Intelligent control algorithm for high-altitude grid-connected cabinet

Abstract

The invention belongs to the technical field of intelligent control of high-altitude power equipment, and particularly provides an intelligent control algorithm for a high-altitude grid-connected cabinet, which comprises a data acquisition module, a low-pressure heat dissipation efficiency adaptive compensation algorithm, a high-altitude multi-source feature fusion fault pre-judgment algorithm, an air temperature sudden change look-ahead temperature control algorithm and feedback iterative optimization. The cabinet can be used in a low-pressure environment, heat dissipation efficiency is improved, and temperature fluctuation in the cabinet is controlled. The method can improve the fault pre-judgment accuracy, reduce the fault false alarm rate, shorten the early warning time and remarkably reduce the high-altitude operation and maintenance cost aiming at the conditions that the traffic and communication in the high-altitude area are not changed, the maintenance cost is large and the loss influence is large. According to the method, a temperature control algorithm is adopted, different countermeasures can be adopted according to actual conditions to reduce temperature changes aiming at the fact that the temperature change rate of a high-altitude area is large, meanwhile, condensation risk pre-judgment is introduced, the condensation occurrence rate is reduced, and part corrosion and short circuit faults are avoided.

Description

Technical Field

[0001] This application belongs to the field of intelligent control technology for high-altitude power equipment, and more specifically, it relates to an intelligent control algorithm for grid-connected cabinets at high altitudes. Background Technology

[0002] High-altitude areas (≥2000m) have unique geographical environments, resulting in low air pressure, extreme diurnal temperature variations (up to 30℃ or more), strong ultraviolet radiation, and a high risk of lightning strikes. The combined effect of these environmental factors poses multiple severe challenges to the long-term stable operation of grid-connected cabinets, becoming a key bottleneck restricting the reliability of high-altitude power systems.

[0003] In terms of heat dissipation control, low air pressure is a core factor affecting the heat dissipation efficiency of grid-connected cabinets. As altitude increases, atmospheric pressure decreases non-linearly, leading to a significant reduction in the air convection heat dissipation coefficient. For every 1000m increase in altitude, heat dissipation efficiency decreases by 10-15%. However, conventional temperature control algorithms use only a single temperature parameter as the control input, completely ignoring the non-linear impact of air pressure on heat dissipation efficiency. This results in a disconnect between the control strategy and actual heat dissipation requirements—either insufficient heat dissipation compensation causes overheating and aging of components inside the cabinet, or excessive heat dissipation leads to energy waste, making it impossible to achieve a balance between heat dissipation effect and energy consumption.

[0004] In terms of fault prediction, the fault causes of grid-connected cabinets at high altitudes exhibit significant multi-factor synergistic characteristics. The interaction of low air pressure, drastic temperature differences, and condensation makes the fault mechanism more complex. Meanwhile, the inconvenient transportation and limited maintenance conditions in high-altitude areas lead to difficulties in fault sample collection and data scarcity. Traditional fault prediction algorithms rely on single electrical parameters and fixed thresholds, making it difficult to adapt to the nonlinear changes in fault characteristics under high-altitude conditions. They generally suffer from low prediction accuracy, high false alarm rates, and poor early warning timeliness. Once a fault occurs, the inability to provide early warning often leads to maintenance delays, resulting in wider power supply interruptions. Furthermore, maintenance costs in high-altitude areas are significantly higher than in plains areas, further amplifying the losses from faults.

[0005] Regarding temperature variation adaptation, the dramatic diurnal temperature variations characteristic of high-altitude regions present grid-connected cabinets with a dual dilemma of "overheating during the day and condensation at night." Conventional temperature control algorithms employ a "passive response" mode, only initiating regulation after the cabinet's internal temperature reaches a set threshold. This makes it impossible to predict temperature change trends in advance and hinders the coordinated control of heat dissipation and condensation prevention. Failure to pre-cool during the day results in components operating at high temperatures for extended periods, while sudden temperature drops at night can easily lead to condensation due to the large temperature difference between the inside and outside of the cabinet. This condensation can corrode internal electrical components, causing secondary faults such as decreased insulation performance and short circuits.

