A method and system for dynamic compensation of supply voltage for integrated fuel pump

Abstract

The application discloses a kind of integrated fuel pump power supply voltage dynamic compensation method and system, it is related to voltage regulating field, comprising: obtaining the real-time operating parameter of fuel pump, at least including load current, speed feedback value and power supply line impedance;Based on the real-time operating parameter, through the voltage drop prediction model of pre-training, generate predicted voltage drop value;According to the predicted voltage drop value and power supply end reference output voltage, calculate dynamic compensation coefficient;Through the dynamic compensation coefficient, feedforward compensation is carried out to power supply voltage, generate compensated drive voltage, and with the compensated drive voltage control fuel pump operation.The application solves the problem that the power supply voltage fluctuation of fuel pump in the prior art is too large, which leads to unstable operation of fuel pump and reduced performance.

Description

Technical Field

[0001] This invention relates to the field of voltage regulation, and more specifically to a method and system for dynamic compensation of the power supply voltage of an integrated fuel pump. Background Technology

[0002] In automotive engine fuel supply systems, the fuel pump assembly is a key component. Traditional fuel pump assemblies without integrated controllers use fuel pressure regulators to regulate fuel pressure.

[0003] Based on mechanical control methods, it is impossible to accurately meet the engine's fuel requirements under different operating conditions when fuel pressure fluctuates significantly. When the engine load changes abruptly, the fuel pressure is difficult to stabilize quickly within a suitable range, leading to unstable engine operation. This not only affects the vehicle's power performance but also increases fuel consumption and exhaust emissions. Summary of the Invention

[0004] This application provides a method and system for dynamic compensation of the power supply voltage of an integrated fuel pump, which addresses the problem in the prior art where excessive fluctuations in the power supply voltage of the fuel pump lead to unstable operation and reduced performance.

[0005] In view of the above problems, this application provides a method and system for dynamic compensation of the power supply voltage of an integrated fuel pump.

[0006] In a first aspect, this application provides a method for dynamic compensation of the power supply voltage of an integrated fuel pump, the method comprising:

[0007] Obtain the real-time operating parameters of the fuel pump, which include at least the load current, speed feedback value, and power supply line impedance;

[0008] Based on the real-time operating parameters, a predicted voltage drop value is generated using a pre-trained voltage drop prediction model;

[0009] Calculate the dynamic compensation coefficient based on the predicted voltage drop and the reference output voltage at the power supply end;

[0010] The power supply voltage is fed forward by the dynamic compensation coefficient to generate a compensated drive voltage, and the fuel pump is controlled by the compensated drive voltage.

[0011] Secondly, the present invention provides a dynamic compensation system for the power supply voltage of an integrated fuel pump, the system comprising:

[0012] The operating parameter acquisition module is used to acquire the real-time operating parameters of the fuel pump, which include at least the load current, speed feedback value, and power supply line impedance.

[0013] The voltage drop prediction module is used to generate a predicted voltage drop value based on the real-time operating parameters and a pre-trained voltage drop prediction model.

[0014] The dynamic compensation coefficient acquisition module is used to calculate the dynamic compensation coefficient based on the predicted voltage drop and the reference output voltage at the power supply end.

[0015] The operation control module is used to perform feedforward compensation on the power supply voltage through the dynamic compensation coefficient, generate a compensated drive voltage, and control the operation of the fuel pump with the compensated drive voltage.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0017] This application provides a method and system for dynamic compensation of the power supply voltage of an integrated fuel pump. First, real-time operating parameters of the fuel pump, including at least load current, speed feedback value, and power supply line impedance, are acquired to provide basic parameter data for subsequent voltage compensation. Second, based on the real-time operating parameters, a branch selection layer and two voltage drop prediction branches are constructed, trained separately, and finally integrated to obtain a voltage drop prediction model, thereby generating accurate predicted voltage drop values ​​and improving the model's prediction accuracy and generalization ability. Third, based on the predicted voltage drop value and the reference output voltage at the power supply end, a dynamic compensation coefficient for the voltage is calculated to facilitate subsequent voltage compensation. By determining whether compensation is needed, the dynamic compensation coefficient of the voltage drop curve requiring compensation is identified, improving the accuracy of compensation. Finally, the power supply voltage is feedforward compensated using the dynamic compensation coefficient to generate a compensated drive voltage. The compensated drive voltage is then used to control the operation of the fuel pump motor, improving voltage compensation and achieving stable operation of the fuel pump motor voltage. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a dynamic compensation method for the power supply voltage of an integrated fuel pump provided in this application.

[0019] Figure 2 This is a schematic diagram of the structure of a dynamic compensation system for the power supply voltage of an integrated fuel pump provided in this application.

[0020] In the attached diagram, the components represented by each number are as follows:

[0021] The module includes: 11 for obtaining operating parameters, 12 for predicting voltage drop, 13 for obtaining dynamic compensation coefficient, and 14 for operating control. Detailed Implementation

[0022] This application provides a method and system for dynamic compensation of the power supply voltage of an integrated fuel pump, which specifically solves the problem in the prior art where excessive fluctuations in the power supply voltage of the fuel pump lead to unstable operation and reduced performance of the fuel pump.

[0023] The present invention will now be described in detail with reference to the accompanying drawings.

[0024] Example 1, as Figure 1 As shown, this application provides a method for dynamic compensation of the power supply voltage of an integrated fuel pump, the method comprising:

[0025] S10: Obtain the real-time operating parameters of the fuel pump, wherein the real-time operating parameters include at least the load current, speed feedback value and power supply line impedance;

[0026] In this embodiment of the application, the fuel pump for which data needs to be collected is first determined. Then, the real-time operating parameters of the fuel pump during multiple operating cycles during continuous and stable operation are obtained. The collected operating parameters include at least the load current, speed feedback value and power supply line impedance.

