A motor energy-saving control method based on load prediction

Abstract

The application relates to the technical field of motor control and industrial energy saving, and particularly discloses a motor energy-saving control method based on load prediction, which comprises the following steps: constructing a three-level time scale load prediction architecture, obtaining motor load working condition parameters of each layer of the load prediction architecture, obtaining a global efficiency MAP (Maximum A Posteriori Probability) graph of a motor to be controlled, taking motor speed and output torque as coordinates, searching for a target speed with the highest efficiency in a corresponding load interval in the global efficiency MAP graph according to the motor load working condition parameters output by a long-term scale prediction layer and a medium-term scale prediction layer, taking the target speed as a motor speed base value, generating a speed pre-adjustment instruction, superimposing the motor speed base value on an instantaneous torque demand, combining the global efficiency MAP graph to perform global loss optimal fine adjustment feedback, determining a control instruction of a motor driver, introducing a predicted value feedforward compensation in a control loop, and performing feedback logic control on a motor state according to the motor load working condition parameters output by the long-term scale prediction layer.

Description

Technical Field

[0001] This invention relates to the fields of motor control and industrial energy-saving technology, specifically to a motor energy-saving control method based on load prediction. Background Technology

[0002] With the continuous improvement of industrial automation, motor systems have become one of the most energy-consuming terminal energy-consuming devices in industrial production. To reduce the energy consumption of motor operation, existing technologies for the high-efficiency operation control of asynchronous motors, permanent magnet synchronous motors, and drive systems under variable load conditions typically employ variable frequency speed regulation, high-efficiency motor replacement, and energy-saving control strategies based on simple load feedback. Common motor energy-saving control methods mainly include: By linearly extrapolating current load or short-term historical data, load changes within the next few seconds can be predicted to adjust motor output. However, due to the short prediction window, this method cannot anticipate medium- to long-term load trends caused by production plans or operating condition changes, leading to frequent motor operation in inefficient zones. Furthermore, the adjustment process itself incurs additional losses. While achieving efficient operation at a fixed speed or torque and making minor adjustments when the load changes, this method cannot consider the dynamic impact of load changes on the optimal efficiency point, especially in scenarios with large and intermittent load fluctuations, resulting in limited energy-saving effects. Existing methods, although incorporating motor efficiency maps, often only perform lookup optimization under steady-state conditions, lacking the ability to proactively predict and plan for load change trends. Moreover, specific losses such as stator copper losses, rotor iron losses, and mechanical losses are not included in the optimization process. Summary of the Invention

[0003] The purpose of this invention is to provide a motor energy-saving control method based on load prediction, and to solve the following technical problems: How to achieve a comprehensive energy-saving control method that integrates load prediction results across long, medium, and short time scales, combines the overall efficiency characteristics of the motor for global optimization, and can dynamically switch the motor's operating state.

[0004] The objective of this invention can be achieved through the following technical solutions: A motor energy-saving control method based on load prediction, the method comprising: Step 1: Construct a three-level time-scale load forecasting architecture and obtain the motor load condition parameters of each level of the load forecasting architecture; the three-level time-scale includes a long-term forecasting layer, a medium-term forecasting layer, and a short-term forecasting layer. Step 2: Obtain the overall efficiency MAP of the motor to be controlled. The efficiency MAP uses motor speed and output torque as coordinates to calibrate the total efficiency of the motor at different operating points and the detailed loss values, including at least stator copper loss, rotor iron loss, and mechanical loss. Step 3: Based on the motor load condition parameters output by the long-term and medium-term prediction layers, retrieve the target speed with the highest efficiency within the corresponding load range from the global efficiency MAP map, use it as the base value for motor speed, and generate a speed pre-adjustment command. Step 4: Based on the motor load condition parameters output in the short-term scale prediction layer, superimpose the instantaneous torque demand on the base value of motor speed, and combine it with the global efficiency MAP map to perform global loss optimization feedback. Step 5: Determine the control command for the motor driver based on the global loss optimal fine-tuning feedback, introduce the predicted value feedforward compensation in the control loop, and execute the multi-time-scale linkage feedback logic to control the motor state according to the motor load condition parameters output by the long-term scale prediction layer.

[0005] Preferably, step one includes: The long-term forecasting layer takes production planning and scheduling data, historical operating data and holidays as input, and uses a time-series forecasting algorithm with periodic decomposition to output the load power in the future minutes to hours, which is used for the pre-positioning of the motor to work efficiently. The mid-scale prediction layer takes the measured load values, operating event labels and long-scale prediction results of the most recent N sampling periods as input, and outputs the load torque change curve in the next ten seconds to minutes for planning the base value of motor speed. The short-term prediction layer takes the motor's three-phase current, speed, and output torque in the current sampling period as input and outputs the instantaneous load fluctuations in the hundreds of milliseconds to seconds in the future, which are used for dynamic compensation of the motor's instantaneous torque.

