invention provides a method for zoning and actively controlling the transversely non-uniform temperature field in a rectangular immersed tunnel, including: S1, temperature field division within the tunnel area; specific steps include:

[0009] S11. Preliminarily determine the burner control area, that is, roughly determine the non-uniform temperature field area controlled by the burner based on its location.

[0010] S12. Quantitatively determine the control area, that is, determine the specific size of the control area;

[0011] S2. Active temperature control in each zone; this involves using machine learning technology to train a neural network model to control the burner; specific steps include:

[0012] S21. Using Fluent numerical simulation, construct the correspondence between the power of the three rows of burners on the small tower combustion vehicle and the temperature and initial power of the five temperature control zones, and obtain a sample set;

[0013] S22. Use the sample set to train the machine learning algorithm and build a prediction model for the transverse non-uniform temperature field of the immersed tunnel. Before training the model, the sample set capacity needs to be estimated, and the sample scheme needs to be designed using the orthogonal design method. Reasonable training, validation and test sets need to be generated through sample data analysis and standardized sample set partitioning.

[0014] S3. Temperature prediction error compensation: The average temperature error of the five temperature control zones is calculated every 10 seconds. The purpose is to reduce the prediction error of the burner power and improve the accuracy of the temperature rise curve restoration of the five temperature control zones.

[0015] Preferably, in step S12, the burners are divided into top burners with the nozzle facing a 90° angle to the ground and side burners with the nozzle facing the ground, according to the nozzle angle. The two control the temperature fields of the top and the side respectively.

[0016] The top temperature field determines the range of the controlled high-temperature area based on the power of the top burner. At this time, the temperature control area of ​​the top burner is divided according to the temperature area with high sensitivity, that is, the temperature area that starts to heat up earlier is the temperature control area of ​​the top burner.

[0017] The side temperature field is determined by the burner nozzle scheduling, i.e., the angle between the burner and the horizontal line, to determine the range of the high-temperature area it controls. The principle of dividing the side temperature field is that the control range of the side burner should be focused on the tunnel sidewall as much as possible to avoid providing a large amount of heat to the top and bottom. The entire process is simulated using Fluent software to quantitatively partition the non-uniform temperature field.

[0018] Preferably, in step S22, taking a single burner controlling a single temperature control zone as an example, the control flow of a single burner controlling a single temperature control zone is mapped to three rows of burners controlling five temperature control zones. This is the active control flow that needs to be implemented, and the specific flow is as follows:

[0019] Step 1: Divide the temperature rise curves; Divide the known temperature rise curves of the five temperature control zones ① to ⑤ into 10-second intervals to obtain the start and end temperatures of the five temperature control zones within several time intervals.

[0020] Step 2: Predict burner power; Using a neural network algorithm, with the start and end temperatures of the five zones and the initial power of the three rows of burners (0kW) in the first 10-second time period, predict the required power settings for the left, top, and right burners in the first 10-second time period; using the start and end temperatures of the five zones and the predicted power of the three rows of burners in the first 10-second time period as the initial power, predict the required power settings for the left, top, and right burners in the second 10-second time period; and so on, to obtain the required burner power settings for all time periods.

[0021] Step 3: Input burner power; Before the test begins, input the power required by the three rows of burners for each time period into the burner control system;

[0022] Step 4: Monitor the temperature of the temperature control zones; monitor the temperature changes of the five temperature control zones respectively, and calculate the average deviation between the actual temperature and the target temperature of the five temperature control zones in each time period;

[0023] Step 5: Burner power prediction error compensation; During the test, based on the cumulative temperature difference trend and average temperature difference in each time period, uniformly increase or decrease the burner power predicted by all neural networks.

[0024] Preferably, the specific process of the temperature prediction error compensation step is as follows:

[0025] Step 1: Calculate the average temperature error; calculate the average temperature error between the actual temperature and the target temperature of the five temperature control zones within the first 10-second time period;

[0026] Step 2: Determine the cumulative trend of temperature error and make qualitative adjustments; when the cumulative trend of temperature error increases, the actual temperature is greater than the target temperature, and the burner power values ​​predicted by all neural networks need to be reduced; when the cumulative trend of temperature error decreases, the actual temperature is less than the target temperature, and the predicted values ​​of all neural networks need to be increased.

[0027] Step 3: Adjust the neural network prediction value based on the average temperature error, i.e., make a quantitative adjustment; when the average temperature error is 0 to 50℃, increase the prediction value by 1.1 times; when the average temperature error is -50 to 0℃, decrease the prediction value by 0.9 times; when the average temperature error is 50 to 100℃, increase the prediction value by 1.2 times; when the average temperature error is -100 to -50℃, increase the prediction value by 0.8 times, and so on.

