A primary and secondary fusion set ring network box cooling device dynamic optimization method

Abstract

This invention provides a dynamic optimization method for a primary and secondary integrated ring network box cooling device, belonging to the field of ring network box cooling optimization technology based on deep learning. It constructs a dynamic monitoring dataset including a ring network box thermal field monitoring dataset and a thermal coefficient monitoring dataset. The thermal field dataset is constructed through five-view infrared image stitching and high-density thermocouple calibration, while the thermal coefficient dataset is constructed by combining vibration, stress, and other mechanical data with manual calibration. A thermal coefficient sensing model is constructed, using three-branch feature extraction and adaptive weighted fusion to evaluate the heat generation and heat dissipation coefficients of the ring network box in real time based on force field data. The thermal field sensing model employs an extractor-reconstructor architecture and a cross-layer feature fusion unit to achieve full-scene temperature sensing. A cooling optimization control algorithm is designed, combining thermal field distribution and thermal coefficients, and adaptively adjusting seven cooling device parameters, such as fan speed and liquid cooling flow rate, through multi-objective optimization to achieve refined cooling control.

Description

Technical Field

[0001] This invention belongs to the field of ring network box cooling optimization technology based on deep learning, and particularly relates to a dynamic optimization method for a primary and secondary integrated ring network box cooling device. Background Technology

[0002] As a core component of the power distribution network, the integrated primary and secondary ring main unit (RMU) faces thermal anomalies (such as contact overheating and insulation aging) during long-term operation, directly impacting the safe and stable operation of the network. Traditional monitoring methods rely on single-point temperature measurement and manual inspection, which are single-modal and low-frequency, resulting in limitations such as delayed state perception, ambiguous fault location, and low maintenance efficiency, making it difficult to meet the intelligent equipment requirements of modern power systems. With the development of intelligent sensing technology and edge computing, the research and development of a multimodal thermal and force field prediction and dynamic monitoring system has become a key path to overcome these bottlenecks. This system integrates multiple sensors, including infrared thermal imaging, vibration sensing, and stress monitoring, to construct a coupled thermo-mechanical-electrical multiphysics model, enabling full-dimensional perception and in-depth analysis of the overall thermal field of the RMU, thus solving the problem of insufficient ability of traditional methods to capture complex temperature field characteristics. Based on the automated sensing of the thermal field of the ring main unit, the system further optimizes the control of the cooling equipment of the ring main unit, which helps to upgrade the distribution network to intelligence and provides key technical support for building a safe, reliable, efficient and low-carbon new power system. It has important engineering significance and industry promotion value for ensuring the stability of urban power supply and improving energy utilization efficiency.

[0003] Currently, the methods for predicting the multimodal thermal and force fields of integrated primary and secondary ring network enclosures mainly focus on the following approaches:

[0004] (1) Traditional physical models and single-mode analysis methods: This type of method is based on fundamental theories such as heat transfer and mechanical vibration. It collects single parameters such as temperature and vibration amplitude through single-point sensors and combines them with preset physical formulas and threshold logic to evaluate the state. Its core is to construct an analytical model through manually designed features. Its advantages are that the physical meaning is clear and the computational efficiency is high, making it suitable for preliminary monitoring under steady-state conditions. However, when faced with complex scenarios such as multi-field coupling and nonlinear effects, single-mode data is difficult to fully characterize the equipment state and has insufficient ability to identify thermal and force fields in complex scenarios.

[0005] (2) Multimodal data-driven fusion method: This type of method uses multi-source sensor data and employs statistical analysis or shallow machine learning models to perform feature fusion and pattern recognition. Its core is to convert multimodal data into a unified feature space and combine it with model analysis to achieve prediction of thermal and force fields. However, this method relies too much on manual feature engineering, has limited ability to model deep nonlinear laws, and the generalization ability of the model is limited by data quality and feature selection strategy;

[0006] (3) Transfer learning and cross-task collaborative methods: This type of method transfers the thermal field prediction model of power equipment such as transformers and switchgear to the ring network box state prediction task. By fine-tuning the key layer parameters to adapt to the thermal-mechanical feature distribution of specific equipment, it effectively reduces the annotation cost. However, when this type of method achieves cross-equipment transfer, it needs to solve the domain offset problem caused by differences in sensor layout and different operating conditions. In addition, the weight allocation of the multi-task loss function needs to be finely adjusted in combination with the physical characteristics of the equipment. The complexity and requirements of model transplantation are too high. Summary of the Invention

[0007] To address the above problems, this invention proposes a dynamic optimization method for a primary and secondary integrated ring network box cooling device, comprising the following processes:

[0008] S1, dynamically acquire infrared sensing images and mechanical characteristic data of the ring network box. The infrared sensing images of the ring network box are constructed by stitching together five-view infrared images to form the thermal field monitoring data of the ring network box as the standard input of the model. The mechanical characteristic data of the ring network box are combined to obtain the thermal coefficient monitoring data of the standard input of the model.

[0009] S2, input the ring network box thermal field monitoring data into the constructed and trained ring network box thermal field sensing model, output the thermal field sensing image, and calculate the temperature mean and temperature variance; input the thermal coefficient monitoring data into the constructed and trained ring network box thermal coefficient sensing model, and output the heat generation coefficient and heat dissipation coefficient.

[0010] S3 aims to maximize the cooling efficiency of the ring network box, minimize the cooling consumption of the ring network box, and maximize the service life of the heat dissipation components. The working parameters of the cooling equipment in the cooling control process of the ring network box are used as working variables. A multi-objective optimization function is constructed by combining the mean temperature, temperature variance, heat generation coefficient and heat dissipation coefficient.

[0011] S4 solves the multi-objective optimization function and outputs refined cooling control parameters.

