Hybrid power plant heat dissipation management system and method

Abstract

The application relates to the technical field of energy storage power stations, in particular to a hybrid power station heat dissipation management system and method, which comprises a power generation and energy storage system and a heat dissipation system, and further comprises the heat dissipation system; the power generation and energy storage system comprises a plurality of photovoltaic inverters, gas turbines and battery groups; the heat dissipation system comprises a plurality of air-cooled motors, liquid-cooled heat dissipation components and radiation plates; the photovoltaic inverters, the gas turbines and the battery groups are electrically connected with a control system; the air-cooled motors, the liquid-cooled heat dissipation components and the radiation plates are electrically connected with the control system; data of the power generation and energy storage system are transmitted to the heat dissipation system after being adjusted and optimized by the control system, so that the power generation and energy storage system is subjected to heat dissipation treatment; the hybrid power station heat dissipation management system and method embeds a thermodynamic partial differential equation as a hard constraint into a space-time graph convolution network, designs a discrete-continuous hybrid action space and a dynamic multi-target reward function, completes air cooling / liquid cooling / radiation collaborative optimization, and enables energy consumption to be reduced by 23.8% and temperature difference fluctuation to be reduced by 18.5%.

Description

Technical Field

[0001] This invention relates to the field of energy storage power station technology, specifically to a hybrid power station heat dissipation management system and method. Background Technology

[0002] Hybrid power plants refer to power plants that combine multiple different types of power generation resources or combine power generation and energy storage. The current problems with hybrid power plants are as follows:

[0003] 1. Dynamic fluctuations in heat load: Hybrid power plants (such as solar-thermal-gas complementary, battery-diesel hybrid) often face intermittent heat load fluctuations, making it difficult for traditional heat dissipation systems to match in real time and reducing efficiency.

[0004] 2. Interference from multiple heat sources: Different heat sources (such as photovoltaic inverters, gas turbines, and battery packs) have large differences in thermal characteristics, resulting in conflicting heat dissipation requirements and easily causing local overheating or energy waste.

[0005] 3. Insufficient environmental adaptability: Heat dissipation performance degrades under extreme climates (high temperature, dust storms), and traditional air-cooled / liquid-cooled systems consume a lot of energy and rely on external resources (such as water resources).

[0006] Current patents such as CN202311572568.6 use phase change thermal storage technology to recover and utilize heat from the converter stage of the energy storage system, and adopt a liquid cooling heat dissipation scheme. However, they do not mention the heat dissipation method between different devices in the hybrid power station, resulting in poor heat dissipation effect. Patent CN202211399301.7 uses liquid cooling to dissipate heat from the battery pack, but does not mention the impact of other heat sources such as photovoltaic inverters and gas turbines.

[0007] Therefore, we propose a hybrid power plant heat dissipation management system and method. Summary of the Invention

[0008] One of the technical problems that this application aims to solve is:

[0009] 1. Dynamic fluctuations in heat load: Hybrid power plants (such as solar-thermal-gas complementary, battery-diesel hybrid) often face intermittent heat load fluctuations, making it difficult for traditional heat dissipation systems to match in real time and reducing efficiency.

[0010] 2. Interference from multiple heat sources: Different heat sources (such as photovoltaic inverters, gas turbines, and battery packs) have large differences in thermal characteristics, resulting in conflicting heat dissipation requirements and easily causing local overheating or energy waste.

[0011] 3. Insufficient environmental adaptability: Heat dissipation performance degrades under extreme climates (high temperature, dust storms), and traditional air-cooled / liquid-cooled systems consume a lot of energy and rely on external resources (such as water resources).

[0012] Current patents for soybeans consider liquid cooling for heat dissipation, but give less consideration to the impact of other heat sources such as photovoltaic inverters and gas turbines.

[0013] To address the aforementioned technical problems, this application provides a hybrid power plant heat dissipation management system, comprising a power generation and energy storage system and a heat dissipation system. The system is characterized by further including a control system.

[0014] The power generation and energy storage system includes a number of photovoltaic inverters, gas turbines, and battery packs;

[0015] The heat dissipation system includes a number of air-cooled motors, liquid-cooled heat dissipation components, and radiant panels;

[0016] The photovoltaic inverter, gas turbine, and battery pack are electrically connected to the control system;

[0017] The air-cooled motor, liquid-cooled heat dissipation components, and radiant panels are electrically connected to the control system;

[0018] The control system adjusts and optimizes the data from the power generation and energy storage system, and then transmits it to the heat dissipation system to cool the power generation and energy storage system.

[0019] In some embodiments, a PT100 sensor is arranged on the surface of the photovoltaic inverter to collect surface temperature;

[0020] Type K thermocouples are arranged on the surface of the gas turbine to collect exhaust temperature;

[0021] NTC thermistors are placed at the battery tabs of the battery pack to collect battery temperature;

[0022] Hall encoders are mounted on the surface of the air-cooled motor shaft to collect the motor's rotational speed data;

[0023] An electromagnetic flow meter is installed inside the liquid cooling heat dissipation component to collect the liquid cooling flow rate of the liquid cooling heat dissipation component;

[0024] A photoelectric encoder is arranged on the surface of the radiating plate to collect the opening degree of the radiating plate.

[0025] In some embodiments, the operation flow of the control system is as follows:

[0026] S101. Data Acquisition and Synchronization: Acquire the thermodynamic state, environmental parameters, and control signals of the hybrid power plant. Heat source parameters include: photovoltaic inverter surface temperature, battery cell temperature, and gas turbine exhaust temperature. Cooling system parameters include: air-cooled motor speed, liquid cooling flow rate, and radiant panel opening. Environmental parameters include: ambient temperature and humidity, wind speed, and dust concentration.