[0027] Specifically, the load current is the actual current value driving the fuel pump motor. The magnitude of this current directly reflects the current output mechanical power of the fuel pump; the larger the current, the greater the load on the pump. The speed feedback value is the actual rotational speed of the fuel pump motor rotor, which can be obtained in real time through Hall sensors, back EMF detection, or sensorless algorithms, and directly determines the output flow rate of the fuel pump. The power supply line impedance is the equivalent resistance along the entire electrical path from the output terminal of the power supply to the input terminal of the fuel pump motor winding. This impedance includes wire resistance, connector contact resistance, fuse resistance, relay / switch conduction resistance, and printed circuit board trace resistance, etc.

[0028] After collecting load current, speed feedback value and power supply line impedance, these are integrated as real-time operating parameters to facilitate subsequent compensation calculations.

[0029] In this embodiment, by acquiring the real-time operating parameters of the fuel pump during its specific operation, a data basis is provided for subsequent voltage compensation, ensuring the accuracy of dynamic compensation and avoiding interference from inaccurate data.

[0030] S20: Based on the real-time operating parameters, generate a predicted voltage drop value using a pre-trained voltage drop prediction model;

[0031] In this embodiment, a training dataset for the voltage prediction model is obtained by calculating the real-time operating parameters of the fuel pump. The voltage prediction model is then trained using the training dataset to obtain a voltage prediction model with predictive capabilities, which predicts the voltage of the fuel pump and facilitates subsequent dynamic compensation of the fuel pump voltage.

[0032] Step S20 in the method provided in this embodiment of the invention includes:

[0033] Calculate the load current change rate, speed fluctuation amplitude, and temperature change rate based on the real-time operating parameters.

[0034] The load current change rate, the speed fluctuation amplitude, and the temperature change rate are input into the voltage drop prediction model, and the predicted voltage drop value in the next time window is output.

[0035] In this embodiment, firstly, based on the load current, speed feedback value, and power supply line impedance in the real-time operating parameters, the load current change rate, speed fluctuation amplitude, and temperature change rate are calculated respectively. The load current change rate is the ratio of the load current difference to the time interval in two adjacent real-time operating cycles of the real-time operating parameters, i.e., load current change rate = load current difference / time interval; the speed fluctuation amplitude is the difference between two adjacent speed feedback values; and the temperature change rate is the ratio of the power supply line impedance temperature difference to the time interval in two adjacent real-time operating cycles.

[0036] Secondly, the calculated load current change rate, speed fluctuation amplitude, and temperature change rate are used as a set of input data and input into the voltage drop prediction model. The trained voltage drop prediction model performs inference calculations on the input data and then outputs the predicted voltage drop value in the next time window.

[0037] The method provided in this embodiment of the invention includes the following steps in constructing the voltage drop prediction model:

[0038] Collect historical operation records, and construct a sample operation parameter set based on the historical operation records. Each sample operation parameter in the sample operation parameter set includes the sample load current change rate, the sample speed fluctuation amplitude, and the sample temperature change rate.

[0039] Extract the historical voltage drop curves corresponding to the operating parameters of each sample from the historical operation records. Correct the historical voltage drop curves based on the compensation effect of the historical voltage drop curves. Label the operating parameters of each sample with voltage drop curves according to the correction results to obtain a set of sample voltage drop curves.

[0040] Based on whether the sample voltage drop curve includes an impact compensation stage, the sample operating parameter set and the sample voltage drop curve set are divided into a first sample operating parameter set and a first sample voltage drop curve set, as well as a second sample operating parameter set and a second sample voltage drop curve set.

[0041] Based on the first sample operating parameter set and the first sample voltage drop curve set, a first voltage drop prediction branch is constructed, and based on the second sample operating parameter set and the second sample voltage drop curve set, a second voltage drop prediction branch is constructed.

[0042] The minimum value of the load current change rate in the first sample operating parameter set is used as the branch selection threshold to construct the branch selection layer;

[0043] The branch selection layer, the first voltage drop prediction branch, and the second voltage drop prediction branch are integrated to obtain the voltage drop prediction model.

[0044] In this embodiment, firstly, historical operating records of the fuel pump over a period of time, such as the past two years, are collected. Then, historical operating parameters are extracted from each historical operating record based on these records. These extracted historical operating parameters are then integrated to construct a sample operating parameter set. The sample operating parameter set includes sample operating parameters for each historical operating record, and each set of sample operating parameters includes the sample load current change rate, the sample speed fluctuation amplitude, and the sample temperature change rate.

[0045] Secondly, historical voltage drop curves corresponding to the operating parameters of each sample are extracted from the historical operating records. These historical voltage drop curves represent the voltage change from before to after voltage reduction compensation in the historical operating records. For each historical voltage drop curve, the voltage drop reflected by the curve and the compensation effect obtained by the fuel pump are analyzed.

[0046] Based on the compensation effect of historical voltage drop curves, historical voltage drop curves with poor compensation effects are corrected. After correction, the corrected voltage drop curves are used as labels to associate and annotate the original sample operating parameters. All historical voltage drop curves that need correction are traversed. Then, the historical voltage drop curves that need correction and those that do not need correction are integrated to obtain a sample voltage drop curve set.

[0047] Secondly, it depends on whether the sample voltage drop curve includes an impulse compensation stage, where the impulse compensation stage refers to the transition stage where the voltage changes rapidly with the load current.

[0048] The sample operating parameter set and sample voltage drop curve set are divided into samples. Sample voltage drop curves that include the impact compensation stage are integrated and grouped into one category. Sample voltage drop curves that do not include the impact compensation stage are integrated and grouped into another category. Finally, the first sample operating parameter set and the first sample voltage drop curve set, as well as the second sample operating parameter set and the second sample voltage drop curve set are obtained.

