Current rotating speed mapping control method of liquid cooling system

Abstract

The invention discloses a current rotating speed mapping control method of a liquid cooling system, and relates to the technical field of liquid cooling systems, and the control method comprises the steps: constructing a double-domain model of a motor domain and a thermal load domain, carrying out the online identification of a liquid path impedance parameter in an operation process, and obtaining a current rotating speed mapping model; a target rotating speed instruction is generated in combination with a current-rotating speed mapping reference result and a thermal load flow estimation result, self-adaptive pump speed control is achieved through weight adjustment and smooth constraint driven by residual errors, and meanwhile baseline mapping and a pump family curve are dynamically updated through a memory buffering and batch reestimation mechanism; according to the invention, the self-adaptive capability and the cooling control stability of the liquid cooling system under complex working conditions are obviously improved, and the contradiction between insufficient cooling and energy consumption increase is effectively avoided.

Description

Technical Field

[0001] This invention relates to the field of liquid cooling system technology, and more specifically to a current-speed mapping control method for a liquid cooling system. Background Technology

[0002] With the popularization of new energy vehicles, high-power DC charging piles are widely used in public and private charging scenarios. Due to the high power density during the charging process, the power module and liquid-cooled charging gun generate a large amount of heat during operation. Therefore, the liquid cooling system has become an important component to ensure the safe and stable operation of the charging pile. Existing liquid cooling systems mostly adopt a control method based on the mapping relationship between motor current and speed, and achieve the matching of cooling capacity and charging power through pump speed adjustment.

[0003] Existing liquid cooling systems typically infer hydraulic load by measuring motor current and adjust pump speed using a preset current-speed mapping relationship to achieve a balance between energy saving and cooling. The advantage of this method is that it eliminates the need for additional flow or differential pressure sensors, relying solely on software control to achieve closed-loop regulation of the liquid cooling system, thus making it widely used in charging piles. In related literature, CN116198373A proposes predicting charging time based on the charging pile's output power and battery capacity, and indirectly improving the operation of the liquid cooling system by adjusting the output power; CN119730197A proposes a cascade control method based on feedforward and feedback, adjusting the liquid cooling oil flow rate in advance by predicting the gun wire temperature, thereby mitigating the response lag problem of traditional methods. These solutions have played a role in improving charging efficiency and enhancing cooling capacity, demonstrating the application value of current mapping control methods in liquid cooling systems.

[0004] While existing methods have improved the performance of liquid cooling systems to some extent, they still have shortcomings: the current-speed mapping relies on a fixed relationship, and this mapping is prone to inaccuracy when ambient temperature, coolant viscosity, or pipeline impedance changes. Pump speed regulation cannot accurately reflect cooling demand, easily leading to insufficient cooling or increased energy consumption. Although the feedforward prediction method proposed in existing technologies improves hysteresis, its control effect depends on the accuracy of model parameters. When operating conditions change rapidly or during long-term operation, prediction errors gradually accumulate, potentially causing slow cooling response or overcompensation. Furthermore, indirect control methods based on charging power fail to fully consider the dynamic changes in the characteristics of the liquid cooling loop, making it difficult to maintain stable and reliable cooling performance under complex operating conditions. Therefore, existing methods still have limitations in addressing hysteresis and adaptability issues, making it difficult to simultaneously meet the requirements of high efficiency and stability. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention discloses a current-speed mapping control method for liquid cooling systems, aiming to improve the adaptability and cooling control stability of liquid cooling systems under complex operating conditions, and avoid the contradiction between insufficient cooling and increased energy consumption.

[0006] To achieve the above-mentioned technical effects, the present invention adopts the following technical solution: A current-speed mapping control method for a liquid cooling system includes: Step 1: In each control cycle, collect the pump motor current, speed, bus voltage, charging pile output power, and at least one temperature parameter of the cooling circuit. Normalize the collected data based on the preset model reference to construct the operating condition vector. Step 2: Based on the operating condition vector, establish a mapping relationship between current and speed in the motor domain and introduce the comprehensive impedance parameter of the liquid circuit; establish a mapping relationship between power and cooling flow rate in the thermal load domain. The motor domain and the thermal load domain share the comprehensive impedance parameter of the liquid circuit to obtain a dual-domain characterization model. Step 3: Construct residuals in the motor domain using the recursive least squares identification method, and update the comprehensive impedance parameters of the hydraulic circuit online under the preset feasible region constraints; Step 4: Construct flow residuals in the thermal load domain, and weight and synthesize the motor domain residuals and thermal load domain residuals to generate a cross-domain consistency index and determine the weights between the dual-domain reference quantities. Step 5: Construct a target speed function based on the weights, jointly optimize the current mapping deviation, thermal load matching deviation and speed smoothing factor, and calculate the target speed reference value within the preset speed range; Step 6: Perform rate constraint and anti-saturation shaping on the target speed reference value to obtain the target speed command; Step 7: Input the target speed command into the current inner loop and speed outer loop of the motor controller for execution, and collect the actual speed and actual current. Step 8: Write the operating condition vector, hydraulic circuit integrated impedance parameters, target speed command and actual speed into the buffer, perform batch re-estimation within a preset period, iteratively update the coefficients of the dual-domain characterization model, and enter the next control cycle after the update is completed, repeating the above process.

[0007] Based on the above technical solution, the positive and beneficial effects of the present invention are as follows: By constructing a unified dual-domain modeling structure between the motor domain and the thermal load domain, and identifying impedance parameters online during operation, the problem of traditional current-speed mapping relying on a fixed relationship is avoided. When the operating conditions of the liquid cooling circuit deviate due to changes in temperature, coolant viscosity, or pipeline impedance, the system can dynamically correct the mapping relationship, so that pump speed regulation is no longer limited to the static model, fundamentally improving the adaptability to actual operating conditions.

[0008] This scheme does not rely solely on feedforward prediction in the reference speed generation stage. Instead, it weights and synthesizes the motor domain current-speed reference results with the thermal load domain flow demand under a unified weighting mechanism, and introduces the pump speed from the previous control cycle as a smoothing constraint, ensuring that control commands maintain continuity and consistency under rapidly changing operating conditions. This dynamic weight allocation based on residual drive effectively overcomes the problem of prediction error accumulation during long-term operation and avoids cooling response lag or overcompensation.

[0009] This solution continuously recalibrates the coefficients of the dual-domain representation model during operation through memory buffering and batch reassessment to prevent control deviations caused by long-term model drift. This online self-updating mechanism enables the control method to not only have real-time performance in the short term but also maintain stability in long-term operation, thereby maintaining a balance between the reliability of cooling performance and energy efficiency under complex and changing operating conditions. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 This is a flowchart of the steps of the present invention; Figure 2 This is a schematic diagram of the dual-domain representation model structure of the present invention; Figure 3 This is a block diagram illustrating the principle of residual synthesis and dynamic weight update in step four of this invention. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0012] As one possible embodiment of the present invention: the liquid cooling system consists of a liquid cooling pump, a motor drive module, coolant piping, a radiator, a temperature sensor, a pressure sensor, a charging controller, a data acquisition unit, and host computer control software. The liquid cooling pump is installed at the cooling circuit inlet at the bottom of the charging pile, using a DC brushless motor as the drive source. The motor drive module is integrated with the pump body, ensuring a compact size and efficient heat dissipation. The coolant piping is a loop design, encompassing the inlet pipe, outlet pipe, radiator, cooling plate, and storage tank, and is arranged in the lower part of the charging pile cabinet to minimize pipe length and reduce head loss. The radiator is located on the side wall of the charging pile cabinet, working in conjunction with a fan to exchange heat externally.

[0013] The sensor section includes: a pressure sensor for flow rate calculation installed at the pump outlet; temperature sensors installed at the inlet and outlet of the liquid cooling circuit to measure the coolant temperature difference; and a chip temperature sensor mounted on the cooling plate of the charging module's power unit to collect the device's operating temperature. Motor current and speed are acquired in real-time via a current sampling circuit within the motor drive module and a Hall effect speed sensor. All sensor signals are connected to the data acquisition unit, digitized using an isolated A / D conversion circuit, and transmitted to the central controller via a CAN bus.