[0027] S201. Data Preprocessing and Feature Engineering: Cleaning outlier data, aligning spatiotemporal scales, constructing high-dimensional physical features, using Z-score filtering to process outliers, cross-sensor verification, upsampling environmental parameters from .Hz to Hz using cubic spline interpolation, dividing the power plant into an m×m D grid, generating virtual temperature nodes using inverse distance weighted IDW interpolation, extracting physical features of the data such as heat flux density and heat dissipation efficiency, and extracting statistical features of the data such as temperature variance and maximum temperature difference;

[0028] S301, Spatiotemporal Thermal Field Prediction: Predicts the thermal field distribution in the next few minutes, embeds thermodynamic and physical constraints, constructs a graph structure, builds a graph structure of internal equipment, inter-equipment equipment, and environmental interaction, and aggregates the hierarchical features of internal equipment, inter-equipment equipment, and environmental interaction, and forces the prediction results to satisfy the residual of the heat conduction equation;

[0029] S401, Multi-objective strategy optimization: Generate a coordinated control strategy for air cooling / liquid cooling / radiation, balance energy consumption, temperature difference, and equipment lifespan, design a hybrid action space, construct a dynamic reward function, and consider factors such as equipment energy consumption, equipment temperature, and equipment stress.

[0030] S501, Online Adaptive Compensation: Quickly adapts to sudden environmental changes such as sandstorms and extreme high temperatures. If the rate of change of ambient temperature is greater than ℃ / min or the sandstorm concentration is greater than μg / m, adaptive adjustment and compensation are performed. Meta-learning is used for rapid fine-tuning, constructing a fine-tuning dataset, calculating meta-gradients, and updating parameters step by step.

[0031] S601, Control Execution and Feedback: Executes control commands and collects feedback data to form a closed-loop optimization. Adjusts the air-cooling speed, liquid-cooling pump frequency, and radiant plate angle through the PLC controller. When the maximum temperature of the equipment exceeds the critical value, it forces the maximum cooling power. Feedback data is collected and recorded to track actual temperature changes and energy consumption.

[0032] In some embodiments, ambient temperature and humidity data are collected by an SHT sensor, wind speed data are collected by an ultrasonic anemometer, and dust concentration is collected by a laser scattering particulate sensor.

[0033] All data is synchronized to edge computing nodes via industrial Ethernet, with timestamps aligned to the millisecond level and stored as a time-series database.

[0034] The collected dataset becomes the original dataset D. raw ={T i (t),ω(t),q(t),θ(t),E env (t)}, where T i Let E be the temperature of the i-th sensor. env These are environmental parameters.

[0035] In some embodiments, Z-score filtering is used to process outliers in the data. The mean and standard deviation of the temperature data are calculated over a sliding window of seconds, and outliers with |z| greater than 3 are removed using the following formula:

[0036] Where u window (t) represents the mean within the window, σ window (t) represents the standard deviation within the window;

[0037] The standard for cross-sensor verification is that if the temperature difference between adjacent sensors is consistently >15℃ for more than 10 seconds, it is marked as abnormal.

[0038] The formula for generating virtual temperature nodes using inverse distance weighted IDW interpolation is as follows: Where d i The distance from the grid point to the i-th sensor is the Euclidean distance, and k is the number of nearest neighbor sensors (default k=5).

[0039] The formula for calculating heat dissipation density is: Where k is the thermal conductivity. It is a temperature gradient;

[0040] The formula for calculating heat dissipation efficiency is: Where q cool For effective heat dissipation, P fan For air-cooled power consumption, P pump For liquid cooling power consumption, P rad For radiated power consumption;

[0041] After data processing, the cleaned and aligned spatiotemporal feature tensor is obtained. Where N is grid The number of grids, T=300, which is the time step within 5 minutes, and C=10, which includes temperature, heat flux, wind speed, etc.

[0042] In some embodiments, the electrical graph structure construction includes the microscale, i.e., the interior of the device, and the edge weight calculation formula is as follows: Where x i x j Let |x| represent the coordinates of points i and j in the spatial grid. i -x j || represents the Euclidean distance between the two nodes;

[0043] At the mesoscale, i.e., between devices, the formula for calculating edge weights is: Q i Q j Let i and j be the heat source power densities at grid points i and j;

[0044] The macro-scale refers to environmental interaction, and the formula for calculating edge weights is as follows: Where T i ,T jThe temperature time series for grid points i and j;

[0045] The formula for hierarchical aggregation is: Where D s Degree matrix D s,ii =∑ j A s,ij Used to normalize the adjacency matrix Let A be the trainable weight matrix corresponding to the l-th layer. s H represents the adjacency matrix corresponding to the micro / meso / macro scales. (l) Let σ be the node feature matrix of the l-th layer, and σ be the activation function such as ReLU, introducing nonlinearity.

[0046] Residual calculation formula for heat conduction equation Where λ is the physical loss weight, ρ is the material density, and c p q is the specific heat capacity, k is the thermal conductivity, and q is the thermal conductivity. cool,i Effective heat dissipation;

[0047] Spatiotemporal thermal field prediction yields the predicted thermal field The format is a grid point temperature matrix.

[0048] In some embodiments, the input data for multi-objective strategy optimization is the predicted thermal field. and device status s t Such as battery SOC and gas turbine load.

[0049] The hybrid motion space includes air-cooled speed ω, liquid-cooled flow rate q, and radiation opening θ;

[0050] The air-cooling speed ω is discrete data, ranging from {0, 800, 1200, 1600} RPM. The physical constraint condition for the air-cooling speed ω is when T... max When the temperature is ≥70℃, ω=0 should not be selected;

[0051] The liquid cooling flow rate q is continuous data, and its value ranges from [0.2, 2.5] m / s. The physical constraint on the liquid cooling flow rate q is that the rate of change of flow rate is limited to a certain value.