[0049] Furthermore, based on the first sample operating parameter set and the first sample voltage drop curve set, a machine learning algorithm is used to construct a first voltage drop prediction branch. This first voltage drop prediction branch can be constructed using a multilayer perceptron (MLP), a feedforward neural network consisting of an input layer, one or more hidden layers, and an output layer. The input layer receives the input data and has 3 nodes; the hidden layers perform a nonlinear transformation on the input data, with the first hidden layer having 64 nodes, the second hidden layer having 32 nodes, and the output layer having 50 nodes, using ReLU as the activation function; the output layer uses a linear activation function. A batch normalization layer is added after each hidden layer to accelerate convergence, and a dropout rate of 0.2 is set to prevent overfitting.

[0050] Specifically, the first sample running parameter set and the first sample voltage drop curve set are used as the input data for the first voltage drop prediction branch. The input data are divided into a training set and a validation set in an 8:2 ratio. The training set is then used to train the model.

[0051] The loss function uses mean squared error, and the batch loss is the average loss of all samples within a batch. The optimizer uses the Adam algorithm, with an initial learning rate set to 0.001. The batch size is set to 32. During training, the training set order is randomly shuffled in each epoch, and batches are fed into the model for forward propagation to calculate the predicted output. Then, backpropagation is used to calculate the gradient of the loss with respect to the weights and biases of each layer. Subsequently, the Adam optimizer updates the network parameters.

[0052] If the validation set loss does not decrease for 10 consecutive rounds, training is stopped and the model parameters that minimize the validation loss are restored. After training, the weight matrix and bias vectors of each layer of the optimal model are saved for real-time inference by the fuel pump controller.

[0053] A second voltage drop prediction branch is constructed based on the second set of operating parameters and the second set of voltage drop curves. This second voltage drop prediction branch is also constructed using a multilayer perceptron (MLP), and its structure consists of an input layer, one or more hidden layers, and an output layer. The training process of the second voltage drop prediction branch is basically similar to that of the first voltage drop prediction branch, employing a simplified network structure. Because the second branch processes steady-state or slowly varying processes without an impulse compensation phase, its output voltage drop curve is monotonically non-decreasing or flat.

[0054] Specifically, the multilayer perceptron structure of the second voltage drop prediction branch is the same as that of the first voltage drop prediction branch, with 3 nodes in the input layer, 32 nodes in a single hidden layer, and 50 nodes in the output layer. The hidden layer uses ReLU activation, and the output layer uses a linear activation function. Similarly, a batch normalization layer and a dropout layer are added after the hidden layers.

[0055] The loss function remains the mean squared error. The optimizer uses the Adam algorithm with an initial learning rate of 0.002. The batch size is set to 64. The data preprocessing method is the same: the second sample running parameter set and the second sample voltage drop curve set are divided into training and validation sets in an 8:2 ratio.

[0056] During training, the training set order is randomly shuffled in each epoch, and training is performed in batches. Forward propagation calculates the predicted output, backpropagation calculates the gradient, and the Adam optimizer updates the parameters. The training epochs are set to 80 epochs, employing early stopping: training is stopped and the optimal model is restored if the loss does not decrease after five consecutive epochs. Finally, the model parameters of the second branch are saved. After the two branches are trained independently, they are integrated with the branch selection layer to form a complete voltage drop prediction model.

[0057] Furthermore, the first sample operating parameter set is traversed, and the minimum value of the sample load current change rate in the first sample operating parameter set is counted as the branch selection threshold. When the load current change rate reaches or exceeds the branch selection threshold, there may be a working condition that requires impact compensation. Therefore, by using the minimum value of the sample load current change rate, it is ensured that in real-time inference, any working condition that reaches or exceeds the threshold will be routed to the first branch that is specifically for handling impact compensation.

[0058] Then, based on the branch selection threshold, a branch selection layer is constructed. This branch selection layer is a logical judgment unit, which contains a conditional judgment statement used for conditional judgment and branch selection.

[0059] Finally, the branch selection layer, the first voltage drop prediction branch, and the second voltage drop prediction branch are integrated according to the parallel arrangement of the branch selection layer, the first voltage drop prediction branch, and the second voltage drop prediction branch to obtain the voltage drop prediction model. After the input data passes through the branch selection layer, only one branch can be selected.

[0060] If the first branch is selected, the first voltage drop prediction branch performs forward inference and outputs a predicted voltage drop curve that includes the impulse compensation stage. If the second branch is selected, the second voltage drop prediction branch performs forward inference and outputs a predicted voltage drop curve that does not include the impulse compensation stage. Subsequently, the voltage drop prediction model uses the predicted voltage drop curves output by the branches as the final output of the entire model for subsequent calculation of the dynamic compensation coefficients.

[0061] The method provided in this embodiment of the invention includes dividing the sample operating parameter set and the sample voltage drop curve set into samples based on whether the sample voltage drop curve includes an impact compensation stage, including:

[0062] Traverse the set of sample voltage drop curves to obtain the target sample voltage drop curve, and obtain the target sample operating parameters corresponding to the target sample voltage drop curve;

[0063] The target sample voltage drop curve is differentiated to obtain the target slope sequence;

[0064] If there are negative values ​​in the target slope sequence, the target sample operating parameters and the target sample voltage drop curve are respectively assigned to the first sample operating parameter set and the first sample voltage drop curve set;

[0065] If all values ​​in the target slope sequence are non-negative, the target sample operating parameters and the target sample voltage drop curve are respectively assigned to the second sample operating parameter set and the second sample voltage drop curve set.

[0066] In this embodiment of the application, firstly, the sample voltage drop curve set is traversed to obtain the target sample voltage drop curve of the sample voltage drop curve set, and the target sample operating parameters corresponding to the target sample voltage drop curve are extracted.

[0067] Secondly, the derivative of the target sample voltage drop curve in the sample voltage drop curve set is calculated. The derivative is used to analyze the slope of the tangent line at a given point. Common methods include explicit and implicit derivatives. Since voltage drop curves are typically composed of discrete sampling points, numerical difference methods, such as forward and backward difference or central difference, are used to calculate the approximate derivative value at each time point. These derivative values ​​are then arranged in chronological order to form the target slope sequence corresponding to the target sample voltage drop curve.

