An Adaptive Temperature Rise Experimental Chamber Based on Deep Learning and Its Method

Abstract

This invention relates to a deep learning-based adaptive temperature rise experimental chamber and its method. The experimental chamber includes: a retractable steel frame, a foldable membrane wall, a temperature acquisition module, and a deep learning temperature control module. The retractable steel frame comprises a skeleton structure consisting of multiple longitudinal columns and transverse connecting rods, as well as a motor drive unit. The foldable membrane wall covers the exterior of the retractable steel frame, forming the experimental chamber. A cooling fan is installed on the foldable membrane wall. The temperature acquisition module is installed inside the experimental chamber to collect temperature data. The deep learning temperature control module uses the temperature data fed back by the temperature acquisition module to predict the optimal control sequence through a deep learning network model to control the operation of the cooling fan. Compared with existing technologies, this invention can adapt to different equipment models, can be quickly deployed, and has high-precision temperature control capabilities.

Description

Technical Field

[0001] This invention relates to the field of substation installation structures, and in particular to an adaptive temperature rise test chamber based on deep learning and its method. Background Technology

[0002] As core power distribution equipment in urban power grid load centers, prefabricated substations' thermal stability directly impacts the reliability of smart grid power supply. With global warming and the increasing frequency of extreme heat events, prefabricated substations face severe thermal failure risks when operating in high-temperature environments. The IEC 62271-202 standard specifies a normal ambient temperature limit of 40°C. When the installation site temperature exceeds this range, targeted high-temperature tests are required to ensure the effectiveness of insulation performance and heat dissipation design. However, current mainstream high-temperature environmental testing heavily relies on large, fixed environmental laboratories. These laboratories are costly to construct, have lengthy testing cycles, and often require weeks or even months from booking and transportation to completion. The high cost and time commitment mean that such tests are typically only "type tests," and only a very small percentage of samples can be tested during the mass production of prefabricated substations, making it difficult to guarantee consistent quality across mass production.

[0003] More importantly, existing high-temperature test chambers are all of fixed dimensions and cannot adapt to prefabricated substations of different models and sizes. When testing equipment of different specifications is required, the test chamber must be rebuilt or modified, which is time-consuming and labor-intensive, as illustrated by the invention disclosed in CN107230942A, which describes a safe, heat-dissipating, and waterproof prefabricated high and low voltage substation. Furthermore, traditional test chamber temperature control systems often employ PID control strategies, which, when faced with strongly coupled and multi-interference systems, often exhibit large overshoot, long settling times, or even oscillation and instability, making it difficult to achieve high-precision temperature control. Therefore, there is an urgent need for a prefabricated substation high-temperature test device that can adapt to different equipment models, can be quickly deployed, and possesses high-precision temperature control capabilities. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art by providing a deep learning-based adaptive temperature rise experimental chamber and method that can adapt to different types of equipment, can be quickly deployed, and has high-precision temperature control capabilities.

[0005] The objective of this invention can be achieved through the following technical solutions: An adaptive temperature rise experimental chamber based on deep learning includes: a retractable steel frame, a foldable thin-film chamber wall, a temperature acquisition module, and a deep learning temperature control module; The retractable steel reinforcement frame includes a frame structure consisting of multiple longitudinal columns and transverse connecting rods, as well as a motor drive unit. The longitudinal columns consist of multiple sleeved steel pipes, with a telescopic structure between adjacent steel pipes, and are driven by the motor drive unit. The foldable membrane cabin wall covers the outside of the retractable steel frame to form an experimental cabin for installing a prefabricated substation inside; a cooling fan is installed on the foldable membrane cabin wall, and the cooling fan is controlled by the deep learning temperature control module. The temperature acquisition module is installed inside the experimental chamber and is used to collect temperature data inside the experimental chamber. The deep learning temperature control module is connected to the cooling fan and the temperature acquisition module respectively. It is used to predict the optimal control sequence based on the temperature data fed back by the temperature acquisition module through a pre-built deep learning network model, so as to control the operation of the cooling fan.

[0006] Furthermore, the telescopic structure is an electric push rod or a lead screw and nut mechanism, and the motor drive unit is connected to each electric push rod or lead screw and nut mechanism to control the length, width and height dimensions of the telescopic steel skeleton. The transverse connecting rod is disposed between two adjacent longitudinal columns, used to connect the adjacent longitudinal columns and to extend and retract synchronously with the extension and retraction of the longitudinal columns; The bottom of the longitudinal column is equipped with casters.

