Current compensation circuit, control method and device for current compensation circuit
By introducing an LSTM current compensation circuit into an electrolytic capacitor-free air conditioning system, active prediction and feedforward compensation for load surges are achieved, solving the problems of bus voltage overshoot and response lag, and improving system stability and device lifespan.
Patent Information
- Application Number
- CN202611449688.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-17
- Publication Date
- 2026-11-17
AI Technical Summary
Electrolytic capacitor-free air conditioning systems suffer from severe bus voltage overshoot during sudden load changes. Traditional passive control systems exhibit delayed response, leading to frequent system protection actions and increased stress on components.
A current compensation circuit based on a long short-term memory network (LSTM) is adopted. By predicting load changes, a feedforward compensation amount is generated. Combined with the speed control unit and the drive unit, the inverter output power is actively adjusted to suppress bus voltage fluctuations.
It significantly suppressed DC bus voltage overshoot in electrolytic capacitor-free air conditioner outdoor units, reduced bus voltage fluctuation amplitude, and improved system reliability and hardware lifespan.
Smart Images

Figure CN123122337A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of motor drive control technology, and more specifically, to a current compensation circuit, a control method for the current compensation circuit, and a device. Background Technology
[0002] With the popularization of green home appliance concepts and increasingly stringent energy efficiency requirements, air conditioner inverter outdoor units are developing towards higher reliability, longer lifespan, and miniaturization. In traditional inverter air conditioner drive circuits, large-capacity electrolytic capacitors are typically used on the DC bus side to buffer power fluctuations and suppress voltage ripple. However, electrolytic capacitors have significant physical defects, such as easy evaporation of electrolyte leading to capacity decay, easy bulging and leakage under high-temperature environments, short service life, and difficulty in recycling. In contrast, the "electrolytic capacitor-free" solution, which uses film capacitors to replace electrolytic capacitors, has advantages such as small size, light weight, high temperature resistance, long lifespan, and no risk of electrolyte leakage. It can improve the integration, energy efficiency ratio, and operational stability of air conditioning systems under high temperature and high humidity environments, and represents an important development trend in future home appliance power management.
[0003] While electrolytic capacitor-free solutions offer advantages in hardware reliability, their significantly reduced energy storage capacity on the DC bus side leads to a substantial decrease in the system's ability to buffer against load surges. During the operation of an electrolytic capacitor-free air conditioner compressor, sudden load changes (such as compressor start-up / shutdown, rapid frequency adjustments, or drastic environmental changes) can easily cause large overshoots or drops in the DC bus voltage. Existing electrolytic capacitor-free air conditioner control technologies often employ "passive response" strategies, such as setting fixed power or current thresholds to trigger frequency reduction protection, or using traditional harmonic injection and proportional resonant control to suppress bus voltage oscillations. However, these traditional methods rely primarily on feedback correction after disturbances occur, resulting in delayed control response and insufficient predictive capabilities. They cannot adjust the control output in advance before load changes, leading to significant bus voltage fluctuations. This, in turn, triggers system protection actions, increases noise, and intensifies stress on power devices, ultimately affecting the overall reliability and lifespan of the system.
[0004] Long Short-Term Memory (LSTM) networks, as a special type of recurrent neural network, exhibit excellent performance in predicting time-varying signals due to their powerful ability to capture long-term dependencies in time-series data. Given that the bus voltage of capacitor-less air conditioners is extremely sensitive to changes in the q-axis current (i.e., torque current), accurate prediction of the q-axis current trend at the next moment can achieve feedforward compensation for the control output. Currently, there is no technology to directly apply the timing prediction capabilities of LSTM to the feedforward control of capacitor-less compressors to solve the problem of bus voltage overshoot caused by sudden load changes.
[0005] There is currently no effective solution to the above problems. Summary of the Invention
[0006] This invention provides a current compensation circuit, a control method for the current compensation circuit, and an apparatus to at least solve the technical problems in the related art, such as the bus voltage being sensitive to load changes and the traditional passive control response being lagging due to the weak energy storage buffering capacity of the lack of electrolytic capacitors, which leads to increased stress on system protection or devices.
[0007] According to one aspect of the present invention, a current compensation circuit is provided, comprising: a compensation unit, a speed control unit, and a drive unit, wherein the compensation unit includes a network model, a first calculation module, and a first gain module; the input of the network model is a multi-dimensional timing input vector; the output of the network model is connected to the positive input of the first calculation module; and the output of the first calculation module is connected to the first gain module. The speed control unit includes a second calculation module, a regulator, and a third calculation module; the output of the second calculation module is connected to one end of the regulator; the other end of the regulator is connected to the first input of the third calculation module; and the output of the first gain module is connected to the second input of the third calculation module. The drive unit includes a second gain module, a conversion module, a modulation module, and an inverter; the output of the third calculation module is connected to one end of the second gain module; the other end of the second gain module is connected to one end of the conversion module; the other end of the conversion module is connected to one end of the modulation module; and the other end of the modulation module is connected to the inverter.
[0008] Optionally, the network model includes an input layer, an output layer, and a hidden layer, wherein the number of neurons in the input layer is a target number, the number of neurons in the output layer is one, used to output the predicted current value, and the number of neurons in the hidden layer is determined based on a preset mathematical formula and the number of neurons in the input layer and the output layer.
[0009] According to another aspect of the present invention, a control method for a current compensation circuit is also provided, comprising: generating a multi-dimensional timing input vector based on multi-dimensional timing data within a preset time period, wherein the multi-dimensional timing data includes current data and frequency data; inputting the multi-dimensional timing input vector into the network model and performing gain calculation to obtain a compensation current; calculating the compensation current to generate a quadrature-axis given current; generating a drive pulse signal based on the quadrature-axis given current, and inputting the drive pulse signal into the inverter to control the operation of the motor.
[0010] Optionally, generating the multidimensional time-series input vector based on multidimensional time-series data within a preset time period includes: collecting a first preset number of current data and frequency data within each time window of the preset time period, wherein the number of time windows is equal to half of the target number; extracting the maximum value of the current data within each time window to obtain a current maximum value sequence; calculating the average value of the frequency data within each time window to obtain an electrical frequency average value sequence; and concatenating the current maximum value sequence with the electrical frequency average value sequence to obtain the multidimensional time-series input vector.
[0011] Optionally, inputting the multidimensional time-series input vector into the network model and performing gain calculation to obtain the compensation current includes: processing the multidimensional time-series input vector using the network model to generate the current prediction value for the next time window; calculating the difference between the current prediction value and the maximum current value of the current time window using the first calculation module to obtain the difference current; and processing the difference current using the first gain module to obtain the compensation current.
[0012] Optionally, the calculation of the compensation current to generate the quadrature-axis given current includes: using the second calculation module to calculate the difference between the preset speed and the actual speed to obtain the speed difference; using the regulator to process the speed difference to obtain the output current; using the third calculation module to calculate the output current and the compensation current to obtain the original given current; and using the second gain module to process the original given current to obtain the quadrature-axis given current.
[0013] Optionally, generating a drive pulse signal based on the quadrature-axis given current and inputting the drive pulse signal into the inverter to control the motor operation includes: processing the quadrature-axis given current and the direct-axis given current using the conversion module to obtain a first voltage component and a second voltage component, wherein the quadrature-axis given current is orthogonal to the direct-axis given current; and modulating the first voltage component and the second voltage component to obtain the drive pulse signal.
[0014] Optionally, the control method for the current compensation circuit further includes: acquiring historical operating data to form an original dataset arranged in chronological order, wherein each data point in the original dataset corresponds to the maximum current value and the average electrical frequency value within a time window; using a second preset number of consecutive maximum current values and average electrical frequency values as an input sequence to obtain corresponding current prediction values, wherein the second preset number is equal to the number of time windows; moving the sliding window backward by one time window, repeating the above selection process until the original dataset is traversed to generate target samples; and dividing the target samples into a first sample set, a second sample set, and a third sample set according to chronological order and a preset ratio.