[0012] Preferably, the dataset used to train the thermal field sensing model of the ring network box is constructed in the following ways:

[0013] Infrared images were acquired from the front, back, left, right, and top surfaces of the cubic ring network box. The five infrared images from different perspectives were then stitched together to form a complete infrared sensing image of the ring network box. ;

[0014] A high-density thermocouple array was uniformly arranged on the surface of the ring network box, and the thermocouples were fixed to the surface of the ring network box with high-temperature resistant tape. Then, the correspondence between the physical coordinates of the thermocouples and the pixel coordinates of the infrared image was established using a checkerboard calibration plate, and the actual temperature values ​​collected by the thermocouples were used as the true value labels for their corresponding pixels. Subsequently, the local weighted regression difference method was used to supplement the temperature labels for pixels that lacked true temperature labels, and finally, the calibrated thermal field sensing image of the ring network box was obtained. This data is then used as the output data for predicting the thermal field of the ring network box.

[0015] Preferably, the dataset used to train the thermal coefficient sensing model of the ring network box is constructed in the following ways:

[0016] Accelerometers are deployed at the four corners of the ring main unit's switch operating mechanism linkage, busbar support insulator base, cable termination, gas-insulated wall shell, and cabinet base to collect corresponding vibration frequency data. and vibration amplitude data Deploy grating strain sensors to collect stress distribution Data is collected by deploying tensile and compressive sensors to gather corresponding tensile force data. and stress data Deploy resistance strain gauges to collect stress and deformation data The mechanical characteristic data of the ring network box collected above are combined to obtain a set of input data in the ring network box heat generation coefficient prediction dataset. ;

[0017] Data is input for each group of heat production coefficient predictions through manual judgment. Add heat production coefficient label and heat dissipation coefficient label The two coefficient data are then combined as the output data in the ring network box heat generation coefficient prediction dataset. .

[0018] Preferably, the ring network box thermal field sensing model includes an infrared image feature extractor, a thermal field sensing image reconstructor, and a cross-layer feature fusion unit;

[0019] The overall feature extraction path of the ring network box thermal field sensing model consists of seven infrared image feature extractors stacked sequentially and connected end to end. The thermal field image reconstruction path of the thermal field sensing model consists of six thermal field sensing image reconstructors stacked sequentially and connected end to end. The outputs of the second and third extractors are sent to the first cross-layer feature fusion unit for processing, and then the output of the feature fusion unit is used as a supplementary input to the second and third reconstructors. At the same time, the outputs of the fourth and fifth extractors are sent to the second cross-layer feature fusion unit for processing, and then the output of the second feature fusion unit is used as a supplementary input to the fourth and fifth reconstructors. The outputs of the sixth and seventh extractors are sent to the third cross-layer feature fusion unit for processing, and then the output of the third feature fusion unit is used as a supplementary input to the sixth and seventh reconstructors.

[0020] Preferably, the infrared image feature extractor has the following specific structure:

[0021] First, two 3x3 convolutional layers are used to perform preliminary feature extraction on the image, followed by a ReLU activation function to achieve non-linear activation of the features. The resulting features are then fed into a global feature perception pathway to extract global thermal field features of the infrared image, and into a detail feature perception pathway to extract local thermal field variation features. In the global feature perception pathway, three sequentially connected 5x5 convolutional layers are used, followed by a normalization layer for feature standardization, and then an average pooling layer for average pooling. In the detail feature perception pathway, four 3x3 convolutional layers are used, followed by a normalization layer for feature standardization, and then a max pooling layer for max pooling. A fully connected layer then fuses the global and local thermal field features, followed by a final feature extraction operation using a 3x3 convolutional layer, and finally, simgoid activation is used for feature activation.

[0022] Preferably, the thermal field-sensing image reconstructor has the following specific structure:

[0023] First, feature reconstruction upsampling is performed using two deconvolutional layers with a kernel size of 3*3, constructing two feature reconstruction pathways with different kernel sizes. The first feature reconstruction pathway includes three sequentially connected deconvolutional layers with a kernel size of 5*5, followed by a BN normalization layer and a Dropout layer to enhance the image reconstruction robustness of the thermal field-aware image reconstructor. The second feature reconstruction pathway includes three sequentially connected deconvolutional layers with a kernel size of 3*3, followed by a BN normalization layer and an unpooling layer. Finally, a fully connected layer and a Mish activation function are used to perform final feature fusion processing on the features of the two feature reconstruction pathways.

[0024] Preferably, the cross-layer feature fusion processor first performs feature extraction processing through two sequentially connected 3*3 convolutional layers, followed by non-linear activation of the features using a ReLU activation function. Then, it performs feature upsampling processing through two interconnected 3*3 deconvolutional layers. Finally, it uses a max-pooling layer and a ReLU activation function to process the features again to obtain the fused ring network box thermal field sensing image. .

[0025] Preferably, the ring network box thermal coefficient sensing model includes a ring network box multi-layer feature sensing path, a ring network box multi-layer feature fusion unit, and a ring network box thermal coefficient generation head;

[0026] The multi-layer feature sensing path of the ring network box first transmits the data. The data is fed into three different feature extraction pathways to complete the feature extraction operation. The network layer architecture of the first feature extraction pathway consists of three sequentially connected fully connected layers, with a ReLU activation function added after each fully connected layer. Features are obtained after processing by the first feature extraction pathway. The network architecture for the second feature extraction pathway remains three sequentially connected fully connected layers, but a new layer is added after each fully connected layer. Activation function, data Features are obtained after processing by the second feature extraction pathway. The network architecture of the third feature extraction pathway consists of three sequentially connected fully connected layers, with an additional layer added after each fully connected layer. Activation function, data Features are obtained after processing through the third feature extraction pathway. ;

[0027] The ring network box multi-layer feature fusion unit is used to fuse the features extracted from the multi-layer feature perception path of the ring network box and adaptively assign weights to different features.