[0007] Furthermore, the foldable membrane bulkhead includes a membrane substrate, folded patterns pre-set on the membrane substrate, and an edge sealing strip. The foldable membrane bulkhead is installed outside the retractable steel frame via a moving guide rail and a fixed frame. The membrane substrate is installed on the moving guide rail and is limited by the fixed frame. The edge sealing strip is located at the connection between the membrane substrate and the moving guide rail and the fixed frame. The folded patterns are arranged parallel to the extension and retraction direction of the retractable steel frame.

[0008] Furthermore, the top of the experimental chamber is provided with multiple heat dissipation vents, and the cooling fan is installed in the corresponding heat dissipation vent; the cooling fan is an axial flow fan. When the cooling fan is turned off, the corresponding heat dissipation vent is in a sealed state; When the cooling fan is turned on, controlled cooling is achieved through forced convection.

[0009] Furthermore, the temperature acquisition module includes temperature sensors and a data acquisition unit. The number of temperature sensors is multiple and they are evenly distributed on the inner wall of the experimental chamber and at key hot spots in the prefabricated substation. The data acquisition device is located in the high-voltage room of the prefabricated substation and is electrically connected to each temperature sensor in the prefabricated substation to collect temperature data within the prefabricated substation.

[0010] The present invention also provides a control method for an adaptive temperature rise experimental chamber based on deep learning as described above, comprising the following steps: The experimental chamber is simulated in advance to obtain simulation data of temperature changes inside the chamber; sparse experimental data is constructed based on the collected temperature data inside the chamber; the deep learning network model is trained using the simulation data and the sparse experimental data to obtain a temperature field reconstruction proxy model. The temperature data collected in real time inside the experimental chamber is input into the temperature field reconstruction proxy model for prediction, and the predicted average temperature of the chamber space is obtained. The predicted result is then compared with the preset target temperature setpoint, and rolling optimization is performed to generate the optimal control sequence. The cooling fan is controlled using an optimal control sequence.

[0011] Furthermore, the expression for the loss function of the deep learning network model during training is as follows: In the formula, The loss function of a deep learning network model. This is the data residual term, used to measure the difference between the prediction results of the deep learning network model and the sparse experimental data; For PDE residuals, it is used to measure the degree to which the prediction results of deep learning network models violate local PDEs; This is the boundary residual term, used to measure the degree to which the prediction results of a deep learning network model violate boundary conditions; This is the power conservation residual term, used to ensure that the solution learned by the deep learning network model satisfies macroscopic physical equilibrium at any time. w data , w PDE , w bou , w cons Weighting coefficients to weigh the importance of each factor.

[0012] Furthermore, the calculation process for the power conservation residual difference term includes: Based on the operating power loss of the prefabricated substation, the total heat capacity of the air and solids inside the experimental chamber, and the total heat exchange between the experimental chamber walls and the environment, a global energy balance equation is constructed. The expression of this global energy balance equation is as follows: In the formula, This refers to the operating power loss of prefabricated substations. The total heat capacity of the air and solids inside the experimental chamber. The average temperature inside the experimental chamber. The total heat exchange between the experimental chamber walls and the environment is expressed as follows: Q wall = U wall · A surf ·( T avg - T 0) In the formula, U wall The equivalent heat transfer coefficient of the bulkhead. A surf To achieve effective heat dissipation surface area, T 0 represents the ambient temperature; In each training batch of the temperature field reconstruction surrogate model, the current average temperature inside the experimental cabin is calculated by integrating the average temperature predicted by the temperature field reconstruction surrogate model. The power conservation residual difference term is calculated, and its expression is as follows: In the formula, The number of sample points in time series t. For the sample Operating loss power of prefabricated substations The error between the calculated value and the actual value.

[0013] Furthermore, the rolling optimization process includes: At each control sampling moment, real-time temperature data inside the experimental chamber is received and input into the temperature field reconstruction proxy model to reconstruct the current complete temperature field distribution inside the chamber. Using this as the initial state, future temperature field distributions are predicted. N p Within a time step, under the assumed future control input sequence { u ( k | k ), u ( k +1| k ), …, u ( k + N p-1 | k Under these conditions, the average temperature inside the experimental chamber... T avg The evolutionary trajectory; Construct a rolling optimization objective function, the expression of which is: In the formula, To optimize the objective function value during rolling, , Used to punish the future N p The predicted average cabin temperature within the step T avg ( k + i | k ) and target set temperature T set The deviation between them, where Q is a positive semi-definite weight matrix. Used to suppress control quantity u This refers to the start / stop status of the cooling fan or drastic changes in its speed; R The weight matrix is ​​positive definite. By adjusting the weight matrix... Q and R Balancing temperature control accuracy with the smoothness of actuator operation; Physical constraints are imposed on configuring the cooling fan for the rolling optimization objective function; At each sampling time k Solve the quadratic programming problem of the rolling optimization objective function with physical constraints of a cooling fan, and obtain the optimal control sequence { u ( k | k ), u ( k +1| k ), …, u ( k + N p-1 | k )}, take the first element of the sequence u ( k | k The current control command is sent to the cooling fan for execution.