[0015] According to another aspect of the present invention, a control device for a current compensation circuit is also provided, comprising: a first generation unit, configured to generate a multi-dimensional timing input vector based on multi-dimensional timing data within a preset time period, wherein the multi-dimensional timing data includes current data and frequency data; a first calculation unit, configured to input the multi-dimensional timing input vector into the network model and perform gain calculation to obtain a compensation current; a second calculation unit, configured to calculate the compensation current to generate a quadrature-axis given current; and a second generation unit, configured to generate a drive pulse signal based on the quadrature-axis given current and input the drive pulse signal into the inverter to control the operation of the motor.
[0016] Optionally, the first generation unit includes: a data acquisition module, configured to acquire a first preset number of current data and frequency data within each time window of the preset time period, wherein the number of time windows is equal to half of the target number; an extraction module, configured to extract the maximum value of the current data within each time window to obtain a current maximum value sequence; a first processing module, configured to calculate the average value of the frequency data within each time window to obtain an electrical frequency average value sequence; and a sequence construction module, configured to concatenate the current maximum value sequence with the electrical frequency average value sequence to obtain the multidimensional time-series input vector.
[0017] Optionally, the first calculation unit includes: a first generation module, used to process the multi-dimensional time-series input vector using the network model to generate the current prediction value for the next time window; a first difference calculation module, used to calculate the difference between the current prediction value and the maximum current value of the current time window using the first calculation module to obtain the difference current; and a compensation calculation module, used to process the difference current using the first gain module to obtain the compensation current.
[0018] Optionally, the second calculation unit includes: a second difference calculation module, used to calculate the difference between the preset speed and the actual speed using the second calculation module to obtain the speed difference; a second processing module, used to process the speed difference using the regulator to obtain the output current; a third processing module, used to calculate the output current and the compensation current using the third calculation module to obtain the original given current; and a fourth processing module, used to process the original given current using the second gain module to obtain the quadrature axis given current.
[0019] Optionally, the second generation unit includes: a fifth processing module, used to process the quadrature-axis given current and the direct-axis given current using the transformation module to obtain a first voltage component and a second voltage component, wherein the quadrature-axis given current is orthogonal to the direct-axis given current; and a modulation module, used to modulate the first voltage component and the second voltage component to obtain the driving pulse signal.
[0020] Optionally, the control device of the current compensation circuit further includes: a data acquisition unit, used to acquire historical operating data and form an original dataset arranged in chronological order, wherein each data point in the original dataset corresponds to the maximum current value and the average electrical frequency value within a time window; a sequence construction unit, used to take a second preset number of consecutive maximum current values and average electrical frequency values as input sequences to obtain corresponding current prediction values, wherein the second preset number is equal to the number of time windows; a third generation unit, used to move the sliding window backward by one time window, repeat the above selection process until the original dataset is traversed, and generate target samples; and a partitioning unit, used to partition the target samples into a first sample set, a second sample set, and a third sample set according to chronological order and a preset ratio.
[0021] According to one aspect of the present invention, an air conditioner outdoor unit is provided, wherein the air conditioner outdoor unit uses the control method of the current compensation circuit described in any one of the above-described embodiments.
[0022] According to one aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium including a stored program, wherein the program executes the control method of the current compensation circuit described in any one of the above embodiments.
[0023] According to one aspect of the present invention, a processor is provided for running a program, wherein the program executes the control method of the current compensation circuit described in any one of the above embodiments.
[0024] According to one aspect of the present invention, a computer program product is provided, including computer instructions that, when executed by a processor, perform the control method for the current compensation circuit described in any one of the above embodiments.
[0025] By applying the technical solution of this application, a multi-dimensional timing input vector is generated based on multi-dimensional timing data within a preset time period. This multi-dimensional timing data includes current data and frequency data. The multi-dimensional timing input vector is input into a network model and gain calculation is performed to obtain a compensation current. The compensation current is then calculated to generate a quadrature-axis given current. Based on the quadrature-axis given current, a drive pulse signal is generated and input into the inverter to control the motor operation. This achieves the goal of introducing LSTM's q-axis current adaptive feedforward compensation into an electrolytic capacitor-free drive system, overcoming the hysteresis limitations of traditional feedback control. This significantly suppresses DC bus voltage overshoot in electrolytic capacitor-free air conditioner outdoor units, reduces bus voltage fluctuations, and improves system reliability and hardware lifespan. Furthermore, it solves the technical problems in related technologies where the weak energy storage buffering capacity of electrolytic capacitor-free systems leads to bus voltage sensitivity to load changes and hysteresis in traditional passive control, resulting in increased stress on system protection or components. Attached Figure Description
[0026] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:
[0027] Figure 1 This is a hardware structure block diagram of a mobile terminal according to an embodiment of the present invention, which describes a control method for a current compensation circuit.
[0028] Figure 2 This is a block diagram of a capacitorless PMSM drive control based on LSTM and q-axis current compensation according to an embodiment of the present invention.
[0029] Figure 3 This is a flowchart of a control method for a current compensation circuit according to an embodiment of the present invention;
[0030] Figure 4 This is a flowchart of a multiplexed circuit signal transmission method according to an embodiment of the present invention;
[0031] Figure 5 This is a graph showing the changes in metrics during the training process of an LSTM model according to an embodiment of the present invention;
[0032] Figure 6 This is a comparison of the q-axis given current before and after compensation according to an embodiment of the present invention;
[0033] Figure 7 This is a simulation result of the bus voltage under sudden load changes without q-axis current compensation according to an embodiment of the present invention;
[0034] Figure 8 This is a simulation result diagram of bus voltage under sudden load changes with q-axis current compensation according to an embodiment of the present invention;
[0035] Figure 9 This is a schematic diagram of the control device for the current compensation circuit according to an embodiment of the present invention.
[0036] The above figures include the following reference numerals:
[0037] 102. Processor; 104. Memory; 106. Transmission device; 108. Input / output device. Detailed Implementation
[0038] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0039] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0040] As described in the background section, related technologies suffer from drawbacks such as weak energy storage and buffering capacity due to the lack of electrolytic capacitors, resulting in bus voltage sensitivity to load changes and delayed response of traditional passive control, thus leading to increased stress on system protection or components. Embodiments of this invention provide a control method and apparatus for a current compensation circuit, an air conditioner outdoor unit, a computer-readable storage medium, a processor, and a computer program product.
[0041] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
[0042] The methods and embodiments provided in this invention can be executed on a mobile terminal, a computer terminal, or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 This is a hardware structure block diagram of a mobile terminal for a current compensation circuit control method according to an embodiment of the present invention. Figure 1 As shown, a mobile terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data are also shown. The mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the mobile terminal described above. For example, the mobile terminal may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.
[0043] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the control method of the current compensation circuit in this embodiment of the invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, thereby implementing the above-described method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the above-described networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. The transmission device 106 is used to receive or send data via a network. Specific examples of the above-described networks may include wireless networks provided by the mobile terminal's communication provider. In one example, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.
[0044] Example 1
[0045] According to an embodiment of the present invention, a method embodiment for a current compensation circuit is provided, comprising: a compensation unit, a speed control unit, and a drive unit, wherein...
[0046] The compensation unit includes a network model, a first calculation module, and a first gain module. The input of the network model is a multi-dimensional time-series input vector. The output of the network model is connected to the positive input of the first calculation module, and the output of the first calculation module is connected to the first gain module. The speed control unit includes a second calculation module, a regulator, and a third calculation module. The output of the second calculation module is connected to one end of the regulator, and the other end of the regulator is connected to the first input of the third calculation module. The output of the first gain module is connected to the second input of the third calculation module. The drive unit includes a second gain module, a conversion module, a modulation module, and an inverter. The output of the third calculation module is connected to one end of the second gain module, and the other end of the second gain module is connected to one end of the conversion module. The other end of the conversion module is connected to one end of the modulation module, and the other end of the modulation module is connected to the inverter.
