A composite heat dissipation method for a transformer suitable for high overload working conditions

Abstract

The application relates to the technical field of transformer heat management and cooling control, and particularly discloses a transformer composite heat dissipation method suitable for high overload working conditions, which first collects high-frequency three-phase currents of a primary side of a transformer and top-layer and bottom-layer insulating oil temperatures, obtains an electro-thermal coupling transient characteristic through transient characteristic extraction and time scale fusion; then predicts a high-overload heat evolution track and a heat transfer delay time window by using a pre-trained model, and quantifies a feedforward cold quantity demand in the time window; then distributes the cold quantity demand into a main pump driving reference and a bypass valve opening instruction according to dynamic capacity of a main frequency pump and response characteristics of an auxiliary cooling bypass; and finally performs closed-loop compensation and pulse modulation on the basis of a deviation between an expected oil temperature change and a real-time top-layer oil temperature, so that the main pump and the auxiliary bypass cooperatively establish a cooling oil flow in advance. The scheme can reduce heat dissipation delay caused by oil temperature feedback, heat transfer inertia and responses of an actuator, and improve heat peak suppression capacity under transient high overload.

Description

Technical Field

[0001] This application relates to the field of transformer thermal management and cooling control technology, and more specifically, to a composite heat dissipation method for transformers suitable for high overload conditions. Background Technology

[0002] With the centralized grid connection of new energy sources, the large-scale charging of electric vehicles, and the increase in impactful industrial loads, transformer loads are characterized by sudden onset, large increases, and high harmonic content. Transient high overloads can significantly increase winding losses in a short period of time, causing the temperature of winding hot spots to rise rapidly. If the generated heat cannot be dissipated in time, it can easily accelerate the aging of insulating paper and the deterioration of insulating oil, reducing the insulation safety margin and service life of the transformer. Therefore, it is necessary to improve heat dissipation capacity through the cooperation of main cooling circuits and auxiliary cooling circuits.

[0003] Existing transformer composite heat dissipation schemes typically collect data on top-layer oil temperature, winding temperature, or load rate, and increase the main oil pump speed or activate the auxiliary cooling circuit after the corresponding parameters reach set thresholds. However, the Joule heating caused by sudden current changes occurs in milliseconds to seconds, and the heat transfer from the windings through solid insulation to the insulating oil to form a measurable oil temperature change requires conduction, convection, and oil circulation processes, exhibiting significant thermal inertia. When oil temperature feedback is used as the primary control basis, the control system only increases cooling output after detecting a temperature rise, by which time hot spots in the windings may have already accumulated rapidly. Simultaneously, the acceleration of the main oil pump must overcome rotational inertia, oil viscosity, and pipeline resistance, further delaying the establishment of effective cooling oil flow and making it difficult to suppress the initial temperature peak during high overload. Furthermore, the main oil pump and auxiliary bypass typically share inlet and outlet oil pipelines. High-speed operation of the main oil pump alters the pipeline dynamic pressure and branch pressure differential, causing the actual flow rate and valve response characteristics of the auxiliary bypass to vary with the main pump's operating conditions. If the existing solution still determines the opening of the auxiliary bypass valve according to fixed parameters, it is easy to cause a discrepancy between the nominal cooling output and the actual cooling capacity delivered. In particular, the auxiliary cooling capacity may be insufficient when the main pump is running at high speed, thereby reducing the effect of the composite heat dissipation system on suppressing transient heat peaks.

[0004] Therefore, an optimized composite heat dissipation solution for transformers suitable for high overload conditions is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a composite heat dissipation method for transformers suitable for high overload conditions.

[0006] According to one aspect of this application, a composite heat dissipation method for transformers suitable for high overload conditions is provided, comprising: S1, transient feature extraction and time-scale fusion of the high-frequency three-phase current sequence of the primary side of the transformer, the real-time temperature value of the top insulating oil and the real-time temperature value of the bottom insulating oil are performed to obtain the electrothermal coupling transient feature tensor; S2, based on a pre-trained spatiotemporal thermal momentum mapping model, performs electrothermal cross-domain mapping and heat transfer hysteresis identification on the electrothermal coupling transient feature tensor to obtain the spatial virtual thermal impulse trajectory sequence and heat transfer hysteresis time window prediction value characterizing the high overload heat evolution. S3, perform preemptive feedforward cooling demand time-series quantization on the spatial virtual thermal impulse trajectory sequence to obtain the preemptive feedforward cooling demand matrix. S4, based on the dynamic capability curve of the main variable frequency pump and the step response model of the auxiliary cooling bypass, performs composite hardware asynchronous decoupling and control benchmark allocation on the preemptive feedforward cooling demand matrix to obtain the main pump over-frequency drive benchmark signal and the auxiliary bypass valve opening benchmark command. S5 calculates the expected oil temperature change based on the main pump overfrequency drive reference signal and the auxiliary bypass valve opening reference command. It then uses the deviation between the expected oil temperature change and the real-time temperature value of the top insulating oil to perform closed-loop compensation and pulse modulation to drive the main frequency converter pump and the auxiliary bypass valve to collaboratively establish a preemptive cooling oil flow within the predicted value of the heat transfer hysteresis time window.

[0007] Compared with existing technologies, this application provides a composite heat dissipation method for transformers under high overload conditions. It utilizes the timing characteristic that changes in electrical quantities precede changes in oil temperature during high overload, integrates transient transformer load information and insulating oil thermal state, predicts the heat evolution trend and heat transfer lag time, and quantifies the required cooling capacity before a significant rise in measurable oil temperature. It coordinates the main frequency converter pump and auxiliary cooling bypass to establish cooling oil flow in advance to compensate for temperature feedback lag, heat transfer inertia, and main pump start-up lag. Simultaneously, based on the impact of the main pump's operating status on the auxiliary bypass flow rate and response characteristics in the shared pipeline, it dynamically corrects the cooling capacity distribution relationship between the main pump and bypass valves, and performs closed-loop adjustments based on actual oil temperature changes. This reduces the deviation between nominal cooling output and actual cooling capacity delivery, improving the ability to suppress heat peaks in the initial stage of transient high overload. Attached Figure Description

[0008] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0009] Figure 1 This is a flowchart of a composite heat dissipation method for transformers applicable to high overload conditions according to an embodiment of this application; Figure 2 This is a data flow diagram of a composite heat dissipation method for transformers applicable to high overload conditions according to an embodiment of this application; Figure 3 This is a flowchart of step S4 of a composite heat dissipation method for transformers under high overload conditions according to an embodiment of this application; Figure 4 This is a flowchart of step S43 of a composite heat dissipation method for transformers under high overload conditions according to an embodiment of this application; Figure 5 This is a flowchart of step S5 of a composite heat dissipation method for transformers applicable to high overload conditions according to an embodiment of this application. Detailed Implementation

[0010] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0011] To address the problems of existing composite heat dissipation methods that rely on the oil temperature to rise before starting cooling, the lag in establishing effective cooling oil flow, and the deviation in auxiliary bypass cooling capacity delivery caused by the high-speed operation of the main pump changing the pressure difference of the shared pipeline network, this application proposes a composite heat dissipation method for transformers suitable for high overload conditions. This scheme first extracts transient features and fuses time scales of the high-frequency three-phase current, top oil temperature, and bottom oil temperature on the primary side of the transformer. Based on this, it identifies the evolution trajectory of high overload heat and the heat transfer hysteresis time window. Then, before the oil temperature rises significantly, it calculates the thermal overshoot within the hysteresis time window, converts and discretizes it into feedforward cooling demand. Then, according to the low-frequency fundamental component and high-frequency transient component of the cooling demand, it generates the main pump overfrequency drive reference signal and the auxiliary bypass valve opening reference command, respectively. When generating the bypass valve command, it quantifies the hydraulic resistance coupling effect of the main pump speed on the bypass pressure difference and valve response online, and corrects the bypass control parameters accordingly. Finally, it combines the deviation between the expected oil temperature change and the real-time top oil temperature to perform closed-loop compensation and pulse modulation on the pump valve command, so that the main pump and the auxiliary bypass can establish a cooling oil flow that matches the actual cooling demand in advance and in coordination within the predicted hysteresis time window.

[0012] Figure 1 This is a flowchart of a composite heat dissipation method for transformers applicable to high overload conditions, according to an embodiment of this application. Figure 2 A schematic diagram of data flow for a composite heat dissipation method for transformers under high overload conditions according to an embodiment of this application is shown below. Figure 1 and Figure 2As shown, the transformer composite heat dissipation method applicable to high overload conditions according to an embodiment of this application includes the following steps: S1, extracting transient features and time-scaled fusing the high-frequency three-phase current sequence of the primary side of the transformer, the real-time temperature value of the top insulating oil, and the real-time temperature value of the bottom insulating oil to obtain an electrothermal coupling transient feature tensor; S2, based on a pre-trained spatiotemporal thermal momentum mapping model, performing electrothermal cross-domain mapping and heat transfer hysteresis identification on the electrothermal coupling transient feature tensor to obtain a spatial virtual thermal impulse trajectory sequence characterizing the heat evolution under high overload and a predicted value of the heat transfer hysteresis time window; S3, performing preemptive feedforward cooling demand timing on the spatial virtual thermal impulse trajectory sequence. Quantization is performed to obtain the preemptive feedforward cooling demand matrix; S4, based on the dynamic capability curve of the main variable frequency pump and the step response model of the auxiliary cooling bypass, the preemptive feedforward cooling demand matrix is ​​subjected to composite hardware asynchronous decoupling and control benchmark allocation to obtain the main pump over-frequency drive benchmark signal and the auxiliary bypass valve opening benchmark command; S5, based on the main pump over-frequency drive benchmark signal and the auxiliary bypass valve opening benchmark command, the expected oil temperature change is calculated, and the deviation between the expected oil temperature change and the real-time temperature value of the top insulating oil is used for closed-loop compensation and pulse modulation to drive the main variable frequency pump and the auxiliary bypass valve to collaboratively establish preemptive cooling oil flow within the predicted value of the heat transfer hysteresis time window.

[0013] Specifically, in step S1, transient features are extracted and time-scaled fused from the high-frequency three-phase current sequence of the transformer primary side, the real-time temperature value of the top insulating oil, and the real-time temperature value of the bottom insulating oil to obtain an electrothermal coupling transient feature tensor. It should be understood that, due to the simultaneous operation of multiple charging devices in a centralized electric vehicle charging station, the transformer primary side current undergoes a step change in a short period. Rectifier and power conversion equipment also introduce harmonic currents. Furthermore, winding losses require heat conduction through solid insulation and convection through insulating oil to manifest as oil temperature changes. Therefore, the primary side current transient and insulating oil temperature respectively reflect the heat source input and the current thermal state, and the sampling frequency and response speed of the two types of data differ. Therefore, in the technical solution of this application, by extracting transient features and time-scaled fused from the high-frequency three-phase current sequence of the transformer primary side, the real-time temperature value of the top insulating oil, and the real-time temperature value of the bottom insulating oil to obtain an electrothermal coupling transient feature tensor, the intensity of load mutation, the degree of harmonic-induced additional heating, and the vertical thermal distribution state of the oil tank are organized into unified time-series data. This avoids incorrect combinations of data from different sampling times, enabling subsequent spatiotemporal thermal momentum mapping models to identify the generation trend of high overload heat and its evolution relative to the current oil temperature state under the same time reference.

