Load prediction driven superconducting switch station refrigeration adaptive regulation method

Abstract

The present application relates to the field of smart grid technology, and relates to a load prediction driven superconducting switch station refrigeration adaptive regulation method, comprising: obtaining a future load current prediction sequence of a power grid, and current equipment temperature and refrigeration parameters; converting the predicted current into a predicted thermal load based on an alternating current loss model; generating a pre-adjustment urgency index by weighting the thermal load change rate, thermal inertia time constant, temperature margin and power margin; calculating the feedforward refrigeration power increment when the index exceeds the limit and adjusting the refrigeration output in advance; collecting the actual temperature response after adjustment, back-calculating the actual thermal load, and comparing and correcting the loss coefficient and thermal inertia time constant. The present application solves the technical problem that, in the refrigeration regulation of a superconducting switch station, a traditional refrigeration system relies on current temperature feedback for adjustment, lacks the ability to predict future load changes, and causes large temperature fluctuations of superconducting equipment and lagged refrigeration response.

Description

Technical Field

[0001] This invention relates to the field of smart grid technology, and in particular to a superconducting switchyard cooling adaptive control method driven by load forecasting. Background Technology

[0002] Superconducting switchyards are a new type of power facility in power systems that uses superconducting equipment to achieve current transmission and fault limitation. Their core equipment needs to operate below a critical temperature to maintain its superconducting state. During operation, superconducting equipment generates AC losses and heat leakage, requiring a dedicated cooling system to continuously remove heat and maintain the equipment temperature within a safe range. Currently, the control of the cooling system in superconducting switchyards mainly adopts a temperature feedback-based regulation method. This involves collecting the real-time temperature of the superconducting equipment and comparing it with a set target temperature. When the temperature rises, the cooling output is increased; when the temperature falls, the cooling output is decreased to maintain a relatively stable equipment temperature.

[0003] However, the above-mentioned control methods have significant shortcomings in practical applications. The grid load changes dynamically over time, and the heat load of the superconducting equipment changes synchronously with fluctuations in the load current. However, temperature feedback is a lagging signal; the cooling system only begins to respond and adjust after a change in heat load causes a change in equipment temperature. This results in significant temperature fluctuations in the superconducting equipment during rapid load changes, and the temperature margin is quickly depleted. Furthermore, this passively following control method leads to frequent start-stop cycles of the cooling unit, affecting not only the stability and energy efficiency of the cooling system but also making it difficult to reserve cooling capacity in advance before load changes, thus hindering the safe and stable operation of the superconducting equipment under high load fluctuation scenarios. Summary of the Invention

[0004] This invention provides a load prediction-driven adaptive control method for cooling in superconducting switch stations. It improves upon existing cooling control methods that rely solely on feedback adjustment based on current temperature and setpoints, without considering the impact of grid load changes on the thermal load of superconducting equipment. This results in cooling power failing to keep up with the load during rapid increases, leading to the rapid depletion of the temperature margin of the superconducting equipment, while excessive cooling causes energy waste during load decreases.

[0005] The present invention discloses the following technical solution: This invention provides a load prediction-driven adaptive control method for cooling in superconducting switchyards, the method comprising: Obtain the future load current prediction sequence provided by the power grid dispatching system, as well as the current temperature and cooling system operating parameters of the superconducting equipment; Based on a pre-calibrated AC loss model, the load current prediction sequence is converted into a predicted heat load sequence; Based on the predicted heat load sequence, the predicted heat load change rate is calculated, and combined with the current thermal inertia time constant of the refrigeration system and the current temperature margin of the superconducting equipment, a pre-adjustment urgency index is generated. When the pre-adjustment urgency index exceeds the preset trigger threshold, the feedforward cooling power increment is obtained based on the predicted heat load change rate and the thermal inertia time constant, and the output power of the cooling unit is adjusted in advance. After the adjustment is executed, the actual temperature response of the superconducting device is collected, and the difference between the predicted heat load sequence and the actual heat load is compared to update and correct the loss coefficient of the AC loss model and the thermal inertia time constant.

[0006] One or more technical solutions provided in this invention have at least the following technical effects or advantages: The technical solution of this invention provides a load prediction-driven adaptive control method for cooling in superconducting switch stations. First, by acquiring the future load current prediction sequence provided by the power grid dispatching system, as well as the current temperature of the superconducting equipment and the operating parameters of the cooling system, it realizes the advance perception of the power grid load change trend and the current thermal state of the superconducting equipment. This provides a forward-looking data foundation for cooling control and solves the problem that traditional cooling systems rely solely on real-time temperature feedback and cannot predict load changes, leading to control lag.

[0007] Furthermore, by using a pre-calibrated AC loss model, the load current prediction sequence is converted into a predicted heat load sequence, realizing a quantitative mapping from grid load to superconducting equipment heat load. This transforms load prediction information into heat load prediction results that can be used for refrigeration control, solving the problem that there is no quantitative correlation between load current and heat load, and that they cannot be directly used for refrigeration planning.

[0008] Furthermore, by calculating the predicted heat load change rate based on the predicted heat load sequence and combining it with the current thermal inertia time constant of the refrigeration system and the current temperature margin of the superconducting equipment, a pre-adjustment urgency index is generated, realizing a comprehensive quantitative assessment of the urgency of heat load changes. This solves the problem that the refrigeration system cannot determine when early intervention is needed and when conventional adjustment alone can meet the requirements.

[0009] Furthermore, when the pre-adjustment urgency index exceeds the preset trigger threshold, the feedforward cooling power increment is obtained based on the predicted heat load change rate and thermal inertia time constant analysis, and the output power of the cooling unit is adjusted in advance. This realizes the proactive adjustment of cooling power before the actual change in heat load occurs, and utilizes the thermal inertia of the cooling system to reserve cooling capacity in advance, solving the problem of large temperature fluctuations in superconducting equipment caused by the lag in temperature response in traditional feedback control.

[0010] Finally, after the adjustment is executed, the actual temperature response of the superconducting equipment is collected. By comparing the difference between the predicted heat load sequence and the actual heat load, the loss coefficient and thermal inertia time constant of the AC loss model are updated and corrected. This realizes online calibration and closed-loop optimization of the prediction model and system parameters, and solves the problems of model parameters becoming inaccurate due to equipment aging or changes in operating conditions, and the accumulation of prediction errors over time.

[0011] In summary, this invention obtains a predicted load current sequence, converts it into a predicted heat load sequence using an AC loss model, and generates a pre-regulation urgency index by combining the thermal inertia time constant and temperature margin. When the load exceeds the limit, the output power is adjusted in advance based on the predicted heat load change rate, using the feedforward cooling power increment. Furthermore, the model parameters are updated using the difference between the predicted and actual heat loads. This achieves proactive feedforward regulation of cooling power, solving the problems of traditional cooling control lacking load prediction and the lag in temperature response of superconducting equipment. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic flowchart of a load prediction-driven adaptive control method for cooling in a superconducting switch station, provided in an embodiment of the present invention.

[0014] Figure 2 The flowchart illustrates the calculation of the pre-regulation urgency index for the load prediction-driven adaptive control method for cooling in superconducting switch stations provided in this embodiment of the invention. Detailed Implementation

[0015] This invention provides a load prediction-driven adaptive control method for cooling in superconducting switch stations. It addresses the technical problem that during the operation of superconducting switch stations, the grid load fluctuates with electricity demand, and the heat load of superconducting equipment changes dynamically. Existing cooling systems lack the ability to predict and adjust load change trends in advance, making it difficult to adjust cooling power in time before significant changes in heat load. This results in delayed temperature response, decreased temperature uniformity, and high cooling energy consumption in superconducting equipment.

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be noted that the numerical values ​​in the embodiments are for illustrative purposes only and do not constitute a limitation on the present invention.