[0047] It should be noted that, Figure 2 This is a block diagram of a capacitor-free PMSM drive control based on LSTM and q-axis current compensation, such as... Figure 2 As shown, based on the traditional speed outer loop (PI regulator generates basic q-axis current command) and current inner loop, an LSTM prediction compensation branch is introduced. It uses historical speed and current sequence to predict the current change at the next moment. The calculated prediction error is superimposed on the basic current setpoint after gain compensation to generate the final total current reference command. Then, the inverter drives the motor through the current controller and space vector pulse width modulation (SVPWM).
[0048] Optionally, such as Figure 2 As shown, LSTM is a Long Short-Term Memory network model, which is a deep learning algorithm that can capture long-term dependencies in time series data. and The inputs to the LSTM model are a vector composed of the maximum values of the compressor's q-axis current collected in 100ms increments over the first second and a vector composed of the average electrical frequency. Figure 2 middle and ; It is the q-axis current, which is the current component used to generate electromagnetic torque in the vector control of a permanent magnet synchronous motor (PMSM), and its magnitude is directly proportional to the torque output by the motor. It is the maximum value of the compressor q-axis current within the k-th 100ms; It outputs the maximum q-axis current value in the last 100ms; the difference between the two is obtained. ; As compensation, it is added to the speed loop output; This refers to the gain coefficient, which is used to proportionally convert the q-axis current variation predicted by the LSTM into a feedforward compensation value to achieve adaptive compensation for the speed loop output; This usually refers to the actual electrical angular velocity or electrical frequency of the compressor; The average value of the compressor's electrical frequency within a sampling window (e.g., 100ms) is PI, which stands for Proportional-Integral Controller. In vector control of air conditioning compressors, it is used to regulate current or speed errors. The proportional term (P) is used to quickly respond to errors, and the integral term (I) is used to eliminate steady-state errors. It is the actual sampled current of the q-axis, which is the actual value of the motor q-axis current obtained by the sensor in real time and after coordinate transformation. It is used as the feedback signal of the current loop PI controller for closed-loop regulation. The q-axis current setpoint after LSTM feedforward compensation is superimposed by the speed loop PI output and the q-axis current compensation amount predicted by LSTM, and finally sent to the current loop controller as the target current command. It is the gain parameter, where, The phase of the input voltage; This is the q-axis current reference value, used to generate torque; This is the d-axis current reference value, usually set to 0 to achieve maximum torque-to-current ratio control; together, they constitute the target command of the current loop in the vector control of a permanent magnet synchronous motor; current loop control refers to the inner loop control link in the motor control system, based on the q-axis current reference value ( ) and d-axis current reference value ( ) and actual sampling current ( and The error is calculated by the PI controller to determine the required α-β coordinate system voltage command. , The switching signal is generated via space vector pulse width modulation, which ultimately drives the inverter to regulate the motor stator current. and These are the voltage vector components obtained through current loop control; Space Vector Pulse Width Modulation (SVPWM) is a control algorithm for an inverter that converts these continuous voltage quantities into the action time of six basic voltage vectors to drive the compressor permanent magnet synchronous motor (PMSM); The compressor permanent magnet synchronous motor is a synchronous motor with built-in or surface-mounted permanent magnets that uses the rotating magnetic field generated by the three-phase alternating current of the stator to interact with the magnetic field of the rotor permanent magnet to generate torque.
[0049] Optionally, such as Figure 2 As shown, the area outside the dashed line is a control block diagram for a conventional electrolytic capacitor-free driven permanent magnet synchronous motor. The control structure is a dual closed-loop control (inner current loop and outer speed loop). The speed loop PI output... Afterwards, it passes through a gain stage (gain parameter is...) , The q-axis current is obtained by taking the input voltage phase as the reference. This ensures that the fluctuations in the q-axis current reference are consistent with the fluctuations in the input power, thereby indirectly regulating the inverter's output power and improving the power factor to some extent. The q-axis current reference is obtained from this process. Then, the voltage vector components in the stationary two-phase coordinate system (α-β coordinate system) are obtained through current loop control. and Through the space vector pulse width modulation (SVPWM) algorithm, these continuous voltage quantities are converted into the action time of six basic voltage vectors, which ultimately drive the permanent magnet synchronous motor (PMSM) of the compressor.
[0050] Optionally, such as Figure 2 As shown, the area within the dashed line represents the q-axis current feedforward compensation based on LSTM prediction. The input to the LSTM model is a vector composed of the maximum values of the compressor's q-axis current collected in 100ms time steps within the first second. The vector is composed of the average value of the electrical frequency. The input matrix is obtained. After model calculation, the maximum q-axis current value in the last 100ms is obtained. The maximum q-axis current value in the last 100ms Get by doing bad things After gain Resulting in compensated output The specific formula is as follows: , , , , .
[0051] In this embodiment, the compensation unit is a set of functional modules for implementing active feedforward current compensation; the network model refers to a prediction model built based on a Long Short-Term Memory (LSTM) network, whose input is a multi-dimensional time-series input vector; the multi-dimensional time-series input vector is a set of data containing historical time series features, consisting of the compressor q-axis current maximum value sequence and the compressor electrical frequency average value sequence collected within a preset time window; the first calculation module is used to receive the predicted q-axis current maximum value output by the network model and perform a difference operation with the actual q-axis current maximum value at the current moment to calculate the change in q-axis current; the first gain module is used to multiply the q-axis current change output by the first calculation module by a gain coefficient to generate the q-axis current feedforward compensation value; the speed control unit refers to the functional module that generates a base current setpoint and superimposes the feedforward compensation amount with the base value to output the final q-axis current setpoint; the second calculation module is the processing unit in the speed control unit used to convert the speed loop (PI) output into the q-axis current reference setpoint value, which ensures the q-axis current reference setpoint value by performing gain processing on the speed loop output and phase transformation based on the input voltage phase. The fluctuation of the q-axis current reference value is consistent with the fluctuation of the input power to indirectly regulate the inverter output power; the regulator usually refers to a proportional-integral (PI) controller, which is used to perform closed-loop regulation of the q-axis current reference setpoint to eliminate steady-state error; the third calculation module is the unit that performs addition operations, which is used to superimpose the q-axis current reference setpoint output by the regulator with the feedforward compensation value output by the compensation unit to generate the final q-axis current setpoint; the drive unit refers to the entire signal processing and power drive link from current setpoint to motor operation, including current loop control, coordinate transformation, modulation and inversion process; the second gain module is used to perform amplitude scaling or preprocessing on the final q-axis current setpoint; the transformation module refers to the coordinate transformation unit that converts the voltage or current reference quantity in the dq coordinate system into the voltage vector component or SVPWM action time in the stationary two-phase coordinate system (α-β coordinate system); the modulation module is the unit that generates the PWM pulse signal to drive the power switch transistor of the inverter according to the parameters output by the transformation module; the inverter is a power conversion circuit that receives the PWM signal and converts the DC bus voltage into three-phase AC power to drive the permanent magnet synchronous motor.
[0052] As described above, in this embodiment, by setting a compensation unit containing an LSTM network model to predict load changes and generate feedforward compensation, and combining the speed control unit with a second calculation module, a regulator, and a third calculation module to superimpose the reference current and the feedforward compensation, and finally executing the drive through a drive unit containing gain, coordinate transformation, modulation, and inverter functions, the system achieves the purpose of proactively predicting and compensating for q-axis current changes before load changes occur, thereby adjusting the inverter output power in advance to suppress DC bus voltage fluctuations. This achieves the technical effect of effectively reducing the DC bus voltage overshoot amplitude of electrolytic capacitor-free air conditioners, reducing the risk of system protection triggering, and significantly improving the operating stability and hardware lifespan of the compressor and the whole machine under variable load conditions.
