Intelligent mixer multi-mode operation control method and system

Abstract

The application discloses a kind of intelligent stirrer multi-mode operation control method and system, it is related to multi-mode operation control technical field, including the following steps, original data is carried out time alignment processing and fusion processing, obtain fusion multidimensional time series data, and carry out time sequence phase verification to obtain phase-corrected fusion multidimensional time series data, and carry out feature extraction to obtain time sequence statistical feature and spectrum analysis feature, calculate the modal confidence coefficient of each original data, the feature enhancement processing is carried out to time sequence statistical feature and spectrum analysis feature by modal confidence coefficient, obtain enhanced feature value, and arrange to obtain feature vector according to fixed order and input to gradient boosting decision tree, output the confidence degree that stirrer current operating state belongs to each preset control mode, based on confidence degree using multilayer decision strategy is judged to obtain target control parameter, and is converted into electric signal and is issued to motor driver, and the stirrer is regulated and controlled by motor driver.

Description

Technical Field

[0001] This invention relates to the field of multi-mode operation control technology, and in particular to a method and system for multi-mode operation control of an intelligent mixer. Background Technology

[0002] In recent years, with the rapid development of smart kitchen appliances, the multi-functional integration and automated control of smart mixers have become an important trend. Users expect the equipment to automatically identify and process tasks, such as crushing ice, kneading dough, and making milkshakes, and to autonomously switch to the optimal working mode to improve ease of use, processing effect, and energy efficiency. Therefore, researching precise and reliable multi-mode operation control technology is of great significance for enhancing the intelligence level of products and improving user experience.

[0003] However, existing control methods for intelligent mixers still have significant shortcomings. First, when using multiple sensors, such as current data, vibration data, and sound data, for state perception, they often focus on simple timestamp alignment and data stitching, lacking fine verification and compensation for deep temporal phase deviations caused by inconsistent mechanical transmission and sensor responses, which affects the quality of fused data. Second, feature processing usually assigns fixed weights to all modal data, failing to dynamically evaluate and utilize the reliability differences of different sensor signals under specific operating conditions. When a sensor is affected by noise or interference, it can easily lead to overall judgment distortion. Furthermore, the mode decision-making mechanism often relies on the output of a single machine learning model, lacking effective auxiliary decision-making strategies when the model confidence is low. Especially under complex and ambiguous operating conditions, it is prone to misjudgment and cannot incorporate user history habits to achieve personalized adaptation. These shortcomings restrict the accuracy, robustness, and practicality of the control system. Summary of the Invention

[0004] The technical problem addressed by this invention is that existing control methods for intelligent mixers still have significant shortcomings. First, when using multiple sensors, such as current data, vibration data, and sound data, for state perception, they often focus on simple timestamp alignment and data stitching, lacking fine verification and compensation for deep temporal phase deviations caused by inconsistent mechanical transmission and sensor responses, thus affecting the quality of fused data. Second, feature processing typically assigns fixed weights to all modal data, failing to dynamically evaluate and utilize the reliability differences of different sensor signals under specific operating conditions. When a sensor is affected by noise or interference, it can easily lead to overall judgment distortion. Furthermore, the mode decision-making mechanism often relies on the output of a single machine learning model, lacking effective auxiliary decision-making strategies when the model confidence is low. Especially under complex and ambiguous operating conditions, it is prone to misjudgment and cannot incorporate user history habits for personalized adaptation. These deficiencies restrict the accuracy, robustness, and practicality of the control system.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a multi-mode operation control method for an intelligent stirrer, comprising the following steps, Step S1: Perform time alignment and fusion processing on the original data to obtain fused multi-dimensional time series data; perform time-series phase verification on the fused multi-dimensional time series data to obtain phase-corrected fused multi-dimensional time series data; and extract features from the phase-corrected fused multi-dimensional time series data to obtain time-series statistical features and spectral analysis features. Step S2: Calculate the modal confidence coefficient of each original data, and perform feature enhancement processing on the time series statistical features and spectral analysis features through the modal confidence coefficient to obtain enhanced feature values. Arrange the enhanced feature values ​​in a fixed order to obtain the feature vector. Step S3: Input the feature vector into the gradient boosting decision tree, output the confidence level of the current running state of the stirrer belonging to each preset control mode, and use a multi-level decision strategy to make a determination based on the confidence level to obtain the target control parameters; Step S4: The control parameters are converted into electrical signals and sent to the motor driver, which then controls the agitator.

[0006] As a preferred embodiment of the intelligent stirrer multi-mode operation control method of the present invention, step S1 includes steps S101, S102, S103 and S104. Step S101: The raw data includes current data, vibration data, and sound data; Step S102, the specific process of time alignment processing includes, for example, calculating the time offset between each original data and the stirrer master control clock, taking into account the inherent time deviation of the original data during acquisition; The original data are time-shifted and corrected according to the time offset to obtain the corrected timestamp sequence of each original data, which serves as intermediate data for phase pre-alignment. Based on the stirrer master clock, a time stamp sequence with a fixed frequency and equal intervals is generated as a unified timing reference, denoted as the target time stamp sequence; The specific process of fusion processing includes: relocating the pre-aligned intermediate data of each phase to the target timestamp sequence; selecting an interpolation window centered on the target timestamp sequence for each target time point; calculating and normalizing the signal-to-noise ratio of each intermediate data within the interpolation window to obtain the interpolation weight; and performing weighted average interpolation calculation on the values ​​of each intermediate data within the interpolation calculation window based on the interpolation weight to generate fused multi-dimensional time series data.

[0007] As a preferred embodiment of the intelligent stirrer multi-mode operation control method of the present invention, in step S103, based on the rated rotation cycle of the stirrer, a preset time window is slid on the fused multi-dimensional time series data. For the position of each time window, the current data segment within the time window is obtained, and the corresponding reference data segment after a delay of one rated rotation cycle is obtained. Calculate the Dynamic Time Warping (DTW) distance between the current data segment and the reference data segment to obtain the optimal bending path; Based on the optimal curved path, obtain the index mapping relationship between the current data segment and the reference data segment; Calculate the normalized DTW distance of the optimal curved path as a temporal alignment metric; If the normalized DTW distance is less than the preset alignment threshold, the timing phase verification of the current time window is determined to be successful. Otherwise, the current time window verification is deemed to have failed, and the index mapping relationship is extracted. Based on the index mapping relationship, the current data segment is subjected to non-linear time resampling and alignment to obtain the corrected current data segment. After traversing the positions of all time windows, the corrected current data segments corresponding to each time window are merged in chronological order to obtain phase-corrected fused multi-dimensional time series data. Step S104: Extract time series statistical features and spectral analysis features from the phase-corrected fused multi-dimensional time series data.