[0006] Most existing industrial control algorithms are designed for plains environments and do not fully consider the specific characteristics of high-altitude environments. They cannot simultaneously address the three core challenges: heat dissipation degradation due to low air pressure, inaccurate fault prediction caused by multi-factor coupling, and poor adaptability due to drastic temperature changes. These challenges severely restrict the operational reliability and maintenance efficiency of grid-connected cabinets in high-altitude areas. Therefore, developing a targeted intelligent control algorithm to fill the technological gap in precise control of grid-connected cabinets in high-altitude areas has become an urgent need to ensure the stable operation of high-altitude power systems. Summary of the Invention

[0007] Based on the above-mentioned technical problems, this application provides an intelligent control algorithm for grid-connected cabinets at high altitudes to solve the problems in the prior art, such as the nonlinear decay of heat dissipation efficiency caused by low air pressure at high altitudes, the conventional temperature control algorithm not considering the influence of air pressure, which is prone to insufficient heat dissipation or excessive energy consumption, the complex causes of faults at high altitudes and the scarcity of fault samples, the conventional fault prediction algorithm relying on a single parameter and a fixed threshold, which has the problems of low accuracy and high false alarm rate, and the drastic temperature difference between day and night at high altitudes, the conventional temperature control algorithm is a passive response mode and cannot predict temperature changes in advance, making it difficult to achieve heat dissipation and anti-condensation control in a coordinated manner.

[0008] To achieve the above objectives, the technical solution adopted in this application is: to provide an intelligent control algorithm for high-altitude grid-connected cabinets, comprising the following steps:

[0009] The S1 data acquisition module includes environmental parameters, electrical parameters, and equipment parameters. The sampling frequency is 1Hz, and the data is stored after adaptive wavelet denoising processing.

[0010] The S2 low-pressure heat dissipation efficiency adaptive compensation algorithm includes constructing a high-altitude-specific air pressure heat dissipation efficiency coupling model and optimizing the heat dissipation strategy based on model predictive control (MPC).

[0011] The S3 high-altitude multi-source feature fusion fault prediction algorithm includes multi-source heterogeneous data preprocessing, transfer learning and ensemble learning fault prediction models, and dynamic threshold early warning mechanism;

[0012] S4 temperature change anti-theft temperature control algorithm, including temperature trend prediction, phase change material thermal state coupling control and condensation risk prediction;

[0013] S5 feedback iterative optimization.

[0014] Furthermore, the high-altitude-specific air pressure heat dissipation efficiency coupling model in step S2 is based on measured data at altitudes of 2000-5000m (covering air pressure of 55-80kPa and temperature difference of 5-35℃), and uses an improved BP neural network and Gaussian process regression to train a nonlinear fitting model.

[0015] h(P) = a·Pb +c·(T in -T out )+d

[0016] Where: h(P) is the convective heat dissipation coefficient (W / (m²)). 2 ·℃), P is the ambient air pressure (kPa), T_t{in-out} is the temperature difference between inside and outside the cabinet (℃), and a, b, c, d are the fitting parameters for high-altitude scenarios;

[0017] Dynamically calculate the total heat load inside the cabinet:

[0018] Q total =Q elec +h(P)·S·(T in -T out )-Q pcm

[0019] In the formula, Qelec=I2*Relec, Relec is the equivalent resistance of the element, S is the heat dissipation area, and Qpcm is the heat absorbed by the phase change material.

[0020] Furthermore, in step S1, the heat dissipation strategy optimization based on model predictive control adopts model predictive control with the objective function of stabilizing the cabinet temperature at 40-55℃ and minimizing energy consumption, and outputs the ventilation opening degree θ (0-100%), micro-positive pressure air supply Vair, and phase change material auxiliary heat dissipation power.

[0021] Furthermore, the multi-source heterogeneous data preprocessing in step S3 includes normalizing the high-altitude features of the collected data, correcting the insulation resistance and calculating the temperature change rate, and removing high-altitude environmental noise through adaptive wavelet denoising (three decomposition layers, threshold λ = 2lnN, where N is the amount of data).

[0022] Furthermore, the transfer learning and ensemble learning prediction models in step S3 include:

[0023] Domain-adaptive transfer learning aligns fault samples from the plains (source domain) with those from the high-altitude (target domain) through adversarial training, minimizing the inter-domain difference loss.

[0024] L total =L task +α·L domain

[0025] Where Ltask is the fault classification loss (cross-entropy loss), Ldomain is the domain difference loss (MMD distance), and α = 0.3 is the balance coefficient;

[0026] Stacked integration model. At the bottom layer, random forest (50 decision trees) and XGBoost (learning rate 0.1, tree depth 5) are used to extract local features. At the top layer, an attention mechanism LSTM (hidden layer dimension 64, number of iterations 100) is used to focus on key features (gradual change of insulation resistance under low pressure, SPD leakage current fluctuation), and output the fault type (insulation breakdown / heat dissipation failure / condensation corrosion) and the remaining useful life (RUL).