[0014] Furthermore, the experimental chamber is also equipped with auxiliary heating equipment to provide supplementary heat when the prefabricated substation's self-generated heat is insufficient; the start and stop of the auxiliary heating equipment is controlled according to the surplus power of self-generated heat, which is the difference between the heat generated by the equipment and the heat dissipation required to maintain the target temperature. The formula for calculating the self-generated heat surplus power is as follows: In the formula, To generate surplus power from self-generated heat, Pdev This refers to the power loss during operation of prefabricated substations. U wall The equivalent heat transfer coefficient of the experimental chamber wall is given. A surf To achieve effective heat dissipation surface area, T set For the target temperature, T 0 represents the ambient temperature; The start-stop control process of the auxiliary heating equipment includes: When Δ P When the value is >0, the auxiliary heating equipment is not activated; When Δ P When the value is ≤0, start the auxiliary heating equipment.

[0015] Compared with the prior art, the present invention has the following advantages: (1) The present invention designs a retractable steel frame, which can adjust the length, width and height by motor drive, and can adapt to different models and specifications of prefabricated substations. This solves the problem that traditional fixed experimental chambers cannot adapt to multiple models of equipment and significantly improves the versatility of the experimental device.

[0016] (2) The present invention adopts a foldable thin film cabin wall, which can be folded and unfolded with the expansion and contraction of the skeleton through a preset V-shaped crease. While ensuring the airtightness of the cabin, it can achieve portable storage and rapid deployment, and the deployment cycle of a single experiment is greatly shortened.

[0017] (3) The present invention opens several heat dissipation vents on the top of the cabin and installs independently controlled cooling fans, which not only ensures the heat dissipation requirements, but also achieves precise temperature regulation through zone control.

[0018] (4) The present invention constructs a deep learning temperature control module. By reconstructing the thermal field data of the experimental chamber through simulation, the deep learning network model is trained by combining the simulation data and the actual collected sparse experimental data to obtain a temperature field reconstruction proxy model, which is used to predict the average temperature of the chamber space based on the measured temperature data. Furthermore, the accurate and stable control of the chamber temperature is achieved through rolling optimization, which solves the problem of poor temperature control accuracy of traditional PID control under strong interference.

[0019] (5) The present invention introduces a power conservation residual term in the loss function of the thermal field reconstruction unit, which aims to force the temperature field learned by the model to satisfy the macroscopic energy balance of the entire experimental chamber system at any time. It embeds the macroscopic global energy conservation as a physical constraint into the deep learning model, which is different from the traditional PINN that only relies on local PDE. The introduction of this residual term ensures that the deep learning model not only learns the microscopic process of local thermal diffusion, but also strictly follows the macroscopic thermodynamic laws, which significantly improves the prediction accuracy and physical consistency of the model in the sparse data region.