[0052] The radiation aperture θ is discrete data, and the value range of radiation aperture θ is [0, 30%, 60%, 100%]. The physical constraint condition of radiation aperture θ is that when the dust concentration is greater than 50 μg / m, θ≤60%.

[0053] The dynamic reward function formula is R(t) = α(t)R enegery +β(t)R temp +γ(t)R stressWhere α(t), β(t), and γ(t) are dynamically adjusted weights, and the adjustment rules are as follows:

[0054] Where SOC(t) is the current state of charge of the battery;

[0055] in To predict the highest temperature, T safe For safe temperature, T crit This refers to the critical temperature of the equipment.

[0056] stress cum Accumulated stress on equipment, such as the number of times a motor starts and stops. max This represents the maximum stress on the equipment.

[0057] Where R enegery For energy consumption, the calculation formula is: In the formula P fan This represents the real-time power consumption of the air-cooled motor. P is the maximum allowable power consumption of the air-cooled motor. pump This represents the real-time power consumption of the liquid cooling components. P represents the maximum allowable power consumption of the liquid cooling heat dissipation component. rad This represents the real-time power consumption of the radiating plate. This represents the maximum allowable power consumption of the radiating plate.

[0058] Where R temp For the temperature difference term, the calculation formula is: In the formula To predict the highest temperature in the thermal field, T opt The optimal operating temperature of the equipment is given, with 10 in the denominator serving as a scaling factor for the temperature difference to control the sensitivity of the exponential function.

[0059] Where R stress For the lifespan term, the calculation formula is: In the formula, the current control action is the air-cooling speed, liquid-cooling flow rate, and radiation opening, which are the allowable range of the control action, such as the air-cooling speed of 0 to 1600 RPM. The coefficient 1 / 10 is the normalization factor to balance the impact of action changes of different dimensions on lifespan.

[0060] The multi-objective policy update mechanism includes physical rule guidance, namely prohibiting the reduction of liquid cooling flow in high-temperature areas, requiring adjacent air-cooled nodes to adjust their rotation speed synchronously, and an improved near-end policy optimization algorithm. The calculation formula is as follows:

[0061]

[0062] Where θ represents the network parameters of the policy, and η is the learning rate, such as the control parameter update step size. Given the ratio of the probabilities of the new and old strategies, and restricting policy mutation, A b For generalized advantage estimation (GAE), the advantage of an action relative to the average is measured, λ = 0.95. Time-difference weights are controlled, and ε is the shear threshold, set to 0.2 for forced r. b (θ) prevents excessive updates between [.,.].

[0063] After optimization of the multi-objective strategy, the control action a can be obtained. t = [ω,q,θ], where ω is the air-cooled rotation speed, q is the liquid-cooled flow rate, and θ is the radiation opening.

[0064] In some embodiments, online adaptive compensation input real-time data window {X t-w ,...,X t}, where the sliding window length is w = 1000;

[0065] Construct the fine-tuning dataset as D adapt ={X t-w ,X t};

[0066] The formula for calculating the meta-gradient is: Where w is the sliding window length and time steps, w = , To predict temperature, T i real This is the actual temperature. To predict the gradient of temperature with respect to the policy network parameters;

[0067] The formula for calculating single-step parameter updates is: Where η meta The meta-learning rate controls the model's fine-tuning step size, and is set to 0.001.

[0068] Online adaptive compensation is implemented in the input real-time data window {X t-w ,...,X t The updated policy parameters θ' are obtained.

[0069] In some embodiments, the control and execution module inputs control action a. t =[ω,q,θ];

[0070] The maximum critical temperature is T crit -5℃;

[0071] The maximum cooling power is a emergency =(ω max ,q max ,θ max ), ω max q max θ max For the maximum fan speed, liquid cooling flow rate, and radiation opening;

[0072] Control and execution module, input control action a t = [ω,q,θ], to obtain the updated dataset D new Return to S201 and begin the next round of optimization loop.

[0073] In some embodiments, a heat dissipation method for a hybrid power plant is characterized in that: it is applied to the heat dissipation management system of any one of the claims to the present invention, and the ventilation and heat dissipation method includes: collecting the thermodynamic state, environmental parameters and control signals of the hybrid power plant and integrating them into a raw dataset D. raw ={T i (t),ω(t),q(t),θ(t),E env (t)};

[0074] The spatiotemporal feature tensor is obtained by preprocessing the original dataset.

[0075] Through spatiotemporal feature tensors To predict the thermal field distribution in the next few minutes, thermodynamic physical constraints are embedded to obtain the predicted thermal field.

[0076] By inputting the predicted thermal field and equipment states such as battery SOC and gas turbine load, multi-objective strategy optimization is performed to obtain control action a. t =[ω,q,θ], at this time, the air-cooled motor, liquid-cooled heat dissipation components and radiant plates are adjusted through the corresponding control system to dissipate heat and reduce temperature;

[0077] Through the real-time data window {X t-w ,...,X t Perform online adaptive compensation to obtain the updated policy parameters θ';

[0078] By inputting control action a t = [ω,q,θ], to obtain the updated dataset D new The data is then preprocessed for the next round of optimization.

[0079] This invention has at least the following beneficial effects:

[0080] 1. By embedding thermodynamic partial differential equations as hard constraints into a spatiotemporal graph convolutional network, high-precision prediction of the thermal field of a hybrid power plant and fusion of physical laws were achieved, resulting in a reliable prediction model even in small sample scenarios.

[0081] 2. By using discrete-continuous hybrid action space design and dynamic multi-objective reward function, the coordinated optimization of air cooling / liquid cooling / radiation was completed, resulting in a comprehensive performance improvement of 23.8% energy consumption reduction and 18.5% temperature fluctuation reduction.