[0027] Further, the dynamic threshold warning mechanism in step S3 is based on the real-time air pressure P and the temperature difference ΔT = Tin - Tout.

[0028] Further, the temperature trend prediction in step S4 uses a bidirectional LSTM + attention mechanism model. The past 72-hour temperature sequence is input to predict the cabinet temperature Tpred(t) in the next 12 hours. Model structure: input layer dimension 6 (Tin, Tout, P, H, V, ΔTrate), 2 hidden layers (64 neurons in each layer), attention weight ωi = exp(si) / ∑exp(si), where: si is the feature importance score, and the prediction error ≤ ±2°C.

[0029] Further, in the phase change material thermal state coupling control in step S4, the remaining latent heat of the phase change material is calculated in real time

[0030]

[0031] Qtotal is the total latent heat of the phase change material, m is the mass, and cpcm is the specific heat of phase change

[0032] When Tpred(t + 1)>45°C (temperature increase in the next 1 hour): Open the ventilation port for pre-cooling 1 hour in advance, and the opening degree increases from 30% to 60% to maximize the storage of low temperature from the outside

[0033] When Tpred(t + 2)<15°C (temperature decrease in the next 2 hours): Close the ventilation port, and use the Qrem released by the phase change material to maintain the cabinet temperature>10°C

[0034] When DeltaTpred = Tpred-max - Tpred-min>25°C (sharp temperature difference change): Link the dehumidifier to start and stop intermittently (run for 10 min / stop for 20 min) to avoid condensation

[0035] Further, in the condensation risk prediction in step S4, a dew point temperature model is introduced

[0036]

[0037] where, a is 17.27 and b is 237.7. When Tpred(t)<Tdew + 2°C, start the heat preservation strategy in advance

[0038] Furthermore, in step S5, feedback iterative optimization is performed by collecting actuator feedback data in real time (such as the actual value of Tin and the accuracy of fault prediction), updating the coupled model parameters a / b / c / d and the prediction model weights every 24 hours, iteratively optimizing the control strategy, and adapting to the dynamic changes in the high-altitude environment.

[0039] Compared with existing technologies, the beneficial effects of the UAV inspection docking method and device provided in this application are:

[0040] Compared to conventional heat dissipation, this method can be used in low-pressure environments to improve heat dissipation efficiency and control temperature fluctuations within the cabinet. Given the unreliable transportation and communication in high-altitude areas, resulting in higher maintenance costs and greater potential losses, this method can improve fault prediction accuracy, reduce false alarm rates, and provide earlier warnings, significantly lowering maintenance costs in high-altitude regions. This method employs a temperature control algorithm. Considering the rapid temperature changes in high-altitude areas, this method can adopt different countermeasures based on actual conditions to reduce temperature fluctuations. Simultaneously, it incorporates condensation risk prediction to reduce the occurrence of condensation, preventing component corrosion and short-circuit faults. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a schematic diagram of the process of the present invention; Detailed Implementation

[0043] To make the technical problems, technical solutions, and beneficial effects to be solved by this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of this application.

[0044] It should be noted that when a component is referred to as being "fixed to" or "set on" another component, it can be directly on or indirectly on that other component. When a component is referred to as being "connected to" another component, it can be directly connected to or indirectly connected to that other component.

[0045] It should be understood that the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0046] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" or "several" means two or more, unless otherwise explicitly specified.

[0047] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0048] Please refer to the following: Figure 1 As shown below, an intelligent control algorithm for high-altitude grid-connected cabinets provided in this application embodiment will be described. This invention provides an intelligent control algorithm for high-altitude grid-connected cabinets, specifically addressing low-pressure heat dissipation compensation, fault prediction, and temperature difference adaptive control for grid-connected cabinets in environments with altitudes ≥2000m. This method employs a five-layer closed-loop architecture of "perception-modeling-decision-execution-feedback," achieving intelligent management and control of high-altitude grid-connected cabinets through multi-source data acquisition, high-altitude-specific model construction, and dynamic decision control.

[0049] In one embodiment of the present invention, the grid-connected cabinet of a photovoltaic power station at an altitude of 3500m includes a data acquisition module, an adaptive compensation algorithm for low air pressure heat dissipation efficiency, a fault prediction algorithm for high altitude multi-source feature fusion, a forward-looking temperature control algorithm for drastic temperature changes, and feedback iterative optimization.