[0028] Step 4: Repeat steps 1 through 3 until the experiment is stopped.

[0029] The beneficial effects of this invention are as follows:

[0030] 1. This method can be applied to various rectangular tunnels. It proposes a quantitative partitioning and active control algorithm for the transverse non-uniform temperature field, which can realize the active control of the transverse non-uniform temperature field in immersed tunnels. This method can be extended to any rectangular tunnel.

[0031] 2. Due to the limited space and semi-enclosed environment of the tunnel, coupled with the influence of overhead jets, the high-temperature zone is mainly concentrated at the top of the tunnel, resulting in a non-uniform lateral temperature field in the immersed tunnel. The quantitative partitioning strategy proposed in this method can reasonably divide the non-uniform lateral temperature field of the immersed tunnel, allowing each burner to intuitively correspond to its high-temperature influence range.

[0032] 3. The burner power prediction model proposed in this method has undergone rigorous optimization training. It only requires the target temperature curves of each region to be given in advance and input into the model to predict the power required by the burner in each region, thereby realizing the generation of non-uniform temperature fields.

[0033] 4. To avoid prediction errors in neural networks, this method proposes an error compensation strategy to compensate for errors based on their actual magnitude. In our verification model, errors in all regions are effectively compensated, making the model closer to the target curve and effectively achieving active control of non-uniform temperature fields. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a schematic diagram of a single temperature control area provided in an embodiment of the present invention;

[0036] Figure 2 A schematic diagram showing five temperature control zones and three rows of burners;

[0037] Figure 3 A schematic diagram of a small tower combustion vehicle;

[0038] Figure 4 This is a schematic diagram of a narrow immersed tunnel.

[0039] Figure 5 A schematic diagram of a small tower-type combustion vehicle and an immersed tunnel numerical model;

[0040] Figure 6 Preliminary zoning diagram of the transverse non-uniform temperature field;

[0041] Figure 7 This is a schematic diagram of the simulation results for grid sensitivity analysis;

[0042] Figure 8 Diagram showing the top burner in operation;

[0043] Figure 9 This is a diagram showing the layout of the survey lines at the top of the immersed tunnel.

[0044] Figure 10 The temperature rise curve of the top burner under 170kW operating conditions;

[0045] Figure 11 The temperature rise curve of the top burner under 340kW operating conditions;

[0046] Figure 12 Diagrams showing the side burner at different angles to the horizontal plane;

[0047] Figure 13 This is a partial survey line layout diagram for the immersed tunnel side section;

[0048] Figure 14 Temperature cloud maps of the side burner at different angles over 360 seconds;

[0049] Figure 15 This is a temperature rise curve of the side burner at 30°C.

[0050] Figure 16 Temperature contour maps at different power levels with an angle of 30° over 360 seconds;

[0051] Figure 17 This is a schematic diagram of the quantitative zoning results;

[0052] Figure 18 A schematic diagram showing the maximum temperature difference in five regions for 480 samples.

[0053] Figure 19 This is a schematic diagram of the basic structure of a BP neural network;

[0054] Figure 20 A temperature comparison diagram showing the addition of an error compensation mechanism. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] To verify the effectiveness of the active control technology for the non-uniform temperature field in immersed tunnels, the following methodological verification is based on a simulated combustion vehicle running the entire process. Since the implementation and verification process is quite complex, the implementation includes preliminary work such as numerical simulation of rectangular cross-section tunnels using Fluent and building a tower combustion vehicle to simulate combustion conditions. All work will be rigorously simulated to ensure the validity of the verification results.

[0057] This embodiment provides a method for zoning and actively controlling the transverse non-uniform temperature field in a rectangular immersed tunnel, specifically as follows:

[0058] Step 1: Determine the basic simulation and verification parameters such as the combustion vehicle, flamethrower, and tunnel dimensions. (This step is to better demonstrate and verify the specific process of the invention.)

[0059] A small tower-type combustion vehicle is used to carry burners to simulate fire sources at different locations, meeting the requirements of tunnels of different sizes and types. The small tower-type combustion vehicle has a cross-sectional design with a bottom width of 1150mm, a top width of 525mm, and a height of 840mm. Figure 3 As shown in (a). Furthermore, the length of the combustion vehicle is set to 1.1m. To ensure that the small tower combustion vehicle forms a nearly uniform temperature field longitudinally within the immersed tunnel and to ensure that the distance between the two burners is less than or equal to 1200mm, the small tower combustion vehicle uses two rows of burners arranged longitudinally with a spacing of 550mm, as shown in (a). Figure 3 As shown in (b).

