Asynchronous motor control method based on scraper load prediction

Abstract

The invention discloses an asynchronous motor control method based on scraper load prediction, and the method comprises the steps: detecting the characteristics of a multi-source load based on a vibration sensor, current harmonic analysis and mechanical transmission parameters, predicting the probability distribution and confidence interval of a load torque outputted by a network through combining with an LSTM time sequence, and providing a prospective input for a decision. And based on a prediction result, multi-motor energy distribution is realized by utilizing a dynamic weighting function and Nash game optimization, time-varying parameter deviations of rotor resistance and mutual inductance are respectively compensated based on a BP neural network and extended Kalman filtering, and meanwhile, instantaneous overload protection is realized by adopting a two-stage load reduction strategy driven by reinforcement learning. According to the invention, a DTC-SVM algorithm is integrated, and a control instruction is accurately executed based on predicted torque compensation, variable switching frequency adjustment and dual-target optimization in combination with a real-time error detection and hardware protection module. According to the scheme, the dynamic response speed, the energy efficiency and the operation reliability of the asynchronous motor under the periodic impact load are remarkably improved.

Description

Technical Field

[0001] This invention relates to an asynchronous motor control method based on scraper load prediction, and belongs to the technical field of asynchronous motor control methods based on scraper load prediction. Background Technology

[0002] Asynchronous motors, with their simple structure and low maintenance costs, occupy an important position in industrial drives, especially in applications facing complex load conditions such as scraper conveyors and mining mixing equipment. However, these devices generally exhibit non-stationary load characteristics during operation: random fluctuations in the base load caused by uneven material accumulation, and periodic instantaneous impacts generated by the meshing of the scraper chain and guide rails, pose a severe challenge to the dynamic response capability of motor control strategies. Furthermore, the rotor resistance of squirrel-cage asynchronous motors exhibits slow time-varying characteristics due to temperature rise, while the mutual inductance parameters experience rapid time-varying fluctuations due to magnetic saturation effects, further exacerbating the parameter mismatch between the control model and the actual operating conditions.

[0003] Traditional control methods suffer from the following significant drawbacks when addressing the aforementioned challenges: First, insufficient dynamic response capability: Vector control based on accurate motor models is sensitive to parameter changes, and parameter drift easily leads to torque observation errors; while Direct Torque Control (DTC) possesses fast response characteristics, its fixed hysteresis width causes switching frequency fluctuations, affecting torque control accuracy. Second, low efficiency in multi-motor coordination: Traditional master-slave control employs a fixed weight allocation strategy, failing to perceive real-time differences in motor load and temperature rise status. Under sudden shocks, torque distribution lag easily occurs, leading to mechanical stress concentration and decreased system energy efficiency. Third, lack of parameter adaptive capability: Existing technologies lack multi-timescale identification and compensation mechanisms for the time-varying characteristics of motor parameters. The temperature drift of slow time-varying parameters and the magnetic saturation effect of fast time-varying parameters are not effectively decoupled. Slow time-varying parameters, such as rotor resistance, and fast time-varying parameters, such as mutual inductance, accumulate errors, leading to deterioration of control performance.

[0004] To address these issues, existing research has largely focused on optimizing single aspects, such as improving prediction algorithms or parameter identification strategies, rather than establishing a collaborative architecture that integrates load prediction, strategy decision-making, and execution control. Summary of the Invention

[0005] To address the issues of dynamic response lag, rigid multi-motor collaborative control, and low energy efficiency caused by parameter mismatch in existing asynchronous motor drive systems under periodic impact loads, this invention provides an asynchronous motor control method based on scraper load prediction. Based on the integration of a three-level collaborative control architecture and intelligent algorithms, it achieves proactive response to load impacts, dynamic allocation of torque weights, and adaptive compensation of motor parameters, thereby improving operating performance and reliability under complex working conditions.

[0006] Preferably, the asynchronous motor control method based on scraper load prediction of the present invention includes:

[0007] S1, the vibration signal of the scraper load is collected in real time by a vibration sensor, the current harmonic data of the asynchronous motor stator is collected in real time by a current sensor, and the mechanical transmission parameters are collected in real time by an encoder. The impact feature E is obtained by wavelet packet decomposition of the vibration signal. impact The total harmonic distortion (THD) is calculated based on current harmonic data and mapped to the load impact strength F. impact The basic load torque τ is calculated based on mechanical transmission parameters, including the torsional vibration angle of the drive shaft. base And all data, including impact characteristics, load impact intensity and mechanical transmission parameters, are standardized.