[0014] The central controller employs an industrial-grade embedded processor, possessing high real-time performance and floating-point computing capabilities. The controller is positioned in the center of the charging pile cabinet for easy wiring and heat dissipation. The controller's software architecture includes a data acquisition module, a working condition vector construction module, a dual-domain representation model module (motor domain and thermal load domain), a recursive identification module, a cross-domain consistency index calculation module, a target speed function generation module, a rate constraint and anti-saturation processing module, and a buffer batch update module.

[0015] During implementation, such as Figure 1 As shown, the overall data flow framework of this invention is as follows: First, the data acquisition module synchronously acquires motor current, speed, bus voltage, charging power, and cooling circuit temperature parameters in each control cycle. The acquisition cycle is set to 10ms to ensure system response speed. Before entering subsequent calculations, the acquired data is normalized using a preset model reference to eliminate dimensional differences and combined into an operating condition vector. This vector contains the main characteristic parameters describing the current load of the liquid cooling circuit and the motor state.

[0016] After the operating condition vector is constructed, the dual-domain representation model is invoked: on the one hand, in the motor domain, a mapping relationship between current and speed is established based on the comprehensive impedance parameters of the hydraulic circuit, and this relationship is jointly represented by the pump characteristic equation and the motor drive characteristics; on the other hand, in the thermal load domain, a mapping relationship between charging power and cooling flow rate is established, and a pseudo-flow observation model is established with the help of pump family curves. The motor domain and the thermal load domain share the comprehensive impedance parameters of the hydraulic circuit, thus forming a cross-domain coupled dual-domain representation model.

[0017] During model operation, the fluid impedance may change due to temperature, pipe aging, or coolant viscosity variations. Therefore, the system introduces a recursive least squares identification method to update the impedance parameters online. Specifically, the residual between the actual current and the model-predicted current is calculated in the motor domain. An error function is constructed using this residual, and the impedance parameters are updated using recursive least squares, while constraining the update range to within a preset physical feasible region to avoid numerical drift.

[0018] Simultaneously, a cooling flow residual is constructed in the thermal load domain and weighted and synthesized with the motor domain residual to generate a cross-domain consistency index. The cross-domain consistency index is calculated by the residual synthesis module within the controller, and its weighting factor is dynamically adjusted according to the operating status. This consistency index reflects the degree of matching between the motor domain and thermal load domain models under actual operating conditions and is further used to determine the weights between the dual-domain reference quantities.

[0019] Once the weights are determined, the controller enters the target speed function generation stage. This function simultaneously considers three types of constraints: current mapping deviation, thermal load matching deviation, and speed smoothing factor. The controller uses a weighted optimization algorithm to jointly construct the objective function from these three types of deviations and searches for the optimal solution within a preset speed boundary range to obtain the target speed reference value. The search process is implemented using a constraint optimization algorithm, ensuring that the target speed falls within the safe operating range of the pump motor.

[0020] After obtaining the target speed reference value, this value needs to be processed by the rate constraint and anti-saturation shaping module. Specifically, the controller first limits the rate of change of speed to prevent sudden changes in the command; then, it uses an anti-saturation mechanism to correct for cases where the reference value exceeds the pump motor's operating limits, ensuring that the target command is always within a safe and executable range. The signal output from this stage is the target speed command.

[0021] The target speed command is sent to the motor drive module, which uses a PWM control strategy to adjust the motor speed. A closed-loop control system is used between the drive module and the motor. The actual speed fed back by a Hall sensor is compared with the commanded speed. The PI regulator inside the drive module adjusts the PWM duty cycle in real time, ensuring the motor speed closely follows the command.

[0022] In the controller's buffer management module, the operating condition vector, updated impedance parameters, target speed command, and actual speed collected in each cycle are written into the buffer, along with timestamps and sample quality labels. When the buffer data accumulates to a preset threshold, or when a significant drift in the operating state is detected, the system automatically triggers a batch reassessment process. The batch reassessment process includes two levels: local fitting and global fitting. At the local level, the controller uses data within a sliding window to perform weighted least squares fitting to quickly adjust short-term parameters. At the global level, a regularized least squares method is used to fit all buffered data to update the global coefficients of the dual-domain representation model. The updated model coefficients are verified through projection onto the physical feasible region to ensure that the results meet the physical constraints of the actual liquid cooling system before replacing the current model parameters.

[0023] In terms of hardware layout, the controller and motor drive module are placed in different compartments of the charging pile cabinet to reduce electromagnetic interference; the data acquisition unit is centrally installed close to the sensors to shorten wiring distance; the liquid cooling pump is installed at the bottom of the cabinet near the liquid inlet to reduce inlet resistance. All modules communicate via CAN bus at a rate of 500kbps to meet real-time requirements. The initial forgetting factor for the recursive least squares identification within the controller is set to 0.98 to ensure the model's sensitivity to new data; the weighting factor for the cross-domain consistency index adopts a dynamic adjustment mechanism, increasing the weight of the motor domain under low load conditions and increasing the weight of the thermal load domain under high load conditions; the maximum speed change rate of the rate constraint is set to 100 rpm / ms to balance response speed and system stability.

[0024] To facilitate understanding of the present invention, a current-speed mapping control method for a liquid cooling system disclosed in the embodiments of this application will be described in detail. Please refer to [link to relevant documentation]. Figure 1 The diagram shows the steps of a current-speed mapping control method for a liquid cooling system. The steps of this method include: Step 1: In each control cycle, collect the pump motor current, speed, bus voltage, charging pile output power, and at least one temperature parameter of the cooling circuit. Normalize the collected data based on the preset model reference to construct the operating condition vector. In practice, the acquisition module typically consists of a current sensor, a Hall speed sensor, and a bus voltage sampling circuit installed on the pump motor controller. The current sensor can be a sampling circuit based on a shunt resistor or a current transformer based on a fluxgate magnetometer. The speed sensor can be a photoelectric encoder or a back EMF analysis unit. The bus voltage sampling circuit is implemented by combining an isolation voltage divider circuit with an analog-to-digital converter.

[0025] When collecting pump motor current and bus voltage, the triggering condition is limited to the inverter PWM carrier being in the steady-state conduction range. Specifically, during the switching cycle... Internally, once the intersection point between the PWM modulation wave and the carrier wave stabilizes, an ADC sampling synchronization event is triggered, thereby ensuring that the acquired current and voltage values ​​remain consistent with the actual driving state of the motor. This synchronization process can be achieved by triggering the ADC sampling completion flag through the ePWM module built into the DSP. To avoid transient spikes caused by electromagnetic interference, this application further employs oversampling and averaging filtering, that is, acquiring several current values ​​within one carrier cycle and performing averaging processing to reduce quantization noise.

[0026] Regarding temperature parameters, the "temperature parameter" in this application is not a fixed single-point temperature, but rather dynamically selected from multiple preset measurement points, with the following priority order: power module cold plate measurement point > liquid-cooled gun wire measurement point > loop inlet measurement point > loop outlet measurement point. The logical judgment condition is: if the system has a cold plate measurement point, then that data is used first; if no cold plate measurement point exists, the next level measurement point is selected sequentially, and a source label is added to that measurement point during data storage. It should be noted that the "source label" in this application differs from a conventional sensor ID; it not only records the measurement point location but also includes the hardware number of the sampling link and the sampling timestamp, used for subsequent data consistency verification. In single-point configuration scenarios, such as when only a loop inlet measurement point is configured, the data needs to be labeled with the "inlet" source in the vector and appended with a unique measurement point identifier.

[0027] After data acquisition, the system performs multi-channel filtering and anomaly removal on the current, speed, power, and temperature parameters. In this application, "multi-channel filtering" is not a single filter process, but rather a combination of a bandpass filter, a Kalman filter, and a sliding window value filter, which respectively suppress high-frequency noise, system dynamic drift, and outliers. For example, the Kalman filtering process uses the following state equation: in, For a moment The state vector contains estimates of the motor speed and current; These are the actual sampled observations; A, B, and H are the system state transition matrix, input matrix, and observation matrix, respectively. , These are process noise and measurement noise, both assumed to be zero-mean Gaussian distributions. This filter enables real-time correction of outliers during data stream input and generates a consistent sample set. The specific method for outlier removal is as follows: when the observed residual... If the data exceeds three times the standard deviation threshold, mark it as a distorted point and remove it.