[0068] Specifically, assume the target sample voltage drop curve consists of N discrete points V at equal time intervals. A central difference method is used. For both endpoints, forward or backward differencing is used to calculate the slope value at each time point in sequence, thus obtaining the target slope sequence.

[0069] Furthermore, if there are negative values ​​in the target slope sequence, it indicates that the target sample voltage drop curve includes an impact compensation stage. The target sample operating parameters and the target sample voltage drop curve are then assigned to the first sample operating parameter set and the first sample voltage drop curve set, respectively.

[0070] Finally, if all values ​​in the target slope sequence are non-negative, it means that the target sample voltage drop curve does not include the impact compensation stage. The target sample operating parameters and the target sample voltage drop curve are then assigned to the second sample operating parameter set and the second sample voltage drop curve set, respectively.

[0071] The method provided in this embodiment of the invention inputs the load current change rate, the speed fluctuation amplitude, and the temperature change rate into a voltage drop prediction model, and outputs the predicted voltage drop value for the next time window, including:

[0072] The load current change rate, the speed fluctuation amplitude, and the temperature change rate are input into the voltage drop prediction model. The voltage drop prediction model inputs the load current change rate into the branch selection layer to determine whether the load current change rate is greater than or equal to the branch selection threshold.

[0073] If the load current change rate is greater than or equal to the branch selection threshold, the first voltage drop prediction branch is invoked to generate the predicted voltage drop value including the impact compensation stage.

[0074] If the load current change rate is less than the branch selection threshold, the second voltage drop prediction branch is invoked to generate the predicted voltage drop value that does not include the impact compensation stage.

[0075] In this embodiment, the load current change rate, speed fluctuation amplitude, and temperature change rate are first input as input data to the voltage drop prediction model. Then, the voltage drop prediction model inputs the load current change rate into the branch selection layer. Branch selection is performed by determining whether the load current change rate is greater than or equal to the branch selection threshold.

[0076] Secondly, if the load current change rate is greater than or equal to the branch selection threshold, the current operating condition is determined to be a severe impact type. The first voltage drop prediction branch is then invoked, and the complete input data is passed to this branch. The first branch performs its internal forward propagation calculation, using its trained multilayer perceptron structure, sequentially activating the hidden layer and linearly transforming the output layer, ultimately outputting the predicted voltage drop value for the impact compensation stage.

[0077] If the load current change rate is less than the branch selection threshold, the current operating condition is determined to be steady-state or slowly changing, and its severity has not yet reached the level of any historical sample requiring impact compensation. In this case, the second voltage drop prediction branch is invoked, and the complete input feature vector is passed to this branch. The second branch performs its internal forward propagation calculation, and its output does not include the predicted voltage drop value from the impact compensation stage.

[0078] In this embodiment, the load current change rate, speed fluctuation amplitude, and temperature change rate are calculated based on real-time operating parameters. Then, the input data is divided according to the impact compensation stage of the target sample voltage drop curve to obtain training data for two voltage drop prediction branches. Subsequently, a voltage drop prediction model including a branch selection layer and two voltage drop prediction branches is constructed, and each branch is trained separately and finally integrated to obtain the voltage drop prediction model, thereby improving the prediction accuracy. The load current change rate, speed fluctuation amplitude, and temperature change rate are input into the voltage drop prediction model, and the predicted voltage drop value in the next time window is output, providing upstream data for subsequent dynamic voltage compensation, improving the efficiency of voltage compensation, and ensuring the accuracy of compensation.

[0079] S30: Calculate the dynamic compensation coefficient based on the predicted voltage drop and the reference output voltage at the power supply end;

[0080] In this embodiment, the dynamic compensation coefficient for voltage compensation is calculated based on the predicted voltage drop value output by the voltage drop prediction model and the reference output voltage at the power supply end.

[0081] Step S30 in the method provided in this embodiment of the invention includes:

[0082] The formula for calculating the dynamic compensation coefficient is as follows:

[0083] K=(U0-ΔU) / U0, where K is the dynamic compensation coefficient, U0 is the reference output voltage of the power supply end, and ΔU is the predicted voltage drop.

[0084] In this embodiment, since the dynamic compensation coefficient K is equal to the difference between the power supply reference output voltage U0 and the predicted voltage drop ΔU, and then divided by U0, K = (U0 - ΔU) / U0, where K is the dynamic compensation coefficient, U0 is the power supply reference output voltage, and ΔU is the predicted voltage drop.

[0085] Specifically, the dynamic compensation coefficient K is a dimensionless scaling factor used to quantify the magnitude of the adjustment to the supply voltage. This coefficient serves as a multiplier or adjustment basis in subsequent voltage compensation operations, and its value is typically a real number greater than 0. The value of K depends on the relative magnitude of the supply-side reference output voltage U0 and the predicted voltage drop ΔU.

[0086] The reference output voltage U0 at the power supply end is the standard voltage value set under ideal conditions to maintain the normal operation of the fuel pump. This reference value is usually given by the engine control unit or power management system based on the rated operating voltage of the fuel pump and current operating conditions. For example, in a 12V electrical system, U0 can be set to 13.5V.

[0087] The predicted voltage drop ΔU is the predicted value output by the voltage drop prediction model. This value represents the expected voltage loss between the power supply end and the fuel pump motor end due to factors such as load current, line impedance, and inductive effect.

[0088] For example, if the predicted voltage drop is 1.8V and the reference output voltage at the power supply end is 13.5V, the dynamic compensation coefficient K = (13.5V - 1.8V) / 13.5V ≈ 0.87 is obtained.

[0089] In this embodiment, the predicted voltage drop is calculated against the reference output voltage at the power supply end, converting the absolute predicted voltage drop into a relative proportionality coefficient. This facilitates subsequent calculations with the actual power supply voltage in the feedforward compensation strategy. This eliminates the influence of the absolute voltage value, making subsequent voltage compensation more applicable.