[0014] More specifically, in the embodiments of this application, step S1 includes: S11, solving for the transient current mutation rate of the high-frequency three-phase current sequence on the primary side of the transformer to obtain the current mutation rate sequence; S12, performing frequency domain nonlinear harmonic analysis on the high-frequency three-phase current sequence on the primary side of the transformer to obtain the energy distribution matrix of each harmonic; S13, calculating the vertical thermal gradient value between the real-time temperature value of the top insulating oil and the real-time temperature value of the bottom insulating oil; S14, based on a unified timestamp reference, performing time-series alignment and multi-dimensional splicing of the current mutation rate sequence, the energy distribution matrix of each harmonic, and the vertical thermal gradient value to obtain the electrothermal coupling transient feature tensor.

[0015] Specifically, in one example of this application, the transformer is an oil-immersed transformer installed in a centralized charging station for electric vehicles. Wideband current transformers are installed at the three-phase input lines on its primary side, and temperature sensors are installed in the upper oil return area and the lower oil inlet area of ​​the tank. The three-phase current is synchronously acquired by an isolation acquisition module at a sampling frequency of 10 kHz, and the temperature sensors are synchronously acquired at a sampling frequency of 1 Hz. The acquisition module and the controller use the same system clock to write the sampling timestamp. The controller uses a feature update period of 100 milliseconds and saves the data of the most recent 5 seconds, thus forming 50 consecutive feature moments. The above sampling frequency is sufficient to cover the fundamental frequency and the target harmonics. The current transformers, temperature sensors, analog-to-digital converters, and controllers used can all be directly configured and implemented according to their respective ranges and communication interfaces.

[0016] In practice, the controller first performs range conversion on the high-frequency three-phase current sequence on the primary side of the transformer, converting the sampled values ​​from the secondary side of the current transformer into primary side current values, and uses a five-point moving average to filter out high-frequency measurement noise generated by analog-to-digital conversion. Then, it calculates the change in current sampled values ​​of each phase relative to the sampling time interval according to the sampling time sequence, obtaining the current mutation rate sequence. To ensure that the current mutation rate sequence corresponds to a 100-millisecond feature update cycle, the controller extracts the root mean square value of the current mutation rate for phases A, B, and C, as well as the signed value with the largest absolute value, within each update cycle. The former represents the overall intensity of load fluctuation within that cycle, and the latter represents the amplitude and direction of the most drastic current change. When multiple charging devices are started simultaneously, the above processing can record the degree of sudden increase in primary side load before the oil temperature changes significantly.

[0017] Next, the controller uses a 200-millisecond frequency domain analysis window and updates it in 100-millisecond steps, sequentially performing DC component removal, Hanning window weighting, and Fast Fourier Transform on the high-frequency three-phase current sequence of the primary side of the transformer for each phase. The 200-millisecond analysis window contains an integer number of fundamental cycles in the 50 Hz power grid, thereby reducing spectral leakage caused by non-integer cycle truncation. Based on the fundamental frequency, the controller locates the frequency positions corresponding to the 3rd, 5th, 7th, 9th, 11th, and 13th harmonics, extracts the corresponding spectral amplitudes, performs square energy conversion, and then arranges the processing results according to characteristic time, phase, and harmonic order to obtain the energy distribution matrix for each harmonic. This matrix not only reflects the changes in primary side current amplitude but also the additional harmonic losses caused by the power conversion process of the charging equipment, distinguishing the heating differences corresponding to different degrees of current distortion under the same load current.

[0018] Then, the real-time temperature values ​​of the top and bottom insulating oils are read according to the temperature sampling timestamps. The difference between the two values ​​is divided by the fixed vertical distance between the two temperature sensors to obtain the vertical thermal gradient value. An increased vertical thermal gradient value indicates that the heat accumulation in the upper part of the oil tank is stronger than that in the lower cooling oil; a smaller vertical thermal gradient value indicates that the temperature difference between the upper and lower insulating oil layers is lower. In the case of multiple rounds of concentrated charging occurring consecutively, this feature enables the controller to distinguish between the initial load surge when the oil temperature is low and the subsequent load surge when heat has already accumulated inside the oil tank.

[0019] Furthermore, the controller uses the characteristic moments generated every 100 milliseconds as a unified timestamp reference. For the current mutation rate sequence, the controller maps the root mean square value and the signed value with the largest absolute value within the corresponding 100-millisecond period to the characteristic moment of that period; for each harmonic energy distribution matrix, the controller uses the center moment of the frequency domain analysis window as the corresponding characteristic moment; for the vertical thermal gradient value, the controller uses the most recent valid value to maintain it until a new temperature sample value arrives, so that online processing does not depend on future temperature data. After completing the timing alignment, the controller splices the current mutation rate characteristics, each harmonic energy, and the vertical thermal gradient value of each phase in a fixed order of phases A, B, and C, where the vertical thermal gradient value is written into the characteristic position of each phase as a three-phase shared thermal state; then, the mean and standard deviation saved during the model training phase are used to standardize various features, and the data of the most recent 50 characteristic moments are arranged according to the time dimension, phase dimension, and feature dimension to obtain the electrothermal coupling transient feature tensor.

[0020] Specifically, in step S2, based on the pre-trained spatiotemporal thermal momentum mapping model, the electrothermal coupling transient characteristic tensor is subjected to electrothermal cross-domain mapping and heat transfer hysteresis identification to obtain the spatial virtual thermal impulse trajectory sequence and predicted value of the heat transfer hysteresis time window characterizing the evolution of high overload heat. It should be understood that when multiple charging devices are simultaneously put into operation in a centralized electric vehicle charging station, the primary current and harmonic losses of the transformer will change in the millisecond to second range. The heat generated by the windings needs to undergo solid insulation heat conduction, insulating oil heat absorption, and oil convection before it manifests as a measurable oil temperature change. There is a nonlinear mapping relationship and a transfer time difference between the electrical transient and the thermal state. The spatial distribution, growth trend, and available time range for early cooling cannot be determined solely based on the current oil temperature. Therefore, in the technical solution of this application, based on a pre-trained spatiotemporal thermal momentum mapping model, electrothermal coupling transient characteristic tensors are subjected to electrothermal cross-domain mapping and heat transfer hysteresis identification to obtain a spatial virtual thermal impulse trajectory sequence characterizing the high overload heat evolution and a predicted value of the heat transfer hysteresis time window. This converts the current mutation and harmonic energy characteristics that occur before the oil temperature change into information on future heat evolution and determines the time boundary before heat is transferred to the insulating oil and forms an observable temperature rise. In this way, high overload thermal shocks can be identified before the real-time temperature value of the top insulating oil has increased significantly, allowing subsequent steps to quantify the cooling demand in advance within the predicted value of the heat transfer hysteresis time window, reducing the impact of oil temperature feedback lag on the timeliness of cooling control.

[0021] More specifically, in the embodiments of this application, step S2 includes: S21, feeding the electrothermal coupling transient feature tensor into a pre-trained long short-term memory network in time step order to extract multi-timescale temporal dependencies to obtain a cross-domain temporal hidden state vector; S22, performing energy up-dimensional projection and decoupling of inertial features on the cross-domain temporal hidden state vector to obtain a spatial virtual thermal impulse trajectory sequence and flow field hysteresis driving features; S23, using the time domain integral of the flow field hysteresis driving features and the spatial virtual thermal impulse trajectory sequence to calculate the heat transfer critical time to obtain the predicted value of the heat transfer hysteresis time window.

[0022] Specifically, in one example of this application, the pre-trained spatiotemporal thermal momentum mapping model consists of a multi-timescale long short-term memory network, an energy-upgraded projection branch, and an inertial feature decoupling branch. It is pre-trained on a server, and then the trained model parameters, feature-normalized parameters, and network structure files are deployed to the transformer controller. The model training data comes from graded loading tests of transformers in centralized electric vehicle charging stations and electromagnetic, thermal, and fluid co-simulations calibrated by temperature rise tests. The loading tests cover different initial oil temperatures, three-phase load growth rates, load imbalance degrees, and harmonic content. According to the transformer structure, the heat dissipation space is divided into the A-phase winding associated heat zone, the B-phase winding associated heat zone, and the C-phase winding associated heat zone. Training labels for thermal impulse trajectories are established based on the increase in heat power relative to steady-state losses in each heat zone over future time periods. Training labels for heat transfer hysteresis are established based on the time interval from the moment of load abrupt change to the time interval when the top insulating oil temperature change rate continuously exceeds the no-load fluctuation threshold. The training samples are divided into training set, validation set and test set according to the experimental batch. The model parameters are updated by error backpropagation method. The weighted sum of thermal impulse trajectory error and heat transfer hysteresis time error of each thermal zone is used as the training target. Training is stopped when the validation error no longer decreases, thereby obtaining the pre-trained spatiotemporal thermal momentum mapping model.

[0023] In the specific implementation process, firstly, the controller receives the electrothermal coupling transient feature tensor generated in step S1. In this specific example, the electrothermal coupling transient feature tensor contains 50 feature moments formed within the last 5 seconds. Each feature moment includes the three-phase current mutation rate feature, the harmonic energy feature, and the vertical thermal gradient value arranged in a fixed order. The controller uses the feature mean and standard deviation saved during the model training phase to verify the dimensional normalization results of each feature, and restricts feature values ​​that exceed the effective range of the training data to the corresponding boundaries, thereby preventing abnormal model input caused by transient sensor interference. Subsequently, the phase dimension and feature dimension corresponding to each feature moment are expanded into input vectors, while maintaining the temporal order of the 50 input vectors. Then, the 50 input vectors are respectively sent to the short-timescale processing branch and the long-timescale processing branch. The short-timescale processing branch feeds all input vectors sequentially into the first long short-term memory network at 100-millisecond intervals to preserve the rapid changes in current and harmonic energy caused by the synchronous startup of charging equipment. The long-timescale processing branch averages every five adjacent input vectors to form ten input vectors spaced 500 milliseconds apart, which are then fed sequentially into the second long short-term memory network to extract the dependency between continuous load growth and vertical thermal gradient changes within a 5-second timeframe. Both long short-term memory networks update their memory states using fixed gating weights after training. The controller concatenates the hidden states of the two branches at their final moments in a fixed order to obtain a cross-domain temporal hidden state vector. This processing simultaneously preserves short-term electrical impulses and relatively slow changes in the thermal state of the insulating oil, allowing single current spikes and sustained high overloads to form different temporal representations.