[0028] After data preprocessing, the model spectrum benchmark normalization stage begins. It should be noted that the "model spectrum benchmark" in this application differs from a single pump performance curve baseline; rather, it is a reference matrix constructed by combining pump family curve calibration and hydraulic circuit condition baseline. Specifically, the pump family curve calibration is based on the flow-head-power characteristic characterization function given during the pump manufacturing stage, while the hydraulic circuit condition baseline is obtained by fitting steady-state sampling points under different flow resistance conditions during the system commissioning stage. Based on this model spectrum benchmark, the consistency sample set is normalized, and the mathematical expression is as follows: in, Indicates the first The original sampled values ​​of each physical quantity; The reference mean value of the corresponding physical quantity in the type spectrum standard; This represents the reference standard deviation. This normalization formula ensures that parameters of different dimensions are represented within the same order of magnitude, avoiding the impact of numerical scale differences on subsequent optimization processes. The normalized data is then expanded into a load condition vector: in, This is the normalized value of the pump motor current. This is the normalized value of the rotational speed. This is the normalized value of the bus voltage. This is the normalized value of the charging pile's output power. This is the normalized temperature value. This is a source label variable used to identify the data source path. This is a quality identifier that indicates the validity level of the sample (e.g., 0 for invalid, 1 for valid, and 2 for requiring correction).

[0029] It should be noted that the "quality identifier" in this application differs from the traditional data validity bit, and can simultaneously characterize data integrity and consistency. The specific rules are as follows: if data passes all filtering and elimination stages, it is identified as 1; if data passes filtering but has missing data points that have been replaced, it is identified as 2; if data is eliminated due to distortion, it is identified as 0. The specific identifier can be determined according to the actual situation, and there are no restrictions on it.

[0030] In practice, the operating condition vector not only serves as the input for constructing the motor domain residual and the thermal load domain residual, but also acts as the benchmark for parameter updates and cross-domain consistency index calculation.

[0031] Step 2: Based on the operating condition vector, establish a mapping relationship between current and speed in the motor domain and introduce the comprehensive impedance parameter of the liquid circuit; establish a mapping relationship between power and cooling flow rate in the thermal load domain. The motor domain and the thermal load domain share the comprehensive impedance parameter of the liquid circuit to obtain a dual-domain characterization model. This step will output the working condition vector from step one. As the sole input, the motor domain mapping and thermal load domain mapping—observation cross-correction—are completed within the same control cycle. Using the integrated liquid circuit impedance parameter as a coupling link, the output is an intra-domain reference quantity and shared parameter that can be directly invoked for subsequent residual consistency and target speed optimization. It should be noted that the "integrated liquid circuit impedance parameter" in this application differs from the traditional static pipeline resistance coefficient. It is an equivalent aggregate quantity that evolves with operating conditions, comprehensively reflecting factors such as coolant viscosity, local pressure drop, elbow and joint losses, and heat exchanger channel blockage. The pseudo-flow observation model is based on the functional inference of pump family curves and speed and impedance, eliminating the need for physical flow sensors in the liquid circuit. The bridging parameter is used to link the error propagation and scaling of the thermal load domain and pump-side curves, serving cross-domain consistency convergence rather than replacing the impedance parameter itself.

[0032] Please refer to the implementation details. Figure 2 The dual-domain representation model adopts a decomposition structure of "baseline term + impedance correction term" in the motor domain. Using normalized rotational speed and bus voltage as independent variables, low-order polynomials or spline basis functions are selected. ,in The baseline coefficient vector obtained from calibration on the test bench and in the historical buffer; impedance correction term. A piecewise set of impedance functions is used to characterize the current-speed deviation caused by changes in the hydraulic circuit. Modeling different flow ranges: ,in For shared overall impedance parameters of the liquid circuit, The pseudo-flow rate given by the pump family curve, This is a range selector. To avoid abrupt boundary changes during range switching, a smooth transition between adjacent ranges is defined: when... Nearest threshold When, a smoothing factor is introduced. In mixed form and to Apply continuity constraints to guarantee first-order continuity. Threshold set. With transition bandwidth The settings are based on the pump type and fluid circuit design, and can be determined according to the actual situation.

[0033] The pseudo-flow observation model uses pump family curves as its core, employs velocity similarity laws, and embeds impedance parameters as load line modulation quantities. ,in The flow gain obtained from pump calibration. To map the impedance parameter to a function of the equivalent flow loss, it can be taken in piecewise linear or spline form; This is the shape parameter set for the pump family curves. To enhance numerical stability, it is possible to... Parallel calculation of head Intersection with the system curve, then backtrack Example pseudocode is as follows: double psi_Q(double nbar, const Theta&th, const PumpShape&g){ double Qfree = g.gQ * nbar; double dQ = phi(th, g); / / Equivalent flow loss return Qfree - dQ;} Among them, phi accepts The pump shape parameter returns the equivalent loss term; gQ is a constant or lookup value that varies with the pump series configuration.

[0034] In the thermal load domain, power-flow mapping is employed. , This is a set of heat-side coefficients, including the effective area of ​​the heat exchanger, the thermal resistance of the cold plate, and a parameterized expression of the target temperature difference strategy. Considering the impact of differences in temperature measurement source on the demand flow sensitivity, a source correction function is introduced. The output of h is scaled to obtain It should be noted that the "power-flow mapping" in this application is not limited to analytical expressions, but can also be implemented by two-dimensional tabular interpolation or radial basis function networks. The interpolation method and kernel width can be selected according to the project and are not limited thereto.

[0035] Dual-domain sharing As a bridge participating in cross-correction, the flow residual is first calculated. At the same time, the motor domain residuals are preserved. Both are used to drive bridge parameters. The recursive update enables the power-side mapping and pump curve prediction to be shared. It converges under the condition of convergence. To distinguish The responsibilities, Dominant load line location, Scaling and biasing of error propagation between dominant domains are used, and the update law employs weighted recursive least squares: in For bridging regression vectors, Let covariance matrix be the variance matrix. Forgetting factor, It is a stable term; Quality labeling from step one With source tags It is mapped to suppress the effects of temporal degradation or abnormal sample replacement; This is the residual balancing coefficient, used to balance the dimensions and magnitudes of the residuals in the motor domain and the thermal load domain. The iteration trigger condition is defined as follows: If any condition is met, perform at most L inner loop updates within the current control cycle, until... Or it may reach the upper limit of iterations, L. Threshold. Upper limit L and The scheduling rules are determined through calibration.

[0036] Piecewise modeling of the impedance correction term uses an interval library. Choose a differentiable local model within each interval. For the basis function set, For follow For a monotonically changing family of coefficients, to ensure the continuity of the function values ​​and first derivatives at interval junctions, a continuity constraint is introduced: In engineering implementation, constrained least squares can be used to obtain the initial values ​​offline. During the online phase, only through Its amplitude can be modulated without disrupting continuity. For simplification in the field, a Logistic mixer can also be used for soft switching near the threshold.

[0037] A "bridging consistency operator" is also introduced for cross-calibration between the thermal load domain and the pump family curves. ,Will It applies to intermediate quantities in either side of the model. For example, it corrects the power-side mapped output: ,coefficient Depend on The analytical mapping is given; or the pseudo-flow rate on the pump side can be fine-tuned: The two paths can be chosen individually or in parallel to adapt to differences in site conditions regarding different pump types or fluid circuit sizes. It should be noted that... The specific form is not limited to linear; quadratic or spline structures can be used depending on stability and computational resources.

[0038] To maintain semantic consistency of the thermal-side mapping in complex temperature source scenarios, a source correction function is defined. For finite set mappings, such as The degree of confidence the system has in the representativeness of each measuring point is given; when When the "sensor switching" or "soft missing" setting is set, make temporary adjustments. To the conservative range. This strategy is consistent with the labeling system in step one, avoiding ambiguity in definitions across steps.