[0053] Therefore, the technical solution provided by the above embodiments of the present invention solves the technical problems in the related art, such as the bus voltage being sensitive to load changes and the traditional passive control response being lagging due to the weak energy storage buffering capacity of the non-electrolytic capacitor, which leads to increased stress on system protection or devices.
[0054] According to the above embodiments of the present invention, the network model includes an input layer, an output layer, and a hidden layer, wherein the number of neurons in the input layer is the target number, the number of neurons in the output layer is one, which is used to output the current prediction value, and the number of neurons in the hidden layer is determined based on a preset mathematical formula and the number of neurons in the input layer and the output layer.
[0055] It should be noted that when constructing the LSTM model, as can be seen from the dataset creation, there are 20 neurons in the input layer and 1 neuron in the output layer. Generally, a single-layer hidden layer structure can approximate a nonlinear function with arbitrary precision when the number of neurons is sufficient. Therefore, a neural network with one hidden layer is chosen, and the number of hidden layer neurons is determined according to the following formula, with k set to 4, resulting in 18 hidden layer neurons. The formula can be expressed as: Where m is the number of neurons in the input layer and n is the number of neurons in the output layer. Since the actual model is greatly affected by the characteristics of the components and the load, setting a parameter k facilitates model adjustment. If the value of k is too large, random search may require multiple iterations to converge, resulting in extremely low efficiency. If the value of k is too small, it may fail to capture the rapid response details when the air conditioner is turned on and off. Adjustments should be made based on the actual model training.
[0056] As can be seen from the above, in this embodiment of the invention, an LSTM prediction model is constructed by dynamically determining the number of hidden layer neurons based on a preset mathematical formula. This model can accurately capture the long-term temporal dependency between q-axis current and electrical frequency and achieve high-precision current prediction. At the same time, it can effectively avoid the problems of low training efficiency caused by network redundancy or insufficient feature capture caused by overly simple structure. It can achieve the effect of advanced adaptive compensation for compressor control output under load change scenarios and significantly suppress DC bus voltage overshoot.
[0057] Example 2
[0058] According to an embodiment of the present invention, a method embodiment for controlling a current compensation circuit is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.
[0059] Figure 3 This is a flowchart of a control method for a current compensation circuit according to an embodiment of the present invention, such as... Figure 3 As shown, the method includes the following steps:
[0060] Step S302: Generate a multi-dimensional time series input vector based on multi-dimensional time series data within a preset time period, wherein the multi-dimensional time series data includes current data and frequency data.
[0061] In this embodiment, the preset time refers to a sampling window period with a fixed length of 100ms, which is used to limit the time span of data acquisition; the multidimensional time series data specifically refers to the compressor q-axis current data and compressor electrical frequency data collected within the preset time window, wherein the current data is specifically taken as the maximum value among all sampling points within the window, and the frequency data is taken as the average value within the window, which serves as the original feature sequence reflecting the motor operating status.
[0062] It should be noted that, Figure 4 This is a flowchart of a signal transmission method for multiplexed circuits. The LSTM model needs to be pre-trained; a simplified flowchart is shown below. Figure 4 As shown, the offline training iterative closed-loop process of the LSTM neural network model from data preparation to performance convergence first collects q-axis current and electrical frequency data during compressor operation. After data preprocessing and dataset establishment, the initial network parameters of the LSTM model are set. Then, the model training phase begins, and the prediction error of the current network is calculated. Finally, error evaluation is performed based on whether the required conditions are met. If not, the process returns to the parameter setting phase for readjustment and continues training until the model accuracy reaches the required level. The specific steps are as follows:
[0063] Optionally, data acquisition uses a 100ms sampling window to collect time-series data containing precise timestamps and calculations of the maximum value of the compressor q-axis current and the average electrical frequency within 100ms. The data is arranged in chronological order, with one set of data collected every 100ms. Assuming N compressor q-axis current and electrical frequency data can be obtained within 100ms, the k-th sampling window can yield: , ,in, and These are the nth q-axis current sampling value and the nth compressor electrical frequency sampling value within the kth sampling window, respectively.
[0064] Optionally, data preprocessing will and Scaling to the [0,1] interval, using min-max normalization, the normalization formula is: Where max and min are the maximum and minimum values in the sample data, respectively; x is the sample data. This is the normalized sample data.
[0065] Optionally, the root mean square error is selected as the model evaluation metric for calculating the model's prediction accuracy. The formula for evaluating the performance of an LSTM model is as follows: ,in, Model predictions; represents the measured value; N is the number of samples. Based on actual needs, the model satisfies... That's all.
[0066] Figure 5 This is a graph showing the changes in metrics during the training of the LSTM model, such as... Figure 5 As shown, during model training, the prediction error decreases with increasing epochs. Particularly in the first 20 epochs, the prediction error significantly decreases, confirming the effectiveness of model parameter optimization. After 30 epochs of training, the error metrics stabilize, with the error eventually dropping below 0.1, demonstrating the model's high prediction accuracy. The trained LSTM model is then applied to the q-axis current compensation algorithm. Simulations simulate load changes by setting a torque surge at 2 seconds, and comparative experiments maintain consistency with the surge conditions. Figure 6 This is a comparison chart of the q-axis current before and after compensation, such as... Figure 6 As shown, after adding the compensation algorithm, the q-axis given current can respond more quickly to sudden load changes. Figure 7 The figure shows the bus voltage simulation results under sudden load changes without q-axis current compensation. Figure 8 The figure shows the simulation results of bus voltage under sudden load changes with q-axis current compensation. Figure 7and Figure 8 Under sudden load changes, the bus voltage overshoots to 469V without the compensation algorithm, while the maximum bus voltage is only 430V after adding the q-axis current compensation algorithm. This shows that a faster q-axis current given response can significantly reduce the bus voltage overshoot value under sudden load changes.
[0067] This method extracts features of the compressor's operating state through a specific time window, constructs an LSTM model, and collects compressor q-axis current and electrical frequency data in each period with a fixed sampling period of 100ms. The maximum value of the q-axis current in that period is used as a feature reflecting the load impact intensity, and the average value of the electrical frequency is used as a reference reflecting the motor speed state, thereby transforming the continuous physical signal into a discrete feature vector with time-dependent relationships.
[0068] By implementing this method, the prediction accuracy of the LSTM model for load change scenarios is improved, and the timeliness of feedforward compensation control is enhanced. By selecting the maximum current value instead of the instantaneous value, prediction deviations caused by signal fluctuations are effectively avoided, ensuring that the model can keenly capture drastic load changes that may lead to bus voltage overshoot. At the same time, the standardized time-series data input format simplifies the model training complexity and improves the convergence speed, enabling the control system to accurately predict future current demands based on historical data in a very short time when load changes occur, and generate compensation current commands in advance. This allows for proactive suppression before voltage overshoot occurs, significantly reducing the fluctuation amplitude of DC bus voltage and ensuring the stable operation of the electrolytic capacitor-free air conditioning system under extreme conditions.
[0069] Step S304: Input the multi-dimensional timing input vector into the network model and perform gain calculation to obtain the compensation current.
[0070] In this embodiment, gain calculation refers to subtracting the predicted q-axis current value at the next time step from the actual q-axis current value at the current time step (or the predicted value at the previous time step) to obtain the current change, and then multiplying this change by a preset gain coefficient (i.e., To obtain the compensation amplitude.
[0071] This method combines the timing prediction capability of LSTM with the proportional gain adjustment mechanism to transform abstract prediction results into specific control compensation quantities. This method is based on the formula... First, the constructed multi-dimensional time-series input vector is fed into the trained LSTM network, which outputs a predicted value of the compressor q-axis current maximum value in the next time window (e.g., 100ms in the future). Then, by calculating the difference between the predicted value and the currently collected q-axis current maximum value, the current change reflecting the load change trend is obtained. Finally, the current change is multiplied by a preset gain coefficient to generate a compensation current with physical meaning.