[0024] Then, the cross-domain temporal hidden state vector is simultaneously input into the energy upscaling projection branch and the inertial feature decoupling branch. The energy upscaling projection branch maps the cross-domain temporal hidden state vector into the prediction results of three hot zones within the next 120 seconds through a fully connected layer, with a prediction time interval of 1 second. Non-negative activation processing is used to eliminate physically meaningless negative thermal impulse values, and the data is then rearranged into a 120×3 data sequence according to the prediction time and hot zone location to obtain a spatial virtual thermal impulse trajectory sequence. Each element in the spatial virtual thermal impulse trajectory sequence represents the virtual thermal impulse rate of the corresponding winding-associated hot zone relative to the steady-state heat at the corresponding future time, thereby spatially distinguishing the heat differences caused by the load imbalance of phases A, B, and C. The inertial feature decoupling branch extracts thermal inertial information characterized by the vertical thermal gradient, load duration, and historical current changes from the cross-domain temporal hidden state vector through an independent fully connected layer, and obtains the flow field hysteresis driving features through bounded activation processing. The parameters of the two branches are jointly determined during the pre-training phase; after model deployment, only forward inference is performed, and the model weights are not changed online.

[0025] Furthermore, the spatial virtual thermal impulse trajectory sequence is discretely integrated in the time domain. The virtual thermal impulse rates of each predicted moment and each winding-related hot zone are weighted and accumulated according to spatial calibration weights to obtain the virtual heat accumulation value from the current moment to each future predicted moment. Simultaneously, the flow field hysteresis driving characteristics are input into the pre-trained hysteresis threshold mapping layer to obtain the critical heat threshold corresponding to the current thermal state of the insulating oil. The controller finds the predicted moment when the virtual heat accumulation value first reaches the critical heat threshold, and determines the time length between this predicted moment and the current moment as the predicted value of the heat transfer hysteresis time window. The above critical time calculation relationship is expressed as:

[0026]

[0027]

[0028]

[0029] in, This represents the cumulative virtual heat value from the current time to the k-th prediction time. This represents the time interval between adjacent prediction moments, which is 1 second in this specific example; Indicates the predicted time sequence number; This indicates the current prediction time number participating in the cumulative calculation; Indicates the winding-associated hot zone number; This indicates the total number of associated hot zones in the winding, which is 3 in this specific example; This represents the spatial calibration weight of the associated hot zone of the j-th winding. The sum of all spatial calibration weights is 1, and it is determined by the temperature rise test. The virtual thermal impulse rate of the j-th winding associated hot zone in the spatial virtual thermal impulse trajectory sequence at the r-th prediction time is represented. This indicates that a non-negative constraint is applied to the virtual thermal impulse rate; This represents the critical heat threshold corresponding to the formation of an observable oil temperature response under the current thermal state; This indicates the hysteresis-driven characteristics of the flow field; and These represent the weight vector and bias determined by the hysteresis threshold mapping layer during the pre-training phase, respectively. This represents the linear mapping result from the flow field hysteresis-driven characteristics to the critical heat threshold; This represents a smooth, non-negative mapping that constrains the critical heat threshold to a positive value; K represents the total number of prediction times, which is 120 in this specific example. This indicates the prediction time number when the virtual heat accumulation value first reaches the critical heat threshold; This represents the predicted value of the heat transfer hysteresis time window. In the above calculation, The virtual thermal impulse rate generated by the three winding associated hot zones in the future is converted into cumulative heat over time, where the spatial calibration weight is used to reflect the difference in contribution of different hot zones to the temperature rise of the top insulating oil. The heat transfer boundary is determined based on the hysteresis-driven characteristics of the flow field formed by the current vertical thermal gradient and load history. Positive value mapping ensures that the boundary conforms to the physical constraints under the heat dimension. This indicates that the predicted heat has reached a state sufficient to overcome the current heat transfer inertia and cause a macroscopic oil temperature response. The more drastic the increase in centralized charging load, the larger the corresponding value of the spatial virtual heat impulse trajectory sequence, the faster the accumulated virtual heat reaches the critical heat threshold, and the shorter the predicted value of the heat transfer hysteresis time window. When the current thermal state of the oil tank corresponds to a strong transfer hysteresis, the hysteresis threshold mapping layer generates a larger critical heat threshold, and the predicted value of the heat transfer hysteresis time window is correspondingly extended. If the critical heat threshold is not reached within the prediction range, the controller sets the upper limit of the prediction range as the predicted value of the heat transfer hysteresis time window, thereby ensuring that the output has a definite value.

[0030] Specifically, in step S3, the spatial virtual thermal impulse trajectory sequence is subjected to preemptive feedforward cooling demand time-series quantization to obtain a preemptive feedforward cooling demand matrix. It should be understood that since the spatial virtual thermal impulse trajectory sequence reflects the evolution of high overload heat during the future prediction period, it cannot be directly used as the driving force for the main variable frequency pump and auxiliary cooling bypass. Furthermore, the increase in heat varies at different prediction times. If the cooling output is determined solely based on the maximum or average predicted value, insufficient cooling capacity in the early part of the hysteresis window and redundant cooling capacity in the later part may occur. Therefore, in the technical solution of this application, the spatial virtual thermal impulse trajectory sequence is further subjected to preemptive feedforward cooling demand time-series quantization to determine the total amount of heat energy that needs to be offset within the predicted value of the heat transfer hysteresis window. This total amount of heat energy is converted into the equivalent cooling power that the composite heat dissipation system can execute, and the cooling power is preferentially released in the early period before the heat generates a significant oil temperature response. In this way, the high overload heat prediction results can be converted into cooling demand with clear power dimensions and execution time, providing a control basis for the main variable frequency pump and auxiliary cooling bypass to establish cooling oil flow in advance, and reducing the delay in heat peak suppression caused by temperature feedback lag and oil flow circulation inertia.

[0031] More specifically, in the embodiments of this application, step S3 includes: S31, using the predicted value of the heat transfer hysteresis time window as the upper boundary of the integration, integrating the thermal overshoot in the constrained time domain of the spatial virtual thermal impulse trajectory sequence to obtain the total thermal energy overshoot accumulation; S32, based on the specific heat capacity and circulating mass of the insulating oil, performing thermodynamic equivalent power conversion on the total thermal energy overshoot accumulation to obtain the equivalent target cooling power; S33, using the predicted value of the heat transfer hysteresis time window as the time domain period constraint, performing nonlinear time-series discretization on the equivalent target cooling power to obtain the preemptive feedforward cooling demand matrix.

[0032] Specifically, in a specific example of this application, the spatial virtual thermal impulse trajectory sequence includes virtual thermal impulse rates for the A-phase winding associated hot zone, the B-phase winding associated hot zone, and the C-phase winding associated hot zone within the next 120 seconds, with a prediction time interval of 1 second, and the unit of each virtual thermal impulse rate is watts; the predicted value of the heat transfer hysteresis time window is output by step S2 and is constrained to between 1 second and 120 seconds. In implementation, firstly, the controller determines the predicted value of the heat transfer hysteresis time window as the upper limit of integration of the spatial virtual thermal impulse trajectory sequence, and extracts trajectory data within this time range starting from the current moment; when the predicted value of the heat transfer hysteresis time window is not an integer multiple of the prediction time interval, the controller performs linear interpolation on the last two adjacent prediction points to make the endpoint of the last integration interval consistent with the predicted value of the heat transfer hysteresis time window. Subsequently, the controller applies non-negative constraints to the virtual thermal impulse rates of the three winding-related hot zones, removes the negative values ​​representing heat fallback, and calculates the thermal overshoot within each prediction interval using the trapezoidal integral method. These calculations are then aggregated according to spatial calibration weights to obtain the total cumulative thermal overshoot. The calculation relationship is as follows:

[0033] in, This represents the total accumulated thermal energy overshoot, expressed in joules. This indicates the total number of associated hot zones in the winding, which is 3 in this specific example; Indicates the winding-associated hot zone number; represents the spatial calibration weight of the associated hot zone of the j-th winding; N represents the total number of integration intervals within the predicted value of the heat transfer hysteresis time window; k represents the sequence number of the integration interval. and These represent the virtual thermal impulse rates at the start and end points of the k-th integral interval for the associated hot zone of the j-th winding, respectively. This indicates that the positive portion of the virtual thermal impulse rate is retained; This represents the duration of the k-th integration interval, and the sum of the durations of all integration intervals equals the predicted value of the heat transfer hysteresis time window. The trapezoidal integral part of the formula converts the continuously changing virtual thermal impulse rate into thermal energy within each time interval, while the spatial calibration weight reflects the difference in the contribution of heat generated by different winding-related hot zones to the overall thermal state of the tank. When the three-phase charging load is unbalanced, the virtual thermal impulse rate corresponding to the higher load increases, and its contribution to the total thermal energy overshoot accumulation increases accordingly. Taking the calculation result of a heat transfer hysteresis time window prediction value of 40 seconds as an example, the weighted integral result of the three winding-related hot zones within this time window is 1.968 MJ, then the controller determines 1.968 MJ as the total thermal energy overshoot accumulation of this round of high overload process.

[0034] Next, the pre-stored specific heat capacity and circulation mass of the insulating oil are read. The specific heat capacity uses product parameters corresponding to the type of insulating oil used in the transformer. The circulation mass is obtained by multiplying the effective oil circuit volume participating in forced oil circulation by the density of the insulating oil at the design temperature, and is calibrated through an oil circuit circulation test at rated pump speed. The controller first converts the total thermal energy overshoot accumulation into the equivalent temperature rise of the circulating insulating oil without increased cooling output, and then calculates the total equivalent target cooling power required by the composite heat dissipation system based on the predicted value of the heat transfer hysteresis time window and the cold energy transfer efficiency. The calculation relationship is as follows:

[0035]

[0036] in, This represents the equivalent temperature rise of the insulating oil corresponding to the total thermal energy overshoot accumulation, in Kelvin. This indicates the cumulative amount of total thermal energy overshoot; This indicates the circulating mass of the insulating oil, expressed in kilograms. This indicates the specific heat capacity of insulating oil, expressed in joules per kilogram (Kelvin). This represents the total equivalent target cooling power, expressed in watts. This represents the heat transfer efficiency from the mechanical output of the pump and valve to the effective heat dissipation power of the insulating oil, and its value is greater than 0 and not greater than 1. This represents the predicted value of the heat transfer hysteresis time window, in seconds. (The formula contains...) This represents the equivalent heat capacity participating in this round of oil circulation, used to convert the total heat energy overshoot accumulation into the temperature rise of the insulating oil; This indicates the equivalent heat that needs to be offset by the composite heat dissipation system; This is used to convert the equivalent heat into the average cooling power to be delivered within the hysteresis window, and to compensate for the heat transfer losses caused by the oil circuit, radiator, and actuator. In this specific example, the specific heat capacity of the insulating oil is 1900 joules per kilogram of Kelvin, the circulating mass is 3000 kilograms, the heat transfer efficiency is 0.82, the total thermal overshoot accumulation is 1.968 megajoules, and the predicted heat transfer hysteresis window value is 40 seconds. Thus, the equivalent temperature rise of the insulating oil is approximately 0.345 Kelvin, and the total equivalent target cooling power is 60 kilowatts.