[0008] S2, based on the environmental parameter embedding vector, inputs all data into the LSTM time series prediction network to predict the load prediction results, including the load torque, the probability distribution range of the load torque within a preset time period, and the confidence interval.

[0009] S3, based on the load prediction results, uses a dynamic weighting function and a Nash game model to generate energy allocation weights for multiple asynchronous motors, and issues control commands containing energy allocation weights;

[0010] S4, the actual rotor resistance value of the asynchronous motor is updated using a BP neural network, and the mutual inductance L is updated in real time using an extended Kalman filter. m .

[0011] Prioritize, S5, when the load torque predicted by S2 exceeds the rated threshold, triggers a two-stage load reduction strategy of flywheel energy storage and hydraulic braking, generating independent protection signals including switching frequency adjustment, flywheel energy storage start-up, and hydraulic braking trigger;

[0012] S6, based on the DTC-SVM algorithm, the control command of step S3, the actual rotor resistance value, the time-varying deviation of the mutual inductance, and the independent protection signal of step S5, adjusts the switching frequency and optimizes the actual output torque of the asynchronous motor.

[0013] Preferred, step S1 includes:

[0014] A triaxial MEMS accelerometer positioned at the scraper chain link acquires the vibration signal of the scraper load. Wavelet packet decomposition is used to extract the impact characteristics E within a preset frequency band. impact ;

[0015] Real-time analysis of the stator current i of an asynchronous motor using sliding window FFT sThe harmonic distortion rate (THD) is used to calculate and output the load impact intensity (F) using a pre-established empirical formula and a mapping model between the harmonic distortion rate and the load impact intensity. impact ;

[0016] The torsional vibration angle θ of the transmission shaft is calculated based on the mechanical transmission parameters. Using a pre-constructed torque-torsion angle transfer function, the base load torque τ is output. base .

[0017] Prior to step S2, based on the environmental parameter embedding vector, all data is input into the LSTM time series prediction network to predict the load prediction results, including load torque, the probability distribution interval of load torque within a preset future time period, and the confidence interval.

[0018] The LSTM time-series prediction network comprises a bidirectional long short-term memory network layer, an environmental parameter embedding layer, an attention mechanism module, and a fully connected output layer connected in sequence.

[0019] The bidirectional long short-term memory network layer is used to receive the standardized temporal feature vector X. seq ∈R 20×5 The time-series feature vector contains the impact features E of 20 historical time series. impact Load impact strength F impact and base load torque τ base The temporal feature encoding h with a dimension of 20×128 is utilized by the bidirectional long short-term memory network layer output. seq ;

[0020] The environmental parameter embedding layer maps the ambient temperature T and humidity RH into a 16-dimensional environmental feature vector h. env The activation function in the environmental parameter embedding layer is ReLU;

[0021] The attention mechanism module is used to fuse temporal feature encoding based on a query-key-value architecture. seq With environmental feature vector h env Obtain the fusion feature h fusion ;

[0022] Using a fully connected output layer to fuse features h fusion After global average pooling, the predicted mean μ and standard deviation σ of the output load torque are obtained.

[0023] Calculate the confidence interval [τ] of the network output load torque. min ,τ max ]:

[0024] τ min =μ-k·σ,τ max =μ+κ·σ,

[0025] In the formula, τmin To predict the lower limit of load torque, τ max κ is the standard deviation adjustment parameter used to predict the upper limit of the load torque.

[0026] Prior to step S3, based on the load prediction results, energy allocation weights for multiple asynchronous motors are generated using a dynamic weighting function and a Nash game model, and control commands containing these energy allocation weights are issued, including:

[0027] Constructing a dynamic weight function:

[0028]

[0029] In the formula, μ i Let σ be the mean of the i-th prediction. i Let ω be the standard deviation of the i-th prediction. i It is the weight vector of the i-th asynchronous motor, k i =γT wind Where γ is the temperature influence factor, T ωind For winding temperature;

[0030] Based on the Nash game optimization model, an objective function including an efficiency factor and a thermal constraint is established:

[0031]

[0032] In the formula, P loss For the system power loss, λ eff , λ thermal T represents the weighting coefficient. limit These are the safe limits for winding temperature;

[0033] The energy allocation weights of multiple asynchronous motors are generated using a dynamic weight function and a Nash game model, and control commands containing the energy allocation weights are issued.