[0050] The data acquisition module can collect:

[0051] Environmental parameters include: ambient air pressure, internal cabinet temperature, and external cabinet temperature;

[0052] Electrical parameters, including: insulation resistance and bus current;

[0053] Equipment status parameters, including the residual latent heat of the phase change material, the opening degree of the vent, and the frequency of micro-positive pressure air supply;

[0054] The raw data was collected at a sampling frequency of 1Hz, and then denoised and normalized to generate a standard dataset.

[0055] The specific sensors selected include a barometric pressure sensor (measurement range 50-110 kPa, accuracy ±0.5 kPa), a temperature and humidity sensor (measurement range -40~85℃, ±0.3℃; 0-100%RH, ±2%RH), an insulation resistance sensor (measurement range 0-1000 MΩ, accuracy ±1%), and a current sensor (measurement range 0-1000 A, accuracy ±0.2%). The actuators selected include an electric vent (opening adjustment accuracy 5%) and a paraffin-based composite phase change material.

[0056] An adaptive compensation algorithm for low-pressure heat dissipation efficiency includes constructing a high-altitude-specific air pressure heat dissipation efficiency coupling model and optimizing the heat dissipation strategy based on model predictive control.

[0057] The high-altitude-specific air pressure heat dissipation efficiency coupling model is based on measured data at altitudes of 2000-5000m (covering air pressure of 55-80kPa and temperature difference of 5-35℃), and uses an improved BP neural network + Gaussian process regression to train the nonlinear fitting model.

[0058] Given the current ambient air pressure (P = 65) kPa, we calculate h(P) = 12.3 W / (m²). 2 ·℃)<15W / (m 2 (℃), the phase change material is activated to absorb heat, the vent opening is adjusted to 80%, and Tin stabilizes at 48℃.

[0059] Dynamic calculation of total heat load inside the cabinet: Equivalent resistance of electrical components Relec = 0.001Ω, Qelec = I 2 ×Relec=300 2 ×0.001=90W; Heat absorption of phase change material Qpcm=m×cpcm×Tpcm-Tphase change=2kg×2100J / (kg·℃)×(48-50)=-8400J=-8.4kJ (the negative sign indicates heat release, and heat absorption actually occurs at Tpcm=52℃); Correct Tpcm=52℃, then Qpcm=2×2100×(52-50)=8400J=8.4kJ; Total heat load Qtotal=90W+13.95×1.2×20.8-8400J / h≈90+13.95×24.96-2.33W≈90+348.2-2.33≈435.87W.

[0060] The heat dissipation strategy optimization based on model predictive control (MPC) adopts model predictive control (MPC) with the objective function of stabilizing the cabinet temperature at 40-55℃ and minimizing energy consumption. The output is the ventilation opening θ (0-100%), the micro-positive pressure air supply Vair, and the auxiliary heat dissipation power of phase change material.

[0061] Objective function: minJ=(λ1×Tin-48)2+λ2×(θ / 100+f / 30) (λ1=0.8, λ2=0.2, target temperature 8℃); Constraints: θ∈[0,100], f∈[5,30]; Solved by gradient descent: θ=80% (opening increases from 30% to 80%), f=20 times / h (air replenishment frequency increases from 10 times / h to 0 times / h); Execution command: Drive the stepper motor of the vent to rotate at a speed of 10° / s, increasing the opening from 30% (corresponding angle 108°) to 80% (corresponding angle 288°); Start the micro-positive pressure air compressor, with air replenishment Vair=0.5×(48-45.8)×10-3=0.0011m 3 / min, each replenishment lasts 10s, with a 3min interval.

[0062] Fault prediction algorithms based on multi-source feature fusion at high altitudes include:

[0063] Step 1: Feature vector construction. Input feature vector X = [P, ΔT, R-normalized, I, Ispd, N, Qrem] to normalize the high-altitude features of the collected data, correct insulation resistance, and calculate the temperature change rate. Remove high-altitude environmental noise through adaptive wavelet denoising (3 decomposition layers, threshold λ = 2lnN, where N is the amount of data).