[0090] S40: The power supply voltage is fed forward and compensated by the dynamic compensation coefficient to generate a compensated drive voltage, and the fuel pump is controlled to operate by the compensated drive voltage.

[0091] In this embodiment, after calculating the dynamic compensation coefficient, the fuel pump control unit enters the voltage output regulation stage. The power supply voltage is fed forward to compensate by the dynamic compensation coefficient to generate a compensated drive voltage. The fuel pump is then controlled by the compensated drive voltage to ensure that the output flow and pressure of the fuel pump are stable and not affected by line voltage drop.

[0092] Step S40 in the method provided in this embodiment of the invention includes:

[0093] The power supply voltage is fed forward by the dynamic compensation coefficient to generate the compensated drive voltage. The compensated drive voltage is obtained by multiplying the dynamic compensation coefficient by the real-time collected actual voltage of the power supply terminal.

[0094] In this embodiment, the fuel pump control unit first reads the actual voltage of the power supply terminal at the current moment in real time through its integrated voltage sampling circuit.

[0095] Then, the supply voltage is feedforward compensated using a dynamic compensation coefficient to generate a compensated drive voltage. The compensated drive voltage is obtained by multiplying the dynamic compensation coefficient by the real-time acquired actual voltage at the supply terminal. The formula for calculating the compensated drive voltage is: Uc = K × U0, where Uc is the compensated drive voltage, K is the dynamic compensation coefficient, and U0 is the reference output voltage at the supply terminal.

[0096] For example, assuming the actual voltage at the current power supply terminal is 14.2V and the dynamic compensation coefficient is 0.87, the compensated driving voltage = 14.2V × 0.87 = 12.354V.

[0097] Step S40 in the method provided in this embodiment of the invention further includes:

[0098] A corresponding PWM drive signal is generated based on the compensated drive voltage;

[0099] The PWM drive signal is output to the fuel pump drive control circuit, and the actual power supply voltage output to the fuel pump motor is adjusted by the drive control circuit.

[0100] In this embodiment, the control unit first sends the compensated drive voltage to the internal power regulation module, such as a boost DC-DC converter or an adjustable PWM signal generator. The power regulation module adjusts the duty cycle of its switching transistor or the pulse width of the PWM signal based on the compensated drive voltage to generate a corresponding PWM drive signal. The PWM drive signal is a square wave signal with a fixed frequency but an adjustable duty cycle, used to control the logic level signal for turning the power switching device on and off. By changing the duty cycle of the PWM signal, the average voltage value output by the power stage can be controlled.

[0101] Secondly, the PWM drive signal is output to the fuel pump's drive control circuit, causing the output voltage of the drive control circuit to quickly reach the compensated drive voltage. The drive control circuit is a power electronic circuit integrated inside the fuel pump, typically including power switching transistors, gate driver chips, freewheeling diodes, filter inductors, and capacitors. It ensures that the actual supply voltage to the fuel pump motor is stabilized within the preset target operating voltage range.

[0102] The preset target operating voltage range is a voltage range that is pre-set based on the fuel pump's design specifications, current operating conditions, and system reliability requirements. The actual supply voltage to the fuel pump motor is expected to remain stable within this range to ensure a stable flow and pressure output from the fuel pump.

[0103] The compensated drive voltage is applied to the input terminal of the fuel pump motor drive circuit. The actual supply voltage output to the fuel pump motor is adjusted by the drive control circuit. The actual supply voltage is the instantaneous or average voltage value directly applied to the two ends of the fuel pump motor winding.

[0104] The method provided in this embodiment of the invention further includes:

[0105] The deviation between the actual output voltage of the fuel pump after compensation and the expected output voltage is collected.

[0106] When the deviation exceeds a preset threshold, the voltage drop prediction model is updated online. The model is incrementally trained using the operating parameters within the most recent time window, and only the parameters of the last layer of the model are updated.

[0107] In this embodiment, firstly, the deviation between the actual output voltage of the fuel pump after compensation and the desired output voltage is collected. The desired output voltage is the theoretically ideal voltage value that the fuel pump motor terminal should reach based on the current operating conditions and a preset target. The desired output voltage is typically equal to the reference output voltage at the power supply end or a set value within the preset target operating voltage range. This value is the target value of the compensation control system.

[0108] The difference between the compensated actual output voltage and the desired output voltage is used as the deviation, which quantifies the magnitude of the error in the current compensation effect. A deviation greater than zero indicates that the actual voltage is higher than the desired output voltage, while a deviation less than zero indicates that the actual voltage is lower than the desired output voltage.

[0109] Secondly, when the deviation exceeds the preset threshold, the voltage drop prediction model is triggered to update online. The model is incrementally trained using the running parameters within the most recent time window, and only the parameters of the last layer of the model are updated.

[0110] Specifically, the preset threshold is a pre-defined upper limit for allowable deviation, which can be determined based on the fuel pump's tolerance to voltage fluctuations, system accuracy requirements, and motor safety range. For example, it can be set to ±0.3V or an absolute deviation of 0.5V. If the deviation does not exceed this threshold, the compensation effect is considered satisfactory; if it exceeds this threshold, it indicates that the current model's prediction accuracy is insufficient, and an update mechanism needs to be triggered.

[0111] Once the triggering conditions are met, the control unit initiates the online update subroutine. This subroutine retrieves the operating parameters from the most recent time window from the internal cache. Each set of records includes: input characteristics: load current change rate, speed fluctuation amplitude, temperature change rate, and the corresponding actual observed values.

[0112] The control unit uses newly acquired time window data to perform incremental training on the currently used voltage drop prediction model. Only the parameters of the last layer of the model are updated, namely the weight matrix and bias vector of the output layer. The parameters of all hidden layers remain unchanged. The specific training process is as follows:

[0113] For each sample within the time window, the input features are forward-propagated to the last hidden layer to obtain the output feature vector of that layer. The dimension of this vector is equal to the number of nodes in the last hidden layer, for example, 32 dimensions. Since the parameters of the hidden layers are fixed, the output feature vector is obtained quickly.