[0060] A 1:2 scale immersed tunnel model was selected, requiring the deployment of a small tower-type combustion engine. Based on the "Highway Tunnel Design Specifications," the dimensions of the model were determined to be 1.5m (length), 4m (width), and 3.5m (height). The cross-section is shown below. Figure 4 As shown.

[0061] Considering the actual engineering requirements, two 281mm diameter exhaust vents were installed on each side of the immersed tunnel, with a burner nozzle diameter of 160mm and a maximum power of 600kW. The small tower-type combustion vehicle was located in the center of the tunnel, 300mm above the tunnel surface. Based on the aforementioned dimensions of the small tower-type combustion vehicle, a numerical model of the small tower-type combustion vehicle and the immersed tunnel was finally established using Space Claim. The specific model is as follows: Figure 5 a- Figure 5 As shown in c.

[0062] Step 2: Dividing the burner control zone in a non-uniform temperature field.

[0063] Preliminary division process:

[0064] When a fire occurs in a tunnel, there is a temperature difference between the central and lateral areas, resulting in a non-uniform temperature field. Based on this temperature field distribution pattern and the aforementioned high-temperature control area of ​​the burners, the transverse non-uniform temperature field of the immersed tunnel is preliminarily divided into zones. This preliminary zoning is based on the approximate control range of burners facing different directions, and the size of the control area is determined indefinitely. The zoning results are as follows: Figure 6 As shown, the five temperature control zones are numbered ①, ②, ③, ④, and ⑤. Zones ①, ③, and ⑤ are directly controlled by the burners; zones ② and ④ are indirectly controlled by both the side and top burners. After the initial zoning of the transverse non-uniform temperature field of the immersed tunnel, the influence of different burner powers on the high-temperature control zone of the top burner needs to be considered to ultimately determine the specific range of temperature control zone ③; the influence of different burner angles on the high-temperature control zone of the side burners needs to be considered to ultimately determine the specific locations and ranges of temperature control zones ① and ⑤; after determining the ranges of temperature control zones ①, ③, and ⑤, the remaining areas are the ranges of temperature control zones ② and ④. The ranges of each zone are quantitatively determined through parameter analysis below.

[0065] Firstly, before quantitative zoning, a grid sensitivity analysis was conducted during the entire operation and verification process. Based on the established numerical model of the transverse non-uniform temperature field of the immersed tunnel, the influence of grid size on calculation accuracy was analyzed. Four grid sizes were selected for numerical simulation and comparative analysis: 0.06m, 0.08m, 0.1m, and 0.2m, corresponding to 320,000, 220,000, 140,000, and 70,000 grids, respectively. Ten measuring lines were evenly arranged at a distance of 1cm from the ceiling at the transverse center section of the immersed tunnel. Three measuring points were arranged on each measuring line, and the average temperature of these three measuring points was taken as the temperature of the measuring line. With the number of grids as the sole variable, all burners were turned on for numerical simulation, and the simulation calculation time was set to 400s.

[0066] Table 1. Average Temperature Error of Survey Lines with Different Grid Quantities

[0067]

[0068] according to Figure 7 According to Table 1, the average temperature error of the survey line is within 5% when the grid size is 0.1m and 0.08m, specifically 3.33% and 0.91%, respectively. Considering both calculation speed and accuracy, the final grid size selected is 0.1m.

[0069] Before quantitatively dividing the high-temperature control zone of the top burner, based on the aforementioned burner arrangement scheme, it was determined that the top burners of the small tower-type combustion vehicle would be arranged in two rows, one in each row, located on the longitudinal centerline of the combustion vehicle and perpendicular to the top of the combustion vehicle, spraying flames directly above the immersed tunnel. Figure 8 As shown.

[0070] Quantitative classification process:

[0071] Next, the high-temperature control zone at the top of the immersed tunnel cross-section by the top burner was further determined, taking into account the influence of the burner's power on the high-temperature zone it controlled. Simultaneously, two top burners were started, with the power of a single burner as the sole variable. Two simulated operating conditions were designed with burner power of 170kW (small fire) and 340kW (large fire), with a calculation time set to 360s.

[0072] To reflect temperature changes in various areas near the inner wall of the immersed tunnel, measuring lines were evenly arranged at 200mm intervals around the tunnel. Three measuring points were placed on each line, and the average temperature of the three points was taken as the temperature of the measuring line. By comparing and analyzing the heating rate and temperature levels of each measuring line, the location of the high-temperature zone was determined. To quantitatively delineate the high-temperature control zone of the top burner, a total of 16 measuring lines were arranged in the top area of ​​the immersed tunnel, numbered from 16 to 31. The specific arrangement of the measuring lines is as follows... Figure 9 As shown.