[0034] Prior to step S4, the actual rotor resistance value is updated using a BP neural network, and the mutual inductance L is updated in real time using an extended Kalman filter. m ,include:

[0035] A backpropagation neural network is used to iteratively update the actual rotor resistance value R at a set period. r (k+1) :

[0036]

[0037] R r (k+1) =R r (k) +ΔR r ,

[0038] In the formula, η is the learning rate, and T e e represents the actual output torque of the asynchronous motor. T For torque tracking error, e T =T ref -T e T ref The target torque reference value (generated by the decision-making level based on load prediction results), ΔR r This is the update increment for the rotor resistance;

[0039] An extended Kalman filter is used to update the mutual inductance L in real time at a set period. m :

[0040]

[0041] In the formula, L m0 τ is the unloaded mutual inductance value. N denoted as the rated load torque of the asynchronous motor, and ∈ is the magnetic saturation compensation coefficient.

[0042] Prioritized, in step S5, when the load torque predicted in S2 exceeds the rated threshold, a two-stage load reduction strategy involving flywheel energy storage and hydraulic braking is triggered, generating independent protection signals including switching frequency adjustment, flywheel energy storage activation, and hydraulic braking triggering, including:

[0043] The defined state space includes load prediction overload, mechanical shock exceeding limits, winding temperature exceeding limits, winding cumulative loss exceeding preset standard thresholds, and action space;

[0044] The predicted overload includes the upper limit of the predicted load torque. The probability is greater than or equal to the pre-set confidence level 1-α;

[0045] Excessive mechanical impact includes impact characteristic E impact >E th E th This refers to the impact energy threshold calibrated based on vibration testing.

[0046] Excessive winding temperature includes winding temperature T wind ≥Preset winding temperature threshold T limit ;

[0047] The operating space includes the frequency conversion load reduction level and the degree of mechanical braking intervention. The frequency conversion load reduction level includes the switching frequency adjustment range, and the degree of mechanical braking intervention includes flywheel energy storage start-up and hydraulic braking trigger.

[0048] The two-stage load reduction strategy includes: if the predicted upper limit of the load torque τ max Exceeding the rated load torque τ of the asynchronous motor NIf the flywheel energy storage device is triggered to absorb the impact energy of the scraper load, and if the scraper load overload is not relieved within a preset time, the hydraulic brake will be activated and the power supply will be cut off.

[0049] Prioritizing step S6, based on the DTC-SVM algorithm, the control command from step S3, the actual rotor resistance value, the time-varying deviation of the mutual inductance, and the independent protection signal from step S5, the switching frequency is adjusted and the actual output torque of the asynchronous motor is optimized, including:

[0050] Introducing compensation coefficient K p and integral compensation coefficient K i Construct the compensation formula:

[0051]

[0052] In the formula, Δτ comp This is the torque compensation amount. Let τ be the predicted mean at time k+1. k This represents the actual output torque of the asynchronous motor at time k. Let τ be the predicted mean at time ki. k-i The actual output torque of the asynchronous motor at time ki;

[0053] If the asynchronous motor speed is ≥ ω th If the asynchronous motor speed is ≤ω, then a preset high switching frequency is used. th Then, a preset low switching frequency ω is used for the asynchronous motor. th The critical speed;

[0054] Construct the objective function:

[0055] min(α·|T ref -T e |+β·P switch ),

[0056] Among them, T ref T is the target torque of the asynchronous motor. e P represents the actual output torque of the asynchronous motor. switch Let α be the switching loss power, β be the torque accuracy weight, and α + β = 1.

[0057] Preferably, the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method described in any of the first aspects.

[0058] Preferably, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any one of the first aspects.

[0059] The beneficial effects achieved by this invention are as follows:

[0060] Thanks to the predictive torque compensation mechanism and variable switching frequency adjustment in step S6, the dynamic response time of this invention is shortened to less than 20ms, compared to 30ms for the traditional DTC method, effectively solving the problem of dynamic response lag in the prior art; the torque balance of multi-asynchronous motors in this invention is improved by 40%, significantly improving the cooperative rigidity phenomenon caused by the traditional fixed weight allocation.