[0008] As a preferred embodiment of the intelligent stirrer multi-mode operation control method of the present invention, step S2 includes steps S201 and S202. Step S201, the calculation process of the modal confidence coefficient includes, for the current data, performing multi-scale permutation entropy (MPE) calculation on the current data in each interpolation window to obtain a set of current multi-scale permutation entropy values, performing normalization calculation on the set of current multi-scale permutation entropy values ​​to obtain the current complexity adjustment coefficient, and multiplying the current complexity adjustment coefficient by a preset sharpening factor to obtain the current exponent value. For vibration data, multi-scale permutation entropy (MPE) is calculated for vibration data within each interpolation window to obtain a set of vibration multi-scale permutation entropy values. The set of vibration multi-scale permutation entropy values ​​is then normalized to obtain a vibration complexity adjustment coefficient. The vibration complexity adjustment coefficient is multiplied by the sharpening factor to obtain the vibration index value. For the audio data, the multi-scale permutation entropy (MPE) is calculated for the audio data in each interpolation window to obtain a set of audio multi-scale permutation entropy values. The set of audio multi-scale permutation entropy values ​​is then normalized to obtain an audio complexity adjustment coefficient. The audio complexity adjustment coefficient is multiplied by the sharpening factor to obtain an audio index value. The sum of the current index, vibration index, and sound index is obtained by adding the current index, vibration index, and sound index together. Divide the current index value by the sum of the exponents to obtain the current mode reliability coefficient; Divide the vibration index value by the sum of the indices to obtain the vibration mode reliability coefficient; Divide the sound index value by the sum of the indices to obtain the sound modality confidence coefficient; The sum of the current mode confidence coefficient, vibration mode confidence coefficient, and sound mode confidence coefficient is 1.

[0009] As a preferred embodiment of the intelligent stirrer multi-mode operation control method of the present invention, in step S202, the time-series statistical features and spectrum analysis features are grouped according to the data source of the time-series statistical features and spectrum analysis features to obtain current feature group, vibration feature group and sound feature group; The data sources include current data, vibration data, and sound data; The current characteristic group includes current time-series statistical characteristics and current spectrum analysis characteristics; The vibration feature set includes vibration time-series statistical features and vibration spectrum analysis features; The sound feature group includes sound temporal statistical features and sound spectral analysis features; Based on the current mode confidence coefficient, vibration mode confidence coefficient, and sound mode confidence coefficient corresponding to the current feature group, vibration feature group, and sound feature group, respectively, feature enhancement processing is performed on all feature values ​​within the same group; The feature enhancement process involves scaling each feature value within the same group using the square root of the modality confidence coefficient corresponding to that group as a scaling factor to obtain a weighted feature value, which is denoted as the enhanced feature value. Wherein, the characteristic value is each of the values ​​of the current time-series statistical characteristic, the current spectrum analysis characteristic, the vibration time-series statistical characteristic, the vibration spectrum analysis characteristic, the sound time-series statistical characteristic, and the sound spectrum analysis characteristic; The enhanced feature values ​​are arranged in a fixed order to obtain a feature vector; The fixed order is: current time sequence statistical characteristics, current spectrum analysis characteristics, vibration time sequence statistical characteristics, vibration spectrum analysis characteristics, sound time sequence statistical characteristics, and sound spectrum analysis characteristics.

[0010] As a preferred embodiment of the intelligent stirrer multi-mode operation control method of the present invention, step S3 includes steps S301, S302, S303, S304 and S305. Step S301: The preset control modes include high-speed ice crushing mode, low-speed stirring mode, dough kneading mode, milkshake making mode and self-cleaning pulse mode. Each control mode has a set of corresponding target control parameters stored in advance, and a control mapping relationship between the control mode and the corresponding target control parameters is established. The target control parameters include the motor speed curve, running time, and steering sequence; Based on the confidence level, a multi-layer decision-making strategy is adopted to determine the target control mode and the corresponding target control parameters.

[0011] As a preferred embodiment of the intelligent stirrer multi-mode operation control method of the present invention, step S302, the multi-layer judgment strategy specifically includes obtaining the confidence level of each preset control mode output by the gradient boosting decision tree, and identifying the highest confidence level and the corresponding control mode among the confidence levels. Step S303: Perform the first-level determination based on the highest confidence level; The first layer of judgment logic is as follows: if the highest confidence level is greater than or equal to the first preset threshold, then the control mode corresponding to the highest confidence level is taken as the target control mode, and the corresponding target control parameters are determined. Otherwise, proceed to the second level of judgment.

[0012] As a preferred embodiment of the intelligent stirrer multi-mode operation control method of the present invention, in step S304, the second-level determination logic is to select the modes with confidence greater than or equal to the second preset threshold as candidate modes from all preset control modes, wherein the second preset threshold is less than the first preset threshold. If one or more candidate modes are selected, the historical frequency comparison rule is executed. The historical frequency comparison rule is to query the historical operation database, obtain the frequency of each candidate mode in the historical operation records within the preset time period, and take the unique candidate mode with the highest frequency as the target control mode. If there are multiple candidate modes with the same highest frequency of occurrence, the candidate mode corresponding to the most recent run is selected from the multiple candidate modes with the same highest frequency of occurrence as the target control mode, and the corresponding target control parameters are determined. If no candidate patterns are selected, or if candidate patterns are selected but a unique target control pattern cannot be obtained based on the historical frequency comparison rules, then proceed to the third level of judgment. Step S305, the logic of the third layer determination is to obtain the target control mode and the corresponding target control parameters according to the preset default rules. The default rules are to take the control mode corresponding to the highest confidence level output by the gradient boosting decision tree as the target control mode and the corresponding target control parameters.

[0013] In a preferred embodiment of the intelligent stirrer multi-mode operation control method described in this invention, step S4 involves calling the control mapping relationship according to the target control mode, obtaining the corresponding target control parameters as a control parameter set, converting the control parameters into electrical signals and sending them to the motor driver, and then controlling the stirrer through the motor driver.

[0014] A multi-mode operation control system for an intelligent mixer includes a fusion module, an enhancement module, a judgment module, and a control module. The fusion module performs time alignment and fusion processing on the original data to obtain fused multi-dimensional time series data. It then performs time-series phase verification on the fused multi-dimensional time series data to obtain phase-corrected fused multi-dimensional time series data. Finally, it extracts features from the phase-corrected fused multi-dimensional time series data to obtain time-series statistical features and spectral analysis features. The enhancement module calculates the modal confidence coefficient of each original data, performs feature enhancement processing on the time series statistical features and spectral analysis features using the modal confidence coefficient, obtains enhanced feature values, and arranges the enhanced feature values ​​in a fixed order to obtain the feature vector; The decision module inputs the feature vector into the gradient boosting decision tree and outputs the confidence level of the current running state of the stirrer belonging to each preset control mode. Based on the confidence level, a multi-level decision strategy is used to make a decision to obtain the target control parameters. The control module converts control parameters into electrical signals and sends them to the motor driver, which then regulates the agitator.