[0006] Preferably, the calibration method for different operating points in the global efficiency MAP in step two includes: During the factory testing or offline calibration phase of the motor, the operating range of the motor is traversed according to the preset speed step size and torque step size. The stator resistance voltage drop, stator current, rotor speed and output mechanical torque are measured at each operating point. The stator copper loss, rotor iron loss, mechanical loss and total efficiency at that operating point are calculated and stored in the controller in the form of a two-dimensional lookup table or a three-dimensional surface.

[0007] Preferably, the retrieval method in the global efficiency MAP in step three includes: Based on the load torque range output by the long-term and medium-term prediction layers Extract the set of all speeds that satisfy the torque constraint from the efficiency MAP. Calculate the average efficiency for the corresponding load range under this set of speeds. Select The maximum speed is used as the target speed. If multiple speeds correspond to the same maximum average efficiency, then the point with the lowest speed is selected as the base value of the motor speed.

[0008] Preferably, the optimal global loss fine-tuning feedback method in step four is as follows: The instantaneous torque fluctuation output by the short-term prediction layer The base torque superimposed on the base speed value To obtain the initial target torque ; Will Input the efficiency MAP plot and calculate the predicted total loss at that operating point. ; like Greater than the preset loss threshold The optimization objective is to minimize the cumulative loss within the prediction window. or Fine-tuning was performed to obtain the adjusted rotational speed. and torque This is output as the feedback result for optimal fine-tuning of global loss.

[0009] Preferably, the objective function for optimizing the global loss-optimized fine-tuning feedback is expressed as:

[0010] in, To optimize the objective function, For stator copper loss, For rotor iron loss, For mechanical wear, As a candidate work site, To predict the time step within the window; The constraints are: , The maximum allowable torque tracking deviation; , This represents the maximum permissible speed adjustment range.

[0011] Preferably, the motor driver control commands determined in step five include: For asynchronous motors, the fine-tuned target speed Convert to inverter output frequency command ,in This represents the number of pole pairs of the motor. The fine-tuned target torque Converted to active component of stator current instruction; The above instructions are used to generate the inverter's switching signals using a space vector pulse width modulation (SVPWM) algorithm.

[0012] Preferably, the calculation method for the predicted value feedforward compensation introduced in the control loop is as follows: In the speed loop controller, the speed change rate output by the mid-scale prediction layer is used. As a feedforward term superimposed on the torque command input, the feedforward compensation amount ,in This is the feedforward gain coefficient; In the torque loop controller, the torque change rate output by the short-term prediction layer is used. This is superimposed on the torque command input as a feedforward term to form a composite control input.

[0013] Preferably, the specific method for controlling the motor state by executing multi-time-scale linkage feedback logic based on the motor load condition parameters output by the long-term scale prediction layer is as follows: When the long-term scale prediction layer continuously outputs future... When the load is consistently below 20% of the rated load for an extended period of time, the motor is controlled to enter a low-power sleep state, maintaining only the minimum excitation current or cutting off the power module output. When the long-term prediction layer detects that the load will be When the load recovers to more than 30% of the rated load after a certain period of time, it should be done in advance. The timer wakes up the motor and executes the preheating control logic, enabling the motor to reach a stable operating temperature before entering the load ramp-up phase.

[0014] The beneficial effects of this invention are: (1) This invention constructs a decoupled architecture of long-term prediction layer, medium-term prediction layer and short-term prediction layer to address three types of control requirements: working condition trend perception, speed base value planning and instantaneous torque compensation. This enables the motor to complete the pre-positioning of the efficient operating point and speed pre-adjustment before the actual load arrives. At the same time, combined with the motor's global efficiency MAP, the speed base value with the highest average efficiency is retrieved within the medium- and long-term prediction window. This avoids the additional adjustment losses caused by the frequent jumps in the instantaneous optimal operating point in traditional schemes, thereby significantly improving the overall operating efficiency of the motor under variable load conditions. (2) By superimposing short-term prediction results with the base value of speed and introducing a global loss optimal fine-tuning mechanism on the basis of the global efficiency MAP, the cumulative sum of stator copper loss, rotor iron loss and mechanical loss is comprehensively considered, breaking through the traditional control limitation that only instantaneous load tracking is the goal; on this basis, combined with long-term prediction results, the motor low-power sleep and preheating control are actively implemented to avoid the extra energy consumption and mechanical stress caused by light load continuous operation and cold start. At the same time, the overshoot and lag of PI regulation are significantly reduced by the feedforward compensation of prediction value. Without changing the hardware topology, the system's comprehensive energy saving level and dynamic response capability can be improved simultaneously by software upgrade.