[0082] 3. By using sliding window meta-learning and edge-cloud collaborative architecture, rapid adaptation to sudden environmental changes was achieved, and real-time response capability for model fine-tuning within 5 minutes was obtained, breaking through the transfer bottleneck of traditional reinforcement learning. Attached Figure Description

[0083] Figure 1 This is a flowchart of the control system operation of the present invention;

[0084] Figure 2 This is a schematic diagram of the overall structural framework of the present invention.

[0085] In the diagram: 1-Power generation and energy storage system; 2-Heat dissipation system; 3-Control system; 11-Photovoltaic inverter; 12-Gas turbine; 13-Battery pack; 21-Air-cooled motor; 22-Liquid-cooled heat dissipation component; 23-Radiant panel. Detailed Implementation

[0086] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0087] Example 1: Please refer to Figure 1-2 The present invention provides a technical solution: a hybrid power plant heat dissipation management system and method, comprising a power generation and energy storage system 1 and a heat dissipation system 2, characterized in that: it further comprises a control system 3;

[0088] The power generation and energy storage system 1 includes a number of photovoltaic inverters 11, gas turbines 12, and battery packs 13;

[0089] The heat dissipation system 2 includes a number of air-cooled motors 21, liquid-cooled heat dissipation components 22, and radiant plates 23;

[0090] The photovoltaic inverter 11, the gas turbine 12, and the battery pack 13 are electrically connected to the control system 3.

[0091] The air-cooled motor 21, the liquid-cooled heat dissipation component 22, and the radiant plate 23 are electrically connected to the control system 3;

[0092] The control system 3 adjusts and optimizes the data of the power generation and energy storage system 1, and then transmits it to the heat dissipation system 2 to dissipate heat from the power generation and energy storage system 1.

[0093] Furthermore, a PT100 sensor is arranged on the surface of the photovoltaic inverter 11 to collect surface temperature;

[0094] K-type thermocouples are arranged on the surface of the gas turbine 12 to collect exhaust temperature;

[0095] An NTC thermistor is placed at the battery tab of the battery pack 13 to collect battery temperature.

[0096] A Hall encoder is arranged on the surface of the shaft of the air-cooled motor 21 to collect the speed data of the air-cooled motor 21;

[0097] An electromagnetic flow meter is installed inside the liquid cooling heat dissipation component 22 to collect the liquid cooling flow of the liquid cooling heat dissipation component 22;

[0098] A photoelectric encoder is arranged on the surface of the radiation plate 23 to collect the opening degree of the radiation plate 23.

[0099] Furthermore, the operation flow of control system 3 is as follows:

[0100] S101, Data Acquisition and Synchronization: Acquire the thermodynamic state, environmental parameters, and control signals of the hybrid power plant. Heat source parameters include: surface temperature of photovoltaic inverter 11, individual cell temperature of battery pack 13, and exhaust temperature of gas turbine 12. Heat dissipation system 2 parameters include: speed of air-cooled motor 21, liquid cooling flow rate, and opening degree of radiant panel 23. Environmental parameters include: ambient temperature and humidity, wind speed, and dust concentration.

[0101] S201, Data Preprocessing and Feature Engineering: Cleaning outlier data, aligning spatiotemporal scales, constructing high-dimensional physical features, using Z-score filtering to process outliers, cross-sensor verification, upsampling environmental parameters from 0.1Hz to 1Hz using cubic spline interpolation, dividing the power plant into a 1m×1m 3D grid, generating virtual temperature nodes using inverse distance weighted IDW interpolation, extracting physical features of the data such as heat flux density and heat dissipation efficiency, and extracting statistical features of the data such as temperature variance and maximum temperature difference;

[0102] S301, Spatiotemporal thermal field prediction: Predict the thermal field distribution in the next 5 minutes, embed thermodynamic physical constraints, construct a graph structure, build a graph structure of internal equipment, inter-equipment, and environmental interaction, and aggregate the hierarchical features of internal equipment, inter-equipment, and environmental interaction, and force the prediction results to satisfy the residual of the heat conduction equation;

[0103] S401, Multi-objective strategy optimization: Generate a coordinated control strategy for air cooling / liquid cooling / radiation, balance energy consumption, temperature difference, and equipment lifespan, design a hybrid action space, construct a dynamic reward function, and consider factors such as equipment energy consumption, equipment temperature, and equipment stress.

[0104] S501: Online adaptive compensation: Quickly adapts to sudden environmental changes such as sandstorms and extreme high temperatures. If the rate of change of ambient temperature is greater than 2℃ / min or the sandstorm concentration is greater than 200μg / m3, adaptive adjustment and compensation are performed. Meta-learning is used for rapid fine-tuning. A fine-tuning dataset is constructed, meta-gradients are calculated, and parameters are updated step by step.

[0105] S601, Control Execution and Feedback: Executes control commands and collects feedback data to form a closed-loop optimization. Adjusts the air-cooling speed, liquid-cooling pump frequency, and radiant plate angle through the PLC controller. When the maximum temperature of the equipment exceeds the critical value, it forces the maximum cooling power, collects feedback data, and records the actual temperature changes and energy consumption.

[0106] Furthermore, ambient temperature and humidity data were collected using an SHT35 sensor, wind speed data were collected using an ultrasonic anemometer, and dust concentration was collected using a laser scattering particulate matter sensor.

[0107] All data is synchronized to edge computing nodes via industrial Ethernet, with timestamps aligned to the millisecond level and stored as a time-series database.

[0108] The collected dataset becomes the original dataset D. raw ={T i (t),ω(t),q(t),θ(t),E env (t)}, where T i Let E be the temperature of the i-th sensor. env These are environmental parameters.