[0015] The beneficial effects of this invention are as follows: Based on the time-series phase verification using Dynamic Time Warping (DTW) distance, it fundamentally solves the problem of deep temporal misalignment caused by mechanical delays or response differences in multi-source sensor data, achieving high-precision data fusion and providing a reliable data foundation for subsequent analysis. Simultaneously, based on the modal reliability dynamic evaluation mechanism using multi-scale permutation entropy (MPE), it quantifies the signal quality and stability of each sensing mode, such as current, vibration, and sound, in real time, and performs adaptive feature enhancement accordingly. This effectively suppresses the negative impact of noise or interference modes, significantly improving the robustness of input feature representation. Furthermore, the multi-layered judgment strategy—including direct judgment with high model confidence, historical frequency assistance, and default rule fallback—greatly enhances decision reliability under complex and fuzzy conditions. It not only avoids misjudgments that may occur with a single model but also intelligently integrates with user habits to achieve personalized adaptation. Ultimately, this ensures that the mixer can accurately and stably switch to the optimal working mode automatically in various practical application scenarios, significantly improving the equipment's intelligence level, control precision, and user experience. Attached Figure Description

[0016] Figure 1This is a flowchart illustrating the steps of a multi-mode operation control method for an intelligent stirrer, as provided in one embodiment of the present invention.

[0017] Figure 2 This is a basic flowchart of a multi-mode operation control system for an intelligent stirrer provided in one embodiment of the present invention. Detailed Implementation

[0018] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0019] Example 1, referring to Figure 1 As an embodiment of the present invention, a multi-mode operation control method for an intelligent stirrer is provided, comprising the following steps: Step S1: Perform time alignment and fusion processing on the original data to obtain fused multi-dimensional time series data; perform time-series phase verification on the fused multi-dimensional time series data to obtain phase-corrected fused multi-dimensional time series data; and extract features from the phase-corrected fused multi-dimensional time series data to obtain time-series statistical features and spectral analysis features. Step S2: Calculate the modal confidence coefficient of each original data, and perform feature enhancement processing on the time series statistical features and spectral analysis features through the modal confidence coefficient to obtain enhanced feature values. Arrange the enhanced feature values ​​in a fixed order to obtain the feature vector. Step S3: Input the feature vector into the gradient boosting decision tree, output the confidence level of the current running state of the stirrer belonging to each preset control mode, and use a multi-level decision strategy to make a determination based on the confidence level to obtain the target control parameters; Step S4: The control parameters are converted into electrical signals and sent to the motor driver, which then controls the agitator.

[0020] In one embodiment, a complete closed loop from accurate sensing to reliable decision-making is constructed. First, in step S1, the raw data of current, vibration, and sound are time-aligned and fused, and then Dynamic Time Warping (DTW) is used for timing phase verification. This effectively overcomes the timing misalignment problem caused by sensor response delay or mechanical transmission, thus achieving the technical effect of obtaining high-quality, highly synchronous fused data and laying a precise time-domain foundation for subsequent analysis. Subsequently, step S2 introduces a modal reliability coefficient based on multi-scale permutation entropy (MPE) to dynamically evaluate the stability of each sensor signal under the current operating conditions. Based on this, the extracted timing and spectral features are adaptively weighted and enhanced, giving the system the ability to distinguish between true and false features at the feature level, significantly improving the input feature vector. The robustness of the representation allows the model to focus more on data modes with high confidence and suppress noise interference. Then, in step S3, the enhanced feature vector is input into the gradient boosting decision tree (GBDT) for preliminary pattern recognition. This is supplemented by a multi-layer decision strategy that combines direct judgment with high confidence of the model, historical frequency assistance, and default rule fallback. This ensures that the decision-making combines the intelligence of the model with the robustness of practical applications, greatly improving the reliability of decision-making under complex working conditions with fuzzy confidence. It can also integrate user habits to achieve personalized adaptation. Finally, in step S4, the target control parameters obtained from the judgment, such as the speed curve and steering sequence, are converted into drive signals for execution. This completes the process from intelligent analysis to physical control. The entire architecture is interconnected, realizing high-precision, highly adaptive, and highly reliable fully automatic intelligent control of the mixer's multi-mode operation.

[0021] Step S1 includes steps S101, S102, S103 and S104; Step S101: The raw data includes current data, vibration data, and sound data; Step S102, the specific process of time alignment processing includes, for example, calculating the time offset between each original data and the stirrer master control clock, taking into account the inherent time deviation of the original data during acquisition; The original data are time-shifted and corrected according to the time offset to obtain the corrected timestamp sequence of each original data, which serves as intermediate data for phase pre-alignment. Based on the stirrer master clock, a time stamp sequence with a fixed frequency and equal intervals is generated as a unified timing reference, denoted as the target time stamp sequence; The specific process of fusion processing includes: relocating the pre-aligned intermediate data of each phase to the target timestamp sequence; for each target time point in the target timestamp sequence, selecting an interpolation window centered on the target time point; calculating and normalizing the signal-to-noise ratio of each intermediate data within the interpolation window to obtain the interpolation weight; and performing weighted average interpolation calculation on the values ​​of each intermediate data within the interpolation calculation window based on the interpolation weight to generate fused multi-dimensional time series data.

[0022] Step S103: Based on the rated rotation cycle of the stirrer, slide a preset time window on the fused multi-dimensional time series data. For the position of each time window, obtain the current data segment within the time window and obtain the corresponding reference data segment after a delay of one rated rotation cycle. Calculate the Dynamic Time Warping (DTW) distance between the current data segment and the reference data segment to obtain the optimal bending path; Based on the optimal curved path, obtain the index mapping relationship between the current data segment and the reference data segment; Calculate the normalized DTW distance of the optimal curved path as a time alignment metric; If the normalized DTW distance is less than the preset alignment threshold, the timing phase check of the current time window is deemed to have passed. Otherwise, the current time window verification is deemed to have failed, and the index mapping relationship is extracted. Based on the index mapping relationship, the current data segment is subjected to non-linear time resampling and alignment to obtain the corrected current data segment. After traversing the positions of all time windows, the corrected current data segments corresponding to each time window are merged in chronological order to obtain phase-corrected fused multi-dimensional time series data. Step S104: Extract time series statistical features and spectral analysis features from the phase-corrected fused multi-dimensional time series data.