[0015] Of course, any product implementing this invention does not necessarily need to achieve all the advantages described above at the same time. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the steps of a motor energy-saving control method based on load prediction according to the present invention. Figure 2 This is a flowchart of the global loss optimization fine-tuning feedback method in step four of the present invention; Figure 3 This is a control flowchart of the feedback logic for controlling the motor state by executing multi-timescale linkage, which is output by the long-term prediction layer of this invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 As shown, the present invention is a motor energy-saving control method based on load prediction, the method comprising: Step 1: Construct a three-level time-scale load forecasting architecture and obtain the motor load condition parameters of each level of the load forecasting architecture; the three-level time-scale includes a long-term forecasting layer, a medium-term forecasting layer, and a short-term forecasting layer. Step 2: Obtain the overall efficiency MAP of the motor to be controlled. The efficiency MAP uses motor speed and output torque as coordinates to calibrate the total efficiency of the motor at different operating points and the detailed loss values, including at least stator copper loss, rotor iron loss, and mechanical loss. Step 3: Based on the motor load condition parameters output by the long-term and medium-term prediction layers, retrieve the target speed with the highest efficiency within the corresponding load range from the global efficiency MAP map, use it as the base value for motor speed, and generate a speed pre-adjustment command. Step 4: Based on the motor load condition parameters output in the short-term scale prediction layer, superimpose the instantaneous torque demand on the base value of motor speed, and combine it with the global efficiency MAP map to perform global loss optimization feedback. Step 5: Determine the control command for the motor driver based on the global loss optimal fine-tuning feedback, introduce the predicted value feedforward compensation in the control loop, and execute the multi-time-scale linkage feedback logic to control the motor state according to the motor load condition parameters output by the long-term scale prediction layer.

[0020] In the above technical solution, firstly, step one involves constructing a decoupled load prediction architecture with long-term, medium-term, and short-term prediction layers in the controller; then, in step two, by obtaining the motor's global efficiency map (MAP), specifically during the motor's factory testing or offline calibration phase, the operating range of the motor is traversed according to preset speed and torque steps, and the stator resistance voltage drop, stator current, rotor speed, and output mechanical torque are measured at each operating point. The stator copper loss, rotor iron loss, mechanical loss, and total efficiency at that operating point are calculated and stored in the controller in the form of a two-dimensional lookup table or a three-dimensional surface, serving as the basis for subsequent optimization; finally, in step three, efficient operating point detection is achieved based on medium- and long-term predictions. Based on the load torque range (time range including upper and lower time limits) output by the long-term and medium-term prediction layers, the system extracts all speed sets that satisfy torque constraints from the efficiency MAP. It then calculates the average efficiency of the corresponding load range for each speed set, selects the maximum speed corresponding to the average efficiency as the target speed, and if multiple speeds correspond to the same maximum average efficiency, selects the lowest speed as the base value for the motor speed. A speed pre-adjustment command is then generated and sent to the motor driver. Through decoupled prediction at three time scales (long, medium, and short), the system serves efficient operating point pre-positioning, speed base value planning, and instantaneous torque compensation, overcoming the shortcomings of single-time-scale prediction in terms of sudden changes in operating conditions and trend perception. Furthermore, in step four, based on the motor load condition parameters output from the short-term scale prediction layer, the instantaneous torque demand is superimposed on the base value of the motor speed, and global loss optimal fine-tuning feedback is performed in conjunction with the global efficiency MAP map; the load prediction results are deeply integrated with the global efficiency MAP map of the motor, and a global loss optimal fine-tuning mechanism is introduced on the basis of instantaneous load tracking to effectively reduce the cumulative sum of stator copper loss, rotor iron loss and mechanical loss; Finally, step five determines the control command for the motor driver based on the global loss-optimal fine-tuning feedback, introduces predicted value feedforward compensation in the control loop, and executes multi-time-scale linkage feedback logic to control the motor state according to the motor load condition parameters output by the long-term prediction layer. Based on the feedforward compensation based on predicted values ​​introduced in step five in the speed loop and torque loop, the overshoot and lag of PI regulation are significantly reduced, improving the system's adaptability to rapidly changing load conditions. Furthermore, based on the long-term prediction results, active execution of motor sleep and preheating control is achieved, avoiding the additional losses caused by cold starts during continuous operation under light load, and further improving the overall system regulation capability. The control commands are determined by directly converting the fine-tuned speed and torque into inverter frequency commands and stator current commands, so that the optimal fine-tuning results of global loss are truly implemented in motor drive control. In actual industrial applications, only the system software upgrade process based on the standard control link is required, without changing the hardware topology. Furthermore, by combining feedforward compensation, the torque response time can be significantly shortened and the steady-state accuracy under load conditions can be improved.

[0021] As one embodiment of the present invention, step one includes: The long-term forecasting layer takes production planning and scheduling data, historical operating data and holidays as input, and uses a time-series forecasting algorithm with periodic decomposition to output the load power in the future minutes to hours, which is used for the pre-positioning of the motor to work efficiently. The mid-scale prediction layer takes the measured load values, operating event labels and long-scale prediction results of the most recent N sampling periods as input, and outputs the load torque change curve in the next ten seconds to minutes for planning the base value of motor speed. The short-term prediction layer takes the motor's three-phase current, speed, and output torque in the current sampling period as input and outputs the instantaneous load fluctuations in the hundreds of milliseconds to seconds in the future, which are used for dynamic compensation of the motor's instantaneous torque.