[0061] This invention utilizes a BP neural network to compensate for deviations in time-varying parameters, identifies rotor resistance errors of <5%, improves EKF tracking mutual inductance accuracy by 60%, and provides adaptive parameter compensation for step S4.

[0062] The overload protection response time of this invention is 10ms, and the temperature rise of the asynchronous motor is reduced by 15%, making it suitable for long-term stable operation in heavy-duty scenarios such as scraper conveyors.

[0063] This implementation method, based on a three-tier architecture with deep collaboration, achieves intelligent operation throughout the entire process from load prediction to execution control, significantly improving the overall performance of asynchronous motors under periodic impact loads. Attached Figure Description

[0064] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1 It illustrates a three-tiered collaborative architecture diagram comprising the prediction layer, decision-making layer, and execution layer.

[0066] Figure 2 The flowchart illustrates how the prediction layer uses multi-source detection and LSTM time-series prediction network to predict the probability distribution and confidence interval of load torque and determine whether protection is triggered.

[0067] Figure 3 The flowchart illustrates how the decision-making level dynamically allocates weights, optimizes energy distribution, compensates parameters, and adopts a two-stage flywheel / hydraulic load reduction strategy based on prediction results.

[0068] Figure 4 The flowchart illustrates the process of controlling an asynchronous motor based on an improved DTC-SVM algorithm at the execution layer, providing real-time error feedback, and implementing hardware protection by combining flywheel energy storage and hydraulic braking. Detailed Implementation

[0069] See Figure 1 This invention provides an asynchronous motor control method based on scraper load prediction, such as... Figure 1As shown, the three-level collaborative control architecture includes: Figure 2 The prediction layer shown, such as Figure 3 The decision-making level shown, such as Figure 4 The execution layer shown enables intelligent control under complex operating conditions. The specific implementation steps are as follows:

[0070] Step S1: Data Acquisition and Preprocessing.

[0071] Vibration signal processing: A triaxial MEMS accelerometer is installed at the scraper chain link. The triaxial accelerometer can be any type of MEMS vibration sensor, such as the Kistler 8766A, with a range of ±500g. Vibration signals are acquired in real time at a sampling frequency of 10kHz. Wavelet packet decomposition is used to decompose the vibration signal into the first three orders of Intrinsic Mode Functions (IMFs), obtaining the IMF components. The energy value of each IMF component is calculated, and the impact characteristic E is extracted. impact .

[0072] Current harmonic analysis: Stator current i is acquired based on Hall sensors (such as LEM HXS 600-NP, bandwidth 200kHz). s The total harmonic distortion (THD) was calculated using a sliding window fast Fourier transform with a window length of 1 s and an overlap rate of 50%.

[0073] Establish a mapping model between total harmonic distortion (THD) and load impact intensity: F impact = a·THD+b;

[0074] Where a and b are coefficients based on experimental calibration, and the output load impact strength F impact .

[0075] Transmission parameter acquisition: Based on real-time acquisition of the drive shaft speed using an encoder, the torsional vibration angle θ is calculated, and a torque-torsion angle transfer function is constructed.

[0076] Among them, K t Where D is the torsional stiffness, D is the damping coefficient, and τ is the output base load torque. base .

[0077] Data standardization: for E impact F impact τ base Environmental parameters are standardized using a sliding window with a window length of 20 sampling periods.

[0078] Step S2, Environmental Parameter Embedding Layer in Load Prediction Layer:

[0079] Input and output configuration: Environmental parameters are converted into a 16-dimensional feature vector h through an embedding layer.env The embedding layer adopts a fully connected structure with an input dimension of 2, corresponding to ambient temperature T and ambient humidity RH, and an output dimension of 16. The activation function is ReLU, and the weights are initialized using the Xavier method.

[0080] Attention mechanism: Temporal features h seq With environmental characteristics h env Temporal features h are dynamically fused through an attention module. seq Dimension 128, environmental features h env Dimension 16. The calculation process is as follows:

[0081] Feature projection:

[0082] Query vector Q = h seq ·W Q (W Q ∈R 128×64 The weight matrix can be trained.

[0083] Bond vector K = h env ·W K (W K ∈R 16×64 The weight matrix can be trained.

[0084] Value vector V = h seq ·W V (W V ∈R 128×64 ), which can train the weight matrix.