[0037] Then, using the predicted value of the heat transfer hysteresis time window as the complete time domain period, it is divided into continuous cooling capacity distribution intervals according to a 1-second control cycle. The controller uses a preset negative exponential decay function to generate the time-series weights of each distribution interval, giving higher weights to distribution intervals closer to the current moment and lower weights to distribution intervals closer to the end of the heat transfer hysteresis time window; subsequently, all time-series weights are normalized by energy conservation to ensure that the sum of the products of cooling power and corresponding duration for each distribution interval after discretization equals the product of the equivalent target cooling power and the predicted value of the heat transfer hysteresis time window. The corresponding time-series quantization relationship is:

[0038]

[0039]

[0040] in, This represents the time-series weight of the k-th cooling capacity allocation interval; Represents an exponential function; The attenuation coefficient of the preemptive cooling capacity distribution is determined by the transformer temperature rise test and stored in the controller. This represents the time offset of the start time of the k-th cooling capacity allocation interval relative to the current time. This represents the predicted value of the heat transfer hysteresis time window; This represents the equivalent cooling power requirement for the k-th cooling capacity allocation zone; This represents the total equivalent target cooling power. This indicates the index of the cold energy distribution interval in the normalized summation; N represents the total number of cold energy distribution intervals. This represents the duration of the r-th cooling capacity distribution interval; This represents the preemptive feedforward cooling demand matrix; to Indicates the start time of each cooling capacity distribution interval; to This represents the equivalent cooling power demand value corresponding to each cooling capacity allocation interval. The exponential term in the formula concentrates the cooling demand in the early part of the heat transfer lag time window, thus reserving time for the main variable frequency pump to overcome rotational inertia and oil circuit resistance, and for the auxiliary cooling bypass to complete valve opening. The normalized denominator ensures that the increase in cooling capacity in the early part does not change the total cooling energy that needs to be delivered. The first column of the preemptive feedforward cooling demand matrix records the execution time, and the second column records the corresponding equivalent cooling power demand value, enabling subsequent steps to separate low-frequency and high-frequency values ​​along the time dimension. When the predicted heat transfer lag time window value is 40 seconds, the total equivalent target cooling power is 60 kW, and the control cycle is 1 second, the controller generates a 40-row preemptive feedforward cooling demand matrix. After setting the attenuation coefficient to 1.2, the equivalent cooling power demand value in the early part of the matrix is ​​higher than 60 kW and gradually decreases with time, but the sum of the products of the power of each row and the corresponding duration remains at 2.4 MJ.

[0041] Specifically, in step S4, based on the dynamic capability curve of the main variable frequency pump and the step response model of the auxiliary cooling bypass, the preemptive feedforward cooling demand matrix is ​​subjected to composite hardware asynchronous decoupling and control benchmark allocation to obtain the main pump overclocking drive benchmark signal and the auxiliary bypass valve opening benchmark command. It should be understood that, because the main variable frequency pump and the auxiliary cooling bypass have different dynamic response characteristics, the main variable frequency pump needs to overcome the motor's rotational inertia, the viscosity of the insulating oil, and the resistance of the pipeline network. Its speed and circulation flow can only be established gradually according to a set slope, making it suitable for handling basic heat dissipation demands that last for a long time and change gradually. The auxiliary cooling bypass valve has a shorter operating time, making it suitable for compensating for the peak heat dissipation demand that appears rapidly in the early stages of high overload. If the preemptive feedforward cooling demand matrix is ​​directly applied to both types of actuators simultaneously, the main variable frequency pump will be unable to follow rapid changes in cooling capacity, or the auxiliary cooling bypass will maintain a large opening for a long time, resulting in a mismatch in cooling capacity distribution. Therefore, in the technical solution of this application, based on the dynamic capability curve of the main variable frequency pump and the step response model of the auxiliary cooling bypass, a composite hardware asynchronous decoupling and control benchmark allocation is performed on the preemptive feedforward cooling demand matrix to obtain the main pump overfrequency drive benchmark signal and the auxiliary bypass valve opening benchmark command. This allows the cooling tasks undertaken by each actuator to be determined according to the frequency of cooling demand changes and the response capabilities of the two types of actuators. In this way, the main variable frequency pump can handle the low-frequency basic cooling demand, the auxiliary cooling bypass can compensate for the high-frequency peak transient cooling demand, and a composite cooling output adapted to the high overload heat evolution can be formed in advance within the predicted heat transfer hysteresis time window.

[0042] Figure 3 This is a flowchart of step S4 of a composite heat dissipation method for transformers applicable to high overload conditions according to an embodiment of this application. Figure 3As shown, step S4 includes: S41, using a digital low-pass filter to perform frequency domain filtering and subtraction separation on the preemptive feedforward cooling demand matrix to obtain the low-frequency basic heat dissipation demand matrix and the high-frequency peak transient heat dissipation demand matrix; S42, performing speed inverse calculation matching on the low-frequency basic heat dissipation demand matrix to generate the main pump overclocking drive reference signal; S43, performing proportional-derivative feedforward response mapping on the high-frequency peak transient heat dissipation demand matrix to obtain the auxiliary bypass valve opening reference command.

[0043] More specifically, in a concrete example of this application, firstly, when multiple charging devices in a centralized electric vehicle charging station are simultaneously put into operation, causing a step increase in the primary load of the transformer, the controller receives a preemptive feedforward cooling demand matrix. This preemptive feedforward cooling demand matrix records the equivalent cooling power demand value corresponding to each moment within the predicted value of the heat transfer hysteresis time window according to a 1-second control cycle. When the predicted value of the heat transfer hysteresis time window is 40 seconds, the preemptive feedforward cooling demand matrix includes 40 rows of cooling demand data arranged in chronological order. The controller retains the time index of each data row and applies a first-order digital low-pass filter to the equivalent cooling power demand value along the time dimension, extracting the parts with gradual changes and longer durations as a low-frequency basic heat dissipation demand matrix. Then, the preemptive feedforward cooling demand matrix is ​​subtracted from the low-frequency basic heat dissipation demand matrix row by row according to the same time index to obtain the high-frequency peak transient heat dissipation demand matrix. Specifically, the cutoff frequency of the digital low-pass filter is determined based on the mechanical response time constant measured by the step response test of the main inverter pump. The initial state of the filter adopts the effective cooling power corresponding to the actual speed of the main inverter pump before the occurrence of this high overload, so that the high-frequency peak transient heat dissipation demand matrix reflects the change in cooling capacity that needs to be quickly compensated by the auxiliary cooling bypass relative to the current basic output of the main inverter pump. The above frequency domain filtering and subtraction separation are implemented according to the following relationship:

[0044]

[0045]

[0046] in, Represents the discrete filter coefficients of a first-order digital low-pass filter; Represents an exponential function; Represents pi; This indicates the cutoff frequency of the digital low-pass filter; The control period is 1 second in this specific example; k represents the time sequence number in the preemptive feedforward cooling demand matrix. This represents the equivalent cooling power demand value corresponding to the k-th time index in the preemptive feedforward cooling demand matrix; This represents the equivalent cooling power demand value corresponding to the k-th time index of the low-frequency basic heat dissipation demand matrix; This represents the low-frequency basic heat dissipation requirement value for the previous time sequence. This represents the transient cooling power demand value corresponding to the k-th time index of the high-frequency peak transient heat dissipation demand matrix. In the above calculation, This indicates the contribution of current cooling demand to low-frequency basic heat dissipation demand. This indicates the continuation of the output state of the main variable frequency pump at the previous control moment. Together, they simulate the dynamic characteristics of the main variable frequency pump not being able to change the circulating oil flow instantaneously. This represents the transient portion of the total cooling demand of the composite heat dissipation system that exceeds the low-frequency handling capacity of the main variable frequency pump. When the centralized charging load is rapidly brought into operation, the equivalent cooling power demand value of the front section of the preemptive feedforward cooling demand matrix increases rapidly, and the high-frequency peak transient heat dissipation demand matrix correspondingly forms a positive peak. When the cooling demand gradually decreases, the negative value in the high-frequency peak transient heat dissipation demand matrix indicates that the opening of the auxiliary cooling bypass needs to be reduced. This negative value is constrained by the lower limit of the physical opening when mapping subsequent valve commands.

[0047] Next, the low-frequency basic heat dissipation demand matrix is ​​read line by line according to the time index, and the pre-stored dynamic capability curve of the main variable frequency pump is called. The low-frequency basic heat dissipation demand value at each moment is inversely matched to the target speed of the main variable frequency pump. The dynamic capability curve of the main variable frequency pump uses the speed or drive frequency as the independent variable and the corresponding effective cooling power as the dependent variable. The controller performs piecewise linear interpolation between adjacent calibration nodes. When the low-frequency basic heat dissipation demand value is between two adjacent effective cooling power calibration values, the controller performs proportional interpolation between the corresponding two speed calibration values. When the low-frequency basic heat dissipation demand value exceeds the effective cooling power corresponding to the rated speed, the controller continues to perform inverse matching within the allowable transient overclocking range of the equipment. Specifically, the dynamic capability curve of the main variable frequency pump is determined through cooling system commissioning tests. The main oil flow rate, radiator inlet and outlet oil temperature difference, and stable cooling power are recorded at different drive frequencies, and a lookup table is formed according to the drive frequency from low to high. In this specific example, the rated drive frequency of the main variable frequency pump is 50 Hz, the allowable maximum transient drive frequency is 60 Hz, the frequency increment within a single control cycle does not exceed 2 Hz, and the continuous overclocking time does not exceed 30 seconds as specified by the equipment manufacturer. The controller sequentially applies minimum speed constraints, maximum overclocking speed constraints, speed rise slope constraints, and continuous overclocking time constraints to the target speed obtained through reverse matching, and arranges the constrained target speeds according to the time index of the preemptive feedforward cooling demand matrix to generate the main pump overclocking drive reference signal. Thus, the main pump overclocking drive reference signal maintains a time-series correspondence with the low-frequency basic heat dissipation demand matrix and is located within the allowable dynamic operating envelope of the main variable frequency pump.