[0064] Step 2: Transfer learning model inference source domain (plain) model output initial fault probabilities: insulation breakdown 0.15, heat dissipation failure 0.08, condensation corrosion 0.05; domain adaptive adversarial training correction: high altitude air pressure correction coefficient α=0.3, after correction fault probabilities: insulation breakdown 0.15×(1-0.3×(80-65) / 20)=0.15×0.775=0.116; heat dissipation failure 0.08×(1+0.2×ΔT / 20)=0.08×1.208=0.097; condensation corrosion 0.05×(1+0.1×H / 50)=0.05×1.12=0.056.

[0065] Step 3: Dynamic threshold judgment of insulation resistance warning threshold R threshold = 100 × [1 - 0.1 × (80 - 65) / 10 - 0.05 × 20.8 / 10] = 100 × (1 - 0.15 - 0.104) = 74.6 MΩ; the current R normalization = 100 MΩ > 74.6 MΩ, the failure probability is < 85%, RUL calculation: RUL = 100 × (1 - sum failure probability) = 100 × (1 - 0.269) = 73.1 days ≈ 73 days, no warning.

[0066] The proactive temperature control algorithm for drastic temperature changes mainly includes the following steps:

[0067] Step 1: Temperature trend prediction. Input the temperature sequence of the past 72 hours (e.g., t-72h: Tin = 40℃, t-48h: 42℃, t-24h: 45℃, current temperature 45.8℃); the bidirectional LSTM model outputs the predicted values ​​for the next 12 hours: t+1h: 47℃, t+3h: 49℃, t+6h: 50℃ (peak), t+12h: 18℃ (nighttime low).

[0068] Step 2: Thermal state control of phase change material. The residual latent heat of the phase change material, Qrem, is 41.6 kJ. Heat needs to be stored to cope with the peak temperature. Instruction: The phase change material cooling fan is turned off (to maximize heat absorption). Tpcm needs to rise to 55℃ before t+3h. The absorbed heat is Qpcm = 2 × 2100 × (55-50) = 21 kJ, and Qrem = 50-21 = 29 kJ.

[0069] Step 3: Dew point temperature calculation for condensation risk prediction: Tdew=237.7×(17.27×45.8 / (237.7+45.8)+ln(60 / 100)) / (17.27-(17.27×45.8 / (237.7+45.8)+ln(0.6); Calculate the intermediate value: 17.27×45.8 / 283.5≈791.97 / 283.5≈2.79,\(ln(0.6)≈-0.51; Numerical value: 237.7×(2.79-0.51)=237.7×2.28≈542.96; Dew point temperature calculation: Tdew=237.7×(17.27×45.8 / (237.7+45.8)+ln(60 / 100)) / (17.27-(17.27×45.8 / (237.7+45.8)+ln(0.6))); Calculate the intermediate value: 17.27×45.8 / 283.5≈791.97 / 283.5≈2.79, ln(0.6)≈-0.51; Numerical value: 237.7×(2.79-0.51)=237.7×2.28≈542.96; Tdew=38.2.

[0070] Feedback iteration optimization uses real-time acquisition of actuator feedback data. Post-execution feedback data: Tin = 47℃ (target 48℃, deviation -1℃), measured h(P) = 12.3W / (m³). 2 ·℃)(Model predicts 13.95W / (m 2 •℃), deviation 1.65); Parameter update: the coupling model a was adjusted from 0.002 to 0.0018, b was adjusted from 2 to 2.1, and after correction h(P) = 0.0018 × 65 2 -0.3×65+25=8.265-19.5+25=13.765 (closer to the measured value).

[0071] It is understood that the parts in the above embodiments can be freely combined or deleted to form different combined embodiments. The specific contents of each combined embodiment will not be repeated here. After this description, it can be considered that the present invention specification has recorded each combined embodiment and can support different combined embodiments.

[0072] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent control algorithm for high-altitude grid-connected cabinets, characterized in that, Includes the following steps: The S1 data acquisition module includes environmental parameters, electrical parameters, and equipment parameters. The sampling frequency is 1Hz, and the data is stored after adaptive wavelet denoising processing. The S2 low-pressure heat dissipation efficiency adaptive compensation algorithm includes constructing a high-altitude air pressure heat dissipation efficiency coupling model and optimizing the heat dissipation strategy based on model prediction control. The S3 high-altitude multi-source feature fusion fault prediction algorithm includes multi-source heterogeneous data preprocessing, transfer learning and ensemble learning fault prediction models, and dynamic threshold early warning mechanism; S4 temperature change anti-theft temperature control algorithm, including temperature trend prediction, phase change material thermal state coupling control and condensation risk prediction; S5 feedback iterative optimization.