[0017] Examples, as shown in the appendix Figure 1 As shown, this invention provides a load prediction-driven adaptive control method for cooling in superconducting switchyards, the method comprising the following steps: S100: Obtain the future load current prediction sequence provided by the power grid dispatching system, as well as the current temperature and cooling system operating parameters of the superconducting equipment.

[0018] In this step, the cooling control of the superconducting switch station requires advance prediction of future heat load changes, and the heat load originates from the grid load current. It is necessary to obtain the predicted load current sequence for future periods and the current temperature and cooling system operating parameters of the superconducting equipment to provide data input for heat load prediction and cooling pre-regulation.

[0019] Step S100 in the method provided in this embodiment of the invention includes: The historical load current sequence of the superconducting switch station is obtained. The historical load current sequence is the load current value recorded at a preset time interval within a preset time period in the past. The time characteristics of the corresponding time period of the historical load current sequence are obtained. The time characteristics include time identifier, weekday type identifier and date type identifier, wherein the weekday type identifier distinguishes between weekdays and non-weekdays, and the date type identifier distinguishes between ordinary days and public holidays. A load current prediction model is constructed, which takes historical load current sequences and corresponding time features as inputs and future load current prediction sequences as outputs. The model is trained using a sample training set obtained by segmenting the historical load current sequences. The historical load current sequences and time features are input into the load current prediction model, which outputs the future load current prediction sequences. Temperature values ​​from multiple temperature measuring points on the current superconducting device are collected. The maximum value among these temperature values ​​is selected as the current temperature of the superconducting device, and the standard deviation of the temperature values ​​from these multiple temperature measuring points is calculated as the temperature uniformity index of the current superconducting device. Operating parameters of the refrigeration system are collected, including the current output power of each refrigeration unit, the chilled water supply temperature, and the chilled water return temperature.

[0020] Specifically, the historical load current sequence is a sequence composed of load current values ​​recorded at preset time intervals within a preset time period at the superconducting switch station, arranged chronologically. The preset time period is 72 hours, based on the autocorrelation analysis results of the station's historical load current data. The autocorrelation coefficient shows a significant periodic peak at the 72-hour lag, indicating that 72 hours can cover the complete daily and weekly load patterns. The preset time interval is 15 minutes, based on the statistical analysis of the intervals between adjacent historical load current records. Among all historical records, the 15-minute interval has the highest frequency, i.e., the mode interval is 15 minutes. Sampling at this interval can completely preserve the main fluctuation characteristics of load changes.

[0021] The time characteristics include time markers, weekday type markers, and date type markers. The time marker represents the hour and minute of a day; the weekday type marker distinguishes between weekdays and non-weekdays, with a value of 1 indicating a weekday and 0 indicating a non-weekday, and this classification is determined based on the results of a significant test of the difference between the historical load current of the station on weekdays and non-weekdays; the date type marker distinguishes between ordinary days and public holidays, with a value of 1 indicating a public holiday and 0 indicating an ordinary day, and this classification is also determined based on the results of a significant test of the difference between the historical load current of the station on public holidays and ordinary days.

[0022] The load current prediction model is a time-series prediction model that takes historical load current sequences and corresponding time characteristics as input and outputs future load current prediction sequences. It is trained using a sample training set obtained from segmented historical load current sequences to generate predicted load current values ​​for future periods. The temperature uniformity index is the standard deviation of temperatures at multiple temperature measurement points on the superconducting equipment, reflecting the uniformity of temperature across different parts of the equipment; a smaller standard deviation indicates a more uniform temperature distribution. The operating parameters of the refrigeration system include the current output power of each refrigeration unit, the chilled water supply temperature, and the chilled water return temperature.

[0023] In this step, historical load current records for the past 72 hours are first retrieved from the historical operation database of the superconducting switch station. These records are then organized into a historical load current sequence at 15-minute intervals. Simultaneously, a time feature vector is constructed for each record, including a time identifier, a day of the week identifier, and a date identifier. The historical load current sequence is then divided into multiple training samples in chronological order. Each sample takes load current values ​​and time features from several past time points as input and outputs load current values ​​from several future time points, thus constructing a sample training set for the load current prediction model.

[0024] Then, a load current prediction model is constructed. The model inputs historical load current sequences and corresponding time features, and the output is a predicted load current sequence for future periods. The load current prediction model uses a long short-term memory network as its core structure. The input layer contains four input dimensions: load current values ​​and three time features. The intermediate hidden layer contains several long short-term memory units to capture the temporal dependencies of the load current. The output layer outputs the predicted load current sequence for future periods. The mean squared error is used as the loss function, and an adaptive moment estimation optimizer is used for training. After training, the historical load current sequences and corresponding time features are input into the trained load current prediction model to obtain the predicted load current sequence for future periods.

[0025] Before inputting historical load current sequences and time features into the load current prediction model, the load current values ​​are Min-Max normalized, linearly mapping the current values ​​in this dimension to the interval between 0 and 1. The basis for normalization is that the numerical range of load current values ​​is much larger than the value range of time features. If normalization is not performed, the gradient update during model training will be dominated by features with large numerical magnitudes, affecting training stability and prediction accuracy.

[0026] Furthermore, temperature values ​​from multiple temperature measuring points on the current superconducting device are collected. The locations of these measuring points are determined based on the finite element thermal analysis results of the superconducting device, selecting locations with larger temperature gradients that can represent the overall thermal state. The maximum value from the multiple temperature measuring points is selected as the current temperature of the superconducting device. The standard deviation of the temperature values ​​at each measuring point is calculated as an index of temperature uniformity; a smaller standard deviation indicates a more uniform temperature distribution across the device. Simultaneously, operating parameters of the refrigeration system are collected. The current output power of each refrigeration unit is obtained by reading the real-time power output signal from the refrigeration unit controller. The chilled water supply and return temperatures are collected in real-time by temperature sensors installed on the supply and return water pipelines, respectively. The difference between the supply and return water temperatures, multiplied by the chilled water flow rate and then by the specific heat capacity of water, can be used to calculate the actual heat removed.

[0027] For example, the training process of the load current prediction model is as follows: Load current records for 12 consecutive months are collected from the historical operation database of the superconducting switch station, sampled at 15-minute intervals, resulting in approximately 35,040 records. Based on the autocorrelation analysis results of the station's load current data, the input time window length is set to 24 hours, i.e., 96 time points, and the output prediction duration is set to 4 hours, i.e., 16 time points. All records are divided into multiple training samples in chronological order. Each sample takes the load current values ​​and corresponding time characteristics of the 96 consecutive time points as input and the load current values ​​of the subsequent 16 time points as output. The sample set is divided into a training set, a validation set, and a test set in an 8:1:1 ratio.

[0028] The load current prediction model adopts a long short-term memory network structure. The input layer has a dimension of 4, corresponding to the load current value, time marker, weekday type marker, and date type marker, respectively. The hidden layer contains two long short-term memory layers, with 64 hidden units in the first layer and 32 hidden units in the second layer, in order to capture the short-term fluctuations and periodic patterns of the load current. The output layer is a fully connected layer with a dimension of 16, corresponding to the predicted load current values ​​at 16 future time points.

[0029] During training, the optimizer uses adaptive moment estimation with an initial learning rate of 0.001 and a batch size of 64. The loss function is the mean squared error, calculated using the following formula: Where M is the total number of time points in a batch, Let m be the predicted load current value output by the model at the m-th time point. This represents the actual load current value at that point in time. After each training round, the mean square error is calculated on the validation set. Training stops when the mean square error on the validation set does not decrease for 15 consecutive rounds or when the total number of rounds reaches 200.

[0030] For example, in one training cycle, the validation set mean squared error decreased to 0.021 kA at the 87th epoch. 2 If the loss does not reach a new low in the subsequent 15 rounds, an early stop is triggered, and the model weights corresponding to that round are saved as the completed load current prediction model.

[0031] In summary, this step, by acquiring the load current prediction sequence and the current equipment and refrigeration system status parameters, provides a forward-looking data foundation for subsequent heat load prediction and refrigeration power feedforward adjustment.