[0020] (6) The present invention also automatically determines the start and stop of the auxiliary heating equipment based on the surplus power of self-generated heat, and gives priority to using the operating loss of the prefabricated substation itself as a heat source, which significantly reduces the test energy consumption and operating cost. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the structure of an adaptive temperature rise experimental chamber based on deep learning provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a retractable steel frame structure provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the folded and unfolded state of a foldable membrane bulkhead provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the arrangement of a heat dissipation port and a cooling fan provided in an embodiment of the present invention; Figure 5 This is a block diagram of a deep learning temperature control module provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of a control method for an adaptive temperature rise experimental chamber based on deep learning, provided in an embodiment of the present invention. In the diagram, 1. Retractable steel frame, 2. Longitudinal column, 3. Horizontal connecting rod, 4. Motor drive unit, 5. Prefabricated substation, 6. Casters, 7. Foldable membrane bulkhead, 8. Heat dissipation vent, 9. Cooling fan, 10. Temperature acquisition module, 11. Temperature sensor, 12. Data acquisition unit, 13. Deep learning temperature control module, 14. Computer workstation, 15. Computer workstation cabinet door, 16. Computer workstation screen, 17. Computer workstation keyboard, 18. Computer workstation cabinet. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0023] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0024] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0025] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of this invention is usually placed during use. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0026] It should be noted that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0027] Furthermore, terms such as "horizontal" and "vertical" do not imply that components must be absolutely horizontal or suspended, but rather that they can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0028] Example 1 like Figure 1 As shown, this embodiment provides an adaptive temperature rise experimental chamber based on deep learning, including: a retractable steel frame 1, a foldable thin film chamber wall 7, a temperature acquisition module 10, and a deep learning temperature control module 13; The retractable steel reinforcement frame 1 includes a frame structure consisting of multiple longitudinal columns 2 and transverse connecting rods 3, as well as a motor drive unit 4. The longitudinal columns 2 include multiple sleeved steel pipes, and a telescopic structure is provided between adjacent steel pipes, which is driven by the motor drive unit 4. A foldable membrane bulkhead 7 covers the exterior of a retractable steel frame 1 to form an experimental chamber for installing a prefabricated substation 5 inside; a cooling fan 9 is installed on the foldable membrane bulkhead 7, and the cooling fan 9 is controlled by a deep learning temperature control module 13. The temperature acquisition module 10 is installed inside the experimental chamber to collect temperature data inside the experimental chamber. The deep learning temperature control module 13 is connected to the cooling fan 9 and the temperature acquisition module 10 respectively. It is used to predict the optimal control sequence based on the temperature data fed back by the temperature acquisition module 10 through a pre-built deep learning network model, so as to control the operation of the cooling fan 9.

[0029] Specifically, such as Figure 1 As shown, the adaptive temperature rise experimental chamber includes a retractable steel frame 1, which includes multiple longitudinal columns 2, transverse connecting rods 3, and a motor drive unit 4.

[0030] like Figure 2 As shown, the longitudinal column 2 is composed of multiple interconnected steel pipes. Electric push rods or lead screw nut mechanisms are installed between adjacent steel pipes, and the extension and retraction of the column are achieved through the motor drive unit 4. Transverse connecting rods 3 are installed between the columns to connect adjacent columns and extend and retract synchronously with the columns, ensuring the overall stability of the frame. The motor drive unit 4 is electrically connected to each electric push rod or lead screw nut mechanism to control the length, width, and height dimensions of the frame 1.

[0031] The bottom of the longitudinal column 2 is equipped with casters 6 for easy movement and fixation.

[0032] YBM229-40.5 / 2 Taking the 0.72 model prefabricated substation 5 as a reference, the corresponding retractable steel frame 1 has an adjustable frame length range of 2000mm to 4000mm, an adjustable width range of 1500mm to 3000mm, and an adjustable height range of 2000mm to 3500mm, which can cover the size specifications of mainstream prefabricated substations on the market.

[0033] The foldable membrane bulkhead 7 includes a membrane substrate, folded patterns pre-set on the membrane substrate, and edge sealing strips. The foldable membrane bulkhead 7 is installed outside the retractable steel frame 1 via a moving guide rail and a fixed frame. The membrane substrate is installed on the moving guide rail and is limited by the fixed frame. The edge sealing strip is located at the connection between the membrane substrate and the moving guide rail and the fixed frame. The folded patterns are arranged parallel to the extension and retraction direction of the retractable steel frame 1.

[0034] Specifically, such as Figure 3 As shown, the foldable membrane bulkhead 7 covers the exterior of the retractable steel frame 1. The foldable membrane bulkhead 7 includes a high-temperature resistant modified PVB membrane substrate, folded patterns pre-set on the membrane substrate, and edge sealing strips.

[0035] The fold lines are arranged parallel to the direction of skeleton expansion and contraction, forming V-shaped folds, so that the film is in a stacked and contracted state when the skeleton shrinks, such as... Figure 3 As shown in 'a'; when the skeleton extends, it is in a flat state, as... Figure 3As shown in b. This folded structure ensures that the membrane can expand or contract synchronously with the skeleton, avoiding excessive stretching or wrinkle buildup. Edge sealing strips are set at the edges of the membrane chamber walls to form an airtight connection with the skeleton and the ground, ensuring the airtightness of the experimental chamber.

[0036] The top of the experimental chamber is equipped with multiple heat dissipation vents 8, and cooling fans 9 are installed in the corresponding heat dissipation vents 8; the cooling fans 9 are axial flow fans.