[0023] In one embodiment, step S101 is first executed, where multimodal raw data, including current data, vibration data, and sound data, are collected in real time by corresponding sensors to form the raw information basis for analysis. Then, step S102, namely time alignment and fusion processing, is performed. Its design aims to solve the inherent time deviations caused by hardware characteristics, sampling start-up time differences, and different transmission paths, and to establish a unified and accurate time reference. Specific steps include calculating the time offset between each data stream and the master clock (e.g., sound sensors may have a delay of tens of milliseconds, such as a fixed delay of 20-50ms), and performing linear time shift correction to obtain preliminary aligned intermediate data; and generating equally spaced target timestamps at a fixed frequency based on the master clock. The sequence serves as a unified time axis, with a fixed frequency of 100Hz. This setting is based on meeting the sampling requirements of Shannon's sampling theorem for the stirrer's operating signal, which typically has a fundamental frequency less than 50Hz, and also allows for a certain frequency margin to preserve harmonic information. For each target time point, a fixed-duration interpolation window is selected centered on that point, with the window size set to cover ±10 original data points. This setting balances interpolation accuracy and computational real-time performance, ensuring that the window includes sufficient data points for stable signal-to-noise ratio (SNR) estimation. The SNR of each intermediate data point within this window is calculated. The SNR is obtained by calculating the ratio of signal power to background noise power within the interpolation window, where the background noise power is measured beforehand with the stirrer unloaded and the motor stationary. The calibration values ​​obtained are measured, and the SNR of all modes is normalized to obtain the interpolation weight of each mode at that point, with the total weight being 1. Finally, based on this weight, the values ​​of each intermediate data within the window are weighted average interpolated to generate fused multi-dimensional time series data. This process achieves data synchronization and fusion on a unified time axis and enhances the contribution of high-quality signals through signal-to-noise ratio weighting, initially improving the overall data quality and reliability. Then, step S103 is executed, i.e., time series phase verification and correction, to solve the problem of nonlinear time series distortion, such as phase drift, that may still be caused by slight changes in speed or sudden changes in load after linear time shift. The specific steps include, based on the rated rotation period T of the motor, for example, corresponding to a rated speed of 1 minute... When the clock rotates at 3000 revolutions per minute, T is 20 ms. This value is calculated from T = 60 seconds / rated speed (revolutions per minute). A preset time window is slid across the fused data, with a window length set to 5T, or 100 ms. This setting is based on covering several complete working cycles to provide sufficient data length for reliable DTW alignment. The current data segment and the corresponding reference data segment after a delay of one cycle T are extracted from the window. The dynamic time warping (DTW) distance between the current data segment and the reference data segment is calculated, and the optimal warping path that aligns the current data segment with the reference data segment is obtained. The normalized DTW distance, i.e., the DTW distance divided by the sum of the lengths of the two sequences, is calculated as a time alignment metric. An alignment threshold ε is set to 0.15. This empirical threshold, determined based on extensive experimental data, strikes a good balance between allowing normal signal fluctuations and identifying significant phase distortion. If the normalized DTW distance is less than ε, the current time window's timing phase verification is considered successful; otherwise, the verification fails. Based on the index mapping relationship defined by the optimal curved path, the current data segment undergoes nonlinear time resampling, such as stretching or compression, to align its waveform with the reference data segment, resulting in a corrected current data segment. After traversing all sliding time windows, all corrected segments are either left as is or remain unchanged. Data segments that pass through the time windows are merged in chronological order, ultimately yielding phase-corrected fused multi-dimensional time series data. This step effectively detects and corrects deep nonlinear phase errors. Strict alignment of the fused data in the periodic analysis dimension greatly improves the accuracy and consistency of subsequent feature extraction, especially spectral features. Finally, step S104, feature extraction, is performed to extract features characterizing the stirrer's operating state from the high-quality fused data that has undergone strict time alignment and correction. Specifically, time-series statistical features and spectral analysis features are extracted. Time-series statistical features include, for example, mean, variance, peak value, root mean square (RMS), waveform factor, and impulse index, used to describe the time-domain amplitude statistical characteristics of the signal. Spectral analysis features include extracting the main frequency components, such as the amplitude, frequency, and proportion of the fundamental frequency and harmonics, by performing a Fast Fourier Transform (FFT) on the data segment or calculating the power spectral density (PSD), used to describe the frequency-domain energy distribution characteristics of the signal. Steps S101 to S104 together constitute a rigorous data preprocessing and feature engineering pipeline. Through progressive processing of multi-source raw data, from clock synchronization to signal-to-noise ratio fusion to periodic phase calibration, the data input to the subsequent intelligent decision-making module is fundamentally guaranteed to have extremely high temporal consistency and information fidelity. This is the cornerstone prerequisite for the entire control system to achieve high-precision and high-reliability pattern recognition and control.

[0024] Step S2 includes steps S201 and S202; Step S201, the calculation process of the modal confidence coefficient includes, for the current data, performing multi-scale permutation entropy (MPE) calculation on the current data in each interpolation window to obtain a set of current multi-scale permutation entropy values, performing normalization calculation on the set of current multi-scale permutation entropy values ​​to obtain the current complexity adjustment coefficient, and multiplying the current complexity adjustment coefficient by a preset sharpening factor to obtain the current exponent value. For vibration data, multi-scale permutation entropy (MPE) is calculated for vibration data within each interpolation window to obtain a set of vibration multi-scale permutation entropy values. The set of vibration multi-scale permutation entropy values ​​is then normalized to obtain a vibration complexity adjustment coefficient. The vibration complexity adjustment coefficient is multiplied by a sharpening factor to obtain a vibration index value. For the audio data, the multi-scale permutation entropy (MPE) is calculated for the audio data in each interpolation window to obtain a set of audio multi-scale permutation entropy values. The set of audio multi-scale permutation entropy values ​​is then normalized to obtain the audio complexity adjustment coefficient. The audio complexity adjustment coefficient is multiplied by the sharpening factor to obtain the audio index value. Add the current index value, vibration index value, and sound index value to obtain the total index; Divide the current exponent value by the sum of the exponents to obtain the current mode reliability coefficient; Divide the vibration index value by the sum of the exponents to obtain the vibration mode confidence coefficient; Divide the sound index value by the sum of the indices to obtain the sound modal credibility coefficient; The sum of the confidence coefficients for the current mode, vibration mode, and sound mode is 1.

[0025] Step S202: The time-series statistical features and spectral analysis features are grouped according to their data sources to obtain current feature group, vibration feature group and sound feature group; Data sources include current data, vibration data, and sound data; The current characteristic group includes current time-series statistical characteristics and current spectrum analysis characteristics; The vibration characteristic group includes vibration time-series statistical characteristics and vibration spectrum analysis characteristics; The sound feature group includes sound temporal statistical features and sound spectral analysis features; Based on the current mode confidence coefficient, vibration mode confidence coefficient, and sound mode confidence coefficient corresponding to the current feature group, vibration feature group, and sound feature group, respectively, feature enhancement processing is performed on all feature values ​​within the same group; The feature enhancement process involves scaling each feature value within the same group using the square root of the modality confidence coefficient corresponding to that group as a scaling factor to obtain a weighted feature value, which is denoted as the enhanced feature value. Among them, the characteristic values ​​are each of the following: current time sequence statistical characteristics, current spectrum analysis characteristics, vibration time sequence statistical characteristics, vibration spectrum analysis characteristics, sound time sequence statistical characteristics, and sound spectrum analysis characteristics. The enhanced feature values ​​are arranged in a fixed order to obtain the feature vector; The fixed order is: current time sequence statistical characteristics, current spectrum analysis characteristics, vibration time sequence statistical characteristics, vibration spectrum analysis characteristics, sound time sequence statistical characteristics, and sound spectrum analysis characteristics.