[0039] In terms of operational mechanism, step two is executed in the model thread of each control cycle. The sequence is: read... Record, calculate ,based on Selected or mixed interval models are obtained ,synthesis And calculate ;calculate And apply source correction to form Enter based on trigger conditions The inner loop completes bridge convergence; after loop convergence or the upper limit of iterations is reached, the current iteration is frozen. This serves as a reference value that can be used downstream within the domain. To ensure consistency with the overall method, Proceed to step three: residual construction and Online identification and This will be consumed by the cross-domain consistency computation in step four. and Participating in the construction of the hot side term of the objective function, This state is then used as a shared state and is refitted and updated during batch reestimation in step eight.

[0040] Regarding variable domain and value protection, All out-of-bounds errors have been trimmed in step one; and In the physically feasible region and Internal updates, if numerical anomalies occur during correction iterations, trigger projection. Back domain; forgetting factor The residual amplitude can be adaptively scheduled to avoid forgetting too quickly during slow drift. The specific strategy and parameters can be calibrated according to the project and are not limited thereto.

[0041] Step 3: Construct residuals in the motor domain using the recursive least squares identification method, and update the comprehensive impedance parameters of the liquid circuit online under preset feasible region constraints. In implementation, the recursive least squares identification module runs within the real-time control loop of the liquid cooling controller. The controller can be implemented using a DSP, MCU, or FPGA. The sampling period is synchronized with an integer multiple of the inverter PWM carrier period, typically set to 100 μs to 1 ms, but the specific value can be determined based on the system hardware performance and is not limited thereto. Input data includes: actual motor current. The current is collected via a sampling resistor or Hall current sensor; the motor speed n_k is calculated by a rotary encoder or a position-free algorithm; the bus voltage U_dc,k is collected by a voltage divider circuit and an ADC; and the pseudo-flow observation value is also included. The output of the motor domain mapping model, distinct from actual flow sensor data, is a model-calculated value that characterizes the estimated flow rate under sensorless conditions. It should be noted that the "pseudo-flow observation" in this application is not only based on the pump characteristic curve but also incorporates the comprehensive fluid path impedance parameter as a correction term, reflecting the impact of coolant viscosity and pipeline resistance changes on the flow rate.

[0042] In regression modeling, the input vector is constructed. Its domain is: ,in , , These are determined by the motor's limiting speed, the supply voltage range, and the pump's rated flow rate, respectively. The output is the predicted current. ,in This is the vector of integrated impedance parameters of the liquid circuit to be identified. Composed of multiple components, for example ,in The flow resistance coefficient, For viscosity correction factor, This is a local loss compensation factor; unlike a single friction coefficient or pressure drop parameter, here... It is a comprehensive quantity that includes liquid properties, pipeline geometry, and thermal coupling effects.

[0043] Residual is defined as If the prediction is consistent with the actual result, the parameters are kept unchanged; otherwise... Then the parameter update process begins. Threshold This can be set using statistics, such as the mean of residuals. With variance Based on, Alternatively, a fixed value can be set directly through experimental calibration. The specific value can be determined based on the actual situation and is not limited.

[0044] When updating parameters, the Recursive Least Squares (RLS) formula is used: in To estimate the covariance matrix for the parameters, initial values ​​are given. It can be set as a diagonal matrix, such as This indicates that the initial uncertainty is relatively large; The forgetting factor has a value range of 0 < λ ≤ 1, and is usually selected from 0.95 to 0.995 in order to balance the weight of historical data and new data under dynamic operating conditions.

[0045] when When the calculation result exceeds the preset feasible region Ω, projection correction is triggered. Ω is defined as: ,in Determined by the pipeline's ultimate pressure drop and rated flow rate. , Corresponding to the viscosity boundary of the coolant in the operating temperature range, The settings are based on the characteristic limitations provided by the pump manufacturer. In this implementation, the working fluid is an aqueous solution of ethylene glycol. 0.5 mPa·s can be taken. A value of 5 mPa·s can be used. The projection correction follows the principle of minimum distance using the Euclidean norm: This ensures that the updated parameters always fall within a physically reasonable range. This can be achieved through numerical iteration or by using a constrained analytical projection operator; the specific method can be flexibly chosen based on hardware resources.

[0046] To avoid numerical overflow and iterative oscillation, this implementation calibrates the ADC input in the hardware circuit design, mapping the voltage and current quantization range to 0~4095 (12-bit ADC), and runs the RLS core formula in fixed-point mode in the DSP to prevent floating-point operations from consuming too many clock cycles. Covariance matrix It is stored in SRAM and a circular cache is used during each cycle update to avoid data conflicts.

[0047] In actual operation, the parameter update action typically converges within dozens of cycles after being triggered, ensuring that the residual between the predicted current and the actual current in the motor domain model gradually decreases to within a threshold. To further improve stability, the update gain can be adjusted. Introducing an upper limit constraint, i.e. This is to avoid large fluctuations in parameters due to instantaneous anomalies.

[0048] It should be noted that the "residual exceeding threshold triggering update" mechanism in this application has event-driven characteristics, which can only perform updates when there is a significant deviation between the model and reality, thereby reducing the computational burden and improving real-time performance.

[0049] Step 4: Construct flow residuals in the thermal load domain, and weight and synthesize the motor domain residuals and thermal load domain residuals to generate a cross-domain consistency index and determine the weights between the dual-domain reference quantities. In implementation, the cross-domain residual weighting operates within the real-time loop of the liquid cooling system controller. This controller can be a DSP with floating-point capabilities, an MCU with a hardware FPU, or deployed in a SoC or FPGA for high computing power requirements. The sampling period is typically set to 1ms to 10ms to balance dynamic thermal load and fast motor domain response. The inputs to this module are: motor domain residual, derived from the difference between the predicted current and the actual current in the motor domain; and thermal load domain residual, derived from the difference between the predicted flow rate in the thermal load domain and the estimated flow rate calculated from the actual cooling loop heat balance. It should be noted that the thermal load domain residual in this application not only reflects the deviation of the heat balance at the heat exchanger end but also considers the influence of coolant temperature rise and heat transfer coefficient changes, thus providing a more comprehensive characterization of the constraints of the hot-end load on the liquid cooling loop.

[0050] Please refer to the following before proceeding with the composition operation. Figure 3 First, Kalman filtering is applied to the residuals in the motor domain and the thermal load domain, respectively. The Kalman filter state vector can be set as follows: ,in The residual rate of change is the measurement equation. , The process noise covariance matrix Q and the observation noise covariance matrix R are determined experimentally, for example, Q can be a diagonal matrix. R is a constant The Kalman filter update formula is: predict: renew: Where A is the state transition matrix and H is the observation matrix, typically taken as... The smoothed residual sequence obtained through the above filtering is denoted as... and .

[0051] After obtaining the residual sequence, the mean square error of the residuals needs to be calculated. and time-varying correlation coefficient . Calculations are performed using a sliding window, for example, with a window length of N=50 sampling points. ,in The mean. Time-varying correlation coefficient. Used to reflect the degree of correlation between two residual sequences, defined as The range is [−1,1].

[0052] based on , and Introducing residual confidence factor Its value range is (0,1], and it is used to characterize the reliability of the residual. A specific construction method can be... , where γ is the scaling factor. For example, when the residual variance is large and the cross-domain correlation is strong, the confidence factor decreases, indicating that the contribution of the residual to the weight allocation is reduced. The specific function form can be set according to needs and is not limited thereto. Then, dynamic weight calculation is performed using the exponential weighting formula: in, Let be the dynamic weighting coefficient of the residual in the k-th domain, and α be the correlation adjustment factor, ranging from (0,1). For example, in the motor domain, if Smaller, then An increase indicates a high level of confidence in the motor domain residuals. The "dynamic weighting coefficient" in this application has time-varying characteristics, and its calculation depends on the statistical characteristics of the residuals and the confidence factor, thus it can be adjusted in real time according to changes in operating conditions.

[0053] After obtaining the dynamic weighting coefficients, they are introduced into the residual synthesis model. Cross-domain consistency index. Defined as ,in and The corresponding weight coefficients, and satisfying . It can be understood as a dual-domain consistency metric, used to describe the degree of fit between the motor domain and thermal load domain prediction models within the same control cycle.

[0054] In practical implementation, to avoid sudden changes in consistency metrics caused by drastic fluctuations in a single residual, the system can introduce a smoothing coefficient η during the weight update process. The value range is [0.01, 0.2], which can be adjusted according to actual dynamic needs.