[0048] Then, the high-frequency peak transient heat dissipation demand matrix is ​​read according to the same time index, and the pre-stored auxiliary cooling bypass step response model is invoked. The auxiliary cooling bypass step response model takes the valve opening command as input and the effective cooling power delivered by the auxiliary bypass as output, and uses steady-state cooling gain, mechanical response time constant, and action pure delay to describe the process of the auxiliary bypass valve from receiving the command to forming a stable cooling oil flow. In particular, the auxiliary cooling bypass step response model is determined through valve step tests. During the calibration process, the main variable frequency pump is kept at a low-speed idle reference state so that the pressure difference squeeze effect of the main circuit on the auxiliary bypass is negligible. Then, multiple known opening commands are applied to the auxiliary bypass valve in sequence, and the changes in bypass flow rate and bypass heat exchanger inlet and outlet oil temperature difference over time are recorded. The steady-state cooling gain is determined according to the ratio of steady-state effective cooling power to valve opening. The mechanical time constant is determined based on the time required for the auxiliary bypass flow rate or effective cooling power to reach 63.2% of its steady-state change. The pure delay of the action is determined based on the time difference between the time the instruction is applied and the time when the bypass flow begins to change. .

[0049] Furthermore, based on the pure delay of the action, the target demand time in the high-frequency peak transient heat dissipation demand matrix is ​​shifted forward by the corresponding number of control cycles. Then, a proportional-derivative feedforward response mapping is performed on the time-delay-aligned high-frequency peak transient heat dissipation demand value and its adjacent time difference value to generate the auxiliary bypass valve opening reference command. The specific calculation relationship is as follows:

[0050]

[0051]

[0052] in, This indicates the number of control cycles corresponding to pure delay of action; Indicates rounding up; This indicates that the operation of the auxiliary bypass valve is purely delayed; Indicates the control cycle; This represents the k-th high-frequency peak transient heat dissipation demand value after pure delay compensation and execution of non-negative constraints. This represents the corresponding element in the high-frequency peak transient heat dissipation demand matrix; Used to limit the read sequence number after shifting to the matrix range; N represents the total number of time sequence numbers contained in the high-frequency peak transient heat dissipation demand matrix; Used to limit negative cooling demand to zero; This indicates the valve opening percentage corresponding to the kth time sequence of the auxiliary bypass valve opening reference command. This means that the calculation results will be limited to the range of valve physical opening degree from 0 to 1; This represents the steady-state cooling gain corresponding to a unit valve opening in the auxiliary cooling bypass; This represents the mechanical response time constant of the auxiliary bypass valve; This represents the change in high-frequency transient cooling demand between adjacent control moments; when k is the first time number... The effective cooling power corresponding to the actual opening degree of the auxiliary cooling bypass before the occurrence of high overload is used. In the above calculation, d is used to make the auxiliary bypass valve act ahead of the target cooling capacity delivery time, thereby compensating for the pure delay in the action of the valve and bypass oil flow; The proportional mapping section is used to convert the target transient cooling power into the steady-state valve opening. The differential feedforward component compensates for the mechanical inertia of the auxiliary bypass valve, ensuring that the faster the high-frequency peak transient heat dissipation demand rises, the larger the feedforward increment of the valve opening reference command. The opening saturation processing ensures that the auxiliary bypass valve opening reference command does not exceed the valve's physical stroke. Thus, when the concentrated charging load just begins to rise sharply, the high-frequency peak transient heat dissipation demand matrix drives the auxiliary bypass valve to open rapidly in advance, while the low-frequency basic heat dissipation demand matrix drives the main variable frequency pump to gradually enter the rated or transient overclocking state according to the allowable speed slope. As the main variable frequency pump establishes its circulating oil flow and the high-frequency demand decreases, the auxiliary bypass valve opening reference command decreases, and the main variable frequency pump continues to undertake the continuous heat dissipation task.

[0053] Specifically, in the transformer composite heat dissipation control system, the main frequency converter pump and the auxiliary bypass valve work together on the same insulating oil circulation network. In step S43 of the first embodiment, when generating the auxiliary bypass valve opening reference command, the offline calibrated and fixed steady-state cooling capacity gain of the auxiliary cooling bypass is used. and mechanical time constant ,in, This represents the steady-state cooling capacity delivery corresponding to a unit valve opening, 1 / This constitutes a fixed-proportion inversion gain that inverts the high-frequency peak transient heat dissipation demand into the nominal valve opening. This is used to compensate for the mechanical response hysteresis of the auxiliary bypass valve. In this embodiment, the proportional-derivative feedforward response mapping is performed only based on the instantaneous amplitude and time rate of change of the high-frequency peak transient heat dissipation demand matrix to calculate the auxiliary bypass valve opening reference command. The main pump overfrequency drive reference signal is only used as a transmission quantity and does not participate in the calculation of the command. Therefore, it implies the assumption that the fluid dynamic environment of the auxiliary bypass branch remains constant, the unit opening cooling capacity and the valve mechanical response characteristics are not affected by the operating state of the main variable frequency pump.

[0054] However, under high overload conditions at centralized charging stations for electric vehicles, the simultaneous operation of multiple charging devices causes the main variable frequency pump to receive the overfrequency drive reference signal and accelerate from its original speed to an overfrequency state far exceeding the normal operating speed. According to Bernoulli's equation in fluid mechanics, the sharp increase in the flow velocity of insulating oil in the main circuit will significantly increase the overall dynamic pressure of the pipeline system and dynamically reconstruct the effective pressure difference distribution between the main circuit, auxiliary bypass, and other branches. Since the main circuit and auxiliary bypass physically share the same inlet and outlet headers, the increase in the hydraulic resistance of the main circuit will create a back pressure crowding-out effect on the auxiliary bypass. That is, the higher the speed of the main variable frequency pump, the smaller the effective pressure difference between the inlet and outlet of the auxiliary bypass branch that can be used to drive the flow of insulating oil. Therefore, the hydraulic environment corresponding to the offline calibration parameters of the first embodiment is no longer consistent with the actual hydraulic environment when the main variable frequency pump is operating at overfrequency. The direct consequences of this flow field-hydraulic resistance coupling interference relationship are reflected on two levels. First, when the auxiliary bypass valve maintains the same nominal opening, the available driving differential pressure of the auxiliary bypass is squeezed out by the main circuit, resulting in a decrease in the actual amount of cooling oil flowing through the auxiliary bypass. This leads to a reduction in the actual cooling capacity delivered per unit opening of the auxiliary bypass valve. For example, after a sudden increase in centralized charging load, the high-frequency peak transient heat dissipation demand matrix requires the auxiliary bypass to immediately deliver peak cooling capacity. However, the valve opening calculated according to the fixed proportional gain calibrated under offline steady-state conditions can only represent the nominal cooling capacity under the calibrated differential pressure, and cannot guarantee the delivery of the same amount of cooling capacity under overclocking network conditions. Second, the mechanical response characteristics of the auxiliary bypass valve, i.e., the dynamic process from receiving the auxiliary bypass valve opening reference command to reaching the target stroke, will experience inertial drift due to changes in back pressure conditions. When the back pressure increases, increasing the resistance to valve core movement and the resistance to bypass oil flow establishment, the actual response speed of the valve decreases, and the original fixed mechanical time constant cannot truly reflect the dynamic behavior of the valve under varying back pressure conditions.

[0055] The above analysis shows that the first embodiment treats the main pump overclocking drive reference signal as a non-calculated transparent quantity, neglecting the dynamic interference of the main variable frequency pump's physical execution action on the auxiliary bypass cooling capacity delivery efficiency. In a typical high overload scenario where the main variable frequency pump's high-frequency overspeed operation and the auxiliary bypass's transient response occur simultaneously, there is a systematic deviation between the nominal command opening calculated based on a fixed proportional gain coefficient and a fixed mechanical time constant and the actual cooling capacity delivery of the auxiliary bypass valve. This deviation monotonically increases with the increase of the main variable frequency pump's speed. Consequently, the auxiliary bypass may experience under-delivery of cooling capacity at the critical moment when the high-frequency peak transient heat dissipation demand is strongest and when it is most necessary to implement transient heat dissipation compensation, thereby weakening the overall composite heat dissipation system's accuracy in suppressing transient heat peaks.

[0056] To address the aforementioned technical deficiencies, a second embodiment is proposed. Specifically, in step S43 of this application, in the second embodiment, Figure 4This is a flowchart of step S43 of a composite heat dissipation method for transformers applicable to high overload conditions according to an embodiment of this application. Figure 4 As shown, S43 includes: S431, performing online calculation of the flow field liquid resistance coupling interference factor on the main pump overfrequency drive reference signal to obtain the liquid resistance coupling interference factor; S432, performing online adaptive correction of the proportional gain and differential time constant based on the liquid resistance coupling interference factor, and performing coupled compensation proportional differential inversion on the high frequency peak transient heat dissipation demand matrix based on the corrected parameters to generate the auxiliary bypass valve opening reference command.

[0057] Accordingly, in step S431, the flow field liquid resistance coupling interference factor is calculated online from the main pump overclocking drive reference signal to obtain the liquid resistance coupling interference factor. It should be understood that, in a transformer composite heat dissipation control system, the main frequency converter pump and the auxiliary cooling bypass are connected to the same insulating oil circulation network. As the main frequency converter pump speed increases, the flow velocity and dynamic pressure in the main circuit increase, and the effective pressure difference distribution between the main circuit and the auxiliary bypass changes. The higher the current operating speed of the main frequency converter pump, the greater its proportion of the total system flow resources, and the more severe the squeezing of the available drive pressure difference in the auxiliary bypass. The fixed proportional gain coefficient and fixed mechanical time constant used in the first embodiment cannot characterize the above dynamic changes. Therefore, in the technical solution of this application, the liquid resistance coupling interference factor is obtained by calculating the flow field liquid resistance coupling interference factor online from the main pump overclocking drive reference signal, thereby quantifying the interference intensity exerted by the current operating condition of the main frequency converter pump on the fluid dynamic environment of the auxiliary bypass. In this way, the macroscopic speed condition of the main variable frequency pump can be converted into a quantitative disturbance measure of the microscopic hydraulic environment of the auxiliary bypass, providing a physical basis for subsequent online correction of the proportional gain and differential time constant based on the hydraulic resistance coupling interference factor.

[0058] In a specific example of this application, firstly, when multiple charging devices in a centralized electric vehicle charging station are simultaneously put into operation, causing a step increase in the primary load of the transformer, the controller reads the main pump overclocking drive reference signal generated in step S42 according to a 1-second control cycle and extracts the instantaneous speed value corresponding to the current time. The main pump overclocking drive reference signal records the target speed that the main variable frequency pump should reach at each control moment in chronological order, and the controller uses the instantaneous speed value therein as a characterizing quantity of the current operating speed condition of the main variable frequency pump. Specifically, when the main pump overclocking drive reference signal is expressed in terms of the inverter drive frequency, the controller converts the current drive frequency into an instantaneous speed value according to the correspondence between the rated drive frequency and the rated design speed of the main variable frequency pump; when the main pump overclocking drive reference signal is expressed in terms of speed, the controller directly reads the instantaneous speed value. The controller also limits the reading result to between the minimum allowable speed and the maximum transient speed of the main variable frequency pump to avoid abnormal commands affecting the online calculation of the hydraulic resistance coupling interference factor.