[0085] Calculate the attention weight matrix α:

[0086]

[0087] Where, d k =64 is the scaling factor, α is the attention weight matrix, the dimension of the attention weight matrix is ​​20×20, corresponding to 20 historical time steps.

[0088] Calculate the fusion feature h fusion :

[0089] h fusion =α·(h seq ·W V ),

[0090] Among them, W V ∈R 128×64 The final output is the fused feature vector, with a dimension of 20×64, which is the value matrix.

[0091] Data integration and network configuration:

[0092] Multi-source input: Impact features E extracted from vibration signals impact Load strength F of current harmonic mapping impact The fundamental torque τ for calculating mechanical transmission parameters base Ambient temperature T and ambient humidity RH are aligned according to time series to form the input vector X. seq ∈R 20×5 The input vector consists of 20 historical time steps and 5 feature dimensions.

[0093] The hierarchical connection order of the network architecture:

[0094] Bidirectional LSTM layer: A bidirectional LSTM layer (64 units per direction, 128 units in total) is used, with the activation function being tanh. The forget gate bias is initialized to 1.0 to prevent gradient vanishing. The output temporal feature encoding h is... seq ∈R 20×128 .

[0095] Environment embedding layer: Maps [T,RH] to a 16-dimensional vector h env .

[0096] Attention mechanism module: fused with h through QKV architecture seq and h env Output fused features h fusion .

[0097] Output layer: Global average pooling: compresses the temporal dimension, reducing h fusion Transform from 20×64 dimensional to 64 dimensional; Fully connected layer (linear activation): Predicted mean μ and standard deviation σ of output load torque, formula: μ, σ = FC(h fusion ).

[0098] Key parameters and functions: Confidence interval calculation: Generates intervals with a confidence level of 1-α based on μ and σ.

[0099] τ min =μ-κ·σ,τ max =μ+κ·σ,

[0100] Wherein, κ is the standard deviation adjustment parameter (default value is 3, covering the 99% confidence interval).

[0101] Overload trigger condition: Set the load overload trigger coefficient. (i.e., when τ) max >1.5τ N (At that time, the pre-protection strategy is triggered).

[0102] Output application: Predicted mean μ of each motor based on S2 output i and standard deviation σ i Combined with real-time winding temperature T ωindCalculate the torque distribution weight ω i .

[0103] Step S3: Dynamic weight allocation

[0104] Input parameters:

[0105] Predicted mean μ for each motor i and standard deviation σ i (From prediction layer S2);

[0106] Real-time acquisition of winding temperature T wind (Data collected via PT100 sensor).

[0107] Temperature adaptive coefficient calculation:

[0108] k i =γT w , where γ=0.01 is the temperature influence factor (experimental calibration value).

[0109] Weight generation:

[0110] Dynamically assign weights using the Softmax function:

[0111]

[0112] Nash game multi-objective optimization:

[0113] Establish an objective function that includes an efficiency factor and thermal constraints:

[0114]

[0115] Where λ eff =0.6, λ thermal =0.4 is the weighting coefficient, T lim it The winding temperature safety limit (e.g., 130℃ for Class B insulated motors) is set via parameters on the motor nameplate. Based on distributed computing to solve for the Nash equilibrium point, the target torque command for each asynchronous motor is generated, such as... Figure 3 The decision-making process is shown below.

[0116] Command issued:

[0117] Weighting commands are sent to each slave asynchronous motor controller based on the CAN FD bus (transmission rate 5Mbps), with a delay controlled within 10μs.

[0118] Step S4: Online parameter identification and compensation

[0119] Slow time-varying parameters (actual rotor resistance value R) r (k+1) Identification:

[0120] Using a BP neural network:

[0121] The input layer has 3 nodes, which receive torque tracking error e. T Electromagnetic torque T e and the effective value of stator current |i s | Input after normalization;

[0122] Hidden layer: 16 nodes, using the Tanh activation function to extract the nonlinear relationship between input features and resistance;

[0123] Output layer: 1 node, linearly activated to increase the output rotor resistance increment ΔR r And the actual rotor resistance value R r (k+1) The updated formula is: R r (k+1) =R r (k) +ΔR r ,

[0124] In the formula, the learning rate η = 0.01, the backpropagation is performed to update the weights at 100ms intervals, and the loss function is the mean squared error (MSE).