[0032] S200: Based on a pre-calibrated AC loss model, the load current prediction sequence is converted into a predicted heat load sequence.

[0033] In this step, the load current prediction sequence provided by the power grid dispatching system cannot be directly used for cooling control decisions. The load current value needs to be converted into the heat load value corresponding to the superconducting equipment so that the cooling system can make forward-looking adjustments based on the heat load prediction results.

[0034] Step S200 in the method provided in this embodiment of the invention includes: obtaining the loss coefficient and inherent heat leakage value of the AC loss model, wherein the loss coefficient and the inherent heat leakage value are obtained in advance through historical operating data; calculating the square of each load current prediction value in the future time period load current prediction sequence to obtain the current square value; multiplying the current square value by the loss coefficient and adding the inherent heat leakage value to obtain the predicted heat load value corresponding to the load current prediction value; arranging all the predicted heat load values ​​in chronological order to form the predicted heat load sequence.

[0035] Specifically, the AC loss model is a mathematical model used to describe the heat loss generated by superconducting equipment under AC load current. The AC loss model adopts an engineering approximation, assuming that the heat load of the superconducting equipment consists of two parts: AC loss and inherent heat leakage. The AC loss is proportional to the square of the load current, while the inherent heat leakage is a constant value independent of the load current. The model expression is: Predicted heat load value = Square of current × Loss coefficient + Inherent heat leakage value. The loss coefficient is a coefficient in the AC loss model characterizing the heat loss generated by the superconducting equipment under unit current square, while the inherent heat leakage value is the static heat continuously entering the superconducting equipment through environmental heat transfer, heat conduction from the supporting structure, etc., when it is not carrying a load current, and is independent of the load current.

[0036] In this embodiment of the invention, the loss coefficient and inherent heat leakage value are obtained through pre-calibration using historical operating data. During the calibration process, load current and corresponding measured heat load data from the historical operation of the superconducting switchyard are collected. Using the square of the load current as the independent variable and the measured heat load as the dependent variable, a univariate linear equation is fitted using least-squares linear regression. The slope of the current square term obtained from the regression fitting is used as the loss coefficient, and the intercept obtained from the regression fitting is used as the inherent heat leakage value. The approximate linear relationship between AC loss and the current square is an engineering approximation, and the coefficient of the current square term is the main fitting parameter. This calibration method uses historical data covering the operating records of the superconducting switchyard at different load levels, ensuring that the calibration results can reflect the heat loss characteristics of the superconducting equipment under actual operating conditions.

[0037] In this step, the loss factor and inherent heat leakage value are first obtained from the calibration results. Then, each predicted load current value in the future time period load current prediction sequence output by S100 is traversed, and its square is calculated to obtain the current square value. The current square value is multiplied by the loss factor, and then the inherent heat leakage value is added to obtain the predicted heat load value corresponding to that moment. After traversing all the predicted load current values, all the predicted heat load values ​​are arranged in chronological order to form a predicted heat load sequence that corresponds one-to-one with the time points of the load current prediction sequence.

[0038] For example, consider the calibration and prediction process of a superconducting switch station. Twenty-one sets of load current and measured heat load data covering different load levels were collected from the station's historical operating data. Least squares linear regression was performed with the squared load current as the independent variable and the measured heat load as the dependent variable, yielding a slope of 0.35 kW / kA. 2 With an intercept of 12.0 kW, a slope of 0.35 is used as the loss coefficient, and an intercept of 12.0 is used as the inherent heat leakage value.

[0039] For example, if the predicted load current value at a certain moment in the future load current prediction sequence output by S100 is 1.2kA, then the square of the current at that moment is 1.2kA. 2 =1.44kA2 The predicted heat load value is 1.44 × 0.35 + 12.0 = 0.504 + 12.0 = 12.504 kW. The above calculation is performed on each predicted load current value in the predicted load current sequence, and the resulting predicted heat load values ​​are arranged in chronological order to form the predicted heat load sequence.

[0040] In summary, this step converts the load current prediction sequence into a predicted heat load sequence using an AC loss model, achieving a quantitative mapping from grid load to the heat load of superconducting equipment. This provides direct input data for subsequent heat load change trend analysis and refrigeration pre-regulation.

[0041] S300: Based on the predicted heat load sequence, calculate the predicted heat load change rate, and combine it with the current thermal inertia time constant of the refrigeration system and the current temperature margin of the superconducting equipment to generate a pre-adjustment urgency index.

[0042] In this step, trend information needs to be extracted from the predicted heat load sequence, and combined with the current refrigeration system response capability and the thermal state of the superconducting equipment, it is determined whether the current heat load change requires proactive intervention beyond conventional feedback regulation. The calculation process for the pre-regulation urgency index in this step is as follows: Figure 2 As shown.

[0043] Step S300 in the method provided in this embodiment of the invention includes: From the predicted heat load sequence, obtain the current predicted heat load value corresponding to the current time and the next predicted heat load value corresponding to the next time point. Calculate the heat load deviation by subtracting the current predicted heat load value from the next predicted heat load value. Divide the heat load deviation by the time interval between the current time and the next time point to obtain the predicted heat load change rate. Obtain the current thermal inertia time constant of the refrigeration system, where the current thermal inertia time constant of the refrigeration system is the time required for the refrigeration system to change its output power until the chilled water temperature reaches the corresponding steady-state value; calculate the critical temperature of the superconducting device minus the temperature of the current superconducting device to obtain the temperature margin of the current superconducting device; calculate the sum of the rated power of each refrigeration unit minus the sum of the current output power of each refrigeration unit to obtain the power margin of the current refrigeration system; after normalizing the predicted heat load change rate, the reciprocal of the temperature margin of the current superconducting device, the reciprocal of the power margin of the current refrigeration system, and the temperature uniformity index of the current superconducting device, perform weighted calculation to obtain the pre-adjustment urgency index.

[0044] Specifically, the predicted heat load change rate is an indicator that characterizes the magnitude of change in heat load per unit time. The calculation formula is: predicted heat load change rate = (predicted heat load value at the next moment - current predicted heat load value) ÷ time interval, with the unit being kW / min. It is used to reflect the upward or downward trend of future heat load.

[0045] The thermal inertia time constant is the time required for a refrigeration system to change its output power until the chilled water temperature reaches its corresponding steady-state value, measured in minutes. This value reflects the response inertia of the refrigeration system. Specifically, after a step change in the output power of the refrigeration unit, the time elapsed until the chilled water temperature transitions from its initial value to a new steady-state value, and the temperature change first reaches 63.2% of the total steady-state temperature change, is the thermal inertia time constant. This 63.2% value is determined by the step response characteristics of a first-order inertial system: for a first-order inertial system, the response quantity approaches the steady-state value exponentially with time, and the time constant corresponds to the moment when the response quantity reaches 63.2% of the steady-state change. This proportion originates from the exponential response function. ≈0.632.

[0046] Temperature margin is the difference between the critical temperature of a superconducting device and the current temperature of the superconducting device, measured in K. A larger temperature margin indicates that the device is further away from the quench temperature and has a larger allowable range for heat load fluctuations; a smaller temperature margin indicates that the device is closer to the risk of quench and requires more urgent cooling intervention.

[0047] In calculating the reciprocal of the temperature margin, to prevent the temperature margin from approaching zero and its reciprocal from approaching infinity due to the current temperature of the superconducting equipment being close to the critical temperature, thus causing instability in the calculation of the pre-adjustment urgency index, a lower limit cutoff value is set for the temperature margin. The lower limit cutoff value is 1.0K, meaning that when the temperature margin is less than 1.0K, the reciprocal of the temperature margin is calculated based on 1.0K, ensuring that the maximum value of the reciprocal of the temperature margin does not exceed 1.0 / K. The basis for this cutoff value is that when the temperature margin is below 1.0K, the superconducting equipment is already in a dangerous operating state, and further differentiation of subtle margin differences has no significant practical decision-making significance for the urgency index. Simultaneously, the cutoff process ensures that the pre-adjustment urgency index always remains within a stable and finite numerical range.