[0073] The result is that both top burners have a power of 170kW. Figure 10 As shown in the figure, it can be observed that when the simulation time reaches 360s, the temperature of measuring lines 22 to 25 is significantly higher than that of the other measuring lines, while the temperature differences of the other measuring lines are relatively small, all within 10℃, forming a cluster in the graph. In the first 25s of the temperature rise curve, the temperature of measuring lines 22 to 25 is consistently slightly higher than that of the other measuring lines, indicating that measuring lines 22 to 25 begin to heat up before the other measuring lines, and have a higher sensitivity to the heat released by the two top burners. Therefore, when the power of the two top burners is 170kW, the area where measuring lines 22 to 25 are located is determined to be the high-temperature control zone of the top burners.

[0074] Using the same method, simulation results for both top burners with a power output of 340kW are available. Figure 11Comparing the simulation results at 170kW, as the burner power increased, the temperature of measuring lines 22 to 25 remained significantly higher than the other measuring lines, and began to rise in temperature 25 seconds earlier than the other measuring lines. Therefore, the area containing measuring lines 22 to 25 was ultimately selected as the high-temperature control zone for the top burner on the top of the immersed tunnel, and quantitatively divided with measuring lines 21 and 26 as boundaries, with a control range of 1m.

[0075] Next, the control area of ​​the side burner was quantitatively determined. Considering the limitations of actual engineering technology, three conventional flame injection angles were set for the side burner nozzle: 0°, 30°, and 45°. Figure 12 As shown, since the immersed tunnel and the small tower-type combustion vehicle model are symmetrically distributed laterally, it is only necessary to study the high-temperature control area of ​​the left burner of the small tower-type combustion vehicle on the left wall of the immersed tunnel. The high-temperature control area of ​​the right burner can then be obtained by mirroring.

[0076] First, the burner angle is determined. The power of the left burner at a certain angle is used as the only control variable. Two simulation conditions (170kW and 340kW) are designed to study the influence of the side burner power change on the high temperature control area on the side of the immersed tunnel. The calculation time is set to 360s. The specific information of the conditions is shown in Table 2.

[0077] In addition, to verify the rationality of the side high temperature control area obtained by this method, numerical simulations were performed with both the left and right burners turned on. The purpose of working condition 5 in Table 2 is to compare whether the high temperature control areas on both sides are symmetrically distributed.

[0078] Table 2 Simulated operating conditions of the left burner

[0079]

[0080]

[0081] The layout of the measuring lines for the side burners is the same as in Section 3.2. To study the temperature influence range of the small tower-type combustion vehicle's side burners on various areas of the immersed tunnel, 15 measuring lines were arranged on both the left and right sides of the tunnel. The measuring lines on the left side were numbered 1 to 15, and those on the right side were numbered 32 to 46. The specific layout of the measuring lines is as follows: Figure 13 As shown.

[0082] Figure 14Temperature cloud maps at different angles were displayed. When the angle of the burner nozzle on the left was 0°, the cloud map showed that the lower left side of the tunnel was darker, with the high-temperature area located in the lower-middle part of the left side of the tunnel, close to the bottom of the left side of the tunnel. When the angle of the burner nozzle on the left was 30°, the middle part of the left side of the tunnel was the darkest, followed by the upper and top parts. The high-temperature area was located in the middle and began to have some influence on the temperature of the top area. When the angle was 45°, the upper left side of the tunnel was the darkest, and the burner flame almost touched the top of the left side of the tunnel. It had a significant impact on the temperature of the upper-middle and top areas of the left side of the immersed tunnel, and there was some overlap with the effect of the burner on the top of the small tower combustion vehicle.

[0083] Further by Figure 15 It can be seen that when the angle between the left burner nozzle and the horizontal plane is 30°, the temperature of the measuring lines is divided into three gradients. Among them, measuring lines 7 to 10 are located in the first gradient, with temperatures slightly higher than the other measuring lines. Therefore, the high-temperature control area of ​​the burner is located near measuring lines 7 to 10, in the middle of the left side of the tunnel. This indicates that under the 30° temperature rise curve of the side burner, its control area can be quantitatively taken as measuring lines 7 to 10.

[0084] In addition, when the simulation time is 360s, the temperature contour maps for the left burner with power of 170kW and 340kW are as follows: Figure 16 As shown in the diagram, observations reveal that regardless of the burner power, after the left burner has been running for 360 seconds, the outlines of the temperature control zones formed on the left side of the immersed tunnel are roughly the same. Within these identical outlines, the temperature color deepens with increasing power, indicating that changes in the left burner power do not affect the range of the temperature control zones formed within the immersed tunnel, but only the temperature level within each zone. Therefore, in the factor analysis of the side burner, only the burner opening angle needs to be considered.