[0028] The ring network box thermal coefficient generator uses two consecutive fully connected layers to fuse features. Further feature extraction processing is performed, followed by... The activation function performs non-linear activation of the features, and finally... The activation function is used to obtain the predicted heat generation coefficient of the ring network box. and the predicted results of heat dissipation coefficient .

[0029] Preferably, seven key control parameters in the ring network box cooling system were selected as working variables, including the number of working fans fg, fan speed fz, air outlet opening sj, coolant flow rate yn of the liquid cooling system, circulating pump working power wp, temperature difference between the start and stop threshold of the heat dissipation device wc, and number of times the circulating pump starts and stops xc.

[0030] The multi-objective optimization function is shown below:

[0031]

[0032] in The objective function for cooling efficiency is to minimize this objective function in order to reduce the secondary heat introduced by the cooling device of the ring network box during operation; The heat generation coefficient of the ring network box during operation. This refers to the heat dissipation coefficient of the ring main unit during operation. and All were calculated using the heat generation coefficient sensing model of the ring network box. This shows the overall temperature of the ring main unit at this time; This represents the function for calculating the heat dissipation efficiency of the force field. The function representing the thermal field factor for calculating heat dissipation efficiency;

[0033] The objective function for cooling consumption is to minimize this objective function in order to reduce the resources and energy consumed in the actual cooling process; Indicates the consumption cost coefficient;

[0034] The objective function is the service life of the component.

[0035] Preferably, the optimization range is limited for the seven working variables, specifically including:

[0036] That is, the minimum value of the number of fans operating, fg, is The maximum value is ; That is, the minimum fan speed is 10 ... The maximum rotational speed is [value missing]. change; That is, the minimum angle at which the air outlet opens is degrees, the maximum opening angle is That is, the minimum flow rate of the coolant in the liquid cooling system is m / s, maximum flow velocity is m / s; That is, the minimum operating power of the circulating pump is Watts, maximum operating power is watt. That is, the minimum temperature difference between the start and stop threshold of the heat dissipation device is Celsius, maximum is Celsius; That is, the minimum number of start-stop cycles for the circulating pump is The maximum number of start-stop cycles is [number]. Second-rate.

[0037] Compared with the prior art, the beneficial effects brought about by the innovation of this invention include:

[0038] (1) Implementation of the adaptive ring network box cooling control system: This invention achieves accurate monitoring of the heat generation of the ring network box from the perspectives of both the force field and the thermal field by constructing a specific dynamic monitoring dataset and a ring network box thermal coefficient sensing model and a ring network box thermal field sensing model. Furthermore, the working parameters of the cooling device are optimized in real time through a designed cooling device optimization algorithm. This realizes a closed-loop system of high-precision thermal field sensing, thermal coefficient prediction, and cooling strategy optimization, significantly improving the automation and intelligence levels of the ring network box heat dissipation control.

[0039] (2) Design of a thermal field perception model with cross-layer feature fusion: This invention designs a six-layer stacked extractor-reconstructor architecture to realize the overall framework design of the thermal field perception model. A dual-channel feature extraction path and a cross-layer feature fusion device are designed in it to take into account the image reconstruction capabilities of both large-scale and small-scale thermal areas, further improving the robustness and generalization ability of thermal field image reconstruction, and ensuring that good thermal field image perception capabilities can be maintained under different ring network box working scenarios;

[0040] (3) Design of multi-objective optimization adaptive cooling control algorithm: This invention designs a multi-objective optimization function and comprehensively considers the actual thermal field conditions of the ring network box. Combined with ε-Pareto front screening and crossover mutation operation, it realizes global optimization of multiple working variables. This algorithm ensures that optimization effects can be achieved in terms of heat dissipation efficiency, heat dissipation energy consumption and cooling device lifespan, improves the optimality of cooling control parameters, and further ensures the comprehensive cooling capacity of the ring network box cooling process. Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating the overall technical route of the present invention.

[0042] Figure 2 This is a diagram of the thermal field sensing model architecture of the ring network box of the present invention.

[0043] Figure 3 This is a schematic diagram of the infrared image feature extractor architecture of the present invention.

[0044] Figure 4 This is a diagram of the thermal field-sensing image reconstructor architecture of the present invention.

[0045] Figure 5 This is an example image of a ring network cage thermal field sensing image.

[0046] Figure 6 Example chart comparing the prediction accuracy of fan speed control parameters.

[0047] Figure 7Example chart comparing the prediction accuracy of air outlet opening control parameters. Detailed Implementation

[0048] This invention proposes a dynamic optimization method for a primary and secondary integrated ring network box cooling device, such as... Figure 1 Specifically, this includes: constructing a dynamic monitoring dataset containing a ring network cage thermal field monitoring dataset and a thermal coefficient monitoring dataset. The thermal field dataset is constructed through five-view infrared image stitching and high-density thermocouple calibration, while the thermal coefficient dataset is constructed by combining vibration, stress, and other mechanical data with manual calibration; constructing a thermal coefficient sensing model, which uses three-branch feature extraction and adaptive weighted fusion to evaluate the heat generation and heat dissipation coefficients of the ring network cage in real time based on force field data; constructing a thermal field sensing model, which adopts an extractor-reconstructor architecture and a cross-layer feature fusion unit to achieve full-scene temperature sensing; and designing a cooling optimization control algorithm, which combines thermal field distribution and thermal coefficients to adaptively adjust seven cooling device parameters, such as fan speed and liquid cooling flow rate, through multi-objective optimization (cooling efficiency, energy consumption, and component loss), thereby achieving refined cooling control.