[0048] The power margin of the refrigeration system is the sum of the rated power of each refrigeration unit minus the sum of the current output power of each refrigeration unit, with the unit being kW. It reflects the additional cooling capacity that the refrigeration system can currently provide. It is calculated by reading the rated power on the nameplate of each refrigeration unit and the real-time power output signal of the controller.

[0049] In calculating the reciprocal of the power margin, to prevent the power margin from approaching zero and its reciprocal from approaching infinity due to all refrigeration units operating at rated power, thus causing instability in the pre-adjustment urgency index calculation, a lower limit cutoff value is set for the power margin. This lower limit cutoff value is 5% of the total rated power of the refrigeration system. That is, when the power margin is less than 5% of the total rated power of the refrigeration system, the reciprocal of the power margin is calculated based on 5% of the total rated power, ensuring that the maximum value of the reciprocal of the power margin does not exceed 20 ÷ the total rated power of the refrigeration system. The basis for this cutoff value is that when the power margin is less than 5% of the total rated power, the refrigeration system is already close to full load operation. Further differentiation of minor differences in remaining power has little practical significance for the actual decision-making of the urgency index. At the same time, the cutoff process ensures that the pre-adjustment urgency index remains within a stable and finite numerical range.

[0050] The pre-regulation urgency index is a comprehensive index calculated by weighting and normalizing the predicted rate of change in heat load, the reciprocal of the temperature margin, the reciprocal of the power margin, and the temperature uniformity index. It is used to quantify the urgency of pre-regulation for cooling under the current trend of heat load change. A higher pre-regulation urgency index indicates a greater need for immediate feedforward regulation.

[0051] Before calculation, the four input quantities were subjected to Min-Max normalization. The upper and lower bounds of each index were determined by the statistical distribution of the historical operating data of the superconducting switch station: the upper and lower bounds of the predicted heat load change rate were taken from the 99th and 1st quantiles of the predicted heat load change rate in the historical data; the upper and lower bounds of the reciprocal of the temperature margin were taken from the 99th and 1st quantiles of the reciprocal of the temperature margin in the historical data; the upper and lower bounds of the reciprocal of the power margin were taken from the 99th and 1st quantiles of the reciprocal of the power margin in the historical data; and the upper and lower bounds of the temperature uniformity index were taken from the 99th and 1st quantiles of the temperature uniformity index in the historical data. The normalization formula was: Normalized value = (Original value - Lower bound) ÷ (Upper bound - Lower bound). The normalization process ensured that all four indices were in the range of 0 to 1.

[0052] The weighting coefficients for the weighted calculation are determined as follows: The predicted heat load change rate, reciprocal of the temperature margin, reciprocal of the power margin, and temperature uniformity index within 10 minutes prior to multiple sudden heat load events in the station's historical operating data are collected as four independent variables. Whether the corresponding event resulted in temperature or limit exceedance is used as the dependent variable. Logistic regression is performed on the historical event samples to obtain the regression coefficients of the four independent variables. Each regression coefficient is normalized by dividing it by the sum of the four regression coefficients, ensuring that the sum of the four weighting coefficients is 1. The resulting normalized coefficients are the final weights of each index.

[0053] In this step, the predicted heat load value at the current time and the predicted heat load value at the next time point are first read from the predicted heat load sequence. The difference between the two is calculated and divided by the time interval to obtain the predicted heat load change rate. Then, the thermal inertia time constant of the current refrigeration system is read from the system parameter library. The thermal inertia time constant of the refrigeration system is pre-measured and stored through a step response experiment. Subsequently, the difference between the critical temperature of the superconducting device and the current temperature of the superconducting device is calculated to obtain the temperature margin; the sum of the rated power of each refrigeration unit is calculated and the sum of the current output power is subtracted to obtain the power margin. The predicted heat load change rate, the reciprocal of the temperature margin, the reciprocal of the power margin, and the temperature uniformity index are all subjected to Min-Max normalization. Then, the pre-adjustment urgency index is calculated according to the weighted weights and output for subsequent trigger judgment.

[0054] For example, consider the operating data of a superconducting switch station. In the predicted heat load sequence, the predicted heat load value at the current time is 12.5 kW, and the predicted heat load value at the next time point is 13.4 kW, with a time interval of 15 minutes. The predicted heat load change rate is (13.4 - 12.5) ÷ 15 = 0.06 kW / min. The current thermal inertia time constant of the refrigeration system is 20 minutes, measured through a step response experiment. The critical temperature of the superconducting device is 90 K, and the current temperature of the superconducting device is 84 K. The temperature margin is 90 - 84 = 6 K, and the reciprocal of the temperature margin is 1 / 6 ≈ 0.167. The sum of the rated power of the three refrigeration units is 300 kW, and the sum of the current output power is 210 kW. The power margin is 300 - 210 = 90 kW, and the reciprocal of the power margin is 1 / 90 ≈ 0.011. The temperature uniformity index is calculated to be 2.5 K.

[0055] The normalized upper and lower bounds of each indicator are obtained from the statistical distribution of historical operating data. For example, in a certain calculation, the normalized value of the predicted heat load change rate is 0.65, the normalized value of the reciprocal of the temperature margin is 0.48, the normalized value of the reciprocal of the power margin is 0.35, and the normalized value of the temperature uniformity index is 0.60. The weights determined based on logistic regression analysis of historical heat load abrupt events are 0.45, 0.25, 0.20, and 0.10, respectively. The pre-regulation urgency index = 0.45 × 0.65 + 0.25 × 0.48 + 0.20 × 0.35 + 0.10 × 0.60 = 0.2925 + 0.12 + 0.07 + 0.06 = 0.5425.

[0056] In summary, this step generates a pre-adjustment urgency index by integrating multi-dimensional information such as heat load change trends, refrigeration system inertia, equipment temperature margin, and power reserve. This provides a quantitative basis for determining whether to trigger feedforward regulation, which is beneficial for identifying scenarios requiring intervention in advance before the temperature changes significantly.

[0057] S400: When the pre-adjustment urgency index exceeds the preset trigger threshold, the feedforward cooling power increment is obtained based on the predicted heat load change rate and the thermal inertia time constant analysis, and the output power of the cooling unit is adjusted in advance.

[0058] In this step, when the pre-adjustment urgency index exceeds the trigger threshold, it is necessary to determine the amount of cooling power that needs to be increased in advance, and to reasonably allocate this power increment to each cooling unit so that the cooling system can complete the power pre-setting before the heat load changes.

[0059] Step S400 in the method provided in this embodiment of the invention includes: A feedforward power mapping model is constructed, which takes the predicted heat load change rate and the current thermal inertia time constant of the refrigeration system as inputs and the feedforward refrigeration power increment as output. When the pre-adjustment urgency index exceeds a preset trigger threshold, the predicted heat load change rate and the current thermal inertia time constant of the refrigeration system are input into the feedforward power mapping model to obtain the feedforward refrigeration power increment. The feedforward refrigeration power increment is allocated to each of the refrigeration units to adjust the output power of each refrigeration unit in advance.

[0060] Specifically, the feedforward cooling power increment is an additional cooling power value, measured in kW, calculated based on the predicted rate of change of heat load and the thermal inertia time constant, that needs to be added to the current cooling output. This is used to reserve cooling capacity in advance before actual changes in heat load occur. The feedforward power mapping model takes the predicted rate of change of heat load and the current thermal inertia time constant of the cooling system as inputs and the feedforward cooling power increment as output.

[0061] A preset trigger threshold is used to determine whether the current heat load change requires the initiation of feedforward regulation. Its value is determined based on statistical analysis of the pre-regulation urgency index before historical heat load abrupt events at the superconducting switch station. Specifically, the pre-regulation urgency index values ​​within 10 minutes before each heat load abrupt event in the past two years are statistically analyzed, and the lower quartile of all pre-event index values ​​is taken as the preset trigger threshold; in this embodiment, it is 0.6. When the urgency index exceeds the preset trigger threshold, the probability of an impending heat load abrupt event increases significantly, requiring the initiation of feedforward regulation.