[0055] During implementation, since Kalman filtering and dynamic weight calculation require matrix operations, linear algebra libraries can be used for optimization when running on a DSP, while parallel fixed-point arithmetic units are needed to implement matrix multiplication and division when running on an FPGA. In terms of storage, each residual sequence needs to store the most recent N sampling points for mean square error calculation. Typically, N is in the tens to hundreds range, requiring relatively little cache space, which can be handled by on-chip RAM.

[0056] When cross-domain consistency index Exceeding the set threshold At that time, it was believed that the mismatch between the motor domain and thermal load domain models increased, and the weights between the reference quantities of the two domains needed to be adjusted. The setting method can be based on the 3σ principle according to historical data statistics, or by setting a fixed value through experimental calibration. For example, under the conditions of cold start and high thermal shock in liquid cooling systems, The load often increases rapidly, at which point the weights automatically shift towards the thermal load domain, thus ensuring that the cooling strategy follows the actual needs of the hot end.

[0057] It should be noted that the "cross-domain consistency index" in this application is an index with dynamic weight adjustment, which has the dual functions of cross-domain balancing and dynamic adaptation. It can allocate reference weights between the motor domain and the thermal load domain, so that the control method can maintain consistency and coordination under different operating conditions.

[0058] Step 5: Construct a target speed function based on the weights, jointly optimize the current mapping deviation, thermal load matching deviation and speed smoothing factor, and calculate the target speed reference value within the preset speed range; It should be noted that the "target speed function" in this application is not only for single-variable optimization of current or flow rate, but rather by introducing current mapping deviation, thermal load matching deviation and speed smoothing factor into the same optimization framework to form a unified evaluation index under multi-objective coupling, thereby achieving coordinated operation of motor control and liquid cooling thermal management.

[0059] In a specific implementation, the objective function is first defined. Its expression is as follows: in, This refers to the pump motor speed. For current mapping deviation, For heat load flow deviation, The rate of change of rotational speed, The adjustment coefficient is dynamically updated by the weighted fusion algorithm in step four. It should be noted that the current mapping deviation in this application differs from ordinary current error. It is the deviation after mapping the difference between the predicted current and the actual sampled current in the motor domain to the liquid-cooled coupling model, directly reflecting the deviation between the drive current distribution and the pump liquid-cooled impedance characteristics. The heat load flow deviation, on the other hand, is based on the difference between the heat power demand of the liquid-cooled system and the actual pump flow rate, describing the dynamic balance between cooling capacity and heat load. As for the speed smoothing factor, it introduces the square of the speed change rate to suppress mechanical shock and fluid flow disturbance caused by large speed fluctuations. Unlike existing simple speed limits, it is implemented through weighting within the optimization function, belonging to a soft constraint method.

[0060] In the optimization process, to ensure that the target rotational speed remains within the physically permissible range, the following constraints are introduced: in, These are the minimum and maximum permissible speeds of the pump motor, determined by a combination of the physical limits of the liquid cooling system, the safe operating range of the pump, and the rated parameters of the motor. The specific values ​​can be set according to the actual needs of different liquid cooling systems, and there are no fixed limitations.

[0061] Solving for the optimal speed reference value The process employs a constraint optimization method based on the Lagrange multiplier method. By combining the objective function with speed range constraints, a Lagrange function is constructed: in, and These are Lagrange multipliers used to ensure the feasibility of the solution. When the optimal solution lies within the interval, If the solution approaches the boundary, the corresponding multiplier term is activated to ensure that the final solution does not exceed the physically permissible range. In numerical implementation, an iterative gradient descent mechanism is used to update the rotational speed. The gradient calculation formula is: in, The sensitivity coefficient of current deviation with rotational speed can be obtained through online identification or experimental calibration. This is the sensitivity coefficient of flow rate deviation to speed, which is usually derived from the pump characteristic curve; The acceleration due to rotational speed can be calculated from the sampled rotational speed sequence through two differences. In this application, the "sensitivity coefficient" differs from a simple proportional coefficient; it is the first derivative of the deviation function with respect to the rotational speed variable. It dynamically reflects the coupling effect of rotational speed on current and flow rate under different operating conditions and is a dynamic characteristic parameter.

[0062] During the iteration process, if the currently calculated gradient magnitude value Less than the set threshold If the condition is met, the termination condition is triggered, and the current rotational speed is output as a reference value for the target rotational speed. Threshold The accuracy can be set according to different requirements; for example, a larger value can be selected when cooling demand changes rapidly. To accelerate convergence, a smaller value is used under steady-state conditions. To improve accuracy.

[0063] As one possible implementation, in practical applications, an optimization calculation is performed once per control cycle. Specifically, this involves first calculating the calculated values ​​based on the measured motor current, estimated pump flow rate, and load heat demand. and Then, the speed smoothing factor is calculated based on the speed change rate of the previous cycle; finally, the above variables are substituted into the objective function. And calculate the gradient Finally, the candidate solutions are updated using gradient descent until convergence or the threshold condition is met. The final target rotational speed reference value is obtained. It will be used as the input to the controller and transmitted to the pump motor drive unit to achieve closed-loop control.

[0064] It should be noted that the "target speed reference value" in this application is a dynamic value obtained through multi-objective optimization. It not only considers the deviation between current and flow rate but also introduces a speed smoothing factor for comprehensive balance. Therefore, it can simultaneously meet the thermal load requirements of the liquid cooling system and the stability of the motor drive. This reference value is updated in each control cycle to ensure the real-time response of the liquid cooling system under dynamic operating conditions.

[0065] In implementation, step five, based on the weights and parameters provided in the previous steps, completes the mapping optimization from the deviation to the target speed. The deviation between the current domain and the thermal load domain provides the optimization target, dynamic weights ensure that the optimization focus adjusts with changes in operating conditions, the Lagrangian constraint mechanism ensures the physical feasibility of the solution, and gradient iteration ensures the real-time performance of the numerical solution. Thus, the current-speed mapping control method for the liquid cooling system achieves a closed-loop transition from parameter updates to speed optimization in this step, providing a directly applicable target signal for the final pump motor execution layer, thereby forming a complete real-time control link.

[0066] Step Six: Perform rate constraint and anti-saturation shaping processing on the target speed reference value to obtain the target speed command; the processing method for performing rate constraint and anti-saturation shaping on the target speed reference value includes: Receive the target speed reference value output in step five and the actual speed at the previous moment, and read the fluid circuit comprehensive impedance parameters obtained by online identification. Based on the fluid circuit comprehensive impedance parameters and the dynamic pole estimation of the driver, calculate the upper / lower limit rate boundary through an adaptive slope limiter. The trajectory of the out-of-limit segment is shaped by piecewise cubic splines and acceleration and jump constraints are applied to generate a smooth intermediate trajectory. The smooth intermediate trajectory is input into the anti-saturation shaper, which uses a combined mechanism of integral anti-saturation back-calculation and feedforward lag compensation to back-calculate and correct the inner loop integral components. If the duration of the saturation state exceeds a preset threshold, the saturation counter is triggered and the hysteresis damping method is used to suppress the turbulence. The shaped command is projected onto the preset speed range via convex projection and then output after adding a timestamp and synchronization mark.

[0067] It should be clarified that the "rate constraint" in this application is an adaptive rate boundary obtained by jointly estimating the combined impedance of the hydraulic circuit and the poles of the driver, thereby constraining the rate of change of rotational speed at the dynamic level. The "anti-saturation shaper," unlike common anti-integral saturation measures, employs a combined structure of integral anti-saturation back-calculation and feedforward hysteresis compensation, which can eliminate the integral accumulation effect while maintaining response continuity.

[0068] During implementation, the input quantities must first be clearly defined. The inputs for step six include: ① the target rotational speed reference value output from step five. ② The actual rotational speed collected at the previous moment Its acquisition cycle can be consistent with the control cycle, generally set to 1 ms to 5 ms; ③ The liquid circuit comprehensive impedance parameters are updated in real time through the online identification module. It is defined as a combined parameter of flow resistance, inertia, and pump-end characteristic impedance in a liquid-cooled pipeline, and its unit can be expressed as Pa·s / m. 3 ④ Dynamic pole estimation of the driver , represents the characteristic poles of the motor driver in the discrete domain, used to reflect its transient response capability.