[0072] By implementing this method, adaptive adjustment of feedforward compensation is achieved, which improves the response speed to load changes and the ability to suppress bus voltage overshoot. It quantifies the severity of load changes and ensures the consistency of the compensation current with the grid voltage phase. Thus, while indirectly adjusting the inverter output power, it maintains a high power factor, offsets the bus voltage fluctuations caused by control loop delay, effectively reduces the voltage peak of the electrolytic capacitor-less air conditioner outdoor unit under dynamic conditions, and improves the system's operational stability and hardware lifespan.
[0073] Step S306: Calculate the compensation current to generate the quadrature axis given current.
[0074] In this embodiment, the quadrature-axis given current (i.e., the q-axis given current value) is the final current command generated by superimposing and summing the output of the speed loop PI controller with the compensation current. This command is directly used as the reference input of the current loop.
[0075] This method superimposes and fuses the prediction-based feedforward compensation component with the feedback-based feedback control component. First, it obtains the basic q-axis current output calculated by the speed loop PI regulator based on the speed error. ); then, the generated compensation current ( The torque current command is formed by algebraically adding the two commands together.
[0076] By implementing this method, the response speed of the current loop to load changes is significantly improved, the instantaneous overshoot of the DC bus voltage is effectively suppressed, and by introducing feedforward compensation, the control system can immediately reflect the corresponding adjustment in the current setpoint when the actual load changes drastically. This makes up for the response delay caused by the integral action lag of the traditional PI controller, and at the same time, it greatly reduces the peak amplitude of the bus voltage, ensuring the voltage stability of the electrolytic capacitor-free air conditioner outdoor unit under dynamic operating conditions.
[0077] Step S308: Based on the quadrature axis given current, generate a drive pulse signal and input the drive pulse signal into the inverter to control the motor operation.
[0078] In this embodiment, the drive pulse signal is calculated by using a space vector pulse width modulation (SVPWM) algorithm to convert the quadrature axis given current (combined with the d-axis given current value) into voltage vector components in a stationary two-phase coordinate system, and then calculates the action time sequence of the six basic voltage vectors of the inverter, which is used to replace the traditional sinusoidal pulse width modulation signal. The inverter is a power electronic device that converts DC bus power into three-phase AC power and drives the compressor motor. By inputting the drive pulse signal, the stator current of the motor can be precisely controlled to follow the given command.
[0079] This method uses coordinate transformation and SVPWM technology to convert current commands into control signals for the inverter's switching devices. Specifically, the rotating coordinate system (dq coordinate system) command, which includes the quadrature-axis given current (and the direct-axis given current), is first converted into a voltage reference vector in a stationary two-phase coordinate system (α-β coordinate system) through inverse coordinate transformation. Then, the SVPWM algorithm is used to calculate the duration of the six basic voltage vectors corresponding to this voltage vector, generating precise switching timing pulse signals. These pulse signals are directly input to the inverter to control the switching of power devices, thereby efficiently converting the DC bus voltage into three-phase AC power to drive the compressor's permanent magnet synchronous motor.
[0080] By implementing this method, rapid and precise control of motor torque is achieved, effectively suppressing DC bus voltage overshoot and improving system dynamic performance. Because the SVPWM algorithm has higher voltage utilization and lower current harmonic distortion, combined with optimized current commands generated by feedforward compensation, the motor can quickly respond to torque demand changes caused by sudden load shifts and promptly balance DC-side power fluctuations. This fast and smooth current control not only significantly reduces the peak overshoot amplitude of the bus voltage and improves voltage stability, but also reduces motor operating noise and electromagnetic stress, extending the service life of core components in the electrolytic capacitor-free air conditioning system.
[0081] As described above, in this embodiment, a multi-dimensional timing input vector is generated based on multi-dimensional timing data within a preset time period. This multi-dimensional timing data includes current data and frequency data. The multi-dimensional timing input vector is input into the network model and gain is calculated to obtain a compensation current. The compensation current is then calculated to generate a quadrature-axis given current. Based on this quadrature-axis given current, a drive pulse signal is generated and input into the inverter to control the motor operation. This achieves the goal of introducing LSTM's q-axis current adaptive feedforward compensation into the electrolytic capacitor-free drive system, overcoming the hysteresis limitations of traditional feedback control. This results in significantly suppressing DC bus voltage overshoot in the electrolytic capacitor-free air conditioner outdoor unit, reducing bus voltage fluctuation amplitude, and improving system reliability and hardware lifespan.
[0082] Therefore, the technical solution provided by the above embodiments of the present invention solves the technical problems in the related art, such as the bus voltage being sensitive to load changes and the traditional passive control response being lagging due to the weak energy storage buffering capacity of the non-electrolytic capacitor, which leads to increased stress on system protection or devices.
[0083] According to the above embodiments of the present invention, generating a multi-dimensional time-series input vector based on multi-dimensional time-series data within a preset time period includes: collecting a first preset number of current data and frequency data within each time window of the preset time period, wherein the number of time windows is equal to half of the target number; extracting the maximum value of the current data within each time window to obtain a current maximum value sequence; calculating the average value of the frequency data within each time window to obtain an electrical frequency average value sequence; and concatenating the current maximum value sequence and the electrical frequency average value sequence to obtain a multi-dimensional time-series input vector.
[0084] In this embodiment, a time window refers to a discrete time period divided within a preset time period, the number of which is equal to half the target number, used to divide the continuous data stream into discrete sampling units; the first preset number of current data and frequency data refers to multiple raw sampling point data collected within each time window, wherein the current data is used to reflect the instantaneous load intensity, and the frequency data is used to reflect the motor operating speed state; the current maximum value sequence is a sequence formed by extracting the maximum value from all current sampling points in each time window; the electric frequency average value sequence is a sequence obtained by calculating the arithmetic mean of all frequency sampling points within each time window.
[0085] This method uses discretization sampling and feature extraction algorithms to transform continuous high-frequency physical signals into structured time-series feature vectors suitable for LSTM model processing. Specifically, the system first divides the preset total time span into several discrete half-cycle time windows, and within each window, it collects a large amount of raw current (…). ) and frequency sampling points ( Subsequently, for the current data reflecting load impact characteristics, the maximum value within the window is extracted to capture the most drastic load change trends and avoid instantaneous noise interference; the arithmetic mean within the window is calculated to smooth fluctuations and obtain stable speed state indicators; finally, the extracted maximum current value sequence and the frequency average value sequence are concatenated along the feature dimension, i.e. A multi-dimensional time-series input vector containing state information from multiple past time steps is constructed.
[0086] By implementing this method, the signal-to-noise ratio and feature representativeness of the input data were significantly optimized, improving the accuracy of the LSTM model in recognizing load surge patterns. By selecting the maximum current value instead of the instantaneous value, the key peak information of the load surge was effectively preserved, enabling the model to keenly perceive potential voltage overshoot risks. At the same time, the use of the average frequency eliminated misjudgments caused by high-frequency speed fluctuations, ensuring the stability of the speed state representation.
[0087] According to the above embodiments of the present invention, inputting a multi-dimensional time-series input vector into a network model and performing gain calculation to obtain a compensation current includes: processing the multi-dimensional time-series input vector using the network model to generate a current prediction value for the next time window; calculating the difference between the current prediction value and the maximum current value of the current time window using a first calculation module to obtain a difference current; and processing the difference current using a first gain module to obtain a compensation current.
[0088] In this embodiment, the predicted current value for the next time window refers to the estimated maximum value of the q-axis current generated after processing by the network model, corresponding to the next 100ms sampling window, which aims to predict load changes in advance; the maximum current value for the current time window refers to the actual maximum value of the q-axis current collected in the just-ended 100ms time window, which serves as the benchmark reference point for prediction.