[0109] Furthermore, Z-score filtering is used to process outliers. The mean and standard deviation of the temperature data are calculated over a 60-second sliding window, and outliers with |z| greater than 3 are removed using the following formula:

[0110] Where u window (t) represents the mean within the window, σ window (t) represents the standard deviation within the window;

[0111] The standard for cross-sensor verification is that if the temperature difference between adjacent sensors is consistently >15℃ for more than 10 seconds, it is marked as abnormal.

[0112] The formula for generating virtual temperature nodes using inverse distance weighted IDW interpolation is as follows: Where d i The distance from the grid point to the i-th sensor is the Euclidean distance, and k is the number of nearest neighbor sensors (default k=5).

[0113] The formula for calculating heat dissipation density is: Where k is the thermal conductivity. It is a temperature gradient;

[0114] The formula for calculating heat dissipation efficiency is: Where q cool For effective heat dissipation, P fan For air-cooled power consumption, P pump For liquid cooling power consumption, P rad For radiated power consumption;

[0115] After data processing, the cleaned and aligned spatiotemporal feature tensor is obtained. Where N is grid The number of grids, T=300, which is the time step within 5 minutes, and C=10, which includes temperature, heat flux, wind speed, etc.

[0116] Furthermore, the graph structure construction includes the microscopic scale, i.e., the interior of the device, and the edge weight calculation formula is as follows: Where x i x j Let |x| represent the coordinates of points i and j in the spatial grid. i -x j || represents the Euclidean distance between the two nodes;

[0117] At the mesoscale, i.e., between devices, the formula for calculating edge weights is: Q i Q j Let i and j be the heat source power densities at grid points i and j;

[0118] The macro-scale refers to environmental interaction, and the formula for calculating edge weights is as follows: Where T i ,T j The temperature time series for grid points i and j;

[0119] The formula for hierarchical aggregation is: Where D s Degree matrix D s,ii =Σ j A s,ij Used to normalize the adjacency matrix Let A be the trainable weight matrix corresponding to the l-th layer. s H represents the adjacency matrix corresponding to the micro / meso / macro scales. (l) Let σ be the node feature matrix of the l-th layer, and σ be the activation function such as ReLU, introducing nonlinearity.

[0120] Residual calculation formula for heat conduction equation Where λ is the physical loss weight, taken as 0.3, ρ is the material density, and c p q is the specific heat capacity, k is the thermal conductivity, and q is the thermal conductivity. cool,i Effective heat dissipation;

[0121] Spatiotemporal thermal field prediction yields the predicted thermal field The format is a grid point temperature matrix.

[0122] Furthermore, the multi-objective strategy optimizes the input data for predicting the thermal field. and device status s t Such as battery SOC, gas turbine 12 load.

[0123] The hybrid motion space includes air-cooled speed ω, liquid-cooled flow rate q, and radiation opening θ;

[0124] The air-cooling speed ω is discrete data, ranging from {0, 800, 1200, 1600} RPM. The physical constraint condition for the air-cooling speed ω is when T... max When the temperature is ≥70℃, ω=0 should not be selected;

[0125] The liquid cooling flow rate q is continuous data, and its value ranges from [0.2, 2.5] m. 3 / s, the physical constraint for the liquid cooling flow rate q is that the rate of change of the flow rate is limited to...

[0126] The radiation aperture θ is discrete data, and the value range of radiation aperture θ is [0, 30%, 60%, 100%]. The physical constraint condition of radiation aperture θ is that when the dust concentration is greater than 50 μg / m3, θ≤60%.

[0127] The dynamic reward function formula is R(t) = α(t)R enegery +β(t)R temp +γ(t)R stress Where α(t), β(t), and γ(t) are dynamically adjusted weights, and the adjustment rules are as follows:

[0128] Where SOC(t) is the current state of charge of the battery;

[0129] in To predict the highest temperature, T safe For safe temperature, T crit This refers to the critical temperature of the equipment.

[0130] stress cum Accumulated stress on equipment, such as the number of times a motor starts and stops. max This represents the maximum stress on the equipment.

[0131] Where R enegery For energy consumption, the calculation formula is: In the formula P fan This refers to the real-time power consumption of the air-cooled motor 21. P is the maximum allowable power consumption of the air-cooled motor 21. pump This refers to the real-time power consumption of the liquid cooling heat dissipation component 22. P represents the maximum allowable power consumption of the liquid cooling heat dissipation component 22. rad This represents the real-time power consumption of the radiating plate 23. The maximum allowable power consumption of the radiating plate 23;

[0132] Where R temp For the temperature difference term, the calculation formula is: In the formula To predict the highest temperature in the thermal field, T opt The optimal operating temperature of the equipment is given, with 10 in the denominator serving as a scaling factor for the temperature difference to control the sensitivity of the exponential function.

[0133] Where R stress For the lifespan term, the calculation formula is: In the formula, the current control action is the air-cooling speed, liquid-cooling flow rate, and radiation opening, which are the allowable range of the control action, such as the air-cooling speed of 0 to 1600 RPM. The coefficient 1 / 10 is the normalization factor to balance the impact of action changes of different dimensions on lifespan.

[0134] The multi-objective policy update mechanism includes physical rule guidance, namely prohibiting the reduction of liquid cooling flow in high-temperature areas, requiring adjacent air-cooled nodes to adjust their rotation speed synchronously, and an improved near-end policy optimization algorithm. The calculation formula is as follows:

[0135]

[0136] Where θ represents the network parameters of the policy, and η is the learning rate (e.g., 0.001) and the step size for controlling parameter updates. Given the ratio of the probabilities of the new and old strategies, and restricting policy mutation, A b For generalized advantage estimation (GAE), the advantage of an action relative to the average is measured, λ = 0.95. Time-difference weights are controlled, and ε is the shear threshold, set to 0.2 for forced r. b (θ) prevents excessive updates between [0.8, 1.2];

[0137] After optimization of the multi-objective strategy, the control action a can be obtained. t = [ω,q,θ], where ω is the air-cooled rotation speed, q is the liquid-cooled flow rate, and θ is the radiation opening.