2. The intelligent control algorithm for high-altitude grid-connected cabinets according to claim 1, characterized in that, The high-altitude-specific air pressure heat dissipation efficiency coupling model in step S2 is based on measured data at altitudes of 2000-5000m, and uses an improved BP neural network and Gaussian process regression to train a nonlinear fitting model. h(P)=a.P b +c·(T in -T out )+d Where: h(P) is the convective heat dissipation coefficient W / (m²) 2 ·℃), P is the ambient air pressure (kPa), T_t{in-out} is the temperature difference between inside and outside the cabinet (℃), and a, b, c, d are the fitting parameters for high-altitude scenarios; Dynamically calculate the total heat load inside the cabinet: Q total =Q elec +h(P)·S·(T in -T out )-Q pcm In the formula, Qelec=I2*Relec, Relec is the equivalent resistance of the element, S is the heat dissipation area, and Qpcm is the heat absorbed by the phase change material.

3. The intelligent control algorithm for high-altitude grid-connected cabinets according to claim 1, characterized in that, The heat dissipation strategy optimization based on model predictive control in step S1 adopts model predictive control, with the objective function being to stabilize the cabinet temperature at 40-55℃ and minimize energy consumption. The output is the ventilation opening θ (0-100%), the micro-positive pressure air supply Vair, and the auxiliary heat dissipation power of the phase change material.

4. The intelligent control algorithm for high-altitude grid-connected cabinets according to claim 1, characterized in that, The multi-source heterogeneous data preprocessing in step S3 includes normalizing the high-altitude features of the collected data, correcting the insulation resistance and calculating the temperature change rate, and removing high-altitude environmental noise through adaptive wavelet denoising.

5. The intelligent control algorithm for high-altitude grid-connected cabinets according to claim 1, characterized in that, The transfer learning and ensemble learning prediction models in step S3 include: Domain-adaptive transfer learning aligns fault samples from plains conditions with those from high-altitude conditions through adversarial training, minimizing the loss due to inter-domain discrepancies. L total =L task +α·L domain Where Ltask is the fault classification loss, Ldomain is the domain difference loss, and α = 0.3 is the balance coefficient; The stacked ensemble model uses random forest and XGBoost to extract local features at the bottom layer and LSTM with attention mechanism at the top layer to focus on key features, outputting fault type and remaining lifetime.

6. The intelligent control algorithm for high-altitude grid-connected cabinets according to claim 1, characterized in that, The dynamic threshold early warning mechanism in step S3 is based on real-time air pressure P and temperature difference ΔT = Tin - Tout.

7. The intelligent control algorithm for high-altitude grid-connected cabinets according to claim 1, characterized in that, The temperature trend prediction in step S4 uses a bidirectional LSTM and attention mechanism model. The model inputs the temperature sequence of the past 72 hours and predicts the temperature inside the cabinet for the next 12 hours. The model structure is as follows: input layer dimension 6, hidden layer 2, attention weight ωi=exp(si) / ∑exp(si), where si is the feature importance score, and the prediction error is ≤±2℃.

8. The intelligent control algorithm for high-altitude grid-connected cabinets according to claim 1, characterized in that, In step S4, the thermal state of the phase change material is coupled and controlled, and the remaining latent heat of the phase change material is calculated in real time. Qtotal represents the total latent heat of the phase change material, m is the mass, and cpcm is the specific heat capacity of the phase change. When Tpred(t+1)>45℃: Open the vents 1 hour in advance to preheat and dissipate heat, increasing the opening from 30% to 60% to maximize the storage of low external temperatures; When Tpred(t + 2) < 15°C: Close the vent and use the Qrem released by the phase change material to maintain the temperature inside the cabinet > 10°C; When DeltaTpred = Tpred - max - Tpred - min > 25°C: Link the dehumidifier to start and stop intermittently to avoid condensation.

9. The intelligent control algorithm for high-altitude grid-connected cabinets according to claim 1, characterized in that, In the prediction of condensation risk in step S4, a dew point temperature model is introduced. Among them, a is 17.27 and b is 237.

7. When Tpred(t) < Tdew + 2°C, the heat preservation strategy is started in advance.

10. The intelligent control algorithm for high-altitude grid-connected cabinets according to claim 1, characterized in that, In the feedback iterative optimization in step S5, the feedback data of the actuator is collected in real time, and the coupling model parameters a / b / c / d and the prediction model weights are updated every 24 hours to iteratively optimize the control strategy and adapt to the dynamic changes of the high altitude environment.

Images