[0089] This method utilizes the temporal prediction capabilities of Long Short-Term Memory (LSTM) networks combined with a feedforward compensation mechanism to transform historical data into proactive control instructions for the future, based on the formula... First, the constructed multi-dimensional time-series input vector is input into the trained LSTM model, which outputs a predicted estimate of the maximum q-axis current value in the next time window (e.g., the next 100ms). Then, the first calculation module subtracts the predicted value from the actual maximum q-axis current value collected in the current time window to obtain the difference current reflecting the load change trend. Finally, the difference current is processed by the first gain module to convert the change in current into a compensation current amplitude.
[0090] By implementing this method, feedforward active adjustment of the control system is achieved, significantly improving response speed and suppressing bus voltage overshoot. By predicting the peak current at the next moment and comparing it with the current actual value, the amplitude and direction of load changes can be accurately identified, generating corresponding compensation current. The application of the gain module ensures that the compensation amount matches the phase of the grid voltage and power fluctuations, enabling the final current command to respond more smoothly and quickly to load changes. This effectively compensates for the response delay caused by integral lag in traditional PI controllers, allowing for adjustment before drastic load changes occur. This significantly reduces the fluctuation amplitude of the DC bus voltage and improves the operational stability of the electrolytic capacitor-free air conditioning system under dynamic conditions.
[0091] According to the above embodiments of the present invention, calculating the compensation current to generate the quadrature-axis given current includes: using a second calculation module to calculate the difference between the preset speed and the actual speed to obtain the speed difference; using a regulator to process the speed difference to obtain the output current; using a third calculation module to calculate the output current and the compensation current to obtain the original given current; and using a second gain module to process the original given current to obtain the quadrature-axis given current.
[0092] In this embodiment, the preset speed refers to the target speed at which the compressor is expected to operate; the actual speed refers to the current real-time speed at which the compressor is operating; the speed difference is the deviation value calculated by subtracting the actual speed from the preset speed, reflecting the current speed tracking error of the motor; the original given current refers to the intermediate current command value obtained by superimposing (summing) the base value q-axis current (i.e., output current) output by the speed loop PI regulator with the feedforward compensation current obtained based on LSTM prediction.
[0093] This method is based on formulas The calculation involves inputting the speed error into the speed loop PI based on the deviation between the preset target speed and the actual operating speed, thereby generating a basic q-axis current output to eliminate steady-state error and dynamic deviation. Subsequently, the third calculation module superimposes and sums this basic output with the compensation current to obtain the original given current containing the predicted correction. Finally, the second gain module performs a final proportional adjustment on the original given current to obtain the final quadrature axis given current command delivered to the current loop.
[0094] By implementing this method, the overall suppression capability of the control system against load disturbances is significantly improved, achieving dual optimization of speed stability and voltage smoothness. This ensures that the motor can accurately track the target speed and maintain basic power output stability. The introduced feedforward compensation current preemptively offsets the torque impact caused by sudden load changes, effectively compensating for the inherent lag of feedback control. This allows for more precise matching of load demand changes, significantly reducing the overshoot of the bus voltage during dynamic processes, reducing the stress on power electronic devices, and improving the operational reliability and energy efficiency of the electrolytic capacitor-free air conditioner outdoor unit under conditions of frequent start-stop or severe load fluctuations.
[0095] According to the above embodiments of the present invention, a drive pulse signal is generated based on the quadrature-axis given current, and the drive pulse signal is input into the inverter to control the motor operation, including: processing the quadrature-axis given current and the direct-axis given current using a conversion module to obtain a first voltage component and a second voltage component, wherein the quadrature-axis given current is orthogonal to the direct-axis given current; and modulating the first voltage component and the second voltage component to obtain the drive pulse signal.
[0096] In this embodiment, the transformation module refers to the coordinate transformation unit, which specifically performs the transformation from the rotating coordinate system (dq axis) to the stationary coordinate system (α-β axis). Its inputs are mutually orthogonal quadrature axis given current (q-axis current given value, reflecting the torque component) and direct axis given current (d-axis current given value, usually used for field control or field weakening control). The output is the voltage vector components in the stationary two-phase coordinate system, namely the first voltage component (α-axis voltage reference value) and the second voltage component (β-axis voltage reference value).
[0097] This method uses coordinate transformation and pulse width modulation (PWM) technology to convert the current command generated by the control algorithm into a switching signal executable by the inverter. Specifically, the transformation module first converts the mutually orthogonal quadrature-axis and direct-axis given currents in the rotating coordinate system into the first and second voltage components in the stationary two-phase coordinate system (α-β coordinate system). Subsequently, these voltage reference components are modulated (SVPWM) to calculate the duration of the six basic voltage vectors of the inverter, generating a precise drive pulse signal sequence, thereby converting DC-side electrical energy into three-phase AC power of specific frequency and amplitude to drive the motor.
[0098] By implementing this method, the motor stator current achieves rapid and accurate tracking of given commands, thereby effectively suppressing bus voltage fluctuations and improving system dynamic performance. Through coordinate transformation, the control system can independently and precisely control the motor's torque and magnetic field components, ensuring that the motor can quickly respond to current command changes caused by feedforward compensation when the load changes abruptly. High-precision pulse modulation reduces current harmonic distortion, switching losses, and electromagnetic noise. This efficient drive control enables the motor torque to timely balance the power fluctuations of the DC bus, significantly reducing the overshoot peak of the bus voltage and ensuring the voltage stability and operational reliability of the electrolytic capacitor-free air conditioning system under dynamic conditions.
[0099] According to the above embodiments of the present invention, the control method of the current compensation circuit further includes: acquiring historical operating data to form an original dataset arranged in chronological order, wherein each data point in the original dataset corresponds to the maximum current value and the average electrical frequency value within a time window; taking a second preset number of consecutive maximum current values and average electrical frequency values as an input sequence to obtain a corresponding current prediction value, wherein the second preset number is equal to the number of time windows; moving the sliding window backward by one time window and repeating the above selection process until the original dataset is traversed to generate a target sample; and dividing the target sample into a first sample set, a second sample set, and a third sample set according to the chronological order and a preset ratio.
[0100] In this embodiment, historical operating data refers to the original records collected during the actual operation of the compressor, including timestamps, q-axis current, and electrical frequency. The original dataset is a sequence formed by organizing these historical operating data in chronological order, where each data point represents a statistical feature within a time window (100ms), specifically including the maximum current value (reflecting the load impact peak) and the average electrical frequency value (reflecting the average speed state) within the window. The sliding window refers to the reading mechanism used to traverse the dataset. After each selection of a set of input sequences and generation of a corresponding predicted output, the window moves backward by the length of one time window, thereby generating a new input-output pair. The target sample refers to all (input sequence, predicted value) paired data extracted from the original dataset through the above sliding window traversal process for model training. The first sample set, the second sample set, and the third sample set refer to the training set, the validation set, and the test set, which are divided into the target samples in chronological order according to a preset ratio, and are used for model parameter learning, hyperparameter adjustment, and final performance evaluation, respectively.
[0101] Optionally, a dataset is created using the maximum q-axis current value and the average compressor electrical frequency within the first 10 100ms intervals as input, and the maximum q-axis current value of the compressor within the next 100ms interval as output. The window is moved forward by 100ms, and this process is repeated to form all samples. The purpose of the first step, data acquisition, is to obtain raw data. Each data point is 100ms apart in time, and each data point needs to be calculated using N data points collected within 100ms (taking the maximum and average values). The third step, creating the dataset, is based on the raw data obtained in the first step, using a sliding window. Each window has 10 data points (n1,...,n10), and the eleventh data point is output as n11. Then, the next window takes data points from the second to the eleventh (n2,...,n11), with n12 as the output, and so on.
[0102] Optionally, the dataset can be partitioned into three parts in chronological order: a training set (70% of the total dataset) for model learning; a validation set (15% of the total dataset) for tuning hyperparameters and monitoring overfitting; and a test set (15% of the total dataset) for final evaluation.