[0085] In summary, in this case, the high-temperature control zone of the top burner was quantitatively divided using survey lines 21 and 26 at the top of the immersed tunnel as boundaries, with a zone width of 1m. (See attached diagram.) Figure 17 Temperature control zone ③ in the text.

[0086] The high-temperature control zone of the side burner is quantitatively divided along survey lines 6 and 11 on the left side of the immersed tunnel, with a zone width of 1m. (See...) Figure 17 Temperature control zone ① is defined in the diagram. The control area of ​​the right-side burner is symmetrical, quantitatively divided along survey lines 36 and 41 on the right side of the immersed tunnel, with a zone width of 1m. (See...) Figure 17 Temperature control zone ⑤ in the middle.

[0087] After quantitatively dividing the specific ranges of temperature control zones ① and ③, the remaining area in the middle is temperature control zone ②, which is the area indirectly controlled by the left burner and the top burner; similarly, temperature control zone ④ can be obtained, which is the area indirectly controlled by the right burner and the top burner.

[0088] Step 3: Establishment of an active control model for non-uniform temperature fields using machine learning.

[0089] Before conducting the fire resistance test of the immersed tunnel, the temperature rise curves of the five temperature control zones and the initial rate (0 kW) of the three rows of burners in the first time period were known. During the test, in order to ensure that the five temperature control zones strictly follow the known temperature rise curves, it is necessary to accurately predict the power required to be set for the three rows of burners on the small tower combustion vehicle based on the known temperature rise curves of the five temperature control zones and the initial power of the three rows of burners, using a neural network algorithm.

[0090] Taking a single burner controlling a single temperature control zone as an example ( Figure 1 First, before the experiment, the target temperature rise curve of the temperature control area was known. Considering the time interval between the temperature acquisition instrument and the burner regulating valve, the target temperature rise curve was divided into several segments with 10-second intervals, and the start and end temperatures of each segment were obtained. Using numerical simulation, sample data of the start and end temperatures of the area, the initial power of the three rows of burners, and the required power setting of the burners were randomly constructed. By learning the pattern of the samples through a neural network, the required power of the burners for each segment was predicted before the experiment based on the known start and end temperatures of the temperature rise curve and the initial power of the three rows of burners, and these powers were input into the burner control system. Finally, a neural network prediction error compensation algorithm was proposed. In the experiment, the predicted value of the neural network was adjusted in real time according to the deviation between the actual temperature and the target temperature of the previous segment.

[0091] Mapping the control flow of a single burner controlling a single temperature control zone to three rows of burners controlling five temperature control zones constitutes the active control flow that needs to be implemented in this chapter. Figure 2 As shown. The specific process is as follows:

[0092] Step 1: Divide the temperature rise curves. Divide the known temperature rise curves of temperature control zones ① to ⑤ into 10-second intervals to obtain the start and end temperatures of the five temperature control zones within several time intervals;

[0093] Step 2: Predict burner power. Using a neural network algorithm, with the start and end temperatures of the five zones and the initial power of the three rows of burners (0kW) in the first 10-second time period, predict the required power settings for the left, top, and right burners in the first 10-second time period; using the start and end temperatures of the five zones and the predicted power of the three rows of burners in the first 10-second time period as the initial power, predict the required power settings for the left, top, and right burners in the second 10-second time period; and so on, to obtain the required burner power settings for all time periods.

[0094] Step 3: Input burner power. Before the test begins, input the power required by the three rows of burners for each time period into the burner control system;

[0095] Step 4: Monitor the temperature of the temperature-controlled zones. Monitor the temperature changes in the five temperature-controlled zones separately, and calculate the average deviation between the actual temperature and the target temperature in each of the five temperature-controlled zones within each time period;

[0096] Step 5: Burner power prediction error compensation. During the test, based on the cumulative temperature difference trend and average temperature difference within each time period, the burner power predicted by all neural networks is uniformly increased or decreased.

[0097] In step 2 above, a neural network model needs to be established to predict the burner power. The model establishment process will be described in detail below.

[0098] First, estimate the sample size using the following empirical formula:

[0099] (1) Based on the network training error estimation formula

[0100]

[0101] n w =l(m+1)+m(n+1) (2)

[0102] In the formula: N is the sample set size; ε is the training error; l is the number of input layer nodes; m is the number of hidden layer nodes; n is the number of output layer nodes.

[0103] (2) Estimation formula for the number of neurons in each layer of a neural network

[0104] N=(2~10)×(lm+mn) (3)

[0105] Estimating the sample set size requires determining the number of hidden layer nodes in the neural network, which necessitates extensive trial and error. Before starting the neural network trial calculations, the number of hidden layer nodes can be roughly estimated using the following formula, based on the number of input layer nodes and output layer nodes:

[0106]

[0107] m = log₂l (5)

[0108]

[0109] In the formula: a is a natural number between 1 and 10.