[0114] The output feature vector is used as input to obtain the predicted value of the output layer. The actual observed voltage drop is used as the label to calculate the prediction error. Gradient descent, such as mini-batch gradient descent, is used to calculate the gradient and update the parameters only on the weight matrix and bias vector. Since the number of parameters in the output layer is much smaller than the number of parameters in the entire network, the computational cost of this incremental training is extremely small and can be completed in real time on a microcontroller, typically requiring only tens of milliseconds to complete one or more update steps.

[0115] After the update is complete, the new output layer parameters replace the old ones, and the model immediately enters online service mode. Subsequent predictions will be based on the updated model. This online update mechanism can be triggered repeatedly throughout the entire lifecycle of the fuel pump, allowing the model to continuously adapt to the time-varying characteristics of the system.

[0116] For example, suppose an electric fuel pump experiences slight wear on its internal motor commutator, leading to increased frictional torque. The control unit detects a deviation exceeding a preset threshold 10 times consecutively, triggering an online update. The system extracts operating parameters from the last 30 seconds, including the actual rate of change of current, speed fluctuations, and temperature changes, along with the corresponding actual voltage drop, and performs incremental training on the currently used first prediction branch. The parameters of the first two layers are fixed, such as 64 and 32 nodes, and only the output layer is updated, such as the weight matrix of 50 output nodes, with a size of 50×32 and a 50-dimensional bias vector. After 5 iterations of mini-batch gradient descent, the parameters of the model's output layer are fine-tuned, resulting in an increase of approximately 0.2V in the predicted voltage on new samples. After the update, subsequent compensation actions are based on the corrected prediction values, and the actual output voltage recovers to 13.45V, with the deviation decreasing to 0.05V, falling within the threshold. The fuel pump continues to operate normally without manual calibration or component replacement.

[0117] In this embodiment, by limiting the generation and application of the PWM drive signal and the voltage regulation function of the drive control circuit, a closed-loop process from voltage command to actual power supply to the motor is completed. Subsequently, an online model update mechanism based on output deviation enables the entire dynamic compensation method to adaptively maintain prediction accuracy during long-term operation, enhancing the system's intelligence and self-maintenance capabilities.

[0118] The embodiments of this application, through the specific implementation methods described above, achieve the following technical effects:

[0119] In this embodiment, the real-time operating parameters of the fuel pump during its operation are first obtained to provide a data basis for subsequent voltage compensation, ensuring the accuracy of dynamic compensation and avoiding interference from inaccurate data.

[0120] Secondly, the load current change rate, speed fluctuation amplitude, and temperature change rate are calculated based on real-time operating parameters. Then, the input data is divided according to the impact compensation stage of the target sample voltage drop curve, resulting in training data for two voltage drop prediction branches. Subsequently, a voltage drop prediction model including a branch selection layer and two voltage drop prediction branches is constructed, and each branch is trained separately and finally integrated to obtain the voltage drop prediction model, improving the prediction accuracy. The load current change rate, speed fluctuation amplitude, and temperature change rate are input into the voltage drop prediction model, and the predicted voltage drop value in the next time window is output, providing upstream data for subsequent dynamic voltage compensation, improving the efficiency of voltage compensation, and ensuring the accuracy of compensation.

[0121] Furthermore, the predicted voltage drop is calculated against the reference output voltage at the power supply end, converting the absolute predicted voltage drop into a relative proportionality coefficient. This facilitates subsequent calculations with the actual power supply voltage in the feedforward compensation strategy. This eliminates the influence of the absolute voltage value, making subsequent voltage compensation more applicable.

[0122] Ultimately, by limiting the generation and application of the PWM drive signal and the voltage regulation function of the drive control circuit, a closed-loop process from voltage command to actual motor power supply was completed. Subsequently, an online model update mechanism based on output deviation enabled the entire dynamic compensation method to adaptively maintain prediction accuracy during long-term operation, enhancing the system's intelligence and self-maintenance capabilities.

[0123] Example 2, as Figure 2 As shown, based on the same inventive concept as the integrated fuel pump power supply voltage dynamic compensation method provided in Embodiment 1, this embodiment of the invention also provides an integrated fuel pump power supply voltage dynamic compensation system, the system comprising:

[0124] The operating parameter acquisition module 11 is used to acquire the real-time operating parameters of the fuel pump, which include at least the load current, speed feedback value and power supply line impedance.

[0125] The voltage drop prediction module 12 is used to generate a predicted voltage drop based on the real-time operating parameters and a pre-trained voltage drop prediction model.

[0126] The dynamic compensation coefficient acquisition module 13 is used to calculate the dynamic compensation coefficient based on the predicted voltage drop and the reference output voltage at the power supply end.

[0127] The operation control module 14 is used to perform feedforward compensation on the power supply voltage through the dynamic compensation coefficient, generate a compensated drive voltage, and control the operation of the fuel pump with the compensated drive voltage.

[0128] In one embodiment, the voltage drop prediction module 12 is used for:

[0129] Calculate the load current change rate, speed fluctuation amplitude, and temperature change rate based on the real-time operating parameters.

[0130] The load current change rate, the speed fluctuation amplitude, and the temperature change rate are input into the voltage drop prediction model, and the predicted voltage drop value in the next time window is output.

[0131] The steps for constructing the voltage drop prediction model include:

[0132] Collect historical operation records, and construct a sample operation parameter set based on the historical operation records. Each sample operation parameter in the sample operation parameter set includes the sample load current change rate, the sample speed fluctuation amplitude, and the sample temperature change rate.

[0133] Extract the historical voltage drop curves corresponding to the operating parameters of each sample from the historical operation records. Correct the historical voltage drop curves based on the compensation effect of the historical voltage drop curves. Label the operating parameters of each sample with voltage drop curves according to the correction results to obtain a set of sample voltage drop curves.