[0059] Next, the controller reads the pre-stored rated design speed of the main variable frequency pump and the hydraulic resistance coupling transfer coefficient between the main circuit and the auxiliary bypass. The rated design speed of the main variable frequency pump is determined according to the equipment nameplate parameters or the dynamic capability curve provided by the manufacturer. The hydraulic resistance coupling transfer coefficient is a non-negative dimensionless parameter and is obtained through offline hydraulic calibration tests. In the offline hydraulic calibration test, the low-speed idling condition used in the calibration of steady-state cooling gain in the first embodiment is used as the low-interference reference state. The auxiliary bypass valve is kept at a set opening, and the main variable frequency pump is run at multiple known speeds. The effective driving pressure difference between the auxiliary bypass inlet and outlet, the auxiliary bypass flow rate, and the effective cooling power are recorded. Then, the relationship between the attenuation degree of the above parameters relative to the low-interference reference state and the square of the main variable frequency pump speed ratio is fitted to obtain the hydraulic resistance coupling transfer coefficient between the main circuit and the auxiliary bypass.

[0060] Then, the instantaneous speed value of the main pump overfrequency drive reference signal at time t is compared with the rated design speed of the main variable frequency pump. The resulting speed ratio is squared, and the squared result is multiplied by the liquid resistance coupling transmission coefficient between the main circuit and the auxiliary bypass to obtain the liquid resistance coupling interference factor online. This calculation is based on the centrifugal pump similarity law. When the pipeline structure and insulating oil properties are within the calibration range, the flow rate change caused by the main pump speed change is proportional to the speed, and the dynamic pressure and equivalent liquid resistance interference caused by the flow rate change are proportional to the square of the speed. The calculation relationship of the liquid resistance coupling interference factor is as follows:

[0061] in, This represents the liquid resistance coupling interference factor obtained by online calculation at time t, which is a dimensionless scalar. The instantaneous rotational speed of the main pump overclock drive reference signal at time t is consistent with the symbol meaning of the main pump overclock drive reference signal in the first embodiment; This represents the rated design speed constant of the main variable frequency pump; This represents the hydraulic resistance coupling transfer coefficient between the main circuit and the auxiliary bypass, obtained in advance through offline hydraulic calibration tests; t represents the current control time. This represents the normalized speed ratio of the current speed of the main variable frequency pump relative to its rated design speed. This represents the amplification terms of dynamic pressure and equivalent hydraulic resistance interference determined according to the centrifugal pump similarity law. In the above calculations, the speed ratio is used to eliminate the dimensional differences between main variable frequency pumps with different rated speed specifications, so that the main pump over-frequency drive reference signal can characterize the main variable frequency pump's occupation of pipeline flow resources on a unified scale; the square term of the speed ratio is used to reflect the nonlinear growth of the main circuit dynamic pressure and differential pressure squeezing effect after the main variable frequency pump speed increases; the hydraulic resistance coupling transmission coefficient between the main circuit and the auxiliary bypass is used to characterize the influence of pipeline structure such as shared inlet and outlet oil headers, pipe diameters, branch lengths, and local resistances on the degree of hydraulic interference transmission; the hydraulic resistance coupling interference factor formed by the combination of the three is used to represent the comprehensive squeezing intensity of the available drive differential pressure of the auxiliary bypass under the current operating state of the main variable frequency pump.

[0062] When the main variable frequency pump is in a low-speed idling state, the speed ratio between the instantaneous speed and the rated design speed is much less than 1. After squaring, the liquid resistance coupling interference factor approaches zero, indicating that the flow field interference of the main variable frequency pump to the auxiliary bypass can be ignored. When the main variable frequency pump is in the rated operating state, the speed ratio is equal to 1, and the liquid resistance coupling interference factor is determined by the liquid resistance coupling transmission coefficient between the main circuit and the auxiliary bypass. When the main variable frequency pump receives the main pump over-frequency drive reference signal and enters the extreme over-frequency state, the speed ratio is greater than 1. Squaring causes the liquid resistance coupling interference factor to increase rapidly with the speed of the main variable frequency pump, thus reflecting the severe squeezing of the available drive pressure difference of the auxiliary bypass by the high pump speed. For example, when the rated design speed of the main variable frequency pump corresponds to a 50 Hz drive frequency and the main pump over-frequency drive reference signal corresponds to a 60 Hz drive frequency, the normalized speed ratio is 1.2, and its square term is 1.44, indicating that the liquid resistance coupling interference transmitted to the auxiliary bypass is amplified proportionally relative to the rated state. Therefore, the hydraulic resistance coupling interference factor is updated based on the main pump overclocking drive reference signal in each control cycle, making the hydraulic resistance coupling interference factor dynamically change with the main variable frequency pump speed, rather than using a fixed hydraulic environment assumption. Before performing proportional-derivative inversion on the high-frequency peak transient heat dissipation demand matrix, the hydraulic resistance coupling interference factor can clearly represent the interference intensity exerted by the main variable frequency pump under the current operating condition on the auxiliary bypass fluid environment, and serve as the input for subsequent online correction of proportional gain and derivative time constant, thus establishing a physical basis for compensating for the decrease in available differential pressure of the auxiliary bypass, the attenuation of cooling capacity per unit opening, and valve response inertial drift.

[0063] Accordingly, in step S432, the proportional gain and differential time constant are adaptively corrected online based on the liquid resistance coupling interference factor, and the high-frequency peak transient heat dissipation demand matrix is ​​coupled and compensated using proportional-differential inversion based on the corrected parameters to generate the auxiliary bypass valve opening reference command. It should be understood that since the main variable frequency pump and the auxiliary cooling bypass work together in the same insulating oil circulation network, when the main variable frequency pump enters high-speed or over-frequency operation, the effective driving pressure difference between the auxiliary bypass inlet and outlet is squeezed, causing a decrease in the actual cooling capacity delivery corresponding to the unit nominal opening degree of the auxiliary bypass valve. Simultaneously, the increased back pressure increases the valve core movement resistance and prolongs the bypass oil flow establishment time. The first embodiment uses a fixed steady-state cooling capacity gain, a fixed proportional inversion gain, and a fixed mechanical time constant, which cannot characterize the changes in the above parameters with the operating state of the main variable frequency pump. Therefore, in the technical solution of this application, the proportional inversion gain and differential time constant are further adaptively corrected online based on the liquid resistance coupling interference factor, and the high-frequency peak transient heat dissipation demand matrix is ​​coupled and compensated using proportional differential inversion based on the corrected parameters to generate the auxiliary bypass valve opening reference command, thereby compensating for the effective differential pressure loss and dynamic response sluggishness of the auxiliary bypass. This allows the nominal command opening degree of the auxiliary bypass valve to adaptively increase with the liquid resistance coupling state of the pipeline network, reducing the deviation between the high-frequency peak transient heat dissipation demand and the actual cooling capacity delivery.

[0064] In a specific example of this application, firstly, when multiple charging devices in a centralized electric vehicle charging station are simultaneously put into operation, the controller reads the liquid resistance coupling interference factor and the high-frequency peak transient heat dissipation demand matrix according to a 1-second control cycle. The liquid resistance coupling interference factor is used to represent the interference intensity exerted by the current main variable frequency pump operating speed on the auxiliary bypass fluid dynamics environment, and the high-frequency peak transient heat dissipation demand matrix records the transient cooling power demand that needs to be quickly compensated by the auxiliary cooling bypass in chronological order. In particular, the controller stores the steady-state cooling gain determined through auxiliary bypass step response tests. and mechanical time constant It also stores the back pressure response inertial sensitivity coefficients determined through valve step tests under different back pressure conditions. The above parameters all correspond to the same auxiliary bypass valve, bypass heat exchanger and insulating oil circulation network.

[0065] Next, the controller calculates the actual cooling capacity delivery corresponding to the unit nominal opening degree of the auxiliary bypass valve under the current pipeline network conditions based on the liquid resistance coupling interference factor. The calculation relationship is as follows:

[0066] in, This represents the actual cooling capacity delivery gain per unit valve opening over time t, expressed in terms of cooling power per unit valve opening. This represents the steady-state cooling capacity gain of the auxiliary cooling bypass, determined through an offline step test. The fluid resistance coupling interference factor represents time t; This represents the attenuation correction term caused by the main variable frequency pump's operating status on the auxiliary bypass cooling capacity delivery capability; t represents the current control time. In the above calculation, The cooling power delivered per unit opening of the auxiliary bypass valve under offline calibrated hydraulic conditions, in the denominator This is used to characterize the degree to which the available drive differential pressure in the auxiliary bypass decreases due to the high-speed operation of the main variable frequency pump. When the liquid resistance coupling interference factor increases, The decrease indicates that when the auxiliary bypass valve maintains the same nominal opening, both its actual cooling oil flow rate and cooling capacity delivery capacity decrease.

[0067] Then, the adaptive proportional inversion gain after coupling correction is calculated based on the actual cooling capacity delivery gain per unit opening degree. Since the auxiliary bypass valve opening reference command requires the nominal valve opening degree to be calculated in reverse based on the target cooling power, the adaptive proportional inversion gain is the reciprocal of the actual cooling capacity delivery gain per unit opening degree, and its calculation relationship is as follows:

[0068] in, The adaptive proportional inversion gain after time t is adjusted by coupling, and the unit is the cooling power per unit valve opening. This indicates the actual cooling capacity delivery gain per unit opening degree; This indicates the steady-state cooling capacity gain of the auxiliary cooling bypass. This represents the liquid resistance coupling interference factor. In the above calculation, This is used to invert the high-frequency peak transient heat dissipation demand into the nominal valve opening required under current hydraulic conditions. When the main variable frequency pump speed increases and the hydraulic resistance coupling interference factor increases, the actual cooling capacity delivery gain per unit opening decreases, while the adaptive proportional inversion gain increases synchronously. The controller outputs a larger nominal valve opening for the same high-frequency peak transient heat dissipation demand to compensate for the insufficient actual flow in the auxiliary bypass caused by the occupancy of available drive pressure differential. Therefore, the decrease in the actual cooling capacity delivery gain per unit opening and the increase in the nominal valve opening form an inverse compensation relationship.