[0026] In one embodiment, step S2 is executed to dynamically evaluate the quality and reliability of different sensor data under the current operating conditions and assign differentiated weights to subsequent features; an adaptive filtering mechanism is constructed so that when a certain sensing mode is affected by noise or interference, its decision weight can be automatically reduced to improve the robustness of the system; step S201 calculates the modal reliability coefficient by quantifying the complexity of each modal signal to evaluate its stability. The more ordered and stable the signal, the higher the reliability. The specific steps include performing multi-scale permutation entropy (MPE) calculation on the data in each interpolation window in step S102 for current data, vibration data, and sound data. The MPE calculation process is as follows: first, multi-scale coarsening is performed. For the original time series, the scale parameter τ is set from 1 to a maximum value. For example, τ = 1, 2, ..., 20 is set, and the maximum value is set to 20. The setting is based on covering the signal characteristic time scale that may occur during the operation of the stirrer caused by load changes. The algorithm considers both speed and computational efficiency. For each τ, a coarse-grained sequence is constructed by averaging τ consecutive data points. Then, the permutation entropy PE is calculated for each scale of the coarse-grained sequence. When calculating PE, the embedding dimension m needs to be set, for example, m=5. This value must satisfy the condition that m! is less than the length of the time series to ensure the stability of the permutation pattern statistics and the time delay L. Usually, L is set to 1. The setting of L is based on the most direct sequential relationship between the original sampling points of the captured signal, which is the most commonly used and effective setting when calculating permutation entropy. Specifically, the time series is reconstructed into a phase space to obtain multiple subsequences of length m. The values ​​in each subsequence are sorted by size to obtain its permutation pattern, for example [2,1,3]. The probability distribution of all permutation patterns is statistically analyzed. Finally, the permutation entropy value at this scale is calculated according to the Shannon entropy formula. The permutation entropy values ​​at all scales are combined into a set, that is, a multi-scale permutation entropy value set. Then, the set is normalized: each entropy value in the set is calculated using the formula A linear mapping is applied to the interval [0,1], where, Here, H represents the entropy value arranged at multiple scales, and H represents the original entropy value. and The minimum and maximum entropy values ​​of the mode across all scales are respectively taken as the minimum and maximum values. Then, the average of the normalized entropy values ​​across all scales is taken to obtain the complexity adjustment coefficient of the mode. The lower the complexity adjustment coefficient, the more ordered and stable the signal. Next, a preset sharpening factor, for example, set to 10, is multiplied by the complexity adjustment coefficient of each mode to obtain the current index, vibration index, and sound index values, respectively. Finally, the current index, vibration index, and sound index values ​​are added to obtain the total index, and the index value of each mode is divided by the minimum and maximum values ​​of the modal. The confidence coefficients for the current mode, vibration mode, and sound mode are summed to obtain a total of 1. This achieves real-time and quantitative evaluation of the quality of multi-source sensor data. Step S202 involves feature enhancement and vector construction. Using the calculated confidence coefficients, features from different modes are weighted differently to construct feature vectors that better reflect the actual operating state. Specifically, the time-series statistical features and spectral analysis features extracted in step S104 are grouped according to their data sources (current data, vibration data, and sound data) to form current feature groups. The system first identifies vibration feature groups and then sound feature groups. Based on the corresponding modal reliability coefficients calculated in step S201, feature enhancement processing is performed on all feature values ​​within the same group. Each feature value is multiplied by the square root of the corresponding modal reliability coefficient. Using the square root instead of directly multiplying the coefficient smooths out the influence of weights, avoiding excessive suppression of low-reliability modes. This achieves a balance between enhancing the dominance of reliable signals and preserving the potential information of all modes, resulting in enhanced feature values. Finally, all enhanced feature values ​​are processed in a fixed order, such as current time-series statistical features, current spectrum analysis features, vibration time-series statistical features, vibration spectrum analysis features, and sound time-series statistical features. The features and sound spectrum analysis features are arranged and combined to form the final feature vector used for model input. A quality-weighted feature representation is constructed, which makes the final feature vector more focused on the sensing modal with high credibility, effectively suppressing the interference of noise or abnormal modalities, and improving the overall signal-to-noise ratio and discrimination ability of the feature data input to the machine learning model. Steps S201 and S202 together constitute a dynamic feature quality evaluation and optimization method. In complex real-world working environments, a more stable signal source is automatically identified and trusted through a data-driven approach, enabling the entire intelligent perception to have inherent anti-interference and adaptive capabilities, thus laying a solid foundation for subsequent high-precision pattern recognition.

[0027] Step S3 includes steps S301, S302, S303, S304 and S305; Step S301: The preset control modes include high-speed ice crushing mode, low-speed stirring mode, dough kneading mode, milkshake making mode and self-cleaning pulse mode. Each control mode has a set of corresponding target control parameters stored in advance, and a control mapping relationship between the control mode and the corresponding target control parameters is established. The target control parameters include the motor speed curve, running time, and steering sequence; Based on confidence level, a multi-level decision-making strategy is adopted to determine the target control mode and the corresponding target control parameters.

[0028] Step S302, the multi-layer decision strategy specifically includes obtaining the confidence level of each preset control mode output by the gradient boosting decision tree, and identifying the highest confidence level and the corresponding control mode. Step S303: Perform the first-level determination based on the highest confidence level; The first layer of judgment logic is that if the highest confidence level is greater than or equal to the first preset threshold, the control mode corresponding to the highest confidence level is taken as the target control mode, and the corresponding target control parameters are determined. Otherwise, proceed to the second level of judgment.

[0029] Step S304, the second-level judgment logic is to select the modes with confidence greater than or equal to the second preset threshold as candidate modes from all preset control modes, wherein the second preset threshold is less than the first preset threshold; If one or more candidate modes are selected, the historical frequency comparison rule is executed. The historical frequency comparison rule is to query the historical operation database, obtain the frequency of each candidate mode in the historical operation records within the preset time period, and take the unique candidate mode with the highest frequency as the target control mode. If there are multiple candidate modes with the same highest frequency of occurrence, the candidate mode corresponding to the most recent run is selected from the multiple candidate modes with the same highest frequency of occurrence as the target control mode, and the corresponding target control parameters are determined. If no candidate patterns are selected, or if candidate patterns are selected but a unique target control pattern cannot be obtained based on the historical frequency comparison rules, then proceed to the third level of judgment. Step S305, the logic of the third layer determination is to obtain the target control mode and the corresponding target control parameters according to the preset default rules. The default rule is to take the control mode corresponding to the highest confidence level output by the gradient boosting decision tree as the target control mode and the corresponding target control parameters.