[0037] Specifically, such as Figure 4 As shown, several heat dissipation vents 8 are opened on the top of the experimental chamber, and a cooling fan 9 is installed at each vent 8. The heat dissipation vents 8 are circular openings. In this embodiment, five circular heat dissipation vents are opened, each with a diameter of 300mm, and the total area occupies approximately 20% of the top area of ​​the chamber. The cooling fans 9 are axial fans, and each cooling fan is independently connected to the deep learning temperature control module 13 to achieve zoned control. When the cooling fans are off, the heat dissipation vents are in a sealed state to prevent heat loss; when the cooling fans are on, controlled heat dissipation is achieved through forced convection to regulate the temperature inside the chamber.

[0038] The temperature acquisition module 10 includes temperature sensors 11 and data acquisition unit 12. There are multiple temperature sensors 11, which are evenly distributed on the inner wall of the experimental chamber and at key hot spots in the prefabricated substation 5. The data acquisition unit 12 is located in the high-voltage room of the prefabricated substation 5 and is electrically connected to each temperature sensor 11 of the prefabricated substation 5 to collect temperature data in the prefabricated substation 5.

[0039] In this embodiment, the temperature sensor is a PT100 platinum resistance temperature sensor with a measurement accuracy of ±0.1℃ and a sampling frequency of 1Hz, which can reflect the temperature distribution inside the cabin and the operating status of the equipment in real time.

[0040] like Figure 1 As shown, the deep learning temperature control module 13 is built into the computer workstation 14, which includes the computer workstation cabinet door 15, the computer workstation screen 16, the computer workstation keyboard 17, and the computer workstation cabinet 18.

[0041] like Figure 5 As shown, the deep learning temperature control module 13 includes a thermal field reconstruction unit, a prediction control unit, and a heat source management module.

[0042] The thermal field reconstruction unit is used to embed the residuals of the local heat conduction partial differential equation (PDE) and the global energy conservation equation as constraints into a loss function. A temperature field reconstruction surrogate model is trained using simulation data and sparse experimental data. The loss function of the thermal field reconstruction unit is constructed as a weighted sum of the data residuals, the PDE residuals, the boundary residuals, and the power conservation residuals. Among them, data residuals Measuring the difference between network predictions and sparse experimental data; PDE residual term Measuring the degree to which network predictions violate local PDEs; boundary residuals Measuring the degree to which network predictions violate boundary conditions; power conservation residuals and differences. Ensure that the solution learned by the neural network strictly satisfies macroscopic physical equilibrium at all times; w data , w PDE , w bou , w cons Weighting coefficients to weigh the importance of each factor.

[0043] This invention innovatively introduces a power conservation residual term. The aim is to force the temperature field learned by the model to satisfy the macroscopic energy balance of the entire experimental chamber system at all times. The specific calculation process is as follows: Based on the operating loss power of prefabricated substations P dev Total heat capacity of air and solids inside the experimental chamber C total Total heat exchange between bulkhead and environment Q wall Construct the global energy balance equation: In the formula, T avg The average temperature inside the cabin; Q wall = U wall · A surf · ( T avg - T 0), of which U wall The equivalent heat transfer coefficient of the bulkhead. A surf To achieve effective heat dissipation surface area, T 0 represents the ambient temperature. In each batch of model training, the current temperature is calculated by integrating the spatial temperature field predicted by the surrogate model. T avg and d T avg / d t To calculate : In the formula N t The number of sample points in the time series is denoted as . The introduction of this residual term ensures that the deep learning model not only learns the microscopic process of local heat diffusion, but also strictly follows the macroscopic thermodynamic laws, significantly improving the model's prediction accuracy and physical consistency in sparse data regions.

[0044] The predictive control unit reconstructs the surrogate model based on the trained temperature field and uses a rolling optimization strategy to generate the optimal control commands for the cooling fan 9. The specific implementation steps are as follows: 1. State Acquisition and Prediction: At each control sampling time... k The predictive control unit receives current temperature data from a limited number of measuring points within the cabin, fed back by the temperature acquisition module 10, and inputs it into the temperature field reconstruction proxy model. This model reconstructs the current complete cabin temperature field distribution and uses this as the initial state to predict future... N p Within a time step, for a series of hypothetical future control input sequences { u ( k | k ), u ( k +1| k ), …, u ( k + N p-1 | k The average temperature inside the cabin T avg The evolutionary trajectory of .