[0049] The invention will be further described below with reference to specific embodiments.

[0050] I. Construction of Dynamic Monitoring Dataset for Ring Network Boxes

[0051] This invention aims to achieve dynamic monitoring of the thermal field of a ring main unit from a thermal perspective. Furthermore, the ring main unit exhibits different heat generation coefficients under different mechanical states, which affect its overall heat dissipation performance. Since there is a relatively small data coupling between mechanical and thermal data, to improve the subsequent optimization of the ring main unit's heat dissipation device parameters, a ring main unit thermal field prediction dataset and a ring main unit heat generation coefficient prediction dataset were specifically constructed during the dynamic monitoring dataset construction phase.

[0052] 1. Method for constructing a thermal field prediction dataset for ring network boxes

[0053] To improve the spatial resolution and efficiency of subsequent thermal field prediction, this invention uses infrared sensing images of the ring network cage as input data for the ring network cage thermal field prediction dataset. Selecting infrared sensing images of the ring network cage as input data reduces the complexity of subsequent data calibration and improves the comprehensiveness of the ring network cage's temperature sensing.

[0054] It should be noted that the present invention will perform further image calibration processing on the infrared sensing images of the ring network box, transforming the infrared images of the ring network box into actual thermal field sensing images of the ring network box, and using the transformed thermal field sensing images of the ring network box as the output data of the thermal field prediction dataset of the ring network box.

[0055] (1) Infrared image sensing of the ring network box: When collecting infrared images of the ring network box, it was difficult to collect effective infrared sensing images due to the fixed installation and ground obstruction on the bottom surface of the ring network box. Therefore, infrared images were collected from the front, back, left, right and top surfaces of the cubic ring network box, and the infrared images from the five different perspectives were stitched together to form an overall infrared sensing image of the ring network box. ;

[0056] (2) Ring-type enclosure infrared image calibration: As the ring-type enclosure is an electrified device, to ensure the accuracy of temperature data acquisition and the safety of the acquisition process, this invention carried out the ring-type enclosure infrared image calibration operation under laboratory conditions: First, a high-density thermocouple array was uniformly arranged on the surface of the ring-type enclosure. To ensure the accuracy of temperature sensing, the distance between any two thermocouples was less than 1 cm, and the thermocouples were fixed to the surface of the ring-type enclosure with high-temperature resistant tape. Then, the correspondence between the physical coordinates of the thermocouples and the pixel coordinates of the infrared image was established using a checkerboard calibration board, and the actual temperature value acquired by the thermocouples was used as the true value label of its corresponding pixel. Then, the local weighted regression difference method was used to supplement the temperature labels of the pixels that lacked true temperature labels, and finally the calibrated ring-type enclosure thermal field sensing image was obtained. This data is then used as the output data for predicting the thermal field of the ring network box.

[0057] 2. Method for constructing a dataset for predicting the heat generation coefficient of a ring network box

[0058] During actual operation of a ring main unit, the vibration state of its internal structure, stress and strain distribution, and changes in the tension and resistance at the contact interfaces all lead to changes in contact resistance and thermal conductivity, thereby altering the heat dissipation and heat generation capabilities of the ring main unit. This invention defines a heat dissipation coefficient. A heat generation coefficient was defined to evaluate the heat dissipation capacity of the ring main unit. To evaluate the heat generation capacity of the ring network cage. Due to the heat generation coefficient... and heat dissipation coefficient Since the data cannot be directly measured by equipment, this invention utilizes in-depth analysis of the collected mechanical data of the ring main unit by professional maintenance personnel to obtain the corresponding heat generation coefficient. and heat dissipation coefficient .

[0059] (1) In this embodiment, acceleration sensors are deployed on the switch operating mechanism linkage, busbar support insulator base, cable terminal head, gas-insulated wall shell, and four corners of the cabinet base of the ring main unit to collect corresponding vibration frequency data. and vibration amplitude data Deploy grating strain sensors to collect stress distribution Data is collected by deploying tensile and compressive sensors to gather corresponding tensile force data. and stress data Deploy resistance strain gauges to collect stress and deformation data The mechanical characteristic data of the ring network cage collected above are combined to obtain a set of input data for the ring network cage heat generation coefficient prediction dataset. ;

[0060] (2) Regarding the mechanical data acquisition method of the ring network box described above, in order to ensure the accuracy of the subsequent assessment of the heat generation coefficient of the ring network box, mechanical operating status data of the ring network box is collected under three different states: overload operation, normal operation, and underload operation to ensure the integrity of the data acquisition.

[0061] (3) Based on the data acquisition environment described above, actual data acquisition is carried out, and the data for each group of heat production coefficient prediction is input by professional personnel through manual judgment. Add a heat production coefficient label and heat dissipation coefficient label The two coefficient data are then combined as the output data in the ring network box heat generation coefficient prediction dataset. .

[0062] Based on the described method for constructing a thermal field prediction dataset for ring network cages, this embodiment collects thermal field prediction data for ring network cages under different outdoor temperature conditions, and a total of [data missing] data were collected. Predicted thermal field data for ring network cages, and the collected data The thermal field prediction data of the ring network cage are combined to complete the thermal field prediction dataset of the ring network cage. The construction.

[0063] Based on the described method for constructing a dataset for predicting the heat generation coefficient of a ring network enclosure, this invention has collected data under different operating conditions of the ring network enclosure. Predicted heat generation coefficient data for ring network cages, and the collected data The heat generation coefficient prediction data of the ring network cage are combined to complete the heat generation coefficient prediction dataset of the ring network cage. The construction of.