[0030] Step S4: According to the target control mode, call the control mapping relationship, obtain the corresponding target control parameters as the control parameter set, convert the control parameters into electrical signals and send them to the motor driver, and regulate the stirrer through the motor driver.

[0031] In one embodiment, steps S3 and S4 are executed to construct a multi-layered, highly fault-tolerant intelligent decision-making process to handle complex situations where the confidence level of the machine learning model output is fuzzy. This process integrates data-driven intelligent recognition with rule-based logical safeguards to ensure that the control system can make reasonable, reliable, and user-friendly decisions under any operating condition, thereby improving user experience and system reliability. Step S301 involves pre-setting mode and parameter mapping to establish clear output targets for all possible operating modes, transforming the intelligent recognition results into specific physical control commands. Specific steps include pre-defining multiple control modes, such as high-speed ice-crushing mode, low-speed stirring mode, dough-kneading mode, milkshake-making mode, and self-cleaning pulse mode, and defining specific control commands for each mode. A set of corresponding target control parameters is pre-stored, including motor speed curve, running time, and steering sequence. The motor speed curve describes the law of speed change over time, and the running time and steering sequence describe the alternating order of forward and reverse rotation. A mapping relationship between the control mode and the target control parameters is established and stored in the controller's non-volatile memory. Step S302: Confidence acquisition. Specifically, the feature vector generated in step S2 is input into the trained Gradient Boosting Decision Tree (GBDT) model. The model outputs the confidence score, a probability value or score, indicating that the current stirring state belongs to each preset control mode. The sum is not necessarily 1. In actual judgment, this confidence score can be directly used for comparison and threshold judgment. Subsequently, identification... And record the highest confidence level and its corresponding control mode; Step S303, the first-level judgment (high confidence direct decision) achieves fast and direct automatic decision-making when the model judgment is very confident, ensuring response speed. The specific steps include setting a first preset threshold, for example, setting it to 0.85. The basis for setting this threshold is that during the model training and validation phase, when the output confidence level is higher than this value, its classification accuracy is close to 100%, which can be regarded as a certain judgment. Perform the first-level judgment. If the highest confidence level is greater than or equal to the first preset threshold (0.85), then directly determine the control mode corresponding to this highest confidence level as the target control mode and jump to step S4 for execution. Otherwise, enter the second-level judgment. This step realizes that in most cases... Under clear and typical working conditions, it can instantly complete accurate pattern recognition and switching; Step S304, the second-level judgment (historical frequency-assisted decision-making), when the model's own judgment is not confident enough, introduces the user's historical behavior data as prior knowledge, so that the decision result is more in line with the user's personal habits and realizes personalized intelligence. The specific steps include, first, setting a second preset threshold, for example, setting it to 0.60, and the first preset threshold is less than the second preset threshold. The basis for setting the second preset threshold is to filter out all possible patterns that cannot be ignored, to avoid missing the correct option. Its value is lower than the first preset threshold to form a decision buffer. Then, the second-level judgment logic is executed to filter out patterns with a confidence level greater than or equal to the second preset threshold (0.60) from all preset patterns.If pattern 60) is selected as a candidate pattern, and at least one candidate pattern is selected, a historical frequency comparison rule is executed. Specifically, the historical operation database is queried, which continuously records the final execution pattern and timestamp of each mixing operation. The frequency of each candidate pattern is calculated within a preset time period (e.g., the last 30 days) from the current moment. This frequency is set to balance the timeliness of user habits with the stability of statistics, reflecting recent user preferences while avoiding excessive fluctuations in frequency statistics due to insufficient data. The single candidate pattern with the highest frequency is determined as the target control pattern. If multiple candidate patterns have the same frequency, [further action is taken]. If all are at their highest values, then the most recently run mode (i.e., the mode with the latest timestamp) is selected as the target control mode, and its corresponding control parameters are determined. If no candidate mode is selected (i.e., all confidence levels are below the second preset threshold), or if a candidate mode is selected but a unique target cannot be obtained based on the historical frequency comparison rules (e.g., the database is empty or the frequency cannot be compared), then the third-level decision is entered. This step effectively solves the decision-making dilemma when the model is uncertain. By incorporating user habits, the decision-making becomes more humanized and practical, improving the rationality of decision-making in complex and fuzzy scenarios. Step S305, the third-level decision (default rule as a fallback) serves as the entire... The safety net of the decision-making process ensures a definite output under any abnormal or marginal conditions, preventing crashes or infinite loops. Specifically, it executes a preset default rule, identifying the control mode corresponding to the highest confidence level of the gradient boosting decision tree output in step S302 as the target control mode and determining its corresponding control parameters. This default rule is designed so that, when no better decision-making basis is available, the model's output, even with low confidence, still represents the best guess based on the training data. Using this as the final output guarantees the integrity of the function. This step ensures the completeness and absolute reliability of the decision logic and is part of the entire intelligent decision-making process. The final guarantee of the process; Step S4, control execution, transforms the results of intelligent decision-making into physical world actions without loss, completing the control closed loop. Specific steps include: according to the target control mode determined by steps S303, S304, or S305, querying the mapping relationship established in step S301 to obtain the corresponding target control parameter set, including speed curve, duration, and steering sequence. Subsequently, the main controller converts the target control parameters into specific pulse width modulation (PWM) signals and digital switching signals, and sends them to the motor driver. The motor driver precisely controls the operation of the stirrer motor according to these electrical signals, ultimately achieving the regulation of the stirrer. Steps S3 and S4 together constitute a collaborative decision-making and control execution system that is model-driven, rule-assisted, history-based, and provides a safety net. Through a hierarchical and progressive judgment strategy, it organically combines the flexibility and intelligence of machine learning models, the empirical value of historical data, and the absolute reliability of deterministic rules. This overcomes the risk of misjudgment that may occur when relying solely on machine learning models at low confidence levels, and significantly improves the overall decision-making robustness, user adaptability, and operational safety of the smart mixer in real and ever-changing kitchen environments.