[0103] This method constructs a training dataset for an LSTM model based on a time-series sliding window mechanism to map historical operating data to future state predictions. Specifically, it first extracts data containing timestamps, q-axis current, and electrical frequency from the compressor's actual operating records, and then calculates the maximum current and average frequency for each window within a fixed time window (e.g., 100ms) to form a time-sorted original feature sequence. Subsequently, using the sliding window technique, it takes a second set of historical window data consecutively as the input sequence and the current prediction value of the next window as the label to generate paired training samples. By sliding the window forward step by step through the entire dataset, it generates a target sample set containing rich temporal dependencies, which is then divided into training, validation, and test sets according to a specific ratio to provide standardized learning samples for the model.
[0104] By implementing this method, the temporal consistency and representativeness of the LSTM model training data were ensured, significantly improving the model's prediction accuracy and generalization ability. By extracting the maximum current and average frequency as features, instantaneous noise interference was effectively removed, while retaining key dynamic information reflecting load mutations and speed changes. The sliding window mechanism ensured the continuity and logical correlation of time steps in the input sequence, enabling the model to fully learn the long-term dependence of current and frequency over time. Ultimately, data leakage was avoided. Through the collaborative work of the training, validation, and test sets, the model's hyperparameters were optimized, and its generalization performance was verified, ensuring that the model can make fast and accurate predictions of load mutations in practical applications.
[0105] As described above, the technical solution provided by the above embodiments of the present invention generates a multi-dimensional timing input vector based on multi-dimensional timing data within a preset time period. The multi-dimensional timing data includes current data and frequency data. The multi-dimensional timing input vector is input into a network model and gain is calculated to obtain a compensation current. The compensation current is calculated to generate a quadrature-axis given current. Based on the quadrature-axis given current, a drive pulse signal is generated and input into the inverter to control the motor operation. This achieves the goal of introducing LSTM's q-axis current adaptive feedforward compensation into a capacitor-free drive system, overcoming the hysteresis limitations of traditional feedback control. This significantly suppresses DC bus voltage overshoot in capacitor-free air conditioner outdoor units, reduces bus voltage fluctuations, and improves system reliability and hardware lifespan.
[0106] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.
[0107] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.
[0108] Example 3
[0109] According to embodiments of the present invention, a control device for a current compensation circuit for implementing the control method of the above-described current compensation circuit is also provided. Figure 9 This is a schematic diagram of the control device for the current compensation circuit according to an embodiment of the present invention, such as... Figure 9 As shown, the device includes: a first generation unit 901, a first calculation unit 903, a second calculation unit 905, and a second generation unit 907. The device will now be described in detail.
[0110] The first generation unit 901 is used to generate a multi-dimensional time series input vector based on multi-dimensional time series data within a preset time period, wherein the multi-dimensional time series data includes current data and frequency data.
[0111] The first computing unit 903 is used to input multi-dimensional time-series input vectors into the network model and perform gain calculation to obtain the compensation current.
[0112] The second calculation unit 905 is used to calculate the compensation current and generate the quadrature axis given current;
[0113] The second generation unit 907 is used to generate a drive pulse signal based on the quadrature axis given current, and input the drive pulse signal into the inverter to control the motor operation.
[0114] It should be noted that the first generation unit 901, the first calculation unit 903, the second calculation unit 905, and the second generation unit 907 mentioned above correspond to steps S302 to S308 in the above embodiments. The four units and the corresponding steps implement the same instances and application scenarios, but are not limited to the content disclosed in the above embodiments.
[0115] As can be seen from the above, in the solution described in the above embodiments of the present invention, the first generation unit 901 is used to generate a multi-dimensional timing input vector based on multi-dimensional timing data within a preset time period, wherein the multi-dimensional timing data includes current data and frequency data; the first calculation unit 903 is used to input the multi-dimensional timing input vector into the network model and perform gain calculation to obtain the compensation current; the second calculation unit 905 is used to calculate the compensation current to generate the quadrature-axis given current; and the second generation unit 907 is used to generate a drive pulse signal based on the quadrature-axis given current and input the drive pulse signal into the inverter to control the motor operation. Through the above technical solution, in the control of the compressor of the outdoor unit of an electrolytic capacitor-free air conditioner, the Long Short-Term Memory (LSTM) network is used to predict load changes and perform q-axis current feedforward compensation in advance to suppress DC bus voltage overshoot. This achieves the effect of effectively reducing the bus voltage fluctuation amplitude through LSTM prediction and advance compensation when load changes occur, improving system operation reliability and hardware lifespan, while maintaining a high input power factor.
[0116] Therefore, the technical solution provided by the above embodiments of the present invention solves the technical problems in the related art, such as the bus voltage being sensitive to load changes and the traditional passive control response being lagging due to the weak energy storage buffering capacity of the non-electrolytic capacitor, which leads to increased stress on system protection or devices.
[0117] Optionally, the first generation unit includes: an acquisition module, used to acquire a first preset number of current data and frequency data within each time window within a preset time period, wherein the number of time windows is equal to half of the target number; an extraction module, used to extract the maximum value of the current data within each time window to obtain a current maximum value sequence; a first processing module, used to calculate the average value of the frequency data within each time window to obtain an electrical frequency average value sequence; and a sequence construction module, used to concatenate the current maximum value sequence and the electrical frequency average value sequence to obtain a multi-dimensional time-series input vector.
[0118] Optionally, the first calculation unit includes: a first generation module, used to process the multi-dimensional time-series input vector using a network model to generate the current prediction value for the next time window; a first difference calculation module, used to calculate the difference between the current prediction value and the maximum current value of the current time window using the first calculation module to obtain the difference current; and a compensation calculation module, used to process the difference current using the first gain module to obtain the compensation current.
[0119] Optionally, the second calculation unit includes: a second difference calculation module, used to calculate the difference between the preset speed and the actual speed using the second calculation module to obtain the speed difference; a second processing module, used to process the speed difference using the regulator to obtain the output current; a third processing module, used to calculate the output current and the compensation current using the third calculation module to obtain the original given current; and a fourth processing module, used to process the original given current using the second gain module to obtain the quadrature axis given current.
[0120] Optionally, the second generation unit includes: a fifth processing module, used to process the quadrature-axis given current and the direct-axis given current using the transformation module to obtain a first voltage component and a second voltage component, wherein the quadrature-axis given current and the direct-axis given current are orthogonal; and a modulation module, used to modulate the first voltage component and the second voltage component to obtain a driving pulse signal.
[0121] Optionally, the control device for the current compensation circuit further includes: a data acquisition unit for acquiring historical operating data to form an original dataset arranged in chronological order, wherein each data point in the original dataset corresponds to the maximum current value and the average electrical frequency value within a time window; a sequence construction unit for using a second preset number of consecutive maximum current values and average electrical frequency values as input sequences to obtain corresponding current prediction values, wherein the second preset number is equal to the number of time windows; a third generation unit for moving the sliding window backward by one time window and repeating the above selection process until the original dataset has been traversed to generate target samples; and a partitioning unit for dividing the target samples into a first sample set, a second sample set, and a third sample set according to chronological order and a preset ratio.
[0122] According to another aspect of the present invention, an air conditioner outdoor unit is also provided, wherein the air conditioner outdoor unit uses the control method of the current compensation circuit described above.
[0123] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored program, wherein the program executes the control method of the current compensation circuit of any of the above.
[0124] According to another aspect of the present invention, a processor is also provided, which is used to run a program, wherein the program executes the control method of the current compensation circuit described above during runtime.
[0125] Optionally, in this embodiment, the computer-readable storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any communication device in a group of communication devices.
[0126] Optionally, in this embodiment, the computer-readable storage medium is configured to store program code for performing the following steps: generating a multi-dimensional timing input vector based on multi-dimensional timing data within a preset time period, wherein the multi-dimensional timing data includes current data and frequency data; inputting the multi-dimensional timing input vector into a network model and performing gain calculation to obtain a compensation current; calculating the compensation current to generate a quadrature-axis given current; generating a drive pulse signal based on the quadrature-axis given current, and inputting the drive pulse signal into the inverter to control the motor operation.