[0110] The model has 13 input layer nodes and 3 output layer nodes. Based on formulas (4), (5), and (6), the number of hidden layer nodes in this model is estimated to be a natural number between 4 and 14. Assuming a training error of 10%, the sample set size of this model is estimated to be between 128 and 2510 using formulas (1), (2), and (3).

[0111] Next, orthogonal design was used to design experiments, a highly efficient experimental design method aimed at minimizing the number of experiments while maximizing information acquisition. In the active control process of the transverse non-uniform temperature field in the immersed tunnel, based on the start and end temperatures of five temperature control zones and the initial power of the three rows of burners (13 data points) within a 10-second time period, a neural network algorithm was used to predict the power required by the three rows of burners on the small tower combustion vehicle within that time period (3 data points). Therefore, in the neural network, 13 data points served as input parameters (the start and end temperatures of the five temperature control zones and the initial power of the three rows of burners), and 3 data points served as output parameters (the required power of the three rows of burners). In the numerical simulation, it is necessary to set the power of the three rows of burners and the start temperature of the five temperature control zones, and obtain the end temperature of the five temperature control zones after 10 seconds of combustion, thus obtaining the corresponding input and output samples.

[0112] In Fluent software, the burner on a small tower combustion vehicle can be simulated using the velocity inlet in the boundary conditions. The burner's power mainly depends on two parameters: the velocity inlet and the temperature. Since the flame temperature at the laboratory burner nozzle is approximately 1500°C, the inlet temperature is set to 1800K (approximately 1500°C) in the Fluent simulation to approximate real-world conditions.

[0113] In the sample design, the burner power is altered by changing the velocity inlet; different initial heating rates allow the burner to vary based on different power levels; and different rates of change enable the burner to achieve different power increases within 10 seconds. This design ensures that the burner undergoes different power changes within 10 seconds based on different initial power levels, creating distinct temperature fields in five temperature control zones. This demonstrates the correlation between burner power changes and temperature changes in the five zones on the small tower combustion vehicle.

[0114] The maximum power of the burner on the small tower combustion vehicle is known to be 600kW, corresponding to an inlet velocity of 19m / s and a temperature of 1800K (approximately 1500℃). Therefore, the maximum inlet velocity achievable by the burner in the sample design is 19m / s. The factors and levels of the orthogonal design of the sample scheme are shown in Table 3.

[0115] Table 3 Number and Level Values ​​of Factors

[0116]

[0117] Based on the orthogonal design method, an orthogonal design was performed using the four factors and six levels in the table as the baseline, resulting in a sample size of L. 48 (6 4 = 48. Given that the same power will exhibit different heating effects at different ambient temperature levels, each sample was calculated at 10 different ambient temperature levels: 300K, 400K, 500K, 600K, 700K, 800K, 900K, 1000K, 1100K, and 1200K (27℃, 127℃, 227℃, 327℃, 427℃, 527℃, 627℃, 727℃, 827℃, and 927℃). Therefore, the final sample size is L. 48 (6 4 )×10=480, which is within the range of sample set capacity.

[0118] Next, a comprehensive sample analysis was conducted to ensure the non-uniformity of the temperature in the temperature-controlled areas obtained by the implementation method. Based on the sample scheme, a numerical simulation method was used to obtain 480 samples, and the maximum temperature difference (highest ending temperature - lowest ending temperature) between the five temperature-controlled zones for each sample was calculated. Figure 18 As shown in the figure. Analysis shows that the maximum difference in the ending temperatures of the five temperature regions in most samples is around 50℃, while the maximum difference in the ending temperatures of the five temperature regions reaches 218℃, indicating that the ending temperatures of the five temperature control regions obtained by numerical simulation have a certain degree of non-uniformity.

[0119] The normalization method for neural networks is the extreme value method, which normalizes the data values ​​to [0,1], as shown in the following formula:

[0120]

[0121] To enhance the training effect of the model, this model uses a 7:1.5:1.5 ratio to divide the training set, validation set, and test set, with 336 samples in the training set, 72 samples in the validation set, and 72 samples in the test set.

[0122] The BP neural network is used as the training model, and its structure diagram can be seen. Figure 19The input layer nodes are the start and end temperatures of the five temperature control zones and the initial power of the three rows of burners, totaling 13 nodes; the output layer nodes are the required power of the three rows of burners, totaling 3 nodes. After trial calculations, the number of hidden layer nodes is 11 (Table 4).