[0032] Example 2, refer to Figure 2 In another embodiment of the present invention, which differs from the first embodiment, a multi-mode operation control system for an intelligent stirrer is provided, including a fusion module, an enhancement module, a determination module, and a control module: The fusion module performs time alignment and fusion processing on the original data to obtain fused multi-dimensional time series data. It then performs time-series phase verification on the fused multi-dimensional time series data to obtain phase-corrected fused multi-dimensional time series data. Finally, it extracts features from the phase-corrected fused multi-dimensional time series data to obtain time-series statistical features and spectral analysis features. The enhancement module calculates the modal confidence coefficient of each original data, performs feature enhancement processing on the time series statistical features and spectral analysis features using the modal confidence coefficient, obtains enhanced feature values, and arranges the enhanced feature values ​​in a fixed order to obtain the feature vector; The decision module inputs the feature vector into the gradient boosting decision tree and outputs the confidence level of the current running state of the stirrer belonging to each preset control mode. Based on the confidence level, a multi-level decision strategy is used to make a decision to obtain the target control parameters. The control module converts control parameters into electrical signals and sends them to the motor driver, which then regulates the agitator.

[0033] In one embodiment, a complete closed loop from accurate sensing to reliable decision-making is constructed. First, the raw data of current, vibration, and sound are time-aligned and fused. Dynamic Time Warping (DTW) is then used for timing and phase verification, effectively overcoming timing misalignment issues caused by sensor response delays or mechanical transmission. This achieves the technical effect of obtaining high-quality, highly synchronized fused data, laying a precise time-domain foundation for subsequent analysis. Subsequently, a modal reliability coefficient based on multi-scale permutation entropy (MPE) is introduced to dynamically evaluate the stability of each sensor signal under the current operating conditions. Based on this, the extracted timing and spectral features are adaptively weighted and enhanced. This design aims to endow the system with the ability to distinguish between true and false signals at the feature level, significantly improving the input... The robustness of the feature vector representation allows the model to focus more on data modes with high confidence and suppress noise interference. Then, the enhanced feature vector is input into the Gradient Boosting Decision Tree (GBDT) for preliminary pattern recognition. This is supplemented by a multi-layer decision strategy that combines direct judgment with high confidence, historical frequency assistance, and default rule fallback. This ensures that the decision-making combines the intelligence of the model with the robustness of practical applications, greatly improving the reliability of decision-making under complex working conditions with fuzzy confidence. It can also integrate user habits to achieve personalized adaptation. Finally, the target control parameters obtained from the judgment, such as speed curves and steering sequences, are converted into drive signals for execution. This completes the process from intelligent analysis to physical control. The entire architecture is interconnected, enabling high-precision, highly adaptive, and highly reliable fully automatic intelligent control of the mixer in multiple modes.

[0034] This invention fundamentally solves the problem of deep temporal misalignment caused by mechanical delay or response differences in multi-source sensor data by using time-series phase verification based on dynamic time warping (DTW) distance. This achieves high-precision data fusion and provides a reliable data foundation for subsequent analysis. Simultaneously, based on a multi-scale permutation entropy (MPE) dynamic modal reliability evaluation mechanism, it quantifies the signal quality and stability of each sensing mode, such as current, vibration, and sound, in real time, and performs adaptive feature enhancement accordingly. This effectively suppresses the negative impact of noise or interference modes, significantly improving the robustness of input feature representation. Furthermore, the multi-layered decision-making strategy, which includes direct judgment with high model confidence, historical frequency assistance, and default rule fallback, greatly enhances the reliability of decision-making in complex and fuzzy conditions. This not only avoids misjudgments that may occur with a single model but also intelligently integrates user habits for personalized adaptation. Ultimately, this ensures that the mixer can accurately and stably switch to the optimal working mode automatically in various practical application scenarios, significantly improving the intelligence level, control precision, and user experience of the equipment.

[0035] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0036] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A method for controlling the multi-mode operation of an intelligent stirrer, characterized in that, Includes the following steps, Step S1: Perform time alignment and fusion processing on the original data to obtain fused multi-dimensional time series data; perform time-series phase verification on the fused multi-dimensional time series data to obtain phase-corrected fused multi-dimensional time series data; and extract features from the phase-corrected fused multi-dimensional time series data to obtain time-series statistical features and spectral analysis features. Step S2: Calculate the modal confidence coefficient of each original data, and perform feature enhancement processing on the time series statistical features and spectral analysis features through the modal confidence coefficient to obtain enhanced feature values. Arrange the enhanced feature values ​​in a fixed order to obtain the feature vector. Step S3: Input the feature vector into the gradient boosting decision tree, output the confidence level of the current running state of the stirrer belonging to each preset control mode, and use a multi-level decision strategy to make a determination based on the confidence level to obtain the target control parameters; Step S4: The control parameters are converted into electrical signals and sent to the motor driver, which then controls the agitator.

2. The intelligent stirrer multi-mode operation control method as described in claim 1, characterized in that, Step S1 includes steps S101, S102, S103 and S104; Step S101: The raw data includes current data, vibration data, and sound data; Step S102, the specific process of time alignment processing includes, for example, calculating the time offset between each original data and the stirrer master control clock, taking into account the inherent time deviation of the original data during acquisition; The original data are time-shifted and corrected according to the time offset to obtain the corrected timestamp sequence of each original data, which serves as intermediate data for phase pre-alignment. Based on the stirrer master clock, a time stamp sequence with a fixed frequency and equal intervals is generated as a unified timing reference, denoted as the target time stamp sequence; The specific process of fusion processing includes: relocating the pre-aligned intermediate data of each phase to the target timestamp sequence; selecting an interpolation window centered on the target timestamp sequence for each target time point; calculating and normalizing the signal-to-noise ratio of each intermediate data within the interpolation window to obtain the interpolation weight; and performing weighted average interpolation calculation on the values ​​of each intermediate data within the interpolation calculation window based on the interpolation weight to generate fused multi-dimensional time series data.

3. The intelligent stirrer multi-mode operation control method as described in claim 2, characterized in that, Step S103: Based on the rated rotation cycle of the stirrer, slide a preset time window on the fused multi-dimensional time series data. For the position of each time window, obtain the current data segment within the time window and obtain the corresponding reference data segment after a delay of one rated rotation cycle. Calculate the Dynamic Time Warping (DTW) distance between the current data segment and the reference data segment to obtain the optimal bending path; Based on the optimal curved path, obtain the index mapping relationship between the current data segment and the reference data segment; Calculate the normalized DTW distance of the optimal curved path as a temporal alignment metric; If the normalized DTW distance is less than the preset alignment threshold, the timing phase verification of the current time window is determined to be successful. Otherwise, the current time window verification is deemed to have failed, and the index mapping relationship is extracted. Based on the index mapping relationship, the current data segment is subjected to non-linear time resampling and alignment to obtain the corrected current data segment. After traversing the positions of all time windows, the corrected current data segments corresponding to each time window are merged in chronological order to obtain phase-corrected fused multi-dimensional time series data. Step S104: Extract time series statistical features and spectral analysis features from the phase-corrected fused multi-dimensional time series data.