[0125] Fast time-varying parameters (mutual inductance L) m )track:

[0126] The mutual inductance L is updated in real time with an extended Kalman filter (EKF) at a period of 10ms. m :

[0127]

[0128] In the formula L m0 The no-load mutual inductance value is dynamically adjusted according to the asynchronous motor model. For example, 0.32H is used for a 110kW asynchronous motor. The magnetic saturation compensation coefficient is ∈=0.015 and is determined through bench testing.

[0129] Step 5 (S5) Emergency Response Strategy

[0130] Using the upper limit τ of the load torque confidence interval output by S2 max Combined with the impact energy E detected by S1 impact The system determines whether the two-stage load reduction strategy should be triggered. The triggering logic and parameter definitions for the two-stage load reduction strategy are as follows:

[0131] Overload detection criteria:

[0132] When the predicted load τ max >1.5τ N And the impact energy E impactWhen the load exceeds 0.8J (threshold based on vibration test calibration), a two-stage load reduction strategy is triggered.

[0133] Level 1 response (executed within 10ms):

[0134] Reduce the switching frequency to 60% of the current value (e.g., reduce it from 20kHz to 10kHz at high speeds) to reduce switching losses;

[0135] The flywheel energy storage device (energy storage capacity 500J) is activated to absorb the impact energy and suppress the current peak.

[0136] Level 2 response (if the Level 1 response lasts for t2 = 1000 ms without relieving the overload):

[0137] Disconnect the power supply to the asynchronous motor and simultaneously activate the hydraulic braking module (braking torque 200 N·m) to prevent mechanical damage.

[0138] Step Six (S6): Execution Layer Control Algorithm

[0139] Based on the DTC-SVM algorithm, combined with the weighting command of S3, the compensation parameters of S4, and the independent protection signal of S5, the switching frequency is adjusted and the torque output is optimized. The process is as follows: Figure 4 As shown, it specifically includes:

[0140] Predictive torque compensation:

[0141] Introducing compensation coefficient K p =0.8, K i =0.2, based on the future predicted torque output by S2 With the actual output torque τ detected in real time k Construct the compensation formula:

[0142]

[0143] In the formula, τ is the predicted torque at a past time. k-i This represents the actual output torque at the corresponding moment. This compensation amount is added to the target torque command in real time to offset load prediction deviations.

[0144] Variable switching frequency:

[0145] When the motor speed is ≥1000rpm, a switching frequency of 20kHz is used; when the speed is ≤1000rpm, a switching frequency of 10kHz (critical speed ω) is used. th =1000rpm).

[0146] Receive the energy allocation weight ω from S3 via the CAN FD bus. i Based on the average total load torque μ predicted by S2, the target torque for each motor is calculated:

[0147] T refi =ω i ·μ,

[0148] High-weight motors are automatically matched with high switching frequencies (e.g., 20kHz), while low-weight motors use low switching frequencies (e.g., 10kHz). The critical speed ω... th Set to 1000rpm.

[0149] Bi-objective optimization:

[0150] Objective function: min(α·|T) ref -T e |+β·P switch The initial weights are α = 0.7 and β = 0.3. An optimal voltage vector is generated based on space vector modulation (SVM) to reduce torque ripple (amplitude controlled within 5% of rated torque).

[0151] Effects of implementing the present invention:

[0152] Dynamic response: Thanks to the predictive torque compensation mechanism and variable switching frequency adjustment of S6, the dynamic response time is shortened to less than 20ms (30ms for traditional DTC), effectively solving the problem of dynamic response lag in existing technologies; the torque balance of multi-asynchronous motors is improved by 40%, significantly improving the cooperative rigidity phenomenon caused by traditional fixed weight allocation;

[0153] Parameter compensation: The BP neural network of the present invention identifies rotor resistance error of <5%, and the EKF tracking mutual inductance accuracy is improved by 60%, corresponding to the parameter adaptive compensation in step S4;

[0154] Reliability: The overload protection response time of this invention is 10ms, and the temperature rise of the asynchronous motor is reduced by 15%, making it suitable for long-term stable operation in heavy-duty scenarios such as scraper conveyors.

[0155] This implementation method, based on a three-tier architecture with deep collaboration, achieves intelligent operation throughout the entire process from load prediction to execution control, significantly improving the overall performance of asynchronous motors under periodic impact loads.

[0156] In this embodiment of the application, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the methods described above.