[0022] In the above technical solution, the long-term scale prediction layer is deployed in the host computer or edge computing node, with production planning and scheduling data (MPS), historical working condition data (including seasonal, weekly and daily load curves) and holiday / shift labels as core input variables.

[0023] A time-series forecasting algorithm with periodic decomposition (such as Prophet, STL-LSTM, or improved ARIMA) is used to decompose the original load series into a trend term, a periodic term, and a residual term. These are then modeled separately and reconstructed to output load power forecasts for the next few minutes to hours. This output is mainly used to determine the load level range within a relatively long time window in the future, providing a basis for the macroscopic pre-positioning of the motor's efficient operating point. The mid-scale prediction layer runs within the real-time controller. It takes the measured load values ​​(e.g., current, torque) from the most recent N sampling periods, operating event labels (e.g., "feeding started," "mold change completed"), and long-scale prediction results as input. It employs either Gradient Boosting Regression Tree (GBRT) or Lightweight Long Short-Term Memory (Lite-LSTM) networks to perform rolling predictions of load torque changes over the next ten seconds to minutes, outputting a continuous time-torque curve. The prediction results of this layer are used to plan the base speed of the motor under the upcoming short-to-medium-term operating conditions, so as to avoid frequent adjustments across the high-efficiency zone. The short-term prediction layer is integrated into the control loop of the motor driver, using the motor's three-phase current in the current sampling period. , , Rotation speed and output torque As input, Kalman filtering or a first-order inertial prediction algorithm is used to predict instantaneous load fluctuations in the range of hundreds of milliseconds to seconds. Real-time estimation is performed. The prediction results of this layer are used to dynamically compensate for the instantaneous torque demand of the motor, ensuring dynamic response performance under rapid load changes. By decoupling the prediction results across the three layers in the time dimension and coordinating them in the spatial dimension, a complete set of motor load condition parameters is formed, providing a reliable data foundation for subsequent efficiency optimization and energy-saving control. The three-layer decoupling design (long, medium, and short layers) addresses the control requirements at three levels: operating condition trend perception, speed planning, and instantaneous compensation, avoiding the problem of decreased prediction accuracy of a single model across multiple time scales. Furthermore, the three-layer prediction architecture can be tailored to meet the real-time requirements of the actual system's computing power, specifically by simplifying the short-term layer algorithm or merging the medium- and long-term layers. This ensures applicability not only to high-performance servo systems but also to variable frequency energy-saving projects.

[0024] As one embodiment of the present invention, the calibration method for different operating points in the global efficiency MAP in step two includes: During the factory testing or offline calibration phase of the motor, the operating range of the motor is traversed according to the preset speed step size and torque step size. The stator resistance voltage drop, stator current, rotor speed and output mechanical torque are measured at each operating point. The stator copper loss, rotor iron loss, mechanical loss and total efficiency at that operating point are calculated and stored in the controller in the form of a two-dimensional lookup table or a three-dimensional surface.

[0025] In the above technical solution, during the motor factory testing or on-site offline calibration stage, the safe operating range of the motor is first determined, including the maximum speed. Maximum output torque And the boundary between the continuous operation zone and the short-term overload zone; according to the preset speed step size and torque step size The region is divided into several discrete working points. And iterate through them sequentially; at each work point To ensure stable motor operation, the following parameters are collected: Stator resistance voltage drop (Calculated or directly measured by the difference between stator terminal voltage and back electromotive force), effective value of stator phase current actual rotor speed Output mechanical torque Estimated indirectly through torque sensors or based on the speed-current relationship; Based on the above measurements, calculate the various losses at this operating point: Stator copper loss ,in Stator phase resistance, rotor iron loss: It is obtained from no-load tests or separation methods, and is related to rotational speed and magnetic flux. Including wind friction loss and bearing loss, which were fitted through no-load operation experiments; Input power Output power , Overall efficiency ; Put all work points corresponding , , , The data is stored in the controller in the form of a two-dimensional lookup table. For systems with sufficient computing power, a continuous efficiency MAP model can also be constructed by using three-dimensional surface fitting (such as polynomial surface or neural network interpolation) to improve the accuracy of the lookup table and the efficiency of the interpolation. In step two above, by discretely traversing all typical operating points within the motor's operating range, the efficiency and loss characteristics of the motor under different speed and torque combinations are fully characterized, avoiding misjudgments of the high-efficiency zone by traditional methods. By providing independent data for stator copper loss, rotor iron loss, mechanical loss, and total efficiency, subsequent control strategies can be specifically optimized under different loss-dominant operating conditions (e.g., prioritizing iron loss reduction under light load and copper loss reduction under heavy load). Furthermore, the use of two-dimensional lookup tables or three-dimensional surface models allows for rapid retrieval of efficiency and loss information near any operating point within the control cycle, meeting the needs of real-time energy-saving control. This calibration method has a wide range of applications, suitable for asynchronous motors, permanent magnet synchronous motors, synchronous reluctance motors, etc., and the calibration results differ from existing single control architectures, and can also be directly used for various control architectures such as vector control and direct torque control. The accurate global efficiency MAP is the basis for the global loss optimal fine-tuning feedback in step four, ensuring accurate calculation of cumulative losses within the prediction window and achieving true energy saving maximization.