[0045] 2. Rolling Optimization Objective Function: Solving a constrained optimization problem to determine the optimal control sequence. The objective function used in this invention takes the following form: In the formula: the first term is the temperature tracking term, Used to punish the future N p The predicted average cabin temperature within the step T avg ( k + i | k ) and target set temperature T set The deviation between them, Q is a positive semi-definite weight matrix; the second term is the control increment penalty term. Used to suppress control quantity u The start / stop status or drastic changes in the speed of cooling fan 9 R The weight matrix is ​​positive definite. By adjusting the weight matrix... Q andR This allows for a balance between temperature control accuracy and the smoothness of the actuator's operation.

[0046] 3. Constraint handling: The optimization process must meet the physical constraints of cooling fan 9.

[0047] 4. Solution and Implementation: At each sampling time... k Solving the constrained quadratic programming problem above yields the optimal control sequence { u ( k | k ), u ( k +1| k ), …, u ( k + N p-1 | k )}. Only the first element of the sequence u ( k | k The control command for the current moment is issued to the cooling fan 9 for execution. At the next sampling moment, the system state is updated, and the above "prediction-optimization-implementation" rolling process is repeated.

[0048] The heat source management module automatically determines the start / stop of auxiliary heating equipment based on the surplus self-generated heat power. Auxiliary heating equipment 19 is installed inside the compartment and is used to provide supplementary heat when the prefabricated substation's self-generated heat is insufficient. The surplus self-generated heat power is defined as the difference between the heat generated by the equipment and the heat dissipation required to maintain the target temperature difference. in, P dev This refers to the power loss during operation of prefabricated substations. U wall The equivalent heat transfer coefficient of the experimental chamber wall is given. A surf To achieve effective heat dissipation surface area, T set For the target temperature, T 0 represents the ambient temperature. When Δ P When Δ > 0, it indicates that the equipment's own heat generation is sufficient to offset the heat loss from the bulkhead, and the auxiliary heating equipment will not be activated; when Δ P ≤0 indicates that the equipment's own heat generation is insufficient to maintain the target temperature, and auxiliary heating equipment is activated to provide supplementary heat.

[0049] This embodiment establishes the following: Figure 1The adaptive temperature rise test chamber shown is used for continuous temperature rise tests on different types of prefabricated substations. During the test, a motor-driven telescopic frame is adjusted to the appropriate size, and the foldable membrane chamber wall unfolds to form a sealed chamber. The temperature acquisition module collects the internal temperature data in real time, and the deep learning temperature control module generates control commands based on the collected data to adjust the start and stop of the cooling fan. The experimental results show that this adaptive temperature rise test chamber can quickly adapt to prefabricated substations of different sizes and specifications, and the internal temperature is stably controlled within the target temperature range of ±1.2℃. The temperature control accuracy meets the requirements of IEC 62271-202 standard for the stability of high-temperature type test environments. The deployment cycle of a single test is shortened to less than 24 hours, significantly improving the testing efficiency and versatility.

[0050] Example 2 like Figure 6 As shown, this embodiment provides a control method for an adaptive temperature rise experimental chamber based on deep learning, as described in Embodiment 1, including the following steps: S1: Simulate the experimental chamber in advance to obtain simulation data of temperature changes inside the experimental chamber; construct sparse experimental data based on the collected temperature data inside the experimental chamber; train the deep learning network model using the simulation data and sparse experimental data to obtain the temperature field reconstruction proxy model. S2: Input the real-time collected temperature data inside the experimental chamber into the temperature field reconstruction proxy model for prediction, obtain the predicted average temperature of the chamber space, compare it with the preset target temperature setpoint, perform rolling optimization, and generate the optimal control sequence. S3: Use the optimal control sequence to control the cooling fan 9.

[0051] Specifically, the expression for the loss function of a deep learning network model during training is as follows: In the formula, The loss function of a deep learning network model. This is the data residual term, used to measure the difference between the prediction results of the deep learning network model and the sparse experimental data; For PDE residuals, it is used to measure the degree to which the prediction results of deep learning network models violate local PDEs; This is the boundary residual term, used to measure the degree to which the prediction results of a deep learning network model violate boundary conditions; This is the power conservation residual term, used to ensure that the solution learned by the deep learning network model satisfies macroscopic physical equilibrium at any time. w data , w PDE , w bou ,w cons Weighting coefficients to weigh the importance of each factor.