[0064] The above-constructed ring network box thermal field prediction dataset And the constructed ring network box heat generation coefficient prediction dataset Combined to complete the dynamic monitoring dataset of the ring network box. The construction.

[0065] II. Construction of the Thermal Field Sensing Model for Ring Network Box

[0066] To achieve high-standard generation of thermal field sensing images for ring network cages, this invention designs an infrared image feature extractor-thermal field sensing image reconstructor architecture within the ring network cage thermal field sensing model. The overall architecture diagram of the ring network cage thermal field sensing model is shown below. Figure 2 As shown.

[0067] 1. Infrared Image Feature Extractor Structure Design

[0068] Infrared image feature extractor architecture such as Figure 3 As shown, the infrared image feature extractor mainly extracts the overall infrared sensing image of the ring network box. A stepwise feature extraction process is performed to provide sufficient features for subsequent thermal field sensing image reconstruction. A dual-channel feature extraction pathway is introduced in this feature extractor to further enhance the feature extraction capability of infrared images.

[0069] The specific architecture of the infrared image feature extractor includes: First, two 3x3 convolutional layers are used to perform preliminary feature extraction on the image, and a ReLU activation function is added after the two convolutional layers to achieve non-linear activation of the features. The obtained features are then fed into a global feature perception sub-channel to extract the global thermal field features of the infrared image, and into a detail feature perception sub-channel to extract the local thermal field variation features of the infrared image. In the global feature perception sub-channel, three sequentially connected 5x5 convolutional layers are used, followed by a normalization layer for feature standardization, and then an average pooling layer for average pooling. In the detail feature perception sub-channel, four 3x3 convolutional layers are used, followed by a normalization layer for feature standardization, and then a max pooling layer for max pooling. Then, a fully connected layer is used to perform feature fusion processing on global and local thermal field features, and a 3*3 convolutional layer is used for final feature extraction. Finally, a simgoid activation is used to activate the features, thus completing the construction of the infrared image feature extractor.

[0070] 2. Structural Design of Thermal Field Sensing Image Reconstructor

[0071] Thermal field-aware image reconstructor architecture such as Figure 4As shown, in the construction of the ring network box thermal field sensing model, a thermal field sensing image reconstructor is needed to generate thermal field sensing images. The architecture of the thermal field sensing image reconstructor is as follows: First, feature reconstruction upsampling is performed using two deconvolutional layers with a kernel size of 3*3. Two feature reconstruction paths with different kernel sizes are constructed. Feature reconstruction path one includes three sequentially connected deconvolutional layers with a kernel size of 5*5, followed by BN normalization layers and Dropout layers to enhance the image reconstruction robustness of the thermal field sensing image reconstructor. Feature reconstruction path two includes three sequentially connected deconvolutional layers with a kernel size of 3*3, followed by BN normalization layers and depooling layers to complete the design of feature reconstruction path two. Finally, fully connected layers and the Mish activation function are used to perform final feature fusion processing on the features of the two feature reconstruction paths to complete the design of the thermal field sensing image reconstructor.

[0072] 3. Cross-layer feature fusion device structure design

[0073] To further ensure the reconstruction accuracy of the thermal field reconstructed image, this invention constructs a cross-layer feature fusion processor to enhance the feature fusion processing capability for cross-layer features. The structure of the cross-layer feature fusion processor is as follows: To ensure the consistency of feature levels, the cross-layer feature fusion processor first performs feature extraction processing through two sequentially connected 3*3 convolutional layers, followed by a ReLU activation function to achieve non-linear activation of the features. Then, it performs feature upsampling processing through two interconnected 3*3 deconvolutional layers. Finally, it uses a max pooling layer and a ReLU activation function for further processing to complete the overall design of the cross-layer feature fusion processor.

[0074] Overall Architecture Design of the Ring Network Box Thermal Field Sensing Model: The overall feature extraction path of the ring network box thermal field sensing model consists of seven sequentially stacked infrared image feature extractors connected one after another. The thermal field image reconstruction path of the thermal field sensing model consists of six sequentially stacked thermal field sensing image reconstructors connected one after another. To further establish a direct feature mapping relationship between the extractors and reconstructors, ensuring the image resolution and accuracy of the reconstructed thermal field image, this invention sends the outputs of extractors two and three to a cross-layer feature fusion unit for processing, and then uses the output of the feature fusion unit as supplementary input to reconstructors two and three. Correspondingly: the outputs of extractors four and five are sent to a cross-layer feature fusion unit two for processing, and then use the output of feature fusion unit two as supplementary input to reconstructors four and five. The outputs of extractors six and seven are sent to a cross-layer feature fusion unit three for processing, and then use the output of feature fusion unit three as supplementary input to reconstructors six and seven.

[0075] An infrared sensing image of the entire ring network cage After processing the data from the constructed ring network box thermal field sensing model, the thermal field sensing image of the ring network box calculated by the model can be obtained. .

[0076] III. Construction of the Thermal Coefficient Sensing Model for Ring Coil

[0077] Data processing layer design: For the heat generation coefficient prediction dataset of ring network boxes Input data for predicting heat production coefficient , The data in the ring network box exhibits differences in units and value ranges, increasing the difficulty of predicting the heat generation coefficient. To reduce the difficulty of predicting the heat generation coefficient of the ring network box and increase the prediction accuracy, a BN normalization layer and a data normalization layer were first designed in the ring network box heat generation coefficient perception model to process the input data. Data augmentation is performed to obtain augmented data. .