[0026] As one embodiment of the present invention, the retrieval method in the global efficiency MAP map in step three includes: Based on the load torque range output by the long-term and medium-term prediction layers Extract the set of all speeds that satisfy the torque constraint from the efficiency MAP. Calculate the average efficiency for the corresponding load range under this set of speeds. Select The maximum speed is used as the target speed. If multiple speeds correspond to the same maximum average efficiency, then the point with the lowest speed is selected as the base value of the motor speed.

[0027] In the above technical solution, firstly, the controller receives the output results from the long-term and medium-term prediction layers, and analyzes them to obtain the range of load torque variation within the future prediction window. ;in, To predict the minimum load torque within the window, This is the maximum load torque; Then, in the torque range As a constraint, a search is performed in the stored global efficiency map, traversing all rotational speeds. Extracting the continuous output at this speed to The set of operating points for torque constitutes the set of speeds that satisfy torque constraints:

[0028] Next, regarding Each rotational speed Calculate its average efficiency within the corresponding load range during the prediction window:

[0029] in, To predict the sampling points of the load torque curve at the corresponding time, The number of sampling points. This represents the efficiency value of the corresponding operating point in the efficiency MAP.

[0030] Furthermore, comparison Average efficiency across all speeds The speed with the highest average efficiency is selected as the target speed. :

[0031] If multiple speeds correspond to the same maximum average efficiency, then the point with the lowest speed is selected as the base value of the motor speed to reduce iron loss and mechanical loss, and reduce dynamic loss in the subsequent speed regulation process. Finally, the selected base speed value The speed pre-adjustment command is converted into a motor driver command and issued before the prediction window opens, so that the motor can smoothly transition to the speed before the actual load arrives.

[0032] The above process effectively avoids frequent switching of the motor's operating point due to instantaneous optimality by calculating the average efficiency over the load range instead of the non-instantaneous efficiency, thus reducing regulation losses. Furthermore, by extracting all speeds that meet torque constraints from the efficiency map and prioritizing lower speeds under the same efficiency conditions, it helps reduce rotor iron losses, windage losses, and bearing losses, aligning with the actual operating characteristics of the motor. The search settings based on the torque range ensure the selected base speed value. It can also perfectly cover the load fluctuations of the prediction window content, reducing the pressure on subsequent instantaneous compensation. The average efficiency is used in the calculation. As an evaluation metric, it is simple to calculate, has good real-time performance, and is suitable for online execution in industrial controllers.

[0033] Please see Figure 2 As shown, in one embodiment of the present invention, the global loss optimal fine-tuning feedback method in step four is as follows: The instantaneous torque fluctuation output by the short-term prediction layer The base torque superimposed on the base speed value To obtain the initial target torque ; Will Input the efficiency MAP plot and calculate the predicted total loss at that operating point. ; like Greater than the preset loss threshold The optimization objective is to minimize the cumulative loss within the prediction window. or Fine-tuning was performed to obtain the adjusted rotational speed. and torque This is output as the feedback result for optimal fine-tuning of global loss.

[0034] In the above technical solution, the controller receives the instantaneous torque fluctuation output from the short-term prediction layer. And add it to the base speed value determined in step three. Corresponding basic torque The initial target torque is obtained as follows:

[0035] in, To ensure the minimum continuous torque required for the load to operate normally within the prediction window, the initial target torque is... With target speed Input the global efficiency map, and retrieve the operating point by table lookup or interpolation. The breakdown of losses and total losses are as follows:

[0036] in, For stator copper loss, For rotor iron loss, For mechanical wear and tear; Predict total loss Compared with the preset loss threshold If a comparison is made, ≤ If no fine-tuning is performed, proceed directly with... As the final control trajectory output; if ≤ Then proceed to the fine-tuning and optimization process; With prediction window The optimization problem is established with the goal of minimizing the cumulative loss within the range. Then, using traversal or gradient search methods, the fine-tuned rotational speed that minimizes the cumulative loss is found within the allowable adjustment range. and torque The result is then used as the global loss optimal fine-tuning feedback output.

[0037] By minimizing the cumulative losses within the prediction window, frequent operating point jumps caused by pursuing instantaneous optimal efficiency are avoided, significantly reducing the additional losses in the adjustment process itself. Furthermore, by utilizing independent data on stator copper losses, rotor iron losses, and mechanical losses in the efficiency MAP, the aim is to dynamically balance various losses under different load levels to achieve global optimization rather than local optimization.

[0038] As one embodiment of the present invention, the objective function for global loss optimal fine-tuning feedback is expressed as:

[0039] in, To optimize the objective function, For stator copper loss, For rotor iron loss, For mechanical wear, As a candidate work site, To predict the time step within the window; The constraints are: , The maximum allowable torque tracking deviation; , This represents the maximum permissible speed adjustment range.