[0052] Preferably, the experimental chamber is also equipped with auxiliary heating equipment to provide supplementary heat when the self-generated heat of the prefabricated substation 5 is insufficient; the start and stop of the auxiliary heating equipment is controlled according to the surplus power of self-generated heat, which is the difference between the heat generated by the equipment and the heat dissipation required to maintain the target temperature. The formula for calculating the surplus power from self-generated heat is: In the formula, To generate surplus power from self-generated heat, P dev For the prefabricated substation 5, the operating loss power, U wall The equivalent heat transfer coefficient of the experimental chamber wall is given. A surf To achieve effective heat dissipation surface area, T set For the target temperature, T 0 represents the ambient temperature; The start-up and shutdown control process of auxiliary heating equipment includes: When Δ P When the value is >0, the auxiliary heating equipment is not activated; When Δ P When the value is ≤0, start the auxiliary heating equipment.

[0053] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A deep learning-based adaptive temperature rise experimental chamber, characterized in that, include: The retractable steel frame (1), the foldable membrane bulkhead (7), the temperature acquisition module (10) and the deep learning temperature control module (13) are all included. The retractable steel frame (1) includes a frame structure consisting of multiple longitudinal columns (2) and transverse connecting rods (3) and a motor drive unit (4). The longitudinal columns (2) include multiple sleeved steel pipes, and a telescopic structure is provided between adjacent steel pipes. The retractable steel frame is driven by the motor drive unit (4). The foldable membrane cabin wall (7) covers the outside of the retractable steel frame (1) to form an experimental cabin for installing a prefabricated substation (5) inside; a cooling fan (9) is installed on the foldable membrane cabin wall (7), and the cooling fan (9) is controlled by the deep learning temperature control module (13). The temperature acquisition module (10) is installed inside the experimental chamber and is used to collect temperature data inside the experimental chamber. The deep learning temperature control module (13) is connected to the cooling fan (9) and the temperature acquisition module (10) respectively. It is used to predict the optimal control sequence based on the temperature data fed back by the temperature acquisition module (10) through a pre-built deep learning network model, so as to control the operation of the cooling fan (9).

2. The adaptive temperature rise experimental chamber based on deep learning according to claim 1, characterized in that, The telescopic structure is an electric push rod or a screw nut mechanism. The motor drive unit (4) is connected to each electric push rod or screw nut mechanism to control the length, width and height dimensions of the telescopic steel skeleton (1). The transverse connecting rod (3) is set between two adjacent longitudinal columns (2) to connect the adjacent longitudinal columns (2) and to extend and retract synchronously with the extension and retraction of the longitudinal columns (2); The bottom of the longitudinal column (2) is equipped with casters (6).

3. The adaptive temperature rise experimental chamber based on deep learning according to claim 1, characterized in that, The foldable membrane bulkhead (7) includes a membrane substrate, folded patterns pre-set on the membrane substrate, and edge sealing strips. The foldable membrane bulkhead (7) is installed outside the retractable steel frame (1) via a moving guide rail and a fixed frame. The membrane substrate is installed on the moving guide rail and is limited by the fixed frame. The edge sealing strip is located at the connection between the membrane substrate and the moving guide rail and the fixed frame. The folded patterns are arranged parallel to the extension direction of the retractable steel frame (1).

4. The adaptive temperature rise experimental chamber based on deep learning according to claim 1, characterized in that, The top of the experimental chamber is provided with multiple heat dissipation vents (8), and the cooling fan (9) is installed in the corresponding heat dissipation vent (8); the cooling fan (9) is an axial flow fan; When the cooling fan (9) is turned off, the corresponding heat dissipation port (8) is in a sealed state; When the cooling fan (9) is turned on, controlled heat dissipation is achieved through forced convection.

5. The adaptive temperature rise experimental chamber based on deep learning according to claim 1, characterized in that, The temperature acquisition module (10) includes a temperature sensor (11) and a data acquisition unit (12). There are multiple temperature sensors (11), which are evenly distributed on the inner wall of the experimental chamber and at key hot spots of the prefabricated substation (5). The data acquisition unit (12) is located in the high-voltage room of the prefabricated substation (5) and is electrically connected to each temperature sensor (11) of the prefabricated substation (5) to collect temperature data in the prefabricated substation (5).

6. A control method for an adaptive temperature rise experimental chamber based on deep learning as described in any one of claims 1-5, characterized in that, Includes the following steps: The experimental chamber was simulated in advance to obtain simulation data on temperature changes inside the experimental chamber; Based on the temperature data collected inside the experimental chamber, sparse experimental data were constructed. The deep learning network model was trained using the simulation data and sparse experimental data to obtain a temperature field reconstruction proxy model; The temperature data collected in real time inside the experimental chamber is input into the temperature field reconstruction proxy model for prediction, and the predicted average temperature of the chamber space is obtained. The predicted result is then compared with the preset target temperature setpoint, and rolling optimization is performed to generate the optimal control sequence. The cooling fan (9) is controlled using the optimal control sequence.