[0134] Based on whether the sample voltage drop curve includes an impact compensation stage, the sample operating parameter set and the sample voltage drop curve set are divided into a first sample operating parameter set and a first sample voltage drop curve set, as well as a second sample operating parameter set and a second sample voltage drop curve set.

[0135] Based on the first sample operating parameter set and the first sample voltage drop curve set, a first voltage drop prediction branch is constructed, and based on the second sample operating parameter set and the second sample voltage drop curve set, a second voltage drop prediction branch is constructed.

[0136] The minimum value of the load current change rate in the first sample operating parameter set is used as the branch selection threshold to construct the branch selection layer;

[0137] The branch selection layer, the first voltage drop prediction branch, and the second voltage drop prediction branch are integrated to obtain the voltage drop prediction model.

[0138] Specifically, based on whether the sample voltage drop curve includes an impact compensation phase, the sample operating parameter set and the sample voltage drop curve set are divided into samples, including:

[0139] Traverse the set of sample voltage drop curves to obtain the target sample voltage drop curve, and obtain the target sample operating parameters corresponding to the target sample voltage drop curve;

[0140] The target sample voltage drop curve is differentiated to obtain the target slope sequence;

[0141] If there are negative values ​​in the target slope sequence, the target sample operating parameters and the target sample voltage drop curve are respectively assigned to the first sample operating parameter set and the first sample voltage drop curve set;

[0142] If all values ​​in the target slope sequence are non-negative, the target sample operating parameters and the target sample voltage drop curve are respectively assigned to the second sample operating parameter set and the second sample voltage drop curve set.

[0143] The load current change rate, the speed fluctuation amplitude, and the temperature change rate are input into the voltage drop prediction model, which outputs the predicted voltage drop value for the next time window, including:

[0144] The load current change rate, the speed fluctuation amplitude, and the temperature change rate are input into the voltage drop prediction model. The voltage drop prediction model inputs the load current change rate into the branch selection layer to determine whether the load current change rate is greater than or equal to the branch selection threshold.

[0145] If the load current change rate is greater than or equal to the branch selection threshold, the first voltage drop prediction branch is invoked to generate the predicted voltage drop value including the impact compensation stage.

[0146] If the load current change rate is less than the branch selection threshold, the second voltage drop prediction branch is invoked to generate the predicted voltage drop value that does not include the impact compensation stage.

[0147] In one embodiment, the dynamic compensation coefficient acquisition module 13 is used for:

[0148] The formula for calculating the dynamic compensation coefficient is as follows:

[0149] K=(U0-ΔU) / U0, where K is the dynamic compensation coefficient, U0 is the reference output voltage of the power supply end, and ΔU is the predicted voltage drop.

[0150] Specifically, the power supply voltage is fed forward to compensate by the dynamic compensation coefficient to generate the compensated driving voltage. The compensated driving voltage is obtained by multiplying the dynamic compensation coefficient by the actual voltage of the power supply terminal collected in real time.

[0151] In one embodiment, the operation control module 14 is used to:

[0152] A corresponding PWM drive signal is generated based on the compensated drive voltage;

[0153] The PWM drive signal is output to the fuel pump drive control circuit, and the actual power supply voltage output to the fuel pump motor is adjusted by the drive control circuit.

[0154] In one embodiment, the operation control module 14 is further configured to:

[0155] The deviation between the actual output voltage of the fuel pump after compensation and the expected output voltage is collected.

[0156] When the deviation exceeds a preset threshold, the voltage drop prediction model is updated online. The model is incrementally trained using the operating parameters within the most recent time window, and only the parameters of the last layer of the model are updated.

[0157] Compared to existing technologies, this application first obtains real-time operating parameters during the specific operation of the fuel pump, providing a data basis for subsequent voltage compensation, ensuring the accuracy of dynamic compensation, and avoiding interference from inaccurate data.

[0158] Secondly, the load current change rate, speed fluctuation amplitude, and temperature change rate are calculated based on real-time operating parameters. Then, the input data is divided according to the impact compensation stage of the target sample voltage drop curve, resulting in training data for two voltage drop prediction branches. Subsequently, a voltage drop prediction model including a branch selection layer and two voltage drop prediction branches is constructed, and each branch is trained separately and finally integrated to obtain the voltage drop prediction model, improving the prediction accuracy. The load current change rate, speed fluctuation amplitude, and temperature change rate are input into the voltage drop prediction model, and the predicted voltage drop value in the next time window is output, providing upstream data for subsequent dynamic voltage compensation, improving the efficiency of voltage compensation, and ensuring the accuracy of compensation.

[0159] Furthermore, the predicted voltage drop is calculated against the reference output voltage at the power supply end, converting the absolute predicted voltage drop into a relative proportionality coefficient. This facilitates subsequent calculations with the actual power supply voltage in the feedforward compensation strategy. This eliminates the influence of the absolute voltage value, making subsequent voltage compensation more applicable.

[0160] Ultimately, by limiting the generation and application of the PWM drive signal and the voltage regulation function of the drive control circuit, a closed-loop process from voltage command to actual motor power supply was completed. Subsequently, an online model update mechanism based on output deviation enabled the entire dynamic compensation method to adaptively maintain prediction accuracy during long-term operation, enhancing the system's intelligence and self-maintenance capabilities.

Claims

1. A method for dynamic compensation of the power supply voltage of an integrated fuel pump, characterized in that, The method includes: Obtain the real-time operating parameters of the fuel pump, which include at least the load current, speed feedback value, and power supply line impedance; Based on the real-time operating parameters, a predicted voltage drop value is generated using a pre-trained voltage drop prediction model; Calculate the dynamic compensation coefficient based on the predicted voltage drop and the reference output voltage at the power supply end; The power supply voltage is fed forward by the dynamic compensation coefficient to generate a compensated drive voltage, and the fuel pump is controlled by the compensated drive voltage.