[0078] Design of multi-layer feature sensing path for ring network cages: enhancing data The data is fed into three different feature extraction pathways to complete the feature extraction operation. The network architecture of feature extraction pathway one consists of three sequentially connected fully connected layers, with a ReLU activation function added after each fully connected layer. Features are obtained after processing by the first feature extraction pathway. The network architecture of feature extraction path two still consists of three sequentially connected fully connected layers, but a new layer is added after each fully connected layer. Activation function, data Features are obtained after processing by the second feature extraction pathway. The network architecture of feature extraction pathway three consists of three sequentially connected fully connected layers, with an additional layer added after each fully connected layer. Activation function, data Features are obtained after processing through the third feature extraction pathway. .

[0079] Design of multi-layer feature fusion unit for ring network box: based on the obtained features , , A multi-layer feature fusion unit for ring network cages was designed to obtain fused sensing features. ;

[0080]

[0081] in, This represents the normalization operation function. Adaptively assign weights to three different features. This is represented as the feature scaling factor.

[0082] Design of the thermal coefficient generator for ring network cages: targeting the obtained fused sensing features , will feature The data is fed into the ring network box coefficient generator for further processing to obtain the predicted heat generation coefficient of the ring network box. and the predicted results of heat dissipation coefficient The overall architecture of the ring network box heat generation coefficient generator is as follows: It first fuses features through two consecutive fully connected layers. Further feature extraction processing is performed, followed by... The activation function performs further nonlinear activation of the features, and finally... The design process of the ring network box thermal coefficient generator head is completed after the activation function is applied.

[0083] IV. Implementation of Optimized Algorithm for Cooling Control of Ring Main Unit

[0084] To achieve high-efficiency optimization of the cooling devices in the ring network box, directly feeding the acquired thermal field sensing image into the optimization algorithm would result in excessively high computational complexity. Therefore, this invention utilizes a thermal field image data processing algorithm to obtain the thermal field sensing image of the ring network box. average temperature and temperature variance And using this average temperature and temperature variance As a reference condition parameter for subsequent optimization algorithms;

[0085] This invention aims to maximize the cooling efficiency of the ring main unit, minimize its cooling consumption, and maximize the service life of its heat dissipation components as the model optimization objectives. It establishes a multi-objective optimization problem for the control parameters of the ring main unit's cooling equipment, using the operating parameters of the cooling equipment during the cooling control process as the working variables. The specific process is as follows:

[0086] Determination of working variables: In order to optimize the key control parameters of the ring network box adaptive cooling process, this invention selects seven key control parameters in the ring network box cooling system as working variables. Specifically, these include: number of fans operating (fg), fan speed (fz), air outlet opening (sj), coolant flow rate (yn) of the liquid cooling system, circulating pump operating power (wp), temperature difference between the start and stop thresholds of the heat dissipation device (wc), and number of times the circulating pump starts and stops (xc).

[0087] Multi-objective optimization function design: The multi-objective optimization function designed in this invention is shown below:

[0088]

[0089] in The objective function for cooling efficiency is to minimize the secondary heat introduced by the cooling device during the operation of the ring network box. The control parameters of the cooling device are dynamically adjusted according to the current thermal field state of the ring network box, thereby improving the cooling efficiency of the ring network box during actual cooling. The heat generation coefficient of the ring network box during operation. This refers to the heat dissipation coefficient of the ring main unit during operation. and All of these can be calculated using the heat generation coefficient sensing model of the ring network box. This shows the overall temperature of the ring network box at this time. This represents the function for calculating the heat dissipation efficiency of the force field. This represents the function for calculating the heat dissipation efficiency of the thermal field factor.

[0090] Let cooling consumption be the objective function. Minimizing this objective function helps reduce the resources and energy consumed in the actual cooling process, thereby improving the economic efficiency of the cooling operation. This represents the consumption cost coefficient. Let the objective function be the component lifespan. Minimizing this objective function helps reduce the component wear of the cooling devices during the actual cooling process of the ring network box, thereby further extending the lifespan of the ring network box cooling device and ensuring the robustness and stability of the cooling operation.

[0091] Limiting the range of working variables: To ensure sufficient optimization space and efficiency for the proposed working parameters, and considering the actual ring network box heat dissipation system, the optimization range of the seven proposed working variables is limited. Specifically: That is, the minimum value of the number of fans operating, fg, is The maximum value is ; That is, the minimum fan speed is 10 ... The maximum rotational speed is [value missing]. change; That is, the minimum angle at which the air outlet opens is degrees, the maximum opening angle is That is, the minimum flow rate of the coolant in the liquid cooling system is m / s, maximum flow velocity is m / s; That is, the minimum operating power of the circulating pump is Watts, maximum operating power is watt. That is, the minimum temperature difference between the start and stop threshold of the heat dissipation device is Celsius, maximum is Celsius. That is, the minimum number of start-stop cycles for the circulating pump is The maximum number of start-stop cycles is [number]. Second-rate.

[0092] Ring network cage working variable population initialization: Based on the value range of each working variable described in the specific description, Bayesian sampling method is used to initialize the population. Sampling within its value range This time Each set of working variables corresponds to a population of working variables for a ring network box. The population is obtained during the population initialization phase. , .

[0093] Working variable population mutation iteration:

[0094] (1) Combined with the heat production coefficient obtained from monitoring and heat dissipation coefficient and monitoring results Using the three defined objective functions Calculate the fitness of each variable population in the ring network cage. For the initial calculation, the selected population is the initial population. ;

[0095] (2) Based on the fitness of each working variable population calculated in process (1) above, sort all populations according to their fitness and select the populations located at the first ε-Pareto front. And discard the remaining population;

[0096] (3) For the population Perform crossover and mutation operations to obtain the mutated population. ;

[0097] (4) The mutated population During the process of sending (1), the population fitness is calculated as a new initial population;

[0098] (5) Iterate through process (1) to process (4) k times to allow the population to undergo sufficient mutation and competition, and then... The population obtained in the next iteration The optimal Pareto solution set for the cooling control parameter optimization problem of the ring cage is selected, and the working variable of the population individual with the smallest fitness variance is selected as the final cooling control parameter used in the actual ring cage.