[0040] In the above technical solution, a prediction window is selected. Internal motor speed With output torque As optimization variables, they constitute candidate operating points. The objective function is constructed by minimizing the cumulative total loss within the prediction window. The objective function is as follows:

[0041] in, To optimize the objective function value, For the first Stator copper loss corresponding to each time step; For the first Rotor iron loss corresponding to each time step; for Mechanical loss corresponding to each time step; To predict the time step within the window; This is the total number of steps within the prediction window.

[0042] To ensure the safety and feasibility of the control process, constraints are set, specifically including torque tracking deviation constraints and speed adjustment range constraints; the torque tracking deviation constraint is as follows:

[0043] in, For the initial target torque, This is the actual output torque after fine-tuning. The maximum allowable torque tracking deviation is used to prevent insufficient load supply due to energy-saving optimization. The speed adjustment range constraint is:

[0044] in, The target rotational speed obtained in step three. This is the adjusted rotation speed. The maximum allowable speed adjustment range is used to limit dynamic losses caused by operating point jumps. In addition to torque tracking deviation constraints and speed adjustment range constraints, actual operation also includes safety operation constraints:

[0045] This safety operation constraint is generally to ensure that the optimization results always remain within the allowable operating range of the motor.

[0046] The optimization solution involves using traversal search, gradient descent, or heuristic algorithms (such as particle swarm optimization) to solve the above optimization problem within the constraints, obtaining the fine-tuned optimal rotational speed. and optimal torque This result is then output as the global loss optimization feedback to step five. This optimization framework improves the applicability of various motors and allows for flexible adjustment of constraint thresholds based on actual system requirements. and Furthermore, by outputting the fine-tuned operating point through digital optimization results, energy-saving control is transformed from experience-based adjustment to a model-based quantitative data-driven decision-making process, which improves the system's adjustability and enhances energy-saving effects.

[0047] In one embodiment of the present invention, the motor driver control command determined in step five includes: For asynchronous motors, the fine-tuned target speed Convert to inverter output frequency command ,in This represents the number of pole pairs of the motor. The fine-tuned target torque Convert to stator current active component command ; The above instructions are used to generate the inverter's switching signals using a space vector pulse width modulation (SVPWM) algorithm.

[0048] In the above technical solution, the specific implementation process of the motor driver control command is determined by the feedback result of the global loss optimal fine-tuning in step five. Specifically, firstly, the target speed is converted into the output frequency, and then the target speed is determined after the controller receives the fine-tuning output from step four. For the number of pole pairs of asynchronous motors Convert it into the inverter's output frequency command The calculation formula is: ,in The unit is rpm. The frequency is set to Hz, and this frequency command is input to the inverter control module as the setpoint for the outer speed loop; then, the conversion from target torque to stator current active component is performed by the controller through fine-tuning the target torque. Converted into stator current active component command in vector control system Within the rotor field-oriented control framework, the specific ,in For electromagnetic torque, This is a torque constant, proportional to the rotor flux linkage amplitude; the corresponding value is calculated by ensuring the current rotor flux linkage amplitude is sufficient through online observation or table lookup. ; Finally, the active component of the stator current output by the speed loop is commanded. Command for reactive component of stator current The input current inner-loop PI controller obtains the voltage reference value in a refined coordinate system. , The space vector pulse width modulation (SVPWM) algorithm is used to combine voltage reference values. The voltage vector space of the three-phase inverter is mapped to calculate the conduction time of each power switching device, and finally six PWM drive signals are generated to control the IGBT turn-on and turn-off of the inverter.

[0049] As one embodiment of the present invention, the calculation method for the predicted value feedforward compensation introduced in the control loop is as follows: In the speed loop controller, the speed change rate output by the mid-scale prediction layer is used. As a feedforward term superimposed on the torque command input, the feedforward compensation amount ,in This is the feedforward gain coefficient; In the torque loop controller, the torque change rate output by the short-term prediction layer is used. This is superimposed on the torque command input as a feedforward term to form a composite control input.

[0050] The above technical solution further explains the feedforward compensation in step five: Since the constraints include maximum torque tracking deviation and maximum speed adjustment range limits; the fine-tuned control trajectory is used to determine the motor driver control commands; and for asynchronous motors, by... Converted into frequency output command of the frequency converter, to achieve... Converted to active component of stator current The instruction; finally, the process of generating the inverter switching signal through the SVPWM algorithm requires the simultaneous introduction of predicted value feedforward compensation in the control loop, that is: The speed change rate dn / dt output from the intermediate prediction layer is added as a feedforward term to the speed loop; The torque change rate dT / dt of the short-term prediction layer output is added to the torque loop as a feedforward term to form a composite control input; Speed ​​loop feedforward compensation is implemented by introducing the speed change rate dn / dt from the mid-scale prediction layer as a feedforward signal into the outer speed loop controller, in addition to the conventional PI regulation output. The specific calculation method is as follows:

[0051] in, The feedforward gain coefficient of the speed loop is obtained through offline tuning or online adaptive adjustment. This feedforward compensation is directly superimposed on the output of the speed loop PI controller as the feedforward increment of the torque command, thus offsetting the influence of load change trends on the speed in advance. Torque loop feedforward compensation is implemented by using the torque change rate dT / dt output from the short-term prediction layer as the feedforward signal in the inner torque loop controller.