[0069] Furthermore, the controller performs online adaptive correction of the mechanical time constant of the auxiliary bypass valve based on the hydraulic resistance coupling interference factor. When the main variable frequency pump operates at high speed, causing an increase in the back pressure of the auxiliary bypass, the valve core needs to overcome greater mechanical resistance under the same control signal, thus extending the time required for the auxiliary bypass flow rate to reach the target value. The controller multiplies the hydraulic resistance coupling interference factor by the back pressure response inertial sensitivity coefficient and uses the result as an amplification correction amount for the mechanical time constant to obtain the adaptive differential time constant. The calculation relationship is as follows:

[0070] in, This represents the adaptive differential time constant after time t has been corrected by coupling; This represents the mechanical time constant determined through an auxiliary bypass step response test; The sensitivity coefficient representing the response inertia of the auxiliary bypass valve to back pressure; Indicates the liquid resistance coupling interference factor; This represents the amplification factor of the valve response inertia relative to the offline calibration state. Specifically, the back pressure response inertia sensitivity coefficient is determined through an offline step test. Under the conditions that the initial opening of the auxiliary bypass valve, the insulating oil temperature, and the oil viscosity are maintained within the set range, the main variable frequency pump is operated at multiple known speeds, and a step opening command of the same amplitude is applied to the auxiliary bypass valve. The corresponding mechanical time constant is determined based on the time required for the auxiliary bypass flow rate to reach 63.2% of its steady-state change at each main pump speed. Then, the back pressure response inertia sensitivity coefficient is obtained by fitting the relationship between the mechanical time constant and the liquid resistance coupling interference factor. When the liquid resistance coupling interference factor increases, The synchronous increase enables the controller to apply stronger feedforward compensation to the rate of change of high-frequency cooling demand, thereby offsetting the slow dynamic response of the auxiliary bypass valve caused by the increase of back pressure.

[0071] Furthermore, following the pure delay compensation method in the first embodiment, the target demand moment in the high-frequency peak transient heat dissipation demand matrix is ​​shifted forward by the number of control cycles corresponding to the pure delay of the auxiliary bypass valve's action. Non-negativity constraints are then applied to the shifted matrix components to obtain the delay-compensated high-frequency heat dissipation demand value corresponding to the current control moment. Subsequently, the controller divides the difference between the delay-compensated high-frequency heat dissipation demand value at the current control moment and the previous control moment by the control cycle to determine the instantaneous rate of change of the high-frequency cooling demand. Coupled compensation proportional-differential inversion is then performed based on the adaptive proportional inversion gain and the adaptive differential time constant, with the calculation relationship as follows:

[0072] in, This represents the reference command for opening the auxiliary bypass valve to be verified generated in the k-th control cycle; This represents the adaptive proportional inversion gain after coupling correction at the k-th control moment; This represents the high-frequency peak transient heat dissipation demand after pure delay compensation and non-negative constraints. This represents the high-frequency heat dissipation requirement value for delay compensation corresponding to the previous control cycle. This represents the adaptive differential time constant at the k-th control moment; This indicates the control cycle, which is 1 second in this specific example; k represents the control cycle number. This represents the k-th control moment.

[0073] In the above calculations, The proportional inversion part is used to convert the current high-frequency peak transient heat dissipation demand into the nominal valve opening required under the current pipeline hydraulic conditions. When the hydraulic resistance coupling interference factor increases, the adaptive proportional inversion gain increases, so that the proportional inversion part actively compensates for the attenuation of the unit opening cold capacity delivery capability. This is used to characterize the variation of high-frequency cooling demand between adjacent control cycles. Dividing by the control cycle yields the instantaneous rate of change of high-frequency cooling demand. A larger adaptive differential time constant results in a larger feedforward opening increment applied by the controller during the rise of high-frequency cooling demand, thus overcoming the slow response of the auxiliary bypass valve under fluctuating back pressure in advance. For example, when the steady-state cooling gain calibrated in the first embodiment... 80 kW per unit opening, mechanical time constant 2 seconds, back pressure response inertial sensitivity coefficient When the coefficient of performance (COP) is 0.5 and the current liquid resistance coupling interference factor is 0.432, the actual cooling capacity delivery gain per unit opening decreases to approximately 55.87 kW per unit opening, the adaptive proportional inversion gain increases to approximately 0.0179 kW per unit opening, and the adaptive differential time constant increases to approximately 2.432 seconds. When the delay compensation high-frequency heat dissipation demand is 30 kW and its instantaneous rate of change is 5 kW / s, the reference command for opening the auxiliary bypass valve to be verified is approximately 0.754. Under the same conditions, if the fixed parameters of the first embodiment are still used, the calculated nominal opening is 0.5, and the difference between the two corresponds to the differential pressure loss and response delay compensation caused by the over-frequency operation of the main variable frequency pump.

[0074] Then, a dual-channel safety constraint verification is performed on the auxiliary bypass valve opening reference command and the main pump overfrequency drive reference signal. The controller determines the main circuit flow rate corresponding to the main pump overfrequency drive reference signal based on the main variable frequency pump dynamic capability curve, determines the bypass flow rate corresponding to the auxiliary bypass valve opening reference command based on the auxiliary bypass valve flow characteristic curve, and calculates the pipeline working pressure when both act together based on the pre-calibrated pipeline flow and pressure characteristic curves. In particular, the main circuit flow curve, auxiliary bypass valve flow characteristic curve, pipeline flow and pressure characteristic curve, valve maximum stroke, and pipeline design pressure limit are determined through cooling system commissioning tests, valve flow tests, and pipeline pressure resistance parameters, and then stored in the controller.

[0075] When the auxiliary bypass valve opening reference command to be verified does not exceed the valve's physical maximum stroke and the pipeline working pressure does not exceed the pipeline's design pressure limit, the controller retains the command. When the auxiliary bypass valve opening or pipeline working pressure exceeds the limit, but the pipeline working pressure corresponding to the main pump overfrequency drive reference signal is still within the safe range, the controller keeps the main pump overfrequency drive reference signal unchanged and performs anti-saturation limiting on the auxiliary bypass valve opening reference command to be verified. When the pipeline working pressure corresponding to the main pump overfrequency drive reference signal has exceeded the pipeline's design pressure limit, the controller first limits the main pump overfrequency drive reference signal to the safe operating boundary of the main circuit, and then limits the auxiliary bypass valve opening reference command according to the remaining pipeline carrying capacity. The two verified control quantities together constitute the output after safety constraints, thereby ensuring that the main frequency converter pump, auxiliary bypass valve, and insulating oil circulation pipeline are within the hardware safety envelope while maintaining transient heat dissipation compensation capability.

[0076] After this processing, the auxiliary bypass valve opening reference command is no longer based on an open-loop estimate formed by a fixed steady-state cooling gain and a fixed mechanical time constant, but rather incorporates the coupled adaptive output value of the main variable frequency pump's operating state and the real-time interference information of the auxiliary bypass flow field. The second embodiment does not require changes to the hardware topology of the main variable frequency pump, auxiliary bypass valve, and insulating oil circulation network. It can adjust the nominal valve opening online according to the network's liquid resistance coupling state caused by the main variable frequency pump's speed, thereby reducing the systemic cooling under-delivery deviation caused by the main variable frequency pump's high-speed operation. In high-overload heat dissipation scenarios where the main variable frequency pump and auxiliary cooling bypass operate simultaneously at high intensity, the actual cooling delivery accuracy of the auxiliary bypass is improved. The suppression effect of the composite heat dissipation system on transient heat peaks is no longer reduced due to flow field liquid resistance coupling interference, thus improving the overall temperature control accuracy and insulation safety margin under extreme conditions.

[0077] Specifically, in step S5, based on the main pump overclocking drive reference signal and the auxiliary bypass valve opening reference command, the expected oil temperature change is calculated. Closed-loop compensation and pulse modulation are then performed using the deviation between the expected oil temperature change and the real-time temperature of the top insulating oil to drive the main frequency converter pump and the auxiliary bypass valve to collaboratively establish a preemptive cooling oil flow within the predicted heat transfer hysteresis time window. It should be understood that since the main pump overclocking drive reference signal and the auxiliary bypass valve opening reference command are feedforward control quantities generated based on the predicted cooling demand, their actual heat dissipation effect is also affected by the insulating oil viscosity, ambient temperature, radiator heat exchange efficiency, and actuator response deviation. Directly driving the main frequency converter pump and the auxiliary bypass valve according to the feedforward control quantities will result in a discrepancy between the theoretical and actual cooling capacity. Therefore, in the technical solution of this application, the expected oil temperature change is calculated based on the main pump overfrequency drive reference signal and the auxiliary bypass valve opening reference command. Closed-loop compensation and pulse modulation are then performed using the deviation between the expected oil temperature change and the real-time temperature of the top insulating oil to drive the main frequency converter pump and the auxiliary bypass valve to collaboratively establish a preemptive cooling oil flow within the predicted heat transfer hysteresis time window. This corrects the cooling capacity delivery deviation of the feedforward control quantity in the actual oil circuit. In this way, the auxiliary bypass valve can quickly compensate for high-frequency transient cooling demand, and the main frequency converter pump can gradually establish a continuous circulating oil flow, thereby forming a cooling capacity that matches the high overload heat evolution before the oil temperature rises significantly.

[0078] Figure 5 This is a flowchart of step S5 of a composite heat dissipation method for transformers applicable to high overload conditions according to an embodiment of this application. Figure 5 As shown, step S5 includes: S51, based on the real-time temperature value of the top insulating oil as the current state reference, performing theoretical heat dissipation efficiency integral inversion on the main pump overfrequency drive reference signal and the auxiliary bypass valve opening reference command through a virtual temperature drop observer to generate the expected oil temperature reduction threshold; S52, comparing the temperature difference between the expected oil temperature reduction threshold and the real-time temperature value of the top insulating oil and calculating the dynamic closed-loop gain to obtain the temperature closed-loop compensation coefficient; S53, using the temperature closed-loop compensation coefficient to perform multiplication correction and pulse width modulation on the main pump overfrequency drive reference signal to drive the main frequency converter pump, and performing integral superposition and duty cycle modulation on the auxiliary bypass valve opening reference command to drive the auxiliary bypass valve.