[0099] V. Explanation of Experimental Results

[0100] To clarify the image reconstruction effect of the ring cage thermal field sensing model of this invention, the extracted reconstructed image of the ring cage thermal field is shown below. Figure 5 As shown.

[0101] To verify the superior control performance of the cooling device optimization control algorithm proposed in this invention, performance comparison tests were conducted under four different operating conditions: ring network box under underload, overload, normal operation, and minor fault. The comparison algorithms selected were fuzzy control and a BP neural network-based control algorithm. First, the optimal control parameters for the cooling device under the aforementioned operating conditions were determined by professionals. And record the control parameters predicted by each algorithm. Then, the parameter prediction accuracy is calculated. , This indicates the absolute value operation.

[0102] The algorithm was compared and verified using the air outlet opening control parameters and fan control parameters as examples. Figure 6 and Figure 7 As shown in the figure, the verification results demonstrate that the algorithm proposed in this invention exhibits the highest accuracy in predicting the optimal control parameters under four different ring network box operating modes. This is because the cooling device optimization control algorithm proposed in this invention can fully explore the optimization space of the control parameters and select the optimal cooling device control parameters based on the full exploration of the optimization space, thereby ensuring the optimality of the control parameters.

[0103] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0104] While the specific embodiments of the present invention have been described above, they are not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A dynamic optimization method for a primary and secondary integrated ring network box cooling device, characterized in that, Includes the following processes: S1, dynamically acquire infrared sensing images and mechanical characteristic data of the ring network box. The infrared sensing images of the ring network box are constructed by stitching together five-view infrared images to form the thermal field monitoring data of the ring network box as the standard input of the model. The mechanical characteristic data of the ring network box are combined to obtain the thermal coefficient monitoring data of the standard input of the model. S2, input the ring network box thermal field monitoring data into the constructed and trained ring network box thermal field sensing model, output the thermal field sensing image, and calculate the temperature mean and temperature variance; input the thermal coefficient monitoring data into the constructed and trained ring network box thermal coefficient sensing model, and output the heat generation coefficient and heat dissipation coefficient. S3 aims to maximize the cooling efficiency of the ring network box, minimize the cooling consumption of the ring network box, and maximize the service life of the heat dissipation components. The working parameters of the cooling equipment in the cooling control process of the ring network box are used as working variables. A multi-objective optimization function is constructed by combining the mean temperature, temperature variance, heat generation coefficient and heat dissipation coefficient. S4 solves the multi-objective optimization function and outputs refined cooling control parameters.

2. The dynamic optimization method for a primary and secondary integrated ring network box cooling device as described in claim 1, characterized in that: The datasets used to train the thermal field sensing model of the ring cage can be constructed in the following ways: Infrared images were acquired from the front, back, left, right, and top surfaces of the cubic ring network box. The five infrared images from different perspectives were then stitched together to form a complete infrared sensing image of the ring network box. ; A high-density thermocouple array was uniformly arranged on the surface of the ring network box, and the thermocouples were fixed to the surface of the ring network box with high-temperature resistant tape. Then, the correspondence between the physical coordinates of the thermocouples and the pixel coordinates of the infrared image was established using a checkerboard calibration plate, and the actual temperature values ​​collected by the thermocouples were used as the true value labels for their corresponding pixels. Subsequently, the local weighted regression difference method was used to supplement the temperature labels for pixels that lacked true temperature labels, and finally, the calibrated thermal field sensing image of the ring network box was obtained. This data is then used as the output data for predicting the thermal field of the ring network box.

3. The dynamic optimization method for a primary and secondary integrated ring network box cooling device as described in claim 1, characterized in that: The datasets used to train the thermal coefficient sensing model of the ring cage can be constructed in the following ways: Accelerometers are deployed at the four corners of the ring main unit's switch operating mechanism linkage, busbar support insulator base, cable termination, gas-insulated wall shell, and cabinet base to collect corresponding vibration frequency data. and vibration amplitude data Deploy grating strain sensors to collect stress distribution Data is collected by deploying tensile and compressive sensors to gather corresponding tensile force data. and stress data Deploy resistance strain gauges to collect stress and deformation data The mechanical characteristic data of the ring network box collected above are combined to obtain a set of input data in the ring network box heat generation coefficient prediction dataset. ; Data is input for each group of heat production coefficient predictions through manual judgment. Add heat production coefficient label and heat dissipation coefficient label The two coefficient data are then combined as the output data in the ring network box heat generation coefficient prediction dataset. .

4. The dynamic optimization method for a primary and secondary integrated ring network box cooling device as described in claim 1, characterized in that: The ring network box thermal field sensing model includes an infrared image feature extractor, a thermal field sensing image reconstructor, and a cross-layer feature fusion unit. The overall feature extraction path of the ring network box thermal field sensing model consists of seven infrared image feature extractors stacked sequentially and connected end to end. The thermal field image reconstruction path of the thermal field sensing model consists of six thermal field sensing image reconstructors stacked sequentially and connected end to end. The outputs of the second and third extractors are sent to the first cross-layer feature fusion unit for processing, and then the output of the feature fusion unit is used as a supplementary input to the second and third reconstructors. At the same time, the outputs of the fourth and fifth extractors are sent to the second cross-layer feature fusion unit for processing, and then the output of the second feature fusion unit is used as a supplementary input to the fourth and fifth reconstructors. The outputs of the sixth and seventh extractors are sent to the third cross-layer feature fusion unit for processing, and then the output of the third feature fusion unit is used as a supplementary input to the sixth and seventh reconstructors.