[0052] in, This is the torque loop feedforward gain coefficient. This feedforward compensation is directly superimposed on the torque command input terminal, and together with the torque command output from the speed loop and the feedback from the current loop, it constitutes the load control input. In the composite control input synthesis, the torque command is ultimately input to the current converter. Represented as:

[0053] in, This is the output of the speed loop PI regulator; through the above load control structure, the system can respond in advance when the load changes, significantly reducing overshoot and lag in the dynamic adjustment process.

[0054] The role of feedforward compensation is to enable the system to act in advance before the load change trend is fully reflected in the feedback signal by introducing feedforward compensation based on the predicted value, which greatly shortens the adjustment time. In addition, feedforward compensation can effectively offset the speed drop or overshoot caused by sudden load changes, so that the system can recover to steady state more quickly and reduce energy waste during the adjustment process. Furthermore, by using the speed change rate of the medium-term prediction layer for the speed loop and the speed change rate of the short-term prediction layer for the torque loop, the prediction information and the control loop are accurately matched.

[0055] Please see Figure 3As shown, in one embodiment of the present invention, the specific method for controlling the motor state by executing multi-time-scale linkage feedback logic based on the motor load condition parameters output by the long-term scale prediction layer is as follows: When the long-term scale prediction layer continuously outputs future... When the load is consistently below 20% of the rated load for an extended period of time, the motor is controlled to enter a low-power sleep state, maintaining only the minimum excitation current or cutting off the power module output. When the long-term prediction layer detects that the load will be When the load recovers to more than 30% of the rated load after a certain period of time, it should be done in advance. The timer wakes up the motor and executes the preheating control logic, enabling the motor to reach a stable operating temperature before entering the load ramp-up phase.

[0056] In the above technical solution, the specific implementation method of step five, which involves executing multi-timescale linkage feedback logic to control the motor state based on the motor load condition parameters output by the long-term prediction layer, is as follows: First, the low-power sleep state is determined and executed. Specifically, the controller monitors the future load prediction curve output by the long-term prediction layer in real time. When the predicted curve is detected in the future... Load power over continuous time When all loads are below 20% of the motor's rated load, the system is determined to have entered a continuous light load range, triggering low-power sleep control logic, including: for asynchronous motor frequency converter systems: controlling the inverter to stop PWM output, cutting off the power module drive signal, and only maintaining weak excitation of the DC bus or complete power de-energization; for permanent magnet synchronous servo systems: entering the "holding braking + zero speed lock" mode, only outputting the minimum stator current required to maintain the position; through the above methods, iron loss, wind friction loss and inverter switching loss under light load operation are eliminated; Secondly, the loop and preheating control logic is executed. When the long-term prediction layer detects that the predicted curve will rise to more than 30% of the rated load after a certain time, it is determined that the load is about to enter the ramp-up phase, triggering the early wake-up and preheating logic: before the actual load rises, the power module is re-enabled and the PWM output is restored; during the preheating control period: the motor is controlled to run at low speed or zero speed, and the stator current is gradually increased to 10%-30% of the rated current, so that the winding temperature rises steadily to the stable operating temperature range; after the preheating is completed, the system enters the normal speed regulation and torque output state, ensuring that it has complete dynamic response capability in the early stage of load ramp-up; Finally, in the multi-scale linkage coordination mechanism, the generation and execution of sleep and wake-up commands are linked with the speed base value planning in step three and the global loss fine-tuning in step four: during sleep, the real-time operation of the medium-term and short-term prediction layers is paused, and only the long-term prediction layer is kept running to reduce the power consumption of the controller; after wake-up, the historical data window of the medium-term and short-term prediction models is refreshed immediately to ensure that the prediction results converge quickly to the actual operating conditions.

[0057] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, and therefore described more simply; relevant parts can be referred to the descriptions of the method embodiments.

[0058] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined in this application, they should all fall within the protection scope of the present invention.

Claims

1. A motor energy-saving control method based on load prediction, characterized in that, The method includes: Step 1: Construct a three-level time-scale load forecasting architecture and obtain the motor load condition parameters of each level of the load forecasting architecture; the three-level time-scale includes a long-term forecasting layer, a medium-term forecasting layer, and a short-term forecasting layer. Step 2: Obtain the global efficiency MAP of the motor to be controlled. The efficiency MAP uses the motor speed and output torque as coordinates to calibrate the total efficiency of the motor at different operating points and the detailed loss values, including at least stator copper loss, rotor iron loss and mechanical loss. Step 3: Based on the motor load condition parameters output by the long-term and medium-term prediction layers, retrieve the target speed with the highest efficiency within the corresponding load range from the global efficiency MAP map, use it as the base value for motor speed, and generate a speed pre-adjustment command. Step 4: Based on the motor load condition parameters output in the short-term scale prediction layer, superimpose the instantaneous torque demand on the base value of motor speed, and combine it with the global efficiency MAP map to perform global loss optimization feedback. Step 5: Determine the control command for the motor driver based on the global loss optimal fine-tuning feedback, introduce the predicted value feedforward compensation in the control loop, and execute the multi-time-scale linkage feedback logic to control the motor state according to the motor load condition parameters output by the long-term scale prediction layer.