[0157] In this application embodiment, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0158] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0159] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention described herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not invented herein. The specification and embodiments are to be considered exemplary only.

[0160] The above specific embodiments further illustrate the purpose, technical solution and beneficial effects of this application. It should be understood that the above are only specific embodiments of this application and are not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of this application should be included within the scope of protection of this application.

Claims

1. An asynchronous motor control method based on scraper load prediction, characterized in that, include: S1, the vibration signal of the scraper load is collected in real time by a vibration sensor, the current harmonic data of the asynchronous motor stator is collected in real time by a current sensor, and the mechanical transmission parameters are collected in real time by an encoder. The impact feature E is obtained by wavelet packet decomposition of the vibration signal. impact The total harmonic distortion (THD) is calculated based on current harmonic data and mapped to the load impact strength F. impact The basic load torque τ is calculated based on mechanical transmission parameters, including the torsional vibration angle of the drive shaft. base And all data, including impact characteristics, load impact intensity and mechanical transmission parameters, are standardized. S2, based on the environmental parameter embedding vector, inputs all data into the LSTM time series prediction network to predict the load prediction results, including the load torque, the probability distribution range of the load torque within a preset time period, and the confidence interval. S3, based on the load prediction results, uses a dynamic weighting function and a Nash game model to generate energy allocation weights for multiple asynchronous motors, and issues control commands containing energy allocation weights; S4, the actual rotor resistance value of the asynchronous motor is updated using a BP neural network, and the mutual inductance L is updated in real time using an extended Kalman filter. m .

2. The asynchronous motor control method based on scraper load prediction according to claim 1, characterized in that, include: S5, when the load torque predicted by S2 exceeds the rated threshold, triggers a two-stage load reduction strategy of flywheel energy storage and hydraulic braking, generating independent protection signals including switching frequency adjustment, flywheel energy storage start-up, and hydraulic braking trigger; S6, based on the DTC-SVM algorithm, the control command of step S3, the actual rotor resistance value, the time-varying deviation of the mutual inductance, and the independent protection signal of step S5, adjusts the switching frequency and optimizes the actual output torque of the asynchronous motor.

3. The asynchronous motor control method based on scraper load prediction according to claim 1, characterized in that, Step S1 includes: A triaxial MEMS accelerometer positioned at the scraper chain link acquires the vibration signal of the scraper load. Wavelet packet decomposition is used to extract the impact characteristics E within a preset frequency band. impact ; Real-time analysis of the stator current i of an asynchronous motor using sliding window FFT s The harmonic distortion rate (THD) is used to calculate and output the load impact intensity (F) using a pre-established empirical formula and a mapping model between the harmonic distortion rate and the load impact intensity. impact ; The torsional vibration angle θ of the transmission shaft is calculated based on the mechanical transmission parameters. Using a pre-constructed torque-torsion angle transfer function, the base load torque τ is output. base .

4. The asynchronous motor control method based on scraper load prediction according to claim 1, characterized in that, In step S2, based on the environmental parameter embedding vector, all data is input into the LSTM time series prediction network to predict the load prediction results, including the load torque, the probability distribution interval of the load torque within a preset future time period, and the confidence interval. The LSTM time-series prediction network comprises a bidirectional long short-term memory network layer, an environmental parameter embedding layer, an attention mechanism module, and a fully connected output layer connected in sequence. The bidirectional long short-term memory network layer is used to receive the standardized temporal feature vector X. seq The time series feature vector contains the impact features E of the historical time series. impact Load impact strength F impact and base load torque τ base The temporal feature encoding h is output by a bidirectional long short-term memory network layer. seq ; The environmental parameter embedding layer maps the ambient temperature T and humidity RH into a 16-dimensional environmental feature vector h. env The activation function in the environmental parameter embedding layer is ReLU; The attention mechanism module is used to fuse temporal feature encoding based on a query-key-value architecture. seq With environmental feature vector h env Obtain the fusion feature h fusion ; Using a fully connected output layer to fuse features h fusion After global average pooling, the predicted mean μ and standard deviation σ of the output load torque are obtained. Calculate the confidence interval [τ] of the network output load torque. min ,τ max ]: t min =μ-k·s,τ max =μ+k·s, In the formula, τ min To predict the lower limit of load torque, τ max To predict the upper limit of the load torque, k is the standard deviation adjustment parameter.