[0069] In the rate constraint stage, an adaptive slope limiter is used to calculate the upper and lower limits of the rate of change of rotational speed. The boundary values ​​can be defined as: Where f(·) is the amplitude limiting function, it can be expressed as ,in This refers to the calibration coefficient. It should be clarified here that the boundary of the adaptive slope limiter is dynamically calculated and adjusted in real time according to changes in the fluid circuit impedance.

[0070] If the difference between the target rotational speed reference value and the rotational speed at the previous moment exceeds the rate boundary, the trajectory shaping stage begins. Trajectory shaping uses piecewise cubic spline interpolation to construct a smooth transition curve and introduces acceleration and abrupt change constraints. The acceleration constraint can be expressed as: The jump constraint can be expressed as: ,in , The maximum allowable acceleration and maximum jump values ​​for the system can be set within a range based on the motor's mechanical parameters and the fluid's inertia, for example... Take 5000 rpm / s, Take 2×10^5 rpm / s 2 The control points of a cubic spline curve can be determined by... The two-point construction allows the midpoint to be automatically generated based on velocity boundary conditions. In this way, the originally excessive speed reference signal is replaced with a continuous and differentiable smooth intermediate trajectory, avoiding mechanical shocks and fluid flow disturbances caused by sudden changes in slope.

[0071] The generated smooth intermediate trajectory is then input into the anti-saturation shaper. The anti-saturation shaper's function is to perform back-calculation correction on the integral components of the inner-loop PI controller or other integral components. Specifically, it uses an integral anti-saturation back-calculation mechanism, that is, when the controller output is detected to be close to the upper / lower limit of driver saturation, the integral component is corrected according to... Perform back calculation, where To limit the saturated output, Calculate the output for the controller. The integral coefficient is used to gradually reduce the integral accumulation. Furthermore, to avoid insufficient dynamic response due to simple integral reduction, a feedforward hysteresis compensation module is introduced into the shaper. This module can take the form of… ,in γ is the feedforward lag time constant, and γ is the adjustment parameter. This compensation signal is superimposed on the back calculation signal to ensure the continuity of the shaping output and the rationality of the phase response.

[0072] If the duration of the detected driver being in a saturated state exceeds a preset threshold For example, if the time is 50 ms, a saturation counter is triggered. The counter increments by 1 each time a saturation event exceeding the threshold is detected, and the hysteresis damping method is invoked. In hysteresis damping, the control commands are adjusted to... , where h is the hysteresis amplitude coefficient, and sign(·) is the sign function used to cancel the overshoot at the moment of saturation recovery. It should be noted that the "hysteresis vibration suppression" in this application is different from the common amplitude limiting vibration suppression. It does not weaken the response by low-pass filtering, but stabilizes the transition process by introducing a small-amplitude reverse bias after saturation elimination.

[0073] The shaped target speed command still needs to be limited to the speed range through convex projection calculation before it is output. Within this range, ensure that the physical limits between the motor and pump end are not exceeded. A convex projection can be formally represented as: Furthermore, to ensure consistency of instructions in multi-threaded or distributed control systems, timestamps need to be added to the instructions. The synchronization identifier ID_sync, the timestamp can be generated by the master timer, and ID_sync can be implemented by the frame sequence number in the bus protocol (such as CAN or EtherCAT).

[0074] In implementation, the rate constraint and trajectory shaping module can be deployed in the computing core of an MCU or DSP, with storage resources primarily used to store the spline coefficient matrix and historical rotational speed data. The anti-saturation shaper is embedded in the software stack of the execution unit as a controller, or it can be implemented as a hardware state machine in an FPGA to ensure rapid determination. The saturation counter and hysteresis damping logic can be implemented using a state register and a logic comparator to ensure triggering within milliseconds. Additionally, it should be noted that the comprehensive impedance parameters of the liquid circuit in this application... Unlike conventional pump impedance curves, this is defined as a sum of parameters including multiple factors such as pipeline pressure drop, local loss coefficient, and pump internal friction, and is continuously updated in the online identification module using a recursive least squares method. Therefore, the rate boundary calculation can be adjusted in real time according to operating conditions, ensuring the effectiveness of rate constraints.

[0075] Finally, the target speed command obtained in step six not only serves as the direct input to the driver but also as a reference trajectory in the feedback channel for error calculation in the next cycle, achieving a coordinated closed loop with steps seven and eight. In this way, the target speed reference value is fully processed before entering the execution phase, ensuring signal smoothness, anti-saturation properties, and timing synchronization, thus enabling the current-speed mapping control method of the liquid cooling system to form a complete closed-loop control link.

[0076] Step 7: Input the target speed command into the inner current loop and outer speed loop of the motor controller for execution, and collect the actual speed and actual current. The inner current loop uses the motor stator current as the controlled variable and achieves transient current tracking by rapidly modulating the PWM duty cycle or inverter voltage vector, thereby ensuring the instantaneous response of the motor's electromagnetic torque. The outer speed loop uses the motor speed as the main adjustment object, and its adjustment result serves as the reference input for the inner current loop. The two form a standard inner and outer dual-loop control architecture to achieve stable drive of the liquid-cooled pump under different thermal loads and liquid circuit impedance conditions.

[0077] Specifically, step seven first involves receiving the target speed command. This instruction carries a timestamp and synchronization flag to ensure timing consistency with other system signals in a bus communication environment. The controller generates an outer-loop speed error signal based on this instruction; the speed error is determined by… It means that, among them This represents the actual speed of the motor currently being collected.

[0078] The outer loop speed error signal enters the speed regulator, typically a PI or improved composite regulator, and its output is the reference current of the inner current loop. This current reference quantity is not a single-phase current, but rather the d-axis and q-axis components formed after Clarke and Park transformations, where... Directly determines the electromagnetic torque of the motor. Used for field weakening or flux linkage optimization control. Through this decoupling method, the output of the outer speed loop can directly act on the current component that generates torque, thereby shortening the dynamic link from speed command to torque output.

[0079] After entering the inner current loop, the actual current is measured by the sampling circuit in the motor driver. It should be noted that the "actual current" in this application differs from the current-limiting signal used only for protection detection; it is obtained by sampling the three-phase stator current through a high-speed ADC and performing real-time transformation. , This serves as the feedback input for the inner current loop regulation. The inner current loop operates with a sampling period of tens to hundreds of microseconds. The regulator calculates the required voltage command, which is then used to generate inverter switching signals via a space vector PWM algorithm or carrier modulation, directly acting on the power devices. IGBTs or SiC MOSFET modules can be selected as the power devices, depending on the cooling method and power rating; there are no specific limitations.

[0080] During execution, if the current or speed error is detected to exceed the preset threshold, for example, if the current instantaneously exceeds 1.5 times the rated value or the speed deviation remains greater than the set value... If the condition is not met, the protection logic will be triggered, including current limiting, soft deceleration, or error-based shutdown. These triggering conditions can all be set in this application according to the rated capacity of the liquid cooling system and the motor's safe operating range; there are no restrictions on this.

[0081] It should be noted that "acquisition" in this application does not refer only to a single sampling, but includes continuous data buffering and time-series recording. The controller is internally configured with a FIFO or circular buffer to store the actual current and speed values ​​for the past few cycles. This acquired data is used for real-time closed-loop control and can also be used for weight calculation and deviation function construction in steps four and five, thus forming cross-step data coupling. Through this mechanism, the inner current loop and outer speed loop not only achieve real-time adjustment but also form a closed-loop interconnection with the upper-level mapping optimization algorithm.

[0082] In terms of mathematical logic, the coupling relationship between the inner and outer loops can be described as: the reference current output by the outer loop speed regulator. With inner loop feedback current The difference in current results in an error, and the inner-loop regulator approaches zero error in the high-speed domain. The speed regulator, on the other hand, ensures... and The consistency. Because the electromagnetic torque of the motor is consistent with... The load torque of the liquid-cooled pump is directly proportional to the square of the flow rate. The above dual-loop mechanism can ensure that the motor speed can stably follow the command when the thermal load and flow resistance change dynamically, without overshooting or jittering.