5. The dynamic optimization method for a primary and secondary integrated ring network box cooling device as described in claim 4, characterized in that: The infrared image feature extractor has the following specific structure: First, two 3x3 convolutional layers are used to perform preliminary feature extraction on the image, followed by a ReLU activation function to achieve non-linear activation of the features. The resulting features are then fed into a global feature perception pathway to extract global thermal field features of the infrared image, and into a detail feature perception pathway to extract local thermal field variation features. In the global feature perception pathway, three sequentially connected 5x5 convolutional layers are used, followed by a normalization layer for feature standardization, and then an average pooling layer for average pooling. In the detail feature perception pathway, four 3x3 convolutional layers are used, followed by a normalization layer for feature standardization, and then a max pooling layer for max pooling. A fully connected layer then fuses the global and local thermal field features, followed by a final feature extraction operation using a 3x3 convolutional layer, and finally, simgoid activation is used for feature activation.

6. The dynamic optimization method for a primary and secondary integrated ring network box cooling device as described in claim 4, characterized in that: The specific structure of the thermal field-sensing image reconstructor is as follows: First, feature reconstruction upsampling is performed using two deconvolutional layers with a kernel size of 3*3, constructing two feature reconstruction pathways with different kernel sizes. The first feature reconstruction pathway includes three sequentially connected deconvolutional layers with a kernel size of 5*5, followed by a BN normalization layer and a Dropout layer to enhance the image reconstruction robustness of the thermal field-aware image reconstructor. The second feature reconstruction pathway includes three sequentially connected deconvolutional layers with a kernel size of 3*3, followed by a BN normalization layer and an unpooling layer. Finally, a fully connected layer and a Mish activation function are used to perform final feature fusion processing on the features of the two feature reconstruction pathways.

7. The dynamic optimization method for a primary and secondary integrated ring network box cooling device as described in claim 4, characterized in that: The cross-layer feature fusion processor first performs feature extraction through two sequentially connected 3x3 convolutional layers, followed by non-linear activation of the features using a ReLU activation function. Then, it performs feature upsampling through two interconnected 3x3 deconvolutional layers. Finally, it uses a max-pooling layer and a ReLU activation function to process the features again, resulting in the fused ring network box thermal field sensing image. .

8. The dynamic optimization method for a primary and secondary integrated ring network box cooling device as described in claim 1, characterized in that: The ring cage thermal coefficient sensing model includes a ring cage multi-layer feature sensing path, a ring cage multi-layer feature fusion unit, and a ring cage thermal coefficient generation head. The multi-layer feature sensing path of the ring network box first transmits the data. The data is fed into three different feature extraction pathways to complete the feature extraction operation. The network layer architecture of the first feature extraction pathway consists of three sequentially connected fully connected layers, with a ReLU activation function added after each fully connected layer. Features are obtained after processing by the first feature extraction pathway. The network architecture for the second feature extraction pathway remains three sequentially connected fully connected layers, but a new layer is added after each fully connected layer. Activation function, data Features are obtained after processing by the second feature extraction pathway. The network architecture of the third feature extraction pathway consists of three sequentially connected fully connected layers, with an additional layer added after each fully connected layer. Activation function, data Features are obtained after processing through the third feature extraction pathway. ; The ring network box multi-layer feature fusion unit is used to fuse the features extracted from the multi-layer feature perception path of the ring network box and adaptively assign weights to different features. The ring network box thermal coefficient generator uses two consecutive fully connected layers to fuse features. Further feature extraction processing is performed, followed by... The activation function performs non-linear activation of the features, and finally... The activation function is used to obtain the predicted heat generation coefficient of the ring network box. And the predicted results of heat dissipation coefficient .

9. The dynamic optimization method for a primary and secondary integrated ring network box cooling device as described in claim 1, characterized in that: Seven key control parameters in the ring network box cooling system were selected as working variables, including the number of working fans fg, fan speed fz, air outlet opening sj, coolant flow rate yn of the liquid cooling system, circulating pump working power wp, temperature difference between the start and stop threshold of the heat dissipation device wc, and number of times the circulating pump starts and stops xc. The multi-objective optimization function is shown below: in The objective function for cooling efficiency is to minimize this objective function in order to reduce the secondary heat introduced by the cooling device of the ring network box during operation; The heat generation coefficient of the ring network box during operation. This refers to the heat dissipation coefficient of the ring main unit during operation. and All were calculated using the heat generation coefficient sensing model of the ring network box. This shows the overall temperature of the ring main unit at this time; This represents the function for calculating the heat dissipation efficiency of the force field. The function representing the thermal field factor and heat dissipation efficiency calculation function; For temperature variance; The objective function for cooling consumption is to minimize this objective function in order to reduce the resources and energy consumed in the actual cooling process; Indicates the consumption cost coefficient; The objective function is the service life of the component.

10. The dynamic optimization method for a primary and secondary integrated ring network box cooling device as described in claim 9, characterized in that: The optimization range was limited for seven working variables, specifically including: That is, the minimum value of the number of fans operating, fg, is The maximum value is ; That is, the minimum fan speed is 10 ... The maximum rotational speed is [value missing]. change; That is, the minimum angle at which the air outlet opens is degrees, the maximum opening angle is That is, the minimum flow rate of the coolant in the liquid cooling system is m / s, maximum flow velocity is m / s; That is, the minimum operating power of the circulating pump is Watts, maximum operating power is watt; That is, the minimum temperature difference between the start and stop threshold of the heat dissipation device is Celsius, maximum is Celsius; That is, the minimum number of start-stop cycles for the circulating pump is The maximum number of start-stop cycles is [number]. Second-rate.

Images