[0138] Furthermore, online adaptive compensation inputs real-time data window {X t-w ,...,X t}, where the sliding window length is w = 1000;

[0139] Construct the fine-tuning dataset as D adapt ={X t-w ,X t};

[0140] The formula for calculating the meta-gradient is: Where w is the sliding window length and the number of time steps, w = 1000. To predict temperature, T i real This is the actual temperature. To predict the gradient of temperature with respect to the policy network parameters;

[0141] The formula for calculating single-step parameter updates is: Where η meta The meta-learning rate controls the model's fine-tuning step size, and is set to 0.001.

[0142] Online adaptive compensation is implemented in the input real-time data window {X t-w ,...,X t The updated policy parameters θ' are obtained.

[0143] Furthermore, in the control and execution module, the input control action a is... t =[ω,q,θ];

[0144] The maximum critical temperature is T crit -5℃;

[0145] The maximum cooling power is a emergency =(ω max ,q max ,θ max ), ω max q max θ max For the maximum fan speed, liquid cooling flow rate, and radiation opening;

[0146] Control and execution module, input control action a t = [ω,q,θ], to obtain the updated dataset D new Return to S201 and begin the next round of optimization loop.

[0147] Furthermore, a heat dissipation method for a hybrid power plant, employing a hybrid power plant heat dissipation management system, includes the following steps: collecting the thermodynamic state, environmental parameters, and control signals of the hybrid power plant and integrating them into a raw dataset.

[0148] D raw ={T i (t),ω(t),q(t),θ(t),E env (t)};

[0149] The spatiotemporal feature tensor is obtained by preprocessing the original dataset.

[0150] Through spatiotemporal feature tensors To predict the thermal field distribution over the next 5 minutes, thermodynamic physical constraints are embedded to obtain the predicted thermal field.

[0151] By inputting the predicted thermal field and equipment states such as battery SOC and gas turbine load, multi-objective strategy optimization is performed to obtain control action a. t =[ω,q,θ], at this time, the air-cooled motor 21, liquid-cooled heat dissipation component 22 and radiant plate 23 are adjusted by the control system 3 to dissipate heat and reduce temperature;

[0152] Through the real-time data window {X t-w ,...,X t Perform online adaptive compensation to obtain the updated policy parameters θ';

[0153] By inputting control action a t = [ω,q,θ], to obtain the updated dataset D new The data is then preprocessed for the next round of optimization.

[0154] The following is combined Figures 1-2 Introduction to hybrid power plant thermal management systems and methods:

[0155] First, a PT100 sensor is mounted on the surface of the photovoltaic inverter 11 to collect surface temperature. A K-type thermocouple is mounted on the surface of the gas turbine 12 to collect exhaust temperature. An NTC thermistor is mounted on the battery tabs of the battery pack 13 to collect battery temperature. A Hall encoder is mounted on the surface of the shaft of the air-cooled motor 21 to collect the speed data of the air-cooled motor 21. An electromagnetic flowmeter is mounted inside the liquid cooling heat dissipation assembly 22 to collect the liquid cooling flow rate of the liquid cooling heat dissipation assembly 22. A photoelectric encoder is mounted on the surface of the radiating plate 23 to collect the opening degree of the radiating plate 23. Ambient temperature and humidity data are collected by an SHT35 sensor, wind speed data is collected by an ultrasonic anemometer, and dust concentration is collected by a laser scattering particulate matter sensor. These raw data are transmitted to the control system 3 through the sensors. The control system 3 processes the received data and integrates it into D. raw ={T i (t),ω(t),q(t),θ(t),E env The spatiotemporal feature tensor (t) is obtained by preprocessing the original dataset. Through spatiotemporal feature tensors To predict the thermal field distribution over the next 5 minutes, thermodynamic physical constraints are embedded to obtain the predicted thermal field. By inputting the predicted thermal field and equipment states such as battery SOC and gas turbine load, multi-objective strategy optimization is performed to obtain control action a. t =[ω,q,θ], at this time, the air-cooled motor 21, liquid-cooled heat dissipation component 22 and radiant plate 23 are adjusted by the control system 3 to dissipate heat and reduce temperature, and the real-time data window {X t-w,...,X t Online adaptive compensation is performed to obtain the updated policy parameters θ', which are then input as control actions a. t = [ω,q,θ], to obtain the updated dataset D new The data is preprocessed for the next optimization cycle to obtain a new control action a. t =[ω,q,θ], and the air-cooled motor 21, liquid-cooled heat dissipation component 22 and radiant plate 23 are adjusted by the control system 3 to achieve the effect of heat dissipation and cooling.

[0156] The effectiveness of this method compared to other methods is shown in the table below:

[0157] Daily energy consumption 1450kWh 1102kWh Decreased by 23.8% Maximum temperature difference 17.3℃ 14.1℃ Reduced by 18.5% Equipment stress fluctuation 0.78 0.52 Reduced by 33.3%

[0158] As can be seen from the table above, this method has lower daily energy consumption, lower maximum temperature difference, lower equipment stress fluctuation, and better heat dissipation compared to traditional methods.