5. The asynchronous motor control method based on scraper load prediction according to claim 4, characterized in that, In step S3, based on the load prediction results, energy allocation weights for multiple asynchronous motors are generated using a dynamic weighting function and a Nash game model, and control commands containing these energy allocation weights are issued, including: Constructing a dynamic weight function: In the formula, μ i Let σ be the mean of the i-th prediction. i Let ω be the standard deviation of the i-th prediction. i It is the weight vector of the i-th asynchronous motor, k i =γT wind Where γ is the temperature influence factor, T ωind For winding temperature; Based on the Nash game optimization model, an objective function including an efficiency factor and a thermal constraint is established: In the formula, P loss For the system power loss, λ eff , λ thermal T represents the weighting coefficient. limit These are the safe limits for winding temperature; The energy allocation weights of multiple asynchronous motors are generated using a dynamic weight function and a Nash game model, and control commands containing the energy allocation weights are issued.

6. The asynchronous motor control method based on scraper load prediction according to claim 5, characterized in that, In step S4, the actual rotor resistance value is updated using a BP neural network, and the mutual inductance L is updated in real time using an extended Kalman filter. m ,include: A backpropagation neural network is used to iteratively update the actual rotor resistance value R at a set period. r (k+1) : R r (k+1) =R r (k) +ΔR r , In the formula, η is the learning rate, and T e e represents the actual output torque of the asynchronous motor. T For torque tracking error, e T =T ref -T e T ref ΔR is the target torque reference value. r This is the update increment for the rotor resistance; An extended Kalman filter is used to update the mutual inductance L in real time at a set period. m : In the formula, L m0 τ is the unloaded mutual inductance value. N denoted as the rated load torque of the asynchronous motor, and ∈ is the magnetic saturation compensation coefficient.

7. The asynchronous motor control method based on scraper load prediction according to claim 2, characterized in that, In step S5, when the load torque predicted in S2 exceeds the rated threshold, a two-stage load reduction strategy involving flywheel energy storage and hydraulic braking is triggered, generating independent protection signals that include switching frequency adjustment, flywheel energy storage activation, and hydraulic braking triggering, including: The defined state space includes load prediction overload, mechanical shock exceeding limits, winding temperature exceeding limits, winding cumulative loss exceeding preset standard thresholds, and action space; The predicted overload includes the upper limit of the predicted load torque. The probability is greater than or equal to the pre-set confidence level 1-α; Excessive mechanical impact includes impact characteristic E impact >E th E th This refers to the impact energy threshold calibrated based on vibration testing. Excessive winding temperature includes winding temperature T wind ≥Preset winding temperature threshold T limit ; The operating space includes the frequency conversion load reduction level and the degree of mechanical braking intervention. The frequency conversion load reduction level includes the switching frequency adjustment range, and the degree of mechanical braking intervention includes flywheel energy storage start-up and hydraulic braking trigger. The two-stage load reduction strategy includes: if the predicted upper limit of the load torque τ max Exceeding the rated load torque τ of the asynchronous motor N If the flywheel energy storage device is triggered to absorb the impact energy of the scraper load, and if the scraper load overload is not relieved within a preset time, the hydraulic brake will be activated and the power supply will be cut off.

8. The asynchronous motor control method based on scraper load prediction according to claim 2, characterized in that, In step S6, based on the DTC-SVM algorithm, the control command from step S3, the actual rotor resistance value, the time-varying deviation of the mutual inductance, and the independent protection signal from step S5, the switching frequency is adjusted and the actual output torque of the asynchronous motor is optimized, including: Introducing compensation coefficient K p and integral compensation coefficient K i Construct the compensation formula: In the formula, Δτ comp This is the torque compensation amount. Let τ be the predicted mean at time k+1. k This represents the actual output torque of the asynchronous motor at time k. Let τ be the predicted mean at time ki. k-i The actual output torque of the asynchronous motor at time ki; If the asynchronous motor speed is ≥ ω th If the asynchronous motor speed is ≤ω, then a preset high switching frequency is used. th Then, a preset low switching frequency ω is used for the asynchronous motor. th The critical speed; Construct the objective function: min(α·|T ref -T e |+β·P switch ), Among them, T ref T is the target torque of the asynchronous motor. e P represents the actual output torque of the asynchronous motor. switch Let α be the switching loss power, β be the torque accuracy weight, and α + β = 1.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 8.

Images