[0062] In this step, it is first determined whether the pre-adjustment urgency index exceeds the preset trigger threshold of 0.6. If it does not exceed the threshold, the current cooling output power is maintained unchanged, and the temperature is maintained solely by conventional feedback regulation. If it exceeds the threshold, the current predicted heat load change rate and the current thermal inertia time constant of the cooling system are input into the pre-built and trained feedforward power mapping model, and the model outputs the feedforward cooling power increment. Subsequently, the feedforward cooling power increment is allocated to each cooling unit according to the allocation method described later, thus adjusting the output power of each cooling unit in advance.

[0063] When the feedforward cooling power increment is negative, it indicates that the predicted heat load is decreasing, and the cooling output power should be reduced in advance to avoid overcooling. At this time, the target output power of each cooling unit is the current output power plus the negative allocation increment. When the target output power is lower than the minimum allowable output power of the cooling unit, the target output power is set to the minimum output power to ensure the safe and stable operation of the cooling system.

[0064] When the feedforward cooling power increment exceeds the sum of the available power margins of all cooling units, the output power of each cooling unit is directly adjusted to its rated power. The actual executable feedforward cooling power increment is recorded as the sum of all available power margins. The unexecuted power increments are not processed temporarily and are recalculated and executed in the next round of adjustment based on the latest predicted heat load change rate.

[0065] Step S400 in the method provided in this embodiment of the invention further includes: Historical operating data of the superconducting switch station is acquired. This historical operating data includes the predicted heat load change rate at multiple historical adjustment times, the current thermal inertia time constant of the refrigeration system, and the actual feedforward refrigeration power increment executed at each historical adjustment time. For each historical adjustment time, post-adjustment evaluation indicators are acquired. These post-adjustment evaluation indicators include the temperature uniformity index and temperature margin of the superconducting equipment within a preset time period after adjustment. From all historical adjustment times, historical adjustment times that simultaneously meet the following conditions are selected as valid adjustment times: the temperature uniformity index of the superconducting equipment within the preset time period after adjustment is less than a preset uniformity threshold, and the temperature margin of the superconducting equipment within the preset time period after adjustment is greater than a preset margin threshold. The predicted heat load change rate and the current thermal inertia time constant of the refrigeration system at each effective adjustment moment are used as a set of input samples, and the actual feedforward refrigeration power increment executed at the effective adjustment moment is used as the output label corresponding to the input sample to construct a training sample set; a feedforward power mapping model is constructed. The input layer of the feedforward power mapping model contains two input nodes, which correspond to the predicted heat load change rate and the current thermal inertia time constant of the refrigeration system, respectively. The hidden layer of the feedforward power mapping model uses radial basis functions as activation functions. The output layer of the feedforward power mapping model contains one output node, which outputs the feedforward refrigeration power increment. The feedforward power mapping model is trained using the training sample set. During the training process, the center and width of the radial basis function are determined by the K-means clustering algorithm, and the connection weights from the hidden layer to the output layer are updated by the gradient descent algorithm until the output error of the feedforward power mapping model converges, thus obtaining the trained feedforward power mapping model.

[0066] More specifically, the construction and training process of the feedforward power mapping model is as follows. Preset uniformity thresholds and preset margin thresholds are used to select historical adjustment moments with good adjustment effects. Both thresholds are determined based on statistical analysis of the post-adjustment temperature uniformity and temperature margin of the station's historical adjustment moments: historical adjustment moments are arranged in ascending order of post-adjustment temperature uniformity, and the 60th percentile is taken as the preset uniformity threshold; historical adjustment moments are arranged in ascending order of post-adjustment temperature margin, and the 40th percentile is taken as the preset margin threshold. The radial basis function is the activation function used in the hidden layer of the feedforward power mapping model. Its value depends only on the Euclidean distance between the input and the center; the closer the distance, the larger the output, used to respond to the local features of the input sample.

[0067] For example, operational data from 3600 historical adjustment moments over the past three years were obtained from the station's historical operational database. Each record included the predicted rate of change of heat load, thermal inertia time constant, and actual feedforward cooling power increment. For each historical adjustment moment, the temperature uniformity index and temperature margin of the superconducting equipment within 30 minutes after adjustment were obtained. Records that met the conditions of temperature uniformity index being less than a preset uniformity threshold and temperature margin being greater than a preset margin threshold were selected as valid adjustment moments, resulting in 2500 valid adjustment moments.

[0068] The 2500 input and output samples were divided into a training set of 2000 groups and a test set of 500 groups in an 8:2 ratio. The radial basis function network had 30 hidden layer nodes, and the centers and widths of the 30 radial basis functions were determined using K-means clustering. During training, gradient descent was used to update the connection weights from the hidden layer to the output layer, with a learning rate of 0.01 and a maximum of 500 iterations.

[0069] The training set output error is measured using mean squared error, calculated using the following formula: Where H is the number of samples in a batch. This represents the predicted feedforward cooling power increment of the feedforward power mapping model for the h-th sample. This represents the actual feedforward cooling power increment corresponding to the h-th sample. Training stops when the mean squared error of the training set decreases by less than 0.5% for 50 consecutive iterations or when the maximum number of iterations (500) is reached.

[0070] For example, in a certain training process, the mean squared error of the training set decreased to 0.0061kW in the 300th round. 2 After 50 consecutive rounds, the decrease rate is less than 0.5%, triggering the stopping condition. The model parameters for that round are saved as the completed feedforward power mapping model.

[0071] Step S400 in the method provided in this embodiment of the invention further includes: Obtain the current output power and rated power of each cooling unit. For each cooling unit, subtract the current output power of the cooling unit from its rated power to obtain the available power margin of the cooling unit. Calculate the sum of the available power margins of all cooling units, and divide the available power margin of each cooling unit by the sum of the available power margins to obtain the allocation weight of the cooling unit. Multiply the feedforward cooling power increment by the allocation weight of each cooling unit to obtain the allocation increment of the cooling unit, and add the allocation increment to the current output power of the cooling unit to obtain the target output power of the cooling unit. Adjust the output power of each cooling unit to the corresponding target output power to complete the advance adjustment of the output power of each cooling unit.

[0072] Specifically, the available power margin is the difference between the rated power of each refrigeration unit and its current output power, reflecting the additional output capacity that the refrigeration unit can currently provide, measured in kW. The allocation weight is a proportional coefficient obtained by dividing the available power margin of each refrigeration unit by the sum of the available power margins of all refrigeration units. It is used to allocate the feedforward refrigeration power increment according to the remaining capacity of each unit. The formula for calculating the allocation increment is: Allocation Increment = Feedforward Refrigeration Power Increment × Allocation Weight. The formula for calculating the target output power is: Target Output Power = Current Output Power + Allocation Increment.

[0073] For example, suppose a superconducting switch station is configured with three cooling units, and the feedforward cooling power increment output by the feedforward power mapping model is 12kW. The rated powers of the three cooling units are 100kW, 100kW, and 100kW, respectively, and their current output powers are 60kW, 70kW, and 80kW, respectively. Then, the available power margins of each unit are 40kW, 30kW, and 20kW, respectively, and the sum of the available power margins is 90kW. The allocation weights of each unit are 40÷90≈0.444, 30÷90≈0.333, and 20÷90≈0.222, respectively. The allocation increments of each unit are 12×0.444≈5.33kW, 12×0.333≈4.00kW, and 12×0.222≈2.67kW, respectively, and the target output powers are 65.33kW, 74.00kW, and 82.67kW, respectively. Adjusting the output power of each cooling unit to the corresponding target output power completes the advance adjustment.