[0127] Optionally, in this embodiment, the computer-readable storage medium is configured to store program code for performing the following steps: generating a multi-dimensional timing input vector based on multi-dimensional timing data within a preset time period, including: collecting a first preset number of current data and frequency data within each time window of the preset time period, wherein the number of time windows is equal to half of the target number; extracting the maximum value of the current data within each time window to obtain a current maximum value sequence; calculating the average value of the frequency data within each time window to obtain an electrical frequency average value sequence; and concatenating the current maximum value sequence and the electrical frequency average value sequence to obtain a multi-dimensional timing input vector.
[0128] Optionally, in this embodiment, the computer-readable storage medium is configured to store program code for performing the following steps: inputting a multidimensional time-series input vector into a network model and performing gain calculation to obtain a compensation current, including: processing the multidimensional time-series input vector using the network model to generate a current prediction value for the next time window; calculating the difference between the current prediction value and the maximum current value of the current time window using a first calculation module to obtain a difference current; and processing the difference current using a first gain module to obtain a compensation current.
[0129] Optionally, in this embodiment, the computer-readable storage medium is configured to store program code for performing the following steps: calculating the compensation current to generate a quadrature-axis given current, including: using a second calculation module to calculate the difference between a preset speed and the actual speed to obtain a speed difference; using a regulator to process the speed difference to obtain an output current; using a third calculation module to calculate the output current and the compensation current to obtain an original given current; and using a second gain module to process the original given current to obtain a quadrature-axis given current.
[0130] Optionally, in this embodiment, the computer-readable storage medium is configured to store program code for performing the following steps: generating a drive pulse signal based on the quadrature-axis given current, and inputting the drive pulse signal into the inverter to control the motor operation, including: processing the quadrature-axis given current and the direct-axis given current using a conversion module to obtain a first voltage component and a second voltage component, wherein the quadrature-axis given current is orthogonal to the direct-axis given current; and modulating the first voltage component and the second voltage component to obtain the drive pulse signal.
[0131] Optionally, in this embodiment, the computer-readable storage medium is configured to store program code for performing the following steps: the control method for the current compensation circuit further includes: acquiring historical operating data to form an original dataset arranged in chronological order, wherein each data point in the original dataset corresponds to the maximum current value and the average electrical frequency value within a time window; taking a second preset number of consecutive maximum current values and average electrical frequency values as an input sequence to obtain a corresponding current prediction value, wherein the second preset number is equal to the number of time windows; moving the sliding window backward by one time window and repeating the above selection process until the original dataset is traversed to generate a target sample; dividing the target sample into a first sample set, a second sample set, and a third sample set according to the chronological order and a preset ratio.
[0132] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0133] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0134] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.
[0135] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0136] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0137] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0138] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.
[0139] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A current compensation circuit, characterized by, include: The compensation unit, speed control unit, and drive unit, among which, The compensation unit includes a network model, a first calculation module, and a first gain module. The input of the network model is a multi-dimensional time-series input vector. The output of the network model is connected to the positive input of the first calculation module, and the output of the first calculation module is connected to the first gain module. The speed control unit includes a second calculation module, a regulator, and a third calculation module. The output of the second calculation module is connected to one end of the regulator, the other end of the regulator is connected to the first input of the third calculation module, and the output of the first gain module is connected to the second input of the third calculation module. The drive unit includes a second gain module, a conversion module, a modulation module, and an inverter. The output of the third calculation module is connected to one end of the second gain module, the other end of the second gain module is connected to one end of the conversion module, the other end of the conversion module is connected to one end of the modulation module, and the other end of the modulation module is connected to the inverter.
2. The current compensation circuit of claim 1, wherein, The network model includes an input layer, an output layer, and a hidden layer. The number of neurons in the input layer is the target number, the number of neurons in the output layer is one, and it is used to output the predicted current value. The number of neurons in the hidden layer is determined based on a preset mathematical formula and the number of neurons in the input layer and the output layer.
3. A control method of a current compensation circuit, characterized by, The current compensation circuit described in claim 1 or 2 includes: Based on multi-dimensional time-series data within a preset time period, the multi-dimensional time-series input vector is generated, wherein the multi-dimensional time-series data includes current data and frequency data; The multidimensional time-series input vector is input into the network model and the gain is calculated to obtain the compensation current; The compensation current is calculated to generate the quadrature axis given current; Based on the given quadrature axis current, a drive pulse signal is generated and input to the inverter to control the motor operation.
4. The control method of the current compensation circuit according to claim 3, characterized by, Based on multi-dimensional time-series data within a preset time period, the multi-dimensional time-series input vector is generated, including: Within each time window of the preset time period, a first preset number of current data and frequency data are collected, wherein the number of time windows is equal to half of the target number; Extract the maximum value of the current data within each time window to obtain a sequence of maximum current values; Calculate the average value of the frequency data within each time window to obtain the electrical frequency average value sequence; The multidimensional timing input vector is obtained by concatenating the sequence of maximum current values with the sequence of average electrical frequencies.
5. The control method of the current compensation circuit according to claim 3, wherein The multidimensional time-series input vector is input into the network model and gain calculation is performed to obtain the compensation current, including: The network model is used to process the multidimensional time-series input vector to generate the current prediction value for the next time window; The difference between the predicted current value and the maximum current value in the current time window is calculated using the first calculation module to obtain the difference current. The difference current is processed using the first gain module to obtain the compensation current.
6. The control method for the current compensation circuit according to claim 3, characterized in that, The compensation current is calculated to generate the quadrature-axis given current, including: The second calculation module is used to calculate the difference between the preset speed and the actual speed to obtain the speed difference; The speed difference is processed using the regulator to obtain the output current; The third calculation module is used to calculate the output current and the compensation current to obtain the original given current; The original given current is processed using the second gain module to obtain the quadrature-axis given current.
7. The control method for the current compensation circuit according to claim 3, characterized in that, Based on the given quadrature-axis current, a drive pulse signal is generated, and the drive pulse signal is input to the inverter to control the motor operation, including: The transformation module is used to process the quadrature-axis given current and the direct-axis given current to obtain a first voltage component and a second voltage component, wherein the quadrature-axis given current is orthogonal to the direct-axis given current; The first voltage component and the second voltage component are modulated to obtain the driving pulse signal.
8. The control method of the current compensation circuit according to any one of claims 3 to 7, characterized by, The control method for the current compensation circuit also includes: Historical operating data is acquired to form a raw dataset arranged in chronological order, wherein each data point in the raw dataset corresponds to the maximum current and the average electrical frequency within a time window; The maximum current value and the average electrical frequency of a second preset number of consecutive current values are used as input sequences to obtain the corresponding current prediction value, wherein the second preset number is equal to the number of time windows; Move the sliding window backward by one of the aforementioned time windows, repeat the above selection process, until the original dataset has been traversed and the target sample is generated; The target samples are divided into a first sample set, a second sample set, and a third sample set according to time sequence and a preset ratio.
9. A control device for a current compensation circuit, characterized by The control method applied to the current compensation circuit of claim 3 or 8 includes: The first generation unit is used to generate the multi-dimensional time-series input vector based on multi-dimensional time-series data within a preset time period, wherein the multi-dimensional time-series data includes current data and frequency data. The first calculation unit is used to input the multi-dimensional time-series input vector into the network model and perform gain calculation to obtain the compensation current. The second calculation unit is used to calculate the compensation current and generate the quadrature axis given current; The second generation unit is used to generate a drive pulse signal based on the given cross-axis current, and input the drive pulse signal into the inverter to control the motor operation.
10. An air conditioner outdoor unit characterized by comprising: The air conditioner outdoor unit uses the control method of the current compensation circuit as described in any one of claims 3 to 8.