[0159] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0160] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A hybrid power plant heat dissipation management system, comprising a power generation and energy storage system (1) and a heat dissipation system (2), characterized in that: It also includes the control system (3); The power generation and energy storage system (1) includes a number of photovoltaic inverters (11), gas turbines (12) and battery packs (13). The heat dissipation system (2) includes a number of air-cooled motors (21), liquid-cooled heat dissipation components (22), and radiant plates (23). The photovoltaic inverter (11), gas turbine (12) and battery pack (13) are electrically connected to the control system (3); The air-cooled motor (21), liquid-cooled heat dissipation assembly (22), and radiant plate (23) are electrically connected to the control system (3); The control system (3) adjusts and optimizes the data of the power generation and energy storage system (1) and transmits it to the heat dissipation system (2) to dissipate heat from the power generation and energy storage system (1); The operation flow of the control system (3) is as follows: S101, Data Acquisition and Synchronization: Acquire the thermodynamic state, environmental parameters and control signals of the hybrid power station. The heat source parameters include: the surface temperature of the photovoltaic inverter (11), the temperature of the individual cells of the battery pack (13), and the exhaust temperature of the gas turbine (12). The parameters of the heat dissipation system (2) include: the speed of the air-cooled motor (21), the liquid cooling flow rate, and the opening degree of the radiant plate (23). The environmental parameters include: ambient temperature and humidity, wind speed, and dust concentration. S201. Data Preprocessing and Feature Engineering: Cleaning outlier data, aligning spatiotemporal scales, constructing high-dimensional physical features, using Z-score filtering to process outliers, cross-sensor verification, upsampling environmental parameters with a sampling frequency of 0.1Hz to 1Hz using cubic spline interpolation, dividing the power plant into a 1m×1m 3D grid, generating virtual temperature nodes through inverse distance weighted interpolation, extracting the physical features of the data, namely heat flux density and heat dissipation efficiency, and extracting the statistical features of the data, namely temperature variance and maximum temperature difference; S301, Spatiotemporal thermal field prediction: Predict the thermal field distribution in the next 5 minutes, embed thermodynamic physical constraints, construct a graph structure, build a graph structure of internal equipment, inter-equipment, and environmental interaction, and aggregate the hierarchical features of internal equipment, inter-equipment, and environmental interaction, and force the prediction results to satisfy the residual of the heat conduction equation; S401, Multi-objective strategy optimization: Generate a control strategy for the coordinated operation of air cooling, liquid cooling and radiation, balance energy consumption, temperature difference and equipment life, design a hybrid action space, construct a dynamic reward function, and consider equipment energy consumption, equipment temperature and equipment stress related factors. S501: Online Adaptive Compensation: Quickly adapts to sudden environmental changes such as sandstorms and extreme high temperatures; detects sudden environmental changes, and if the rate of change in ambient temperature is greater than 2... o When C / min or dust concentration is greater than 200 μg / m3, adaptive adjustment and compensation are performed, meta-learning is used for fast fine-tuning, a fine-tuning dataset is constructed, meta-gradient is calculated, and single-step parameter updates are performed. S601, Control Execution and Feedback: Execute control commands and collect feedback data to form a closed-loop optimization. Adjust the air cooling speed, liquid cooling pump frequency, and the angle of the radiant plate (23) through the PLC controller. When the maximum temperature of the equipment exceeds the critical value, force the maximum cooling power, collect feedback data, and record the actual temperature change and energy consumption.

2. The hybrid power plant heat dissipation management system according to claim 1, characterized in that: The photovoltaic inverter (11) has a PT100 sensor arranged on its surface to collect surface temperature; The gas turbine (12) is equipped with K-type thermocouples on its surface to collect exhaust temperature; An NTC thermistor is arranged at the battery tab of the battery pack (13) to collect battery temperature; A Hall encoder is arranged on the surface of the shaft of the air-cooled motor (21) to collect the speed data of the air-cooled motor (21); An electromagnetic flow meter is arranged inside the liquid cooling heat dissipation component (22) to collect the liquid cooling flow of the liquid cooling heat dissipation component (22); A photoelectric encoder is arranged on the surface of the radiating plate (23) to collect the opening degree of the radiating plate (23).

3. The hybrid power plant heat dissipation management system according to claim 1, characterized in that: The ambient temperature and humidity data were collected using an SHT35 sensor, the wind speed data were collected using an ultrasonic anemometer, and the dust concentration was collected using a laser scattering particulate matter sensor. All data is synchronized to edge computing nodes via industrial Ethernet, with timestamps aligned to the millisecond level and stored as a time-series database. The collected dataset becomes the original dataset. ,in Let i be the temperature of the i-th sensor. These are environmental parameters.

4. The hybrid power plant heat dissipation management system according to claim 1, characterized in that: Z-score filtering was used to remove outliers. For a 60-second temperature data period, the mean and standard deviation were calculated using a sliding window, and outliers with |z| greater than 3 were removed. The formula is as follows: ,in The mean within the window. The standard deviation within the window; The standard for cross-sensor verification is that if the temperature difference between adjacent sensors is consistently >15°C for more than 10 seconds, it is marked as abnormal. The formula for generating virtual temperature nodes using inverse distance weighted (IDW) interpolation is as follows: , where d i Let k be the Euclidean distance from a grid point to the i-th sensor, and k be the number of nearest neighbor sensors. The formula for calculating the heat flux density is as follows: , among which is Thermal conductivity, It is a temperature gradient; The formula for calculating the heat dissipation efficiency is as follows: ,in For effective heat dissipation, For air-cooled power consumption, For liquid cooling power consumption, For radiated power consumption; After data processing, the cleaned and aligned spatiotemporal feature tensor is obtained. , of which The number of grids, T=300 (time step within 5 minutes), and C=10, specifically include three types of parameters: temperature, heat flux, and wind speed.