[0074] In summary, this step obtains the feedforward cooling power increment through the feedforward power mapping model and allocates it to each cooling unit according to the proportion of available power margin, realizing the feedforward active adjustment of cooling power. This is beneficial for pre-stocking cooling capacity and reducing temperature fluctuations when the heat load changes.

[0075] S500: After the adjustment is executed, the actual temperature response of the superconducting device is collected, the difference between the predicted heat load sequence and the actual heat load is compared, and the loss coefficient of the AC loss model and the thermal inertia time constant are updated and corrected.

[0076] In this step, the loss coefficient and thermal inertia time constant of the AC loss model may drift during equipment operation due to aging, changes in operating conditions, etc., resulting in a deviation between the predicted heat load and the actual heat load. It is necessary to use the adjusted actual temperature response data to back-calculate the actual heat load, compare the difference between the predicted and actual values, and update and correct the model parameters.

[0077] Step S500 in the method provided in this embodiment of the invention includes: After the adjustment is executed, the actual temperature change sequence of the superconducting device is collected within the preset observation period; based on the actual temperature change sequence and the actual output power of the refrigeration system, the actual heat load sequence within the preset observation period is calculated based on the heat balance equation; the part of the predicted heat load sequence corresponding to the preset observation period is compared with the actual heat load sequence, and the root mean square error between the predicted heat load sequence and the actual heat load sequence is calculated as the heat load prediction deviation. When the predicted heat load deviation exceeds the preset deviation tolerance limit, the actual heat load sequence and the load current value of the corresponding time period are used as new training samples to incrementally train the AC loss model, and update the loss coefficient and inherent heat leakage value; based on the response delay time of the actual temperature change sequence, the current thermal inertia time constant is obtained, and the current thermal inertia time constant is weighted and averaged with the current refrigeration system thermal inertia time constant to obtain the updated refrigeration system thermal inertia time constant.

[0078] Specifically, the actual temperature change sequence is a sequence composed of temperature values ​​from the superconducting equipment collected at fixed sampling intervals within a preset observation period after the adjustment is implemented, arranged in chronological order. The sampling interval is 15 minutes, consistent with the mode of the sampling frequency distribution of the station's historical temperature records. The preset observation period is 4 hours, based on the statistical analysis of the duration of major fluctuations in heat load in the station's historical operating data; the median length of the main periods of continuous increase or decrease in heat load is 4 hours.

[0079] Heat load prediction bias is calculated by comparing the portion of the predicted heat load sequence corresponding to a preset observation period with the actual heat load sequence, and then calculating the root mean square error between the two. This error is used to measure the prediction accuracy of the AC loss model. The preset upper limit of tolerance for bias is set at 1.5 times the mean of the historical statistical distribution of heat load prediction bias at the station. This is based on statistical analysis of historical prediction biases. If the bias exceeds the preset upper limit of tolerance, the model parameters are considered to be significantly inaccurate, and incremental training needs to be triggered.

[0080] In this step, after the adjustment is executed, a preset observation period is entered, and the temperature values ​​of the superconducting equipment are collected at 15-minute sampling intervals to form an actual temperature change sequence. Then, based on the actual temperature change sequence and the actual output power of the refrigeration system, the actual heat load value for each sampling interval within this period is calculated according to the heat balance equation described later, forming an actual heat load sequence. The portion of the predicted heat load sequence corresponding to the observation period is compared point by point with the actual heat load sequence, and the root mean square error is calculated to obtain the heat load prediction deviation. When the heat load prediction deviation exceeds the preset deviation tolerance limit, the actual heat load sequence and the square of the load current for the corresponding period are used as new training samples to incrementally train the AC loss model, updating the loss coefficient and the inherent heat leakage value.

[0081] The incremental training method is as follows: New training samples are merged with the original calibration samples, and the least squares linear regression is re-executed with the square of the load current as the independent variable and the actual heat load as the dependent variable to update the loss coefficient and inherent heat leakage value. If the amount of historical sample data is too large, leading to an increased computational burden, only the samples corresponding to the most recent N adjustments can be retained for rolling window regression, with N being the sample size corresponding to 50 adjustments, to achieve a balance between update response speed and computational cost.

[0082] Simultaneously, the current thermal inertia time constant is obtained based on the step response delay time of the actual temperature change sequence, and then updated after being weighted and averaged with the original thermal inertia time constant.

[0083] Step S500 in the method provided in this embodiment of the invention further includes: The total heat capacity of the superconducting device is obtained, which is the amount of heat absorbed or released by the superconducting device for every unit change in temperature. For each sampling interval within the preset observation period, the temperature value at the next sampling moment in the actual temperature change sequence is obtained by subtracting the temperature value at the previous sampling moment from the temperature value at the next sampling moment to obtain the temperature change. The total heat capacity is multiplied by the temperature change and then divided by the duration of the sampling interval to obtain the heat storage change power of the superconducting device. The actual output power of the cooling system within the sampling interval is obtained, and the actual output power of the cooling system is added to the heat storage change power of the superconducting device to obtain the actual heat load value of the sampling interval. The actual heat load values ​​of all sampling intervals within the preset observation period are arranged in chronological order to form the actual heat load sequence.

[0084] More specifically, total heat capacity is the amount of heat absorbed or released by a superconducting device for every unit change in temperature, measured in kJ / K, and obtained through device design parameters or heat capacity calibration experiments. The formula for calculating the heat storage change power of a superconducting device is: Heat storage change power of superconducting device = Total heat capacity × Temperature change ÷ Sampling interval, reflecting the heat power stored or released by the superconducting device per unit time due to temperature changes.

[0085] The actual heat load is calculated using the formula: Actual Heat Load = Actual Output Power of the Cooling System + Change in Heat Storage Power of the Superconducting Device. The principle is based on the heat balance equation: the heat load of the superconducting device consists of two parts: the heat removed by the cooling system and the change in heat storage within the superconducting device itself. When the device temperature rises, the change in heat storage power is positive; when the device temperature decreases, the change in heat storage power is negative. The total heat capacity is obtained through a heat capacity calibration experiment. In this experiment, an electric heating device with known power is placed inside the superconducting device, and the rate of temperature change over time is recorded. The average ratio of the rate of temperature change to the heating power under different heating powers is taken as the total heat capacity.

[0086] For example, the total heat capacity of a superconducting device is measured to be 800 kJ / K through a heat capacity calibration experiment. Within a preset observation period, a sampling interval of 15 minutes is used. The temperature at the previous sampling moment is 83.5 K, and the temperature at the next sampling moment is 84.1 K, a temperature change of 0.6 K. The heat storage power of the superconducting device is 800 × 0.6 ÷ 15 = 32 kW. The actual output power of the cooling system within this sampling interval is 230 kW, therefore the actual heat load value for this sampling interval is 230 + 32 = 262 kW. The above calculation is performed for each sampling interval within the preset observation period, and the resulting actual heat load values ​​are arranged in chronological order to form an actual heat load sequence.

[0087] Step S500 in the method provided in this embodiment of the invention further includes: obtaining the moment when the output power of the cooling unit undergoes a step change from the actual temperature change sequence as the step start moment; obtaining the temperature value of the superconducting device corresponding to the step start moment, and the total steady-state temperature change corresponding to the step change; obtaining the moment when the temperature of the superconducting device starts to change from the step start moment until the temperature change first reaches a preset proportion of the total steady-state temperature change from the actual temperature change sequence as the step response moment; subtracting the step start moment from the step response moment, and using the difference as the current thermal inertia time constant.

[0088] Specifically, the total steady-state temperature change is the difference between the final steady-state temperature reached by the superconducting device after a step change, from the initial temperature at the start of the step change to the initial temperature at the start of the step change. The real-time acquisition method for the total steady-state temperature change is as follows: at the start of the step change, first obtain the predicted heat load value at that moment, and then read the target output power of the refrigeration system after the step change. Based on the heat balance equation, the steady-state heat load equals the sum of the refrigeration output power and the change in the device's heat storage power, i.e., predicted heat load value = target output power + total heat capacity × (steady-state temperature value - temperature at the start of the step change) ÷ time constant. Substituting the predicted heat load value, target output power, and total heat capacity into the heat balance equation, the steady-state temperature value can be estimated. Subtracting the temperature at the start of the step change from the steady-state temperature value yields the total steady-state temperature change.