7. The method according to claim 6, characterized in that, The expression for the loss function of the deep learning network model during training is as follows: In the formula, The loss function of a deep learning network model. This is the data residual term, used to measure the difference between the prediction results of the deep learning network model and the sparse experimental data; For PDE residuals, it is used to measure the degree to which the prediction results of deep learning network models violate local PDEs; This is the boundary residual term, used to measure the degree to which the prediction results of a deep learning network model violate boundary conditions; This is the power conservation residual term, used to ensure that the solution learned by the deep learning network model satisfies macroscopic physical equilibrium at any time. w data , w PDE , w bou , w cons Weighting coefficients to weigh the importance of each factor.

8. The method according to claim 7, characterized in that, The calculation process for the power conservation residual difference term includes: Based on the operating power loss of the prefabricated substation, the total heat capacity of the air and solids inside the experimental chamber, and the total heat exchange between the experimental chamber walls and the environment, a global energy balance equation is constructed. The expression of this global energy balance equation is as follows: In the formula, This refers to the operating power loss of prefabricated substations. The total heat capacity of the air and solids inside the experimental chamber. The average temperature inside the experimental chamber. The total heat exchange between the experimental chamber walls and the environment is expressed as follows: Q wall = U wall · A surf ·( T avg - T 0) In the formula, U wall The equivalent heat transfer coefficient of the bulkhead. A surf To achieve effective heat dissipation surface area, T 0 represents the ambient temperature; In each training batch of the temperature field reconstruction surrogate model, the current average temperature inside the experimental cabin is calculated by integrating the average temperature predicted by the temperature field reconstruction surrogate model. The power conservation residual difference term is calculated, and its expression is as follows: In the formula, The number of sample points in time series t. For the sample Operating loss power of prefabricated substations The error between the calculated value and the actual value.

9. The method according to claim 6, characterized in that, The rolling optimization process includes: At each control sampling moment, real-time temperature data inside the experimental chamber is received and input into the temperature field reconstruction proxy model to reconstruct the current complete temperature field distribution inside the chamber. Using this as the initial state, future temperature field distributions are predicted. N p Within a time step, under the assumed future control input sequence { u ( k | k ), u ( k +1| k ), …, u ( k + N p-1 | k Under these conditions, the average temperature inside the experimental chamber... T avg The evolutionary trajectory; Construct a rolling optimization objective function, the expression of which is: In the formula, To optimize the objective function value during rolling, , Used to punish the future N p The predicted average cabin temperature within the step T avg ( k + i | k ) and target set temperature T set The deviation between them, where Q is a positive semi-definite weight matrix. Used to suppress control quantity u This refers to the start / stop status of the cooling fan or drastic changes in its speed; R The weight matrix is ​​positive definite. By adjusting the weight matrix... Q and R Balancing temperature control accuracy with the smoothness of actuator operation; Physical constraints are imposed on configuring the cooling fan for the rolling optimization objective function; At each sampling time k Solve the quadratic programming problem of the rolling optimization objective function with physical constraints of a cooling fan, and obtain the optimal control sequence { u ( k | k ), u ( k +1| k ), …, u ( k + N p-1 | k )}, take the first element of the sequence u ( k | k The current control command is sent to the cooling fan for execution.

10. The method according to claim 6, characterized in that, The experimental chamber is also equipped with auxiliary heating equipment, which is used to provide supplementary heat when the prefabricated substation (5) is insufficient in its own heat generation; the start and stop of the auxiliary heating equipment is controlled according to the surplus power of the self-generated heat, which is the difference between the heat generated by the equipment and the heat dissipation required to maintain the target temperature. The formula for calculating the self-generated heat surplus power is as follows: In the formula, To generate surplus power from self-generated heat, P dev For the operating power loss of prefabricated substation (5), U wall The equivalent heat transfer coefficient of the experimental chamber wall is given. A surf To achieve effective heat dissipation surface area, T set For the target temperature, T 0 represents the ambient temperature; The start-stop control process of the auxiliary heating equipment includes: When Δ P When the value is >0, the auxiliary heating equipment is not activated; When Δ P When the value is ≤0, start the auxiliary heating equipment.

Images