5. The hybrid power plant heat dissipation management system according to claim 1, characterized in that: The graph structure construction includes the microscale, i.e., the interior of the device, and the edge weight calculation formula is as follows: ,in , Let i and j be the coordinates of the spatial grid points. The Euclidean distance between the two nodes; At the mesoscale, i.e., between devices, the formula for calculating edge weights is: ,in , Let i and j be the heat source power densities at grid points i and j; The macro-scale refers to environmental interaction, and the formula for calculating edge weights is as follows: ,in The temperature time series for grid points i and j; The formula for hierarchical aggregation is: ,in Degree matrix Used to normalize the adjacency matrix Let l be the trainable weight matrix corresponding to the l-th layer. This is an adjacency matrix, which is divided into three corresponding scales: micro, meso, and macro. Let l be the node feature matrix of the l-th layer. The activation function is ReLU, which introduces nonlinearity. The formula for calculating the residual of the heat conduction equation ,in The physical loss weight is set to 0.

3. For material density, ρ is the specific heat capacity, k is the thermal conductivity. Effective heat dissipation; The spatiotemporal thermal field prediction yields the predicted thermal field. The format is a grid point temperature matrix.

6. The hybrid power plant heat dissipation management system according to claim 1, characterized in that: The multi-objective strategy optimization input data is the predicted thermal field. and equipment status Such as battery SOC, the gas turbine (12) load; The hybrid motion space includes air-cooled rotation speed. Liquid cooling flow rate Radiative aperture ; The air-cooling speed The air-cooling rotation speed is discrete data. The air-cooling speed is in the range of {0, 800, 1200, 1600} RPM. The physical constraints are when When, selection is prohibited. ; The liquid cooling flow rate The liquid cooling flow rate is continuous data. The value range of m is [0.2, 2.5]. 3 / s, the liquid cooling flow rate The physical constraint is that the rate of change of flow is limited to ; The radiation aperture The radiation aperture is discrete data. The value range is [0, 30%, 60%, 100%], and the radiation aperture is... The physical constraint is that when the dust concentration is greater than 50 μg / m3. ; The formula for the dynamic reward function is as follows: ,in , , The weights are dynamically adjusted, and the adjustment rules are as follows: , among which This indicates the battery's current state of charge. ,in To predict the highest temperature, For safe temperature, This refers to the critical temperature of the equipment. ,in To accumulate stress on the equipment, the number of motor start-stop cycles is used as an evaluation indicator. This represents the maximum stress on the equipment. in For energy consumption, the calculation formula is: In the formula The real-time power consumption of the air-cooled motor (21) is... The maximum allowable power consumption of the air-cooled motor (21) is The real-time power consumption of the liquid cooling heat dissipation component (22) is... The maximum allowable power consumption of the liquid cooling heat dissipation component (22) is... The real-time power consumption of the radiating plate (23) is... The maximum allowable power consumption of the radiating plate (23); in For the temperature difference term, the calculation formula is: In the formula To predict the highest temperature in the thermal field, The optimal operating temperature of the equipment is given, with 10 in the denominator serving as a scaling factor for the temperature difference to control the sensitivity of the exponential function. in For the lifespan term, the calculation formula is: The formula includes three types of parameters for the current control action: air cooling speed, liquid cooling flow rate, and radiation opening degree. The allowable range of air cooling speed is 0~1600RPM. The coefficient 1 / 10 is a normalization factor to balance the impact of action changes of different dimensions on lifespan. The multi-objective strategy update mechanism includes physical rule guidance, namely prohibiting the reduction of liquid cooling flow in high-temperature areas, requiring adjacent air-cooled nodes to adjust their rotation speed synchronously, and an improved near-end strategy optimization algorithm, the calculation formula of which is as follows: Among them Policy network parameters, To update the step size for the learning rate control parameters, The ratio of the probability of the new and old strategies is used to restrict policy mutation. Generalized advantage estimation (GAE) measures the advantage of an action relative to the average. Controlling the time difference weights, The shearing threshold is set to 0.

2. Prevent excessively large updates within the range of [0.8, 1.2]. The multi-objective strategy, after optimization, can yield control actions. ,in Air-cooled speed, Liquid cooling flow rate, Radiation aperture.

7. The hybrid power plant heat dissipation management system according to claim 1, characterized in that: The online adaptive compensation input real-time data window The length of the sliding window is w=1000; The construction of the fine-tuning dataset is as follows: ; The formula for calculating the meta-gradient is: Where w is the length of the sliding window, w=1000, To predict temperature, This is the actual temperature. To predict the gradient of temperature with respect to the policy network parameters; The calculation formula for the single-step parameter update is as follows: ,in The meta-learning rate controls the model's fine-tuning step size, and is set to 0.

001. The online adaptive compensation is implemented in the input real-time data window. Get the updated strategy parameters .

8. The hybrid power plant heat dissipation management system according to claim 1, characterized in that: The control and execution module inputs control actions. ; The maximum critical value of the temperature is o C; Maximum control action is , , , For the maximum fan speed, liquid cooling flow rate, and radiation opening; The control and execution module inputs control actions. To obtain the updated dataset Return to S201 and begin the next round of optimization loop.

9. A heat dissipation method for a hybrid power station, characterized in that: The ventilation and heat dissipation method applied to the hybrid power plant heat dissipation management system according to any one of claims 1 to 8 includes: S101. By collecting the thermodynamic state, environmental parameters, and control signals of the mixing station, the data is integrated into a raw dataset. ; S201. The spatiotemporal feature tensor is obtained by preprocessing the original dataset. ; S301, using the spatiotemporal feature tensor To predict the thermal field distribution over the next 5 minutes, thermodynamic physical constraints are embedded to obtain the predicted thermal field. ; S401. By inputting the predicted thermal field, equipment status, and gas turbine (12) load, multi-objective strategy optimization is performed to obtain the control action. At this time, the air-cooled motor (21), liquid-cooled heat dissipation component (22) and radiant plate (23) are adjusted by the control system (3) to dissipate heat and cool down; S501, via real-time data window Perform the online adaptive compensation to obtain the updated policy parameters. ; S601, Control action via input The updated dataset is obtained. The data is then preprocessed for the next round of optimization.

Images