[0089] The preset ratio is 63.2% because the definition of the thermal inertia time constant itself uses the criterion that the response quantity reaches 63.2% of the steady-state change. The S500 needs to measure the actual value of this parameter under the current operating state, and the measurement criterion must be consistent with the definition. If a different ratio is used, the measured result will no longer be the thermal inertia time constant. Therefore, the time difference between the moment when the temperature change first reaches 63.2% of the total steady-state temperature change and the moment of the step change is taken as the current measured value of the thermal inertia time constant.

[0090] The current thermal inertia time constant is calculated as: Current thermal inertia time constant = Step response time - Step start time. The updated thermal inertia time constant is obtained using a weighted average method, and the calculation formula is: Updated refrigeration system thermal inertia time constant = 0.7 × Current refrigeration system thermal inertia time constant + 0.3 × Current thermal inertia time constant. The weights are determined based on the statistical analysis of the historical parameter update effect of this station.

[0091] It should be noted that the objects measured for the thermal inertia time constant differ: the original thermal inertia time constant was measured using the step response of chilled water temperature, reflecting the inherent response inertia of the refrigeration system; the current thermal inertia time constant is measured using the step response of the superconducting device temperature, reflecting the response delay of the complete heat transfer chain from the change in refrigeration output to the change in the superconducting device temperature. This chain includes the series delay of both the chilled water temperature change and the heat storage of the superconducting device itself. Since the temperature of the superconducting device is the direct control target of refrigeration regulation, the current thermal inertia time constant measured using the device temperature response more closely reflects the hysteresis characteristics of the actual regulation process. Therefore, a weighted average of the current thermal inertia time constant and the original thermal inertia time constant is performed to obtain the updated refrigeration system thermal inertia time constant, which takes into account both the inherent characteristics of the refrigeration system and the actual operating characteristics.

[0092] The weighted average weights are 0.7 and 0.3, and their determination method is as follows: the historical parameter update records are backtracked and verified, the candidate weight combinations are traversed, and the temperature overshoot and temperature recovery time in subsequent control after the update of the thermal inertia time constant are used as evaluation indicators. The weight combination that minimizes the temperature overshoot and temperature recovery time is selected. The weights of 0.7 and 0.3 are determined to be the optimal combination after verification by historical data. That is, 70% of the original time constant is retained and 30% of the current measured time constant is included, so as to track the changes in the operating state while maintaining the inherent response characteristics of the system.

[0093] For example, after the adjustment is executed, the output power of the cooling unit undergoes a step change at a certain moment, at 10:00. At this time, the temperature of the superconducting device is 82.0K. After the step change, the temperature of the superconducting device eventually stabilizes at 85.0K, and the total steady-state temperature change is 85.0 - 82.0 = 3.0K. The moment when the temperature change first reaches 3.0 × 63.2% = 1.9K is found in the actual temperature change sequence, i.e., the moment when the temperature reaches 83.9K is 10:20, and the step response time is 10:20. The current thermal inertia time constant is 10:20 - 10:00 = 20 minutes. The original thermal inertia time constant was 18 minutes, and the updated thermal inertia time constant is 0.7 × 18 + 0.3 × 20 = 12.6 + 6 = 18.6 minutes.

[0094] In summary, this step involves calculating the actual heat load by collecting the actual temperature response and updating and correcting the AC loss model and thermal inertia time constant accordingly. This enables online calibration of the prediction model, which helps maintain prediction accuracy and control effectiveness.

[0095] Through the above specific implementation methods, the embodiments of the present invention achieve the following technical effects: This invention proposes a load prediction-driven adaptive control method for cooling in superconducting switchgear. By acquiring a predicted load current sequence, it converts this sequence into a predicted heat load sequence based on an AC loss model, enabling quantitative prediction of future heat load changes in the superconducting equipment. By calculating the predicted heat load change rate and combining it with the thermal inertia time constant of the cooling system, the temperature margin of the superconducting equipment, and the power reserve, a pre-regulation urgency index is generated, quantifying the urgency of cooling pre-regulation. When the urgency index exceeds a trigger threshold, a feedforward power mapping model is used to obtain the incremental cooling power and allocate it according to the available reserve of each unit, achieving proactive power pre-setting before heat load changes. After regulation, the actual heat load is calculated back based on the heat balance equation, and the AC loss model is incrementally trained and the thermal inertia time constant is updated with weights, achieving online closed-loop calibration of the prediction model. This achieves load prediction-driven feedforward control and adaptive parameter correction of cooling power, solving the technical problems of traditional cooling control lacking load prediction capability, lagging temperature response of superconducting equipment, and inability to update prediction parameters online.

[0096] It should be noted that the order of the embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

Claims

1. A method for adaptive cooling control of superconducting switchyards driven by load prediction, characterized in that, include: Obtain the future load current prediction sequence provided by the power grid dispatching system, as well as the current temperature and cooling system operating parameters of the superconducting equipment; Based on a pre-calibrated AC loss model, the load current prediction sequence is converted into a predicted heat load sequence; Based on the predicted heat load sequence, the predicted heat load change rate is calculated, and combined with the current thermal inertia time constant of the refrigeration system and the current temperature margin of the superconducting equipment, a pre-adjustment urgency index is generated. When the pre-adjustment urgency index exceeds the preset trigger threshold, the feedforward cooling power increment is obtained based on the predicted heat load change rate and the thermal inertia time constant, and the output power of the cooling unit is adjusted in advance. After the adjustment is executed, the actual temperature response of the superconducting device is collected, and the difference between the predicted heat load sequence and the actual heat load is compared to update and correct the loss coefficient of the AC loss model and the thermal inertia time constant.

2. The superconducting switchyard cooling adaptive control method driven by load prediction according to claim 1, characterized in that, The acquisition of the future load current prediction sequence provided by the power grid dispatching system, as well as the current temperature and cooling system operating parameters of the superconducting equipment, includes: Obtain the historical load current sequence of the superconducting switch station, wherein the historical load current sequence consists of load current values ​​recorded at preset time intervals within a preset time period in the past; Obtain the time characteristics of the corresponding time period of the historical load current sequence. The time characteristics include a time identifier, a weekday type identifier, and a date type identifier, wherein the weekday type identifier distinguishes between weekdays and non-weekdays, and the date type identifier distinguishes between ordinary days and public holidays. A load current prediction model is constructed. The load current prediction model takes the historical load current sequence and the corresponding time features as input, and the load current prediction sequence for the future period as output. The model is trained using a sample training set obtained by segmenting the historical load current sequence. The historical load current sequence and the time characteristics are input into the load current prediction model, and the load current prediction model outputs the load current prediction sequence for the future period. Collect temperature values ​​from multiple temperature measuring points on the current superconducting device, select the maximum value from the multiple temperature measuring points as the temperature of the current superconducting device, and calculate the standard deviation of the temperature values ​​from the multiple temperature measuring points as the temperature uniformity index of the current superconducting device. Collect the operating parameters of the refrigeration system, which include the current output power of each refrigeration unit, the chilled water supply temperature, and the chilled water return temperature at the current moment.

3. The superconducting switchyard cooling adaptive control method driven by load prediction according to claim 1, characterized in that, The method of converting the load current prediction sequence into a predicted heat load sequence based on a pre-calibrated AC loss model includes: The loss coefficient and inherent heat leakage value of the AC loss model are obtained, and the loss coefficient and inherent heat leakage value are obtained in advance through historical operating data; Calculate the square of each predicted load current value in the future time period load current prediction sequence to obtain the squared current value; Multiply the squared current value by the loss coefficient and add the inherent heat leakage value to obtain the predicted heat load value corresponding to the predicted load current value. All predicted heat load values ​​are arranged in chronological order to form the predicted heat load sequence.