2. The motor energy-saving control method based on load prediction according to claim 1, characterized in that, Step one includes: The long-term forecasting layer takes production planning and scheduling data, historical operating condition data and holidays as input, and uses a time-series forecasting algorithm with periodic decomposition to output the load power in the future minutes to hours, which is used for pre-positioning of the motor to work efficiently. The mid-scale prediction layer takes the measured load values, operating condition event labels, and long-scale prediction results of the most recent N sampling periods as input, and outputs the load torque change curve in the next ten seconds to minutes for planning the base value of motor speed. The short-term prediction layer takes the motor's three-phase current, speed, and output torque in the current sampling period as input and outputs the instantaneous load fluctuations in the hundreds of milliseconds to seconds for dynamic compensation of the motor's instantaneous torque.

3. The motor energy-saving control method based on load prediction according to claim 1, characterized in that, The calibration methods for different operating points in the global efficiency MAP in step two include: During the factory testing or offline calibration phase of the motor, the operating range of the motor is traversed according to the preset speed step size and torque step size. The stator resistance voltage drop, stator current, rotor speed and output mechanical torque are measured at each operating point. The stator copper loss, rotor iron loss, mechanical loss and total efficiency at that operating point are calculated and stored in the controller in the form of a two-dimensional lookup table or a three-dimensional surface.

4. The motor energy-saving control method based on load prediction according to claim 1, characterized in that, The retrieval methods in the global efficiency MAP map in step three include: Based on the load torque range output by the long-term and medium-term prediction layers Extract the set of all speeds that satisfy the torque constraint from the efficiency MAP. Calculate the average efficiency for the corresponding load range under this set of speeds. Select The maximum speed is used as the target speed. If multiple speeds correspond to the same maximum average efficiency, then the point with the lowest speed is selected as the base value of the motor speed.

5. The motor energy-saving control method based on load prediction according to claim 1, characterized in that, The optimal global loss fine-tuning feedback method in step four is as follows: The instantaneous torque fluctuation output by the short-term prediction layer The base torque superimposed on the base speed value To obtain the initial target torque ; Will Input the efficiency MAP plot and calculate the predicted total loss at that operating point. ; like Greater than the preset loss threshold The optimization objective is to minimize the cumulative loss within the prediction window. or Fine-tuning was performed to obtain the adjusted rotational speed. and torque This is output as the feedback result for optimal fine-tuning of global loss.

6. The motor energy-saving control method based on load prediction according to claim 5, characterized in that, The objective function for optimizing the global loss fine-tuning feedback is expressed as: in, To optimize the objective function, For stator copper loss, For rotor iron loss, For mechanical wear, As a candidate work site, To predict the time step within the window; The constraints are: , The maximum allowable torque tracking deviation; , This represents the maximum permissible speed adjustment range.

7. The motor energy-saving control method based on load prediction according to claim 1, characterized in that, The motor driver control commands determined in step five include: For asynchronous motors, the fine-tuned target speed Convert to inverter output frequency command ,in This represents the number of pole pairs of the motor. The fine-tuned target torque Converted to active component of stator current instruction; The above instructions are used to generate the inverter's switching signals using a space vector pulse width modulation (SVPWM) algorithm.

8. The motor energy-saving control method based on load prediction according to claim 1, characterized in that, The calculation method for the predicted value feedforward compensation introduced in the control loop is as follows: In the speed loop controller, the speed change rate output by the mid-scale prediction layer is used. As a feedforward term superimposed on the torque command input, the feedforward compensation amount ,in This is the feedforward gain coefficient; In the torque loop controller, the torque change rate output by the short-term prediction layer is used. This is superimposed on the torque command input as a feedforward term to form a composite control input.

9. The motor energy-saving control method based on load prediction according to claim 1, characterized in that, The specific method for controlling the motor state by executing multi-time-scale linkage feedback logic based on the motor load condition parameters output by the long-term scale prediction layer is as follows: When the long-term scale prediction layer continuously outputs future... When the load is consistently below 20% of the rated load for an extended period of time, the motor is controlled to enter a low-power sleep state, maintaining only the minimum excitation current or cutting off the power module output. When the long-term prediction layer detects that the load will be When the load recovers to more than 30% of the rated load after a certain period of time, it should be done in advance. The timer wakes up the motor and executes the preheating control logic, enabling the motor to reach a stable operating temperature before entering the load ramp-up phase.

Images