[0123] Table 4. Trial calculation results of hidden layer nodes in BP neural network

[0124]

[0125] Following the same trial-and-error method, the initialization method, activation function, training function, error function, training accuracy, and number of training epochs of the BP network were calculated. The final hyperparameter selections for the BP network are shown in Table 5.

[0126] Table 5 BP Neural Network Model Parameter Configuration

[0127]

[0128]

[0129] To test the true predictive ability, or generalization ability, of a trained BP neural network model, a performance evaluation metric is typically used for direct evaluation. This model uses the mean relative error (MRE) as the performance evaluation metric, as shown in formula (8).

[0130]

[0131] In the formula: y i This is a predicted value; This is the actual value.

[0132] The test set is predicted using a pre-trained BP neural network, and the average relative error between the predicted and target results is calculated. Simultaneously, the superiority of the proposed method is verified by comparing the training results of a radial basis function (RBF) model and a radial basis function model optimized by particle swarm optimization (PSO-RBF). The results are shown in Table 6.

[0133] As shown in the table, the average relative error of the BP neural network on the test set is within 10%, indicating that the model performs well on the test set and has excellent generalization ability, meeting the expected accuracy requirements. Compared with RBF and PSO-RBF neural networks, the BP neural network has superior performance. Therefore, the BP neural network was finally selected as the model for predicting the transverse non-uniform temperature field of immersed tunnel.

[0134] Table 6. Average Relative Error of BP Neural Network Test Set

[0135]

[0136] Step 4: Neural network error compensation mechanism.

[0137] To ensure that the actual temperature closely approximates the target temperature during the experiment, a prediction error compensation mechanism is proposed. This mechanism calculates the average temperature error of five temperature control zones every 10 seconds. The specific process of this mechanism is as follows:

[0138] Step 1: Calculate the average temperature error. Calculate the average temperature error between the actual temperature and the target temperature in the five temperature control zones during the first 10-second time period;

[0139] Step 2: Determine the cumulative trend of temperature error (qualitative adjustment). When the cumulative trend of temperature error increases, the actual temperature is greater than the target temperature, and the burner power values ​​predicted by all neural networks need to be reduced; when the cumulative trend of temperature error decreases, the actual temperature is less than the target temperature, and the predicted values ​​of all neural networks need to be increased.

[0140] Step 3: Adjust the neural network prediction value based on the average temperature error (quantitative adjustment). Based on experience, when the average temperature error is 0–50℃, the prediction value is increased by a factor of 1.1; when the average temperature error is -50–0℃, the prediction value is decreased by a factor of 0.9; when the average temperature error is 50–100℃, the prediction value is increased by a factor of 1.2; when the average temperature error is -100–-50℃, the prediction value is increased by a factor of 0.8, and so on.

[0141] Step 4: Repeat steps 1 through 3 until the experiment is stopped.

[0142] Based on the above process, assuming that all five temperature control zones must achieve the temperature of the first 60 seconds of the HC curve, and according to the 10-second division principle, the start and end temperatures of the five temperature control zones and the initial power (0 kW) of the three rows of burners are obtained for six time periods. The power of the three rows of burners at the six time points is predicted by a neural network, and the power is input into Fluent for numerical simulation to obtain the temperature rise curves of the five zones. Tables 7 and 8 show the predicted power values ​​of the burners at the time points before and after adding the error compensation mechanism, respectively.

[0143] Table 7 Predicted values ​​for three-row burners at six time points.

[0144]

[0145] Table 8 Adjustment values ​​for six time points for the three-row burner

[0146]

[0147] As shown in Tables 7 and 8, compared with the unadjusted neural network prediction, the burner power at the 10-second node is increased by 1.1 times because the actual temperature is lower than the target temperature and the deviation is within 50℃ during the first 10-second period. During the remaining periods, the actual temperature is lower than the target temperature, so the burner power is reduced.

[0148] The data was adjusted using the aforementioned error compensation method. Figure 20 The target temperature, pre-adjustment temperature, and post-adjustment temperature of five zones are displayed. It can be observed that after adding the prediction error compensation mechanism, the error between the actual temperature and the target temperature in all five temperature control zones decreases after 60 seconds. At 60 seconds, the errors for temperature control zones ①, ②, ③, ④, and ⑤ are 15%, 4%, -5%, -2%, and 2%, respectively, indicating that the mechanism has a certain compensating effect on the error in predicting the power of the three rows of burners. Zone ① remains the zone with the largest error, indicating that the compensation mechanism can play a role in overall adjustment.

[0149] This concludes the demonstration of the entire invention's operation and verification process.