[0083] In addition, it should be noted that in some application scenarios, model predictive control (MPC) or adaptive regulators can be used to replace traditional PI controllers to deal with parameter drift and nonlinear disturbances. However, the scope of protection of this application is not limited to a specific control algorithm, and the specific application can be determined according to actual needs.

[0084] Step 8: Write the operating condition vector, hydraulic circuit integrated impedance parameters, target speed command and actual speed into the buffer, perform batch re-estimation within a preset period, iteratively update the coefficients of the dual-domain characterization model, and enter the next control cycle after the update is completed, repeating the above process.

[0085] The process of batch re-estimation within a preset period includes: establishing an index matrix in the buffer according to timestamps for the operating condition vector, impedance parameters, target speed command and actual speed, and using the batch least squares algorithm to estimate the overall parameters of the buffer data at the end of the period to obtain the update coefficients of the dual-domain representation model.

[0086] The process of iteratively updating the coefficients of the dual-domain representation model includes: using the coefficient vector output by batch least squares as the initial estimate, performing convergence iteration within the impedance parameter constraint set using the quasi-Newton iteration method, and correcting the coupling term coefficients of the motor domain and the thermal load domain respectively using the Jacobian matrix block update method.

[0087] To reiterate, the dual-domain representation model in this application comprises two coupled parts: a motor domain and a thermal load domain. The former primarily characterizes the electromagnetic dynamics of motor current-speed-torque, while the latter mainly reflects the thermal characteristics of liquid circuit impedance-flow rate-thermal load. Mathematically, a cross-domain mapping relationship is established through coupling coefficient terms, thereby achieving a precise correspondence between current and speed. This model must be updated according to changes in operating conditions during use; otherwise, deviations will occur due to changes in liquid viscosity, pump wear, or drift in driver parameters. Therefore, step eight is set up for periodic batch re-evaluation and iterative correction.

[0088] In the specific implementation process, step eight first receives four types of input quantities: operating condition vector, hydraulic circuit integrated impedance parameters, target speed command, and actual speed, and writes them into the controller's internal buffer. The data writing to the buffer uses a timestamp indexing mechanism, meaning each record includes the signal acquisition time and sequence identifier, thus enabling the establishment of a complete index matrix at the end of the preset period. This index matrix arranges the four types of input quantities into a multidimensional dataset, forming the input samples for the batch least squares algorithm. Specifically, let the period length be... Then all data points within that time period are calculated according to... Arranged in form, among which Including operating vectors and target speed, These are the combined observed values ​​of actual rotational speed and impedance parameters. The batch least squares algorithm fits all data points at the end of the cycle, obtaining the updated coefficient vector of the dual-domain representation model. It should be noted that the "batch least squares" method here is different from the real-time recursive least squares method. Its advantage lies in avoiding the bias caused by instantaneous noise and improving the stability of the estimation by solving in batches periodically.

[0089] After obtaining the initial estimated coefficient vector Subsequently, step eight does not directly apply this result, but instead uses it as the starting point for iterative optimization. Specifically, a quasi-Newton iteration method is used to perform convergence calculations within the impedance parameter constraint set. The constraint set is set based on the physical feasibility conditions of the fluid path; for example, the drag coefficient must be non-negative, the inertia coefficient must not be less than zero, and the coupling term coefficients must be within the experimentally calibrated range. The quasi-Newton iteration method uses a second-order approximation to correct the gradient information, thereby achieving faster convergence within a finite number of iterations. To ensure computational efficiency, the controller only allows a limited number of iterations per cycle. If the convergence condition is not met within the maximum number of iterations, the current optimal result is used as an approximate solution. The convergence criterion can be set as the Euclidean distance between two adjacent coefficient vectors being less than a threshold. The threshold value can be set according to the required calculation accuracy and control cycle length.

[0090] In the iterative update process, a block update method for the Jacobian matrix is ​​introduced. This "block update of the Jacobian matrix" differs from direct inversion of the overall matrix. It divides the model parameters into two sub-blocks: the motor domain and the thermal load domain. Gradients and corrections are calculated separately for each sub-block, and cross-correction is introduced into the coupling terms, thereby reducing computational complexity and enhancing numerical stability. For example, during the motor domain update, the parameters of the thermal load domain are fixed, and new current-speed mapping coefficients are obtained. During the thermal load domain update, the motor domain parameters remain unchanged, and the correlation coefficient of the hydraulic circuit impedance is corrected. Finally, synchronous optimization of the parameters in both domains is achieved through block correction of the coupling terms.

[0091] In practical applications, the trigger condition for step eight is "the preset period arrives," which can be set to perform a batch re-estimation every 500ms or 1s. This timescale is much larger than the sampling period of the current and speed control loops, so it will not affect the stability of real-time control. At the same time, if a significant change in the comprehensive impedance parameter of the liquid circuit is detected (such as exceeding 20% ​​of the most recent estimated value), an additional forced update can also be triggered to ensure the adaptability of the model.

[0092] It should be noted that the "buffer" in this application is not only a logical storage unit, but can also be implemented as a circular array in the MCU memory, or as a high-speed cache in the FPGA or external RAM, to ensure data availability in large-scale parallel computing. The dimension, storage precision, and sampling interval of the data index matrix can all be adjusted according to the actual hardware platform capabilities, and are not limited thereto.

[0093] In implementation step eight, the liquid cooling system not only achieves real-time target speed control within each control cycle, but also re-estimates and optimizes the model parameters at the end of the cycle. This forms a closed-loop architecture of "dual loop + dual domain": the internal and external current / speed dual loops ensure short-term dynamic performance, while the batch iterative updates of the dual-domain representation model ensure long-term adaptability and robustness across operating conditions. Ultimately, this enables the liquid-cooled pump motor to possess both high dynamic response capability and parameter self-adaptation and long-term operational stability under complex operating conditions, thereby effectively improving the overall energy efficiency and reliability of the liquid cooling system.

[0094] It should be noted that the mathematical formulas, derivations, symbol definitions, and parameter calculation methods used in this specification are all for the purpose of further clarifying and demonstrating the technical content of this invention, so that those skilled in the art can more intuitively and accurately understand the working mechanism and technical effects of this invention. These formulas are only used as quantitative expressions or illustrative examples of technical features and do not constitute limiting conditions of the claims of this invention. Those skilled in the art should understand that, without changing the core idea of ​​this invention, the parameter forms, calculation methods, numerical ranges, and even symbol representations involved in the formulas can be equivalently replaced or simplified in engineering according to the actual application environment. The specifics can be determined according to the actual situation, and no limitation is imposed. It should also be emphasized that the formulas in this specification are not theoretical derivations in the style of academic research papers, but rather an engineering description of the embodiments of this invention. Their purpose is to enhance the understandability and implementability of this invention, rather than to increase redundancy and complexity. Those skilled in the art can choose whether to use such quantitative tools when reading this specification, or can achieve the same technical effects through other equivalent methods.

[0095] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these specific embodiments are merely illustrative. Those skilled in the art can omit, substitute, and modify the details of the above methods and systems in various ways without departing from the principles and essence of the present invention. For example, combining the above method steps to perform substantially the same function and achieve substantially the same result according to substantially the same method falls within the scope of the present invention. Therefore, the scope of the present invention is defined only by the appended claims.

Claims

1. A current-speed mapping control method for a liquid cooling system; characterized in that: include: Step 1: In each control cycle, collect the pump motor current, speed, bus voltage, charging pile output power, and at least one temperature parameter of the cooling circuit. Normalize the collected data based on the preset model reference to construct the operating condition vector. Step 2: Based on the operating condition vector, establish a mapping relationship between current and speed in the motor domain and introduce the comprehensive impedance parameter of the liquid circuit; establish a mapping relationship between power and cooling flow rate in the thermal load domain. The motor domain and the thermal load domain share the comprehensive impedance parameter of the liquid circuit to obtain a dual-domain characterization model. Step 3: Construct residuals in the motor domain using the recursive least squares identification method, and update the comprehensive impedance parameters of the hydraulic circuit online under the preset feasible region constraints; Step 4: Construct flow residuals in the thermal load domain, and weight and synthesize the motor domain residuals and thermal load domain residuals to generate a cross-domain consistency index and determine the weights between the dual-domain reference quantities. Step 5: Construct a target speed function based on the weights, jointly optimize the current mapping deviation, thermal load matching deviation and speed smoothing factor, and calculate the target speed reference value within the preset speed range; Step 6: Perform rate constraint and anti-saturation shaping on the target speed reference value to obtain the target speed command; Step 7: Input the target speed command into the current inner loop and speed outer loop of the motor controller for execution, and collect the actual speed and actual current. Step 8: Write the operating condition vector, hydraulic circuit integrated impedance parameters, target speed command and actual speed into the buffer, perform batch re-estimation within a preset period, iteratively update the coefficients of the dual-domain characterization model, and enter the next control cycle after the update is completed, repeating the above process.