4. The intelligent stirrer multi-mode operation control method as described in claim 3, characterized in that, Step S2 includes steps S201 and S202; Step S201, the calculation process of the modal confidence coefficient includes, for the current data, performing multi-scale permutation entropy (MPE) calculation on the current data in each interpolation window to obtain a set of current multi-scale permutation entropy values, performing normalization calculation on the set of current multi-scale permutation entropy values ​​to obtain the current complexity adjustment coefficient, and multiplying the current complexity adjustment coefficient by a preset sharpening factor to obtain the current exponent value. For vibration data, multi-scale permutation entropy (MPE) is calculated for vibration data within each interpolation window to obtain a set of vibration multi-scale permutation entropy values. The set of vibration multi-scale permutation entropy values ​​is then normalized to obtain a vibration complexity adjustment coefficient. The vibration complexity adjustment coefficient is multiplied by the sharpening factor to obtain the vibration index value. For the audio data, the multi-scale permutation entropy (MPE) is calculated for the audio data in each interpolation window to obtain a set of audio multi-scale permutation entropy values. The set of audio multi-scale permutation entropy values ​​is then normalized to obtain an audio complexity adjustment coefficient. The audio complexity adjustment coefficient is multiplied by the sharpening factor to obtain an audio index value. The sum of the current index, vibration index, and sound index is obtained by adding the current index, vibration index, and sound index together. Divide the current index value by the sum of the exponents to obtain the current mode reliability coefficient; Divide the vibration index value by the sum of the indices to obtain the vibration mode reliability coefficient; Divide the sound index value by the sum of the indices to obtain the sound modality confidence coefficient; The sum of the current mode confidence coefficient, vibration mode confidence coefficient, and sound mode confidence coefficient is 1.

5. The intelligent stirrer multi-mode operation control method as described in claim 4, characterized in that, Step S202: The time-series statistical features and spectral analysis features are grouped according to their data sources to obtain current feature group, vibration feature group and sound feature group; The data sources include current data, vibration data, and sound data; The current characteristic group includes current time-series statistical characteristics and current spectrum analysis characteristics; The vibration feature set includes vibration time-series statistical features and vibration spectrum analysis features; The sound feature group includes sound temporal statistical features and sound spectral analysis features; Based on the current mode confidence coefficient, vibration mode confidence coefficient, and sound mode confidence coefficient corresponding to the current feature group, vibration feature group, and sound feature group, respectively, feature enhancement processing is performed on all feature values ​​within the same group; The feature enhancement process involves scaling each feature value within the same group using the square root of the modality confidence coefficient corresponding to that group as a scaling factor to obtain a weighted feature value, which is denoted as the enhanced feature value. Wherein, the characteristic value is each of the values ​​of the current time-series statistical characteristic, the current spectrum analysis characteristic, the vibration time-series statistical characteristic, the vibration spectrum analysis characteristic, the sound time-series statistical characteristic, and the sound spectrum analysis characteristic; The enhanced feature values ​​are arranged in a fixed order to obtain a feature vector; The fixed order is: current time sequence statistical characteristics, current spectrum analysis characteristics, vibration time sequence statistical characteristics, vibration spectrum analysis characteristics, sound time sequence statistical characteristics, and sound spectrum analysis characteristics.

6. The intelligent stirrer multi-mode operation control method as described in claim 5, characterized in that, Step S3 includes steps S301, S302, S303, S304 and S305; Step S301: The preset control modes include high-speed ice crushing mode, low-speed stirring mode, dough kneading mode, milkshake making mode and self-cleaning pulse mode. Each control mode has a set of corresponding target control parameters stored in advance, and a control mapping relationship between the control mode and the corresponding target control parameters is established. The target control parameters include the motor speed curve, running time, and steering sequence; Based on the confidence level, a multi-layer decision-making strategy is adopted to determine the target control mode and the corresponding target control parameters.

7. The intelligent stirrer multi-mode operation control method as described in claim 6, characterized in that, Step S302, the multi-layer decision strategy specifically includes obtaining the confidence level of each preset control mode output by the gradient boosting decision tree, and identifying the highest confidence level and the corresponding control mode among the confidence levels; Step S303: Perform the first-level determination based on the highest confidence level; The first layer of judgment logic is as follows: if the highest confidence level is greater than or equal to the first preset threshold, then the control mode corresponding to the highest confidence level is taken as the target control mode, and the corresponding target control parameters are determined. Otherwise, proceed to the second level of judgment.

8. The intelligent stirrer multi-mode operation control method as described in claim 7, characterized in that, Step S304, the second layer of judgment logic is to select the modes with confidence greater than or equal to the second preset threshold as candidate modes from all preset control modes, wherein the second preset threshold is less than the first preset threshold; If one or more candidate modes are selected, the historical frequency comparison rule is executed. The historical frequency comparison rule is to query the historical operation database, obtain the frequency of each candidate mode in the historical operation records within the preset time period, and take the unique candidate mode with the highest frequency as the target control mode. If there are multiple candidate modes with the same highest frequency of occurrence, the candidate mode corresponding to the most recent run is selected from the multiple candidate modes with the same highest frequency of occurrence as the target control mode, and the corresponding target control parameters are determined. If no candidate patterns are selected, or if candidate patterns are selected but a unique target control pattern cannot be obtained based on the historical frequency comparison rules, then proceed to the third level of judgment. Step S305, the logic of the third layer determination is to obtain the target control mode and the corresponding target control parameters according to the preset default rules. The default rules are to take the control mode corresponding to the highest confidence level output by the gradient boosting decision tree as the target control mode and the corresponding target control parameters.

9. The intelligent stirrer multi-mode operation control method as described in claim 8, characterized in that, Step S4: According to the target control mode, call the control mapping relationship to obtain the corresponding target control parameters as the control parameter set, convert the control parameter set into an electrical signal and send it to the motor driver, and regulate the stirrer through the motor driver.

10. A multi-mode operation control system for an intelligent mixer, applied to a multi-mode operation control method for an intelligent mixer as described in any one of claims 1-9, characterized in that, It includes a fusion module, an enhancement module, a decision-making module, and a control module: The fusion module performs time alignment and fusion processing on the original data to obtain fused multi-dimensional time series data. It then performs time-series phase verification on the fused multi-dimensional time series data to obtain phase-corrected fused multi-dimensional time series data. Finally, it extracts features from the phase-corrected fused multi-dimensional time series data to obtain time-series statistical features and spectral analysis features. The enhancement module calculates the modal confidence coefficient of each original data, performs feature enhancement processing on the time series statistical features and spectral analysis features using the modal confidence coefficient, obtains enhanced feature values, and arranges the enhanced feature values ​​in a fixed order to obtain the feature vector; The decision module inputs the feature vector into the gradient boosting decision tree and outputs the confidence level of the current running state of the stirrer belonging to each preset control mode. Based on the confidence level, a multi-level decision strategy is used to make a decision to obtain the target control parameters. The control module converts control parameters into electrical signals and sends them to the motor driver, which then regulates the agitator.

Images