[0150] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for zoning and active control of transverse non-uniform temperature field in rectangular immersed tunnels, characterized in that, include: S1. Temperature field division within the tunnel area; The specific steps include: S11. Preliminarily determine the burner control area, that is, determine the non-uniform temperature field area controlled by the burner based on its location. S12. Quantitatively determine the control area, that is, determine the specific size of the control area; S2. Active temperature control in each zone; this involves using machine learning technology to train a neural network model to control the burner; specific steps include: S21. Using Fluent numerical simulation, construct the correspondence between the power of the three rows of burners on the small tower combustion vehicle and the temperature and initial power of the five temperature control zones, and obtain a sample set; In step S12, the burners are divided into top burners with the nozzle facing a 90° angle to the ground and side burners with the nozzle facing the ground, according to the nozzle angle. The two types of burners control the temperature fields at the top and sides, respectively. The top temperature field determines the range of the controlled high-temperature area based on the power of the top burner. At this time, the temperature control area of ​​the top burner is divided according to the temperature area with high sensitivity, that is, the temperature area that starts to heat up earlier is the temperature control area of ​​the top burner. The side temperature field is determined by the burner nozzle scheduling, i.e., the angle between the burner and the horizontal line, to determine the range of the high-temperature area it controls. The principle for dividing the side temperature field is that the control range of the side burner should be focused on the tunnel sidewall as much as possible to avoid providing a large amount of heat to the top and bottom. The entire process is simulated using Fluent software to quantitatively divide the non-uniform temperature field. S22. Use the sample set to train the machine learning algorithm and build a prediction model for the transverse non-uniform temperature field of the immersed tunnel. Before training the model, the sample set capacity needs to be estimated, and the sample scheme needs to be designed using the orthogonal design method. Reasonable training, validation and test sets need to be generated through sample data analysis and standardized sample set partitioning. S3. Temperature prediction error compensation: The average temperature error of the five temperature control zones is calculated every 10 seconds. The purpose is to reduce the prediction error of the burner power and improve the accuracy of the temperature rise curve restoration of the five temperature control zones.

2. The method for zoning and active control of transverse non-uniform temperature field in rectangular immersed tunnels as described in claim 1, characterized in that, In step S22, taking a single burner controlling a single temperature control zone as an example, the control flow of a single burner controlling a single temperature control zone is mapped to three rows of burners controlling five temperature control zones. This is the active control flow that needs to be implemented, and the specific flow is as follows: Step 1: Divide the temperature rise curves; Divide the known temperature rise curves of the five temperature control zones ① to ⑤ into 10-second intervals to obtain the start and end temperatures of the five temperature control zones within several time intervals. Step 2: Predict burner power; Using a neural network algorithm, with the start and end temperatures of the five zones and the initial power of the three rows of burners (0kW) in the first 10-second time period, predict the required power settings for the left, top, and right burners in the first 10-second time period; using the start and end temperatures of the five zones and the predicted power of the three rows of burners in the first 10-second time period as the initial power, predict the required power settings for the left, top, and right burners in the second 10-second time period; and so on, to obtain the required burner power settings for all time periods. Step 3: Input burner power; Before the test begins, input the power required by the three rows of burners for each time period into the burner control system; Step 4: Monitor the temperature in the temperature-controlled area; The temperature changes in five temperature control zones were monitored separately, and the average deviation between the actual temperature and the target temperature in the five temperature control zones was calculated for each time period. Step 5: Burner power prediction error compensation; During the test, based on the cumulative temperature difference trend and average temperature difference in each time period, uniformly increase or decrease the burner power predicted by all neural networks.

3. The method for zoning and active control of the transverse non-uniform temperature field in a rectangular immersed tunnel as described in claim 1, characterized in that, The specific process for temperature prediction error compensation is as follows: Step 1: Calculate the average temperature error; calculate the average temperature error between the actual temperature and the target temperature of the five temperature control zones within the first 10-second time period; Step 2: Determine the cumulative trend of temperature error, i.e., make qualitative adjustments; When the cumulative temperature error trend increases, the actual temperature is greater than the target temperature, and the burner power values ​​predicted by all neural networks need to be reduced; when the cumulative temperature error trend decreases, the actual temperature is less than the target temperature, and the burner power values ​​predicted by all neural networks need to be increased. Step 3: Adjust the burner power value predicted by the neural network according to the average temperature error, i.e., quantitative adjustment; when the average temperature error is 0~50℃, the predicted value is increased by 1.1 times; when the average temperature error is -50~0℃, the predicted value is decreased by 0.9 times; when the average temperature error is 50~100℃, the predicted value is increased by 1.2 times; when the average temperature error is -100~-50℃, the predicted value is decreased by 0.8 times, and so on. Step 4: Repeat steps 1 through 3 until the experiment is stopped.

Images