4. The superconducting switchyard cooling adaptive control method driven by load prediction according to claim 1, characterized in that, The step involves calculating the predicted heat load change rate based on the predicted heat load sequence, and combining this with the current thermal inertia time constant of the refrigeration system and the current temperature margin of the superconducting equipment to generate a pre-regulation urgency index, including: From the predicted heat load sequence, obtain the current predicted heat load value corresponding to the current time and the next predicted heat load value corresponding to the next time point. Calculate the heat load deviation by subtracting the current predicted heat load value from the next predicted heat load value. Divide the heat load deviation by the time interval between the current time and the next time point to obtain the predicted heat load change rate. Obtain the current thermal inertia time constant of the refrigeration system, wherein the current thermal inertia time constant of the refrigeration system is the time required for the refrigeration system to change its output power until the chilled water temperature reaches the corresponding steady-state value; Calculate the critical temperature of the superconducting device and subtract the temperature of the current superconducting device to obtain the temperature margin of the current superconducting device; Calculate the sum of the rated power of each refrigeration unit and subtract the sum of the current output power of each refrigeration unit to obtain the current power margin of the refrigeration system; The pre-adjustment urgency index is obtained by normalizing and weighting the predicted heat load change rate, the reciprocal of the current superconducting device temperature margin, the reciprocal of the current cooling system power margin, and the current superconducting device temperature uniformity index.

5. The superconducting switchyard cooling adaptive control method driven by load prediction according to claim 1, characterized in that, When the pre-adjustment urgency index exceeds the preset trigger threshold, the feedforward cooling power increment is obtained based on the predicted heat load change rate and the thermal inertia time constant analysis, and the output power of the cooling unit is adjusted in advance, including: A feedforward power mapping model is constructed, which takes the predicted heat load change rate and the current thermal inertia time constant of the refrigeration system as inputs and the feedforward refrigeration power increment as output. When the pre-adjustment urgency index exceeds the preset trigger threshold, the predicted heat load change rate and the current refrigeration system thermal inertia time constant are input into the feedforward power mapping model to obtain the feedforward refrigeration power increment. The feedforward cooling power increment is allocated to each of the cooling units to adjust the output power of each cooling unit in advance.

6. The superconducting switchyard cooling adaptive control method driven by load prediction according to claim 5, characterized in that, The construction of the feedforward power mapping model includes: Acquire historical operating data of the superconducting switch station. The historical operating data includes the predicted heat load change rate at multiple historical adjustment times, the current thermal inertia time constant of the refrigeration system, and the actual feedforward refrigeration power increment executed at the historical adjustment times. For each historical adjustment moment, obtain the post-adjustment evaluation index corresponding to the historical adjustment moment, wherein the post-adjustment evaluation index includes the temperature uniformity index and temperature margin of the superconducting device within a preset time period after adjustment. From all historical adjustment times, the historical adjustment times that simultaneously meet the following conditions are selected as valid adjustment times: the temperature uniformity index of the superconducting device within the preset time period after adjustment is less than the preset uniformity threshold, and the temperature margin of the superconducting device within the preset time period after adjustment is greater than the preset margin threshold. The predicted heat load change rate and the current thermal inertia time constant of the refrigeration system at each effective adjustment moment are used as a set of input samples, and the actual feedforward refrigeration power increment executed at the effective adjustment moment is used as the output label corresponding to the input sample to construct a training sample set. A feedforward power mapping model is constructed. The input layer of the feedforward power mapping model contains two input nodes, which correspond to the predicted heat load change rate and the current thermal inertia time constant of the refrigeration system, respectively. The hidden layer of the feedforward power mapping model uses radial basis functions as activation functions. The output layer of the feedforward power mapping model contains one output node, which outputs the feedforward refrigeration power increment. The feedforward power mapping model is trained using the training sample set. During the training process, the center and width of the radial basis function are determined by the K-means clustering algorithm, and the connection weights from the hidden layer to the output layer are updated by the gradient descent algorithm until the output error of the feedforward power mapping model converges, thus obtaining the trained feedforward power mapping model.

7. The superconducting switchyard cooling adaptive control method driven by load prediction according to claim 5, characterized in that, The step of allocating the feedforward cooling power increment to each of the cooling units and adjusting the output power of each cooling unit in advance includes: Obtain the current output power and rated power of each cooling unit. For each cooling unit, subtract the current output power of the cooling unit from the rated power of the cooling unit to obtain the available power margin of the cooling unit. Calculate the sum of the available power margins of all refrigeration units, and divide the available power margin of each refrigeration unit by the sum of the available power margins to obtain the allocation weight of the refrigeration unit. Multiply the feedforward cooling power increment by the allocation weight of each cooling unit to obtain the allocation increment of the cooling unit, and add the allocation increment to the current output power of the cooling unit to obtain the target output power of the cooling unit; The output power of each of the cooling units is adjusted to the corresponding target output power, thus completing the advance adjustment of the output power of each of the cooling units.

8. The superconducting switchyard cooling adaptive control method driven by load prediction according to claim 1, characterized in that, After the adjustment is executed, the actual temperature response of the superconducting device is collected, and the difference between the predicted heat load sequence and the actual heat load is compared. The loss coefficient of the AC loss model and the thermal inertia time constant are then updated and corrected, including: After the adjustment is executed, the actual temperature change sequence of the superconducting device is collected within the preset observation period; Based on the actual temperature change sequence and the actual output power of the refrigeration system, the actual heat load sequence within the preset observation period is calculated using the heat balance equation. The portion of the predicted heat load sequence corresponding to the preset observation period is compared with the actual heat load sequence, and the root mean square error between the predicted heat load sequence and the actual heat load sequence is calculated as the heat load prediction deviation. When the predicted heat load deviation exceeds the preset deviation tolerance limit, the actual heat load sequence and the load current value of the corresponding time period are used as new training samples to incrementally train the AC loss model and update the loss coefficient and inherent heat leakage value. Based on the response delay time of the actual temperature change sequence, the current thermal inertia time constant is obtained. The current thermal inertia time constant is then weighted and averaged with the current refrigeration system thermal inertia time constant to obtain the updated refrigeration system thermal inertia time constant.

9. The superconducting switchyard cooling adaptive control method driven by load prediction according to claim 8, characterized in that, The step of calculating the actual heat load sequence within the preset observation period based on the actual temperature change sequence and the actual output power of the refrigeration system, using the heat balance equation, includes: The total heat capacity of the superconducting device is obtained, which is the amount of heat absorbed or released by the superconducting device for every unit change in temperature. For each sampling interval within the preset observation period, the temperature value at the next sampling moment in the actual temperature change sequence is obtained by subtracting the temperature value at the previous sampling moment from the temperature value at the next sampling moment to obtain the temperature change amount; Multiply the total heat capacity by the temperature change and then divide by the duration of the sampling interval to obtain the heat storage change power of the superconducting device. The actual output power of the cooling system within the sampling interval is obtained, and the actual output power of the cooling system is added to the heat storage change power of the superconducting device to obtain the actual heat load value of the sampling interval. The actual heat load values ​​of all sampling intervals within the preset observation period are arranged in chronological order to form the actual heat load sequence.

10. The superconducting switchyard cooling adaptive control method driven by load prediction according to claim 8, characterized in that, The step of obtaining the current thermal inertia time constant based on the response delay time of the actual temperature change sequence includes: From the actual temperature change sequence, the moment when the output power of the refrigeration unit undergoes a step change is obtained as the step start time; Obtain the temperature value of the superconducting device corresponding to the start time of the step change, and the total steady-state temperature change corresponding to the step change; From the actual temperature change sequence, the moment when the temperature of the superconducting device starts to change from the step start moment until the temperature change first reaches a preset proportion of the total steady-state temperature change is obtained as the step response moment. The difference between the step response time and the step start time is used as the current thermal inertia time constant.

Images