2. The current-speed mapping control method for a liquid cooling system according to claim 1, characterized in that: The working principle of step one includes: When collecting pump motor current and bus voltage, the data is synchronized with the inverter PWM carrier steady-state window, and a unified timestamp is assigned to each sample. Temperature parameters are selected from existing measurement points according to priority, namely power module cold plate measurement point, liquid cooling gun line measurement point, loop inlet or outlet measurement point, and the source of the measurement point is marked in the single-point configuration scenario. The collected current, speed, power and temperature parameters are filtered and anomalies are removed through multi-channel filtering to generate a consistent sample set; Based on the calibration of pump family curves and the establishment of a model reference based on the hydraulic circuit operating condition baseline, the consistency sample set is normalized to generate an operating condition vector with source label and quality identifier.

3. The current-speed mapping control method for a liquid cooling system according to claim 1, characterized in that: The process of establishing the dual-domain characterization model in step two includes: in the motor domain, the mapping relationship between current and speed is decomposed into a baseline polynomial term and an impedance correction term, and the impedance correction term is nonlinearly coupled and compensated by the comprehensive impedance parameter of the liquid circuit to characterize the speed and current deviation caused by changes in coolant viscosity and pipeline pressure drop; in the thermal load domain, the mapping relationship between charging pile output power and flow demand is cross-calibrated with the pseudo-flow observation model. The cross-calibration is performed iteratively to converge the difference between the power-side mapped output and the pump characteristic curve prediction value by using the shared comprehensive impedance parameter of the liquid circuit as a bridging variable.

4. The current-speed mapping control method for a liquid cooling system according to claim 3, characterized in that: The nonlinear coupling compensation of the impedance correction term is achieved by constructing an impedance function set. Segmented modeling is performed for different flow ranges, and continuity constraints are used when switching models. Suppress boundary abrupt changes; the cross-correction process includes: calculating the residual between the power-side mapped output and the predicted value of the pump characteristic curve in each iteration, and updating the bridging parameters by recursive least squares weighted update to converge the residual to within a preset threshold.

5. The current-speed mapping control method for a liquid cooling system according to claim 1, characterized in that: The process of step three includes: Within each control cycle, a regression model based on the motor domain is established using a recursive least squares identification algorithm. The regression model uses an input vector consisting of speed, bus voltage, and pseudo-flow observations as independent variables and outputs a predicted current. The residual is obtained by calculating the difference between the actual current and the predicted current; if the residual exceeds the preset statistical threshold, the parameter update is triggered. During the parameter update process, the historical covariance matrix is ​​exponentially weighted and corrected using a forgetting factor mechanism to obtain the updated matrix. Based on the update matrix, the comprehensive impedance parameter vector of the liquid circuit is iteratively calculated, and the update formula is as follows: in, Indicates the first The vector of combined impedance parameters of the periodic liquid circuit; For the first The periodic gain vector is used to correct the parameter estimates; For the first The periodic covariance matrix is ​​used to characterize the uncertainty of parameter estimation; For the first The periodic regression vector is composed of motor speed, bus voltage, and pseudo-flow observations. for The transpose of; For the first The residual of the period is the difference between the actual current and the predicted current. This is a forgetting factor used to adjust the weight of historical data in the parameter update process, with a range of [range missing]. ; If the updated impedance parameters do not satisfy the preset feasible region constraint, the impedance parameters are mapped to the boundary of the parameter feasible region using a projection operator; the projection operator adopts the minimum distance criterion based on the Euclidean norm to project the superboundary parameter vector as follows: in, This represents the feasible region set of the comprehensive impedance parameters of the liquid circuit. This is the corrected impedance parameter vector; This represents the Euclidean L2 norm, used to measure the magnitude of parameter deviation.

6. The current-speed mapping control method for a liquid cooling system according to claim 1, characterized in that: The process of weighted synthesis of the motor domain residual and the thermal load domain residual includes: Kalman filtering is used to suppress noise in the motor domain residuals and thermal load domain residuals to obtain the residual sequence; Based on the mean squared error and time-varying correlation coefficient of the residual sequence, the residual confidence factor is obtained through normalization, and the dynamic weight coefficient is calculated using an exponentially weighted recursive formula. The expression of the exponentially weighted recursive formula is as follows: in Denotes the variance of the residual in the k-th domain. This is the residual correlation adjustment factor; Indicates the index of the residual field; The dynamic weighting coefficients are input into the residual synthesis model to generate a cross-domain consistency index. If the cross-domain consistency index is less than a preset threshold, the weights of the dual-domain reference quantities are updated through the Lagrange multiplier constraint mechanism.

7. The current-speed mapping control method for a liquid cooling system according to claim 1, characterized in that: The engineering process for step five is as follows: The target speed function is constructed using the Lagrange multiplier method, and the current mapping deviation, thermal load matching deviation, and speed smoothing factor are input into the optimization framework; the formula expression of the target function is: in, The objective function value, This refers to the pump motor speed. For current mapping deviation, For heat load flow deviation, The rate of change of rotational speed, These are the adjustment coefficients determined by the dynamic weights output in step four; A constrained optimization algorithm is used to solve for the optimal speed reference value within a preset speed range, and the candidate speed solution is updated through an iterative gradient descent mechanism. Within each control cycle, the solution is based on the real-time updated... and The gradient is calculated using the following formula: in, The sensitivity coefficient representing the change in current deviation with rotational speed. The sensitivity coefficient representing the flow deviation as a function of rotational speed. It is the rotational acceleration; If during the iteration process Less than the threshold If the condition is met, the termination condition is triggered, and the current iteration solution is output as a reference value for the target rotational speed. .

8. The current-speed mapping control method for a liquid cooling system according to claim 1, characterized in that: The processing method for rate constraint and anti-saturation shaping of the target speed reference value includes: Receive the target speed reference value output in step five and the actual speed at the previous moment, and read the fluid circuit comprehensive impedance parameters obtained by online identification. Based on the fluid circuit comprehensive impedance parameters and the dynamic pole estimation of the driver, calculate the upper / lower limit rate boundary through an adaptive slope limiter. The trajectory of the out-of-limit segment is shaped by piecewise cubic splines and acceleration and jump constraints are applied to generate a smooth intermediate trajectory. The smooth intermediate trajectory is input into the anti-saturation shaper, which uses a combined mechanism of integral anti-saturation back-calculation and feedforward lag compensation to back-calculate and correct the inner loop integral components. If the duration of the saturation state exceeds a preset threshold, the saturation counter is triggered and the hysteresis damping method is used to suppress the turbulence. The shaped command is projected onto the preset speed range via convex projection and then output after adding a timestamp and synchronization mark.

9. The current-speed mapping control method for a liquid cooling system according to claim 1, characterized in that: The process of batch re-estimation within a preset period includes: establishing an index matrix in the buffer according to timestamps for the operating condition vector, impedance parameters, target speed command and actual speed, and using the batch least squares algorithm to estimate the overall parameters of the buffer data at the end of the period to obtain the update coefficients of the dual-domain representation model.

10. The current-speed mapping control method for a liquid cooling system according to claim 1, characterized in that: The process of iteratively updating the coefficients of the dual-domain representation model includes: using the coefficient vector output by batch least squares as the initial estimate, performing convergence iteration within the impedance parameter constraint set using the quasi-Newton iteration method, and correcting the coupling term coefficients of the motor domain and the thermal load domain respectively using the Jacobian matrix block update method.

Images