2. The method for dynamic compensation of power supply voltage for an integrated fuel pump according to claim 1, characterized in that, Based on the real-time operating parameters, a predicted voltage drop value is generated using a pre-trained voltage drop prediction model, including: Calculate the load current change rate, speed fluctuation amplitude, and temperature change rate based on the real-time operating parameters. The load current change rate, the speed fluctuation amplitude, and the temperature change rate are input into the voltage drop prediction model, and the predicted voltage drop value in the next time window is output.

3. The method for dynamic compensation of power supply voltage for an integrated fuel pump according to claim 2, characterized in that, The steps for constructing the voltage drop prediction model include: Collect historical operation records, and construct a sample operation parameter set based on the historical operation records. Each sample operation parameter in the sample operation parameter set includes the sample load current change rate, the sample speed fluctuation amplitude, and the sample temperature change rate. Extract the historical voltage drop curves corresponding to the operating parameters of each sample from the historical operation records. Correct the historical voltage drop curves based on the compensation effect of the historical voltage drop curves. Label the operating parameters of each sample with voltage drop curves according to the correction results to obtain a set of sample voltage drop curves. Based on whether the sample voltage drop curve includes an impact compensation stage, the sample operating parameter set and the sample voltage drop curve set are divided into a first sample operating parameter set and a first sample voltage drop curve set, as well as a second sample operating parameter set and a second sample voltage drop curve set. Based on the first sample operating parameter set and the first sample voltage drop curve set, a first voltage drop prediction branch is constructed, and based on the second sample operating parameter set and the second sample voltage drop curve set, a second voltage drop prediction branch is constructed. The minimum value of the load current change rate in the first sample operating parameter set is used as the branch selection threshold to construct the branch selection layer; The branch selection layer, the first voltage drop prediction branch, and the second voltage drop prediction branch are integrated to obtain the voltage drop prediction model.

4. The method for dynamic compensation of power supply voltage for an integrated fuel pump according to claim 3, characterized in that, Based on whether the sample voltage drop curve includes an impact compensation phase, the sample operating parameter set and the sample voltage drop curve set are divided into samples, including: Traverse the set of sample voltage drop curves to obtain the target sample voltage drop curve, and obtain the target sample operating parameters corresponding to the target sample voltage drop curve; The target sample voltage drop curve is differentiated to obtain the target slope sequence; If there are negative values ​​in the target slope sequence, the target sample operating parameters and the target sample voltage drop curve are respectively assigned to the first sample operating parameter set and the first sample voltage drop curve set; If all values ​​in the target slope sequence are non-negative, the target sample operating parameters and the target sample voltage drop curve are respectively assigned to the second sample operating parameter set and the second sample voltage drop curve set.

5. The method for dynamic compensation of power supply voltage for an integrated fuel pump according to claim 4, characterized in that, The load current change rate, the speed fluctuation amplitude, and the temperature change rate are input into the voltage drop prediction model, and the predicted voltage drop value for the next time window is output, including: The load current change rate, the speed fluctuation amplitude, and the temperature change rate are input into the voltage drop prediction model. The voltage drop prediction model inputs the load current change rate into the branch selection layer to determine whether the load current change rate is greater than or equal to the branch selection threshold. If the load current change rate is greater than or equal to the branch selection threshold, the first voltage drop prediction branch is invoked to generate the predicted voltage drop value including the impact compensation stage. If the load current change rate is less than the branch selection threshold, the second voltage drop prediction branch is invoked to generate the predicted voltage drop value that does not include the impact compensation stage.

6. The method for dynamic compensation of power supply voltage for an integrated fuel pump according to claim 1, characterized in that, Based on the predicted voltage drop and the reference output voltage at the power supply end, the dynamic compensation coefficient is calculated, including: The formula for calculating the dynamic compensation coefficient is as follows: K=(U0-ΔU) / U0, where K is the dynamic compensation coefficient, U0 is the reference output voltage of the power supply end, and ΔU is the predicted voltage drop.

7. The method for dynamic compensation of power supply voltage for an integrated fuel pump according to claim 1, characterized in that, The power supply voltage is fed forward by the dynamic compensation coefficient to generate the compensated drive voltage. The compensated drive voltage is obtained by multiplying the dynamic compensation coefficient by the real-time collected actual voltage of the power supply terminal.

8. The method for dynamic compensation of power supply voltage for an integrated fuel pump according to claim 1, characterized in that, Controlling the fuel pump operation with the compensated drive voltage includes: A corresponding PWM drive signal is generated based on the compensated drive voltage; The PWM drive signal is output to the fuel pump drive control circuit, and the actual power supply voltage output to the fuel pump motor is adjusted by the drive control circuit.

9. The method for dynamic compensation of power supply voltage for an integrated fuel pump according to claim 1, characterized in that, Also includes: The deviation between the actual output voltage of the fuel pump after compensation and the expected output voltage is collected. When the deviation exceeds a preset threshold, the voltage drop prediction model is updated online. The model is incrementally trained using the operating parameters within the most recent time window, and only the parameters of the last layer of the model are updated.

10. A dynamic voltage compensation system for an integrated fuel pump, characterized in that, A method for dynamically compensating the power supply voltage of an integrated fuel pump according to any one of claims 1-9, the system comprising: The operating parameter acquisition module is used to acquire the real-time operating parameters of the fuel pump, which include at least the load current, speed feedback value, and power supply line impedance. The voltage drop prediction module is used to generate a predicted voltage drop value based on the real-time operating parameters and a pre-trained voltage drop prediction model. The dynamic compensation coefficient acquisition module is used to calculate the dynamic compensation coefficient based on the predicted voltage drop and the reference output voltage at the power supply end. The operation control module is used to perform feedforward compensation on the power supply voltage through the dynamic compensation coefficient, generate a compensated drive voltage, and control the operation of the fuel pump with the compensated drive voltage.

Images