[0079] More specifically, in a concrete example of this application, firstly, when multiple charging devices in a centralized electric vehicle charging station are simultaneously put into operation, the controller uses the real-time temperature of the top insulating oil when a high overload is identified as the current state reference, and reads the main pump overfrequency drive reference signal and the auxiliary bypass valve opening reference command within the predicted value of the heat transfer hysteresis time window according to a 1-second control cycle. The controller uses a virtual temperature drop observer to map the main pump overfrequency drive reference signal and the auxiliary bypass valve opening reference command to the theoretical effective cooling power of the main circuit and the theoretical effective cooling power of the auxiliary bypass, respectively. Then, it integrates the insulating oil circulation mass and the specific heat capacity of the insulating oil cycle by cycle to calculate the expected oil temperature drop, and subtracts the expected oil temperature drop from the current state reference to generate the expected oil temperature reduction threshold. In particular, the mapping relationship between the control signal and the effective cooling power is obtained through a cooling system calibration test and stored in the controller in the form of a lookup table. The controller uses linear interpolation for adjacent calibration nodes. The formula is expressed as follows:

[0080]

[0081] in, This represents the expected decrease in oil temperature up to the k-th control cycle; and These represent the theoretical effective cooling power mapping relationship between the main circuit and the auxiliary bypass, respectively. Indicates the first The main pump overclocking drive reference signal for each control cycle; Indicates the first The auxiliary bypass valve opening reference command for each control cycle; Indicates the quality of the insulating oil circulation; Indicates the specific heat capacity of insulating oil; Indicates the control cycle; This indicates the expected reduction in the threshold at the desired oil temperature; This represents the real-time temperature of the top insulating oil when a high overload is identified. The sum of the theoretical effective cooling power of the main circuit and the theoretical effective cooling power of the auxiliary bypass represents the theoretical heat dissipation output of the composite heat dissipation system in the current cycle. Its product with the control cycle represents the theoretical cooling capacity delivered. Dividing this by the heat capacity of the insulating oil yields the corresponding expected oil temperature drop.

[0082] Next, the controller reads the real-time temperature value of the top insulating oil in the current control cycle and compares it with the expected oil temperature reduction threshold in the same control cycle. When the real-time temperature value of the top insulating oil is higher than the expected oil temperature reduction threshold, it is determined that the actual cooling capacity delivery is lower than the theoretical cooling capacity delivery; when the deviation between the two decreases, it is determined that the composite heat dissipation output gradually reaches the target. The controller calculates the temperature closed-loop compensation coefficient based on the current temperature deviation and the changes in adjacent control cycles, and limits the temperature closed-loop compensation coefficient within a preset safety range. In particular, the proportional gain, derivative gain, and upper and lower limits of the compensation coefficient are calibrated through transformer temperature rise tests, and the temperature deviation is treated as zero deviation when it is within the measurement error range of the temperature sensor. The formula is expressed as follows:

[0083]

[0084] in, This represents the temperature deviation during the k-th control cycle; This indicates the current real-time temperature value of the top insulating oil. This indicates the expected reduction in the threshold at the desired oil temperature; This represents the temperature closed-loop compensation coefficient; and These represent the proportional gain and the rate of change gain of the temperature deviation, respectively. and These represent the lower and upper limits of the temperature closed-loop compensation coefficient, respectively. This indicates a limiting effect. The proportional term is used to compensate for the current deviation in cooling capacity delivery, while the derivative term is used to respond to the trend of the deviation, so that when the actual oil temperature deviates from the expected boundary at an increased rate, the combined heat dissipation output increases in advance.

[0085] Then, the temperature closed-loop compensation coefficient is multiplied by the main pump overclocking drive reference signal to obtain the corrected main pump target speed, and minimum speed, maximum overclocking speed, and speed change rate limits are applied to it. Subsequently, the main pump target speed is converted into the inverter target frequency, and the inverter drives the main variable frequency pump to gradually increase the speed through pulse width modulation. In particular, the speed range, allowable acceleration rate, and continuous overclocking time of the main variable frequency pump are determined according to the dynamic capability curve of the main variable frequency pump. Further, the deviation of the temperature closed-loop compensation coefficient from the reference value 1 is integrated, and the integral result is superimposed on the auxiliary bypass valve opening reference command, and the superimposed result is limited to the physical opening range of the auxiliary bypass valve. The controller then converts the corrected valve target opening into the drive pulse duty cycle to drive the auxiliary bypass valve to quickly form a bypass cooling oil flow. When the actual oil temperature is higher than the expected boundary, the integral compensation gradually increases; when the actual oil temperature reaches the expected boundary, the integral compensation stops increasing. Therefore, in the early stage of the predicted heat transfer hysteresis time window, the auxiliary bypass valve opens quickly to deliver transient cooling, and the main variable frequency pump establishes a continuous circulating oil flow according to the allowable acceleration rate; as the real-time temperature value of the top insulating oil approaches the expected oil temperature reduction threshold, the temperature closed-loop compensation coefficient returns to the benchmark value, and the main variable frequency pump and the auxiliary bypass valve enter a coordinated operation state that matches the feedforward cooling demand.

[0086] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A composite heat dissipation method for transformers suitable for high overload conditions, characterized in that, include: S1, transient feature extraction and time-scale fusion of the high-frequency three-phase current sequence of the primary side of the transformer, the real-time temperature value of the top insulating oil and the real-time temperature value of the bottom insulating oil are performed to obtain the electrothermal coupling transient feature tensor; S2, based on a pre-trained spatiotemporal thermal momentum mapping model, performs electrothermal cross-domain mapping and heat transfer hysteresis identification on the electrothermal coupling transient feature tensor to obtain the spatial virtual thermal impulse trajectory sequence and heat transfer hysteresis time window prediction value characterizing the high overload heat evolution. S3, perform preemptive feedforward cooling demand time-series quantization on the spatial virtual thermal impulse trajectory sequence to obtain the preemptive feedforward cooling demand matrix; S4, based on the dynamic capability curve of the main variable frequency pump and the step response model of the auxiliary cooling bypass, performs composite hardware asynchronous decoupling and control benchmark allocation on the preemptive feedforward cooling demand matrix to obtain the main pump over-frequency drive benchmark signal and the auxiliary bypass valve opening benchmark command. S5 calculates the expected oil temperature change based on the main pump overfrequency drive reference signal and the auxiliary bypass valve opening reference command. It then uses the deviation between the expected oil temperature change and the real-time temperature value of the top insulating oil to perform closed-loop compensation and pulse modulation to drive the main frequency converter pump and the auxiliary bypass valve to collaboratively establish a preemptive cooling oil flow within the predicted value of the heat transfer hysteresis time window.

2. The composite heat dissipation method for transformers suitable for high overload conditions according to claim 1, characterized in that, Step S1 includes: S11, the transient current mutation rate is solved for the high-frequency three-phase current sequence on the primary side of the transformer to obtain the current mutation rate sequence; S12, frequency domain nonlinear harmonic analysis is performed on the high-frequency three-phase current sequence on the primary side of the transformer to obtain the energy distribution matrix of each harmonic; S13, calculate the vertical thermal gradient between the real-time temperature values ​​of the top insulating oil and the bottom insulating oil. S14, based on a unified timestamp benchmark, performs time-series alignment and multi-dimensional splicing of the current mutation rate sequence, the energy distribution matrix of each harmonic, and the vertical thermal gradient value to obtain the electrothermal coupling transient feature tensor.

3. The composite heat dissipation method for transformers suitable for high overload conditions according to claim 1, characterized in that, Step S2 includes: S21, the electrothermal coupling transient feature tensor is fed into the pre-trained long short-term memory network in time step order to extract multi-time scale temporal dependencies and obtain cross-domain temporal hidden state vectors. S22, energy updimensional projection and inertial feature decoupling are performed on the cross-domain temporal hidden state vector to obtain the spatial virtual thermal impulse trajectory sequence and flow field hysteresis driving features; S23 utilizes the flow field hysteresis driving characteristics and the time-domain integration of the spatial virtual thermal impulse trajectory sequence to calculate the heat transfer critical time and obtain the predicted value of the heat transfer hysteresis time window.

4. The composite heat dissipation method for transformers suitable for high overload conditions according to claim 3, characterized in that, The spatial virtual thermal impulse trajectory sequence includes the virtual thermal impulse rates of the A-phase winding associated thermal zone, the B-phase winding associated thermal zone, and the C-phase winding associated thermal zone at each future predicted time.

5. The composite heat dissipation method for transformers suitable for high overload conditions according to claim 3, characterized in that, The predicted value of the heat transfer hysteresis time window is determined by the time length between the predicted future moment when the virtual heat accumulation value first reaches the critical heat threshold and the current moment.

6. The composite heat dissipation method for transformers suitable for high overload conditions according to claim 1, characterized in that, Step S3 includes: S31, using the predicted value of the heat transfer hysteresis time window as the upper boundary of the integration, the thermal overshoot in the restricted time domain is integrated to obtain the total thermal energy overshoot accumulation. S32, based on the specific heat capacity and circulating mass of insulating oil, performs thermodynamic equivalent power conversion on the total thermal energy overshoot accumulation to obtain the total equivalent target cooling power; S33 uses the predicted value of the heat transfer hysteresis time window as a time-domain periodic constraint to perform nonlinear time-series discretization on the equivalent target total cooling power to obtain the preemptive feedforward cooling demand matrix.

7. The composite heat dissipation method for transformers suitable for high overload conditions according to claim 6, characterized in that, The preemptive feedforward cooling demand matrix includes the start time of each cooling capacity allocation interval and the equivalent cooling power demand value corresponding to each start time.

8. The composite heat dissipation method for transformers suitable for high overload conditions according to claim 1, characterized in that, Step S4 includes: S41 uses a digital low-pass filter to perform frequency domain filtering and subtraction separation on the preemptive feedforward cooling demand matrix to obtain the low-frequency basic heat dissipation demand matrix and the high-frequency peak transient heat dissipation demand matrix. S42 performs inverse speed calculation and matching on the low-frequency basic heat dissipation demand matrix to generate the main pump overclocking drive reference signal. S43 performs proportional-differential feedforward response mapping on the high-frequency peak transient heat dissipation demand matrix to obtain the auxiliary bypass valve opening reference command.

9. The composite heat dissipation method for transformers suitable for high overload conditions according to claim 1, characterized in that, Step S5 includes: S51, based on the real-time temperature value of the top insulating oil as the current state reference, uses a virtual temperature drop observer to perform theoretical heat dissipation efficiency integral inversion on the main pump overfrequency drive reference signal and the auxiliary bypass valve opening reference command to generate the expected oil temperature reduction threshold. S52, compare the temperature difference between the expected oil temperature reduction threshold and the real-time temperature value of the top insulating oil and calculate the dynamic closed-loop gain to obtain the temperature closed-loop compensation coefficient. S53 uses the temperature closed-loop compensation coefficient to perform multiplication correction and pulse width modulation on the main pump overfrequency drive reference signal to drive the main variable frequency pump, and performs integral superposition and duty cycle modulation on the auxiliary bypass valve opening reference command to drive the auxiliary bypass valve.

10. The composite heat dissipation method for transformers under high overload conditions according to claim 8, characterized in that, S43 includes: S431, the flow field liquid resistance coupling interference factor is calculated online from the main pump overfrequency drive reference signal to obtain the liquid resistance coupling interference factor; S432 performs online adaptive correction of proportional gain and differential time constant based on liquid resistance coupling interference factor, and performs coupled compensation proportional-derivative inversion on high frequency peak transient heat dissipation demand matrix based on the corrected parameters to generate auxiliary bypass valve opening reference command.

Images