Alloy window self-adaptive control method and system based on multi-modal data

Abstract

The invention relates to the field of automatic control of doors and windows, in particular to an alloy window self-adaptive control method and system based on multi-modal data, and the method comprises the following steps: a data collection step: collecting facial feature data and posture feature data of a window opener, and an axial pressure value and a pressure change rate of a linkage rod; a timestamp, a windowing frequency and real-time windowing angle data; a windowing user judgment step: a user assistance control step: outputting a target boosting force parameter through a self-adaptive assistance calculation strategy and controlling a linkage rod to give a boosting force; and a visitor assistance control step: calling a window opening behavior model of a corresponding type to obtain a pressure value, calculating a boosting force, and controlling a linkage rod to give the boosting force, and through a mode of performing window opening assistance on a user, the operation smoothness and safety are improved, the service life of equipment is prolonged, and the maintenance cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of automatic control of doors and windows, and in particular to a method and system for adaptively controlling alloy windows based on multimodal data. Background Art

[0002] Traditional window designs often require users to exert considerable force when opening a window. This is especially true during the opening process, as the window's trajectory is typically semicircular. The direction of force applied by the user doesn't exactly align with the window's actual movement. This can lead to jamming and even damage to the window's sliding track due to improper force. Furthermore, window-opening habits and strength vary significantly among users, especially the elderly and children, who may struggle to open the window smoothly due to insufficient strength. This not only impacts the user experience but can also degrade the window's long-term performance.

[0003] While some existing smart window control systems can detect user actions through sensors and provide assistive force, these systems generally lack the ability to accurately identify and adapt to individual user differences. For example, existing systems often fail to perform personalized assistance calculations based on a user's historical window-opening behavior, nor do they offer differentiated assistance strategies for different user groups, such as the elderly and children. Furthermore, existing systems typically employ fixed assistance modes when handling uncommon users, such as visitors, and are unable to dynamically adjust based on the user's real-time operating status, resulting in poor assistance effectiveness.

[0004] Therefore, in order to solve the above problems, the present invention proposes an alloy window adaptive control method and system based on multimodal data. Summary of the Invention

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] An adaptive control method for alloy windows based on multimodal data, characterized by comprising the following steps: a data acquisition step, in which a camera is used to collect facial and body feature data of a person opening the window in real time, a pressure sensor is used to collect the axial pressure value and pressure change rate of the linkage rod in real time, and operation timestamps, window opening frequency, and real-time window opening angle data are recorded;

[0007] The window opening user identification step, when the infrared sensor detects a human approach signal, matches the facial feature data of the current window opener with the facial feature data of users pre-stored in the database through facial feature comparison. If the match is successful, the process enters the user assistance control step; otherwise, the process enters the visitor assistance control step;

[0008] The user power assist control step obtains the axial pressure value, pressure change rate and window opening angle data by calling the user's historical window opening behavior model, outputs the target power assist force parameters through the adaptive power assist calculation strategy, and controls the linkage rod to provide power assist force;

[0009] The visitor assistance control step classifies the window opener into a characteristic type according to the body feature data through a feature classification algorithm, calls the corresponding type of window opening behavior model to obtain the pressure value and calculate the assist force, and controls the linkage rod to provide the assist force.

[0010] As a further improvement of the present invention, the user-assisted control step also includes performing time series matching based on the current real-time system time and the historical window opening behavior model to obtain the window opening angle data of the current window opener at the corresponding time and the axial pressure value and pressure change rate applied by the current window opener to the window linkage rod. The time series matching includes matching the operation timestamp of the window opening operation recorded by the current window opener with the real-time system time of this window opening, and locking the window opening angle data, axial pressure value and pressure change rate in the historical window opening behavior model of the current window opener at the corresponding time.

[0011] As a further improvement of the present invention, it also includes a model optimization step. After each window opening operation is completed, the time characteristic data, pressure change curve and actual window opening angle data of this operation are input into the corresponding user behavior model, and the model parameters are updated through an incremental learning algorithm. The registered user model adopts an individualized parameter update strategy, and the guest user model adopts a group feature optimization strategy.

[0012] As a further improvement of the present invention, the visitor assistance control step also includes establishing a window opening action continuity model through time series analysis of the pressure change rate. When it is detected that the pressure value fluctuates intermittently within a predetermined time and the fluctuation amplitude exceeds a preset threshold, it is determined to be an operation pause state, and an operation continuity evaluation parameter is generated and the window opening instruction is completed according to the operation continuity evaluation parameter.

[0013] As a further improvement of the present invention, the visitor assist control step also includes, when it is determined that the operation is in a pause state, recording the angle difference between the initial pause position and the final stop position, calculating the angle compensation coefficient based on the angle difference, and correcting the visitor assist force value by correcting the angle compensation coefficient.

[0014] As a further improvement of the present invention, the adaptive power assist calculation strategy includes obtaining the historical average power assist force of the current window opener by calculating, and obtaining the current power assist force value based on the average power assist force combined with the real-time pressure change rate and the current axial pressure value.

[0015] As a further improvement of the present invention, the visitor assistance control step also includes a categorized assistance calculation strategy, which includes calculating the assistance value in the visitor assistance control based on the average assistance force baseline value of the population after the feature classification, combined with the real-time pressure change rate and the historical average value of the window opening angle.

[0016] As a further improvement of the present invention, it also includes a human-computer interaction control step. When it is detected that the real-time axial pressure value of the connecting rod is lower than the preset threshold and reaches the preset time, the fully automatic auxiliary mode is started, and the corresponding boost force is given to the connecting rod according to the current window opening user judgment result.

[0017] As a further improvement of the present invention, when the user is classified as an elderly or child group, the second-order derivative of the operating force is monitored in real time. When it is detected that it exceeds a preset safety threshold, the reverse braking mechanism is activated to apply reverse braking force to the window linkage rod.

[0018] An alloy window adaptive control system based on multimodal data, comprising:

[0019] The data acquisition module collects the facial and body feature data of the person opening the window in real time through a camera, collects the axial pressure value and pressure change rate of the linkage rod in real time through a pressure sensor, and records the operation timestamp, window opening frequency and real-time window opening angle data;

[0020] The window opening user identification module detects a human approach signal through the infrared sensor and matches the facial feature data of the current window opener with the facial feature data of users pre-stored in the database through facial feature comparison. If the match is successful, the module enters the user assistance control step; otherwise, the module enters the visitor assistance control step;

[0021] The user power assist control module obtains the axial pressure value, pressure change rate and window opening angle data by calling the user's historical window opening behavior model, outputs the target power assist force parameters through the adaptive power assist calculation strategy, and controls the linkage rod to provide power assist force;

[0022] The visitor assistance control module classifies the window opener into a characteristic type according to the body feature data through a feature classification algorithm, calls the corresponding window opening behavior model to obtain the pressure value and controls the linkage rod to provide an assist force.

[0023] (1) Reduce the force intensity: effectively reduce the force intensity when the user opens the window, avoid mechanical jamming due to deviation in the force direction, and make the window opening operation easier.

[0024] (2) Accurately adapt to different users: Through multimodal data collection and analysis, accurately identify user identities and behavioral habits. For registered users, combine historical window opening behavior models and real-time data to provide personalized assistance, reducing the learning cost of repeated operations; for visitors, classify them according to their body characteristics and call the corresponding type of window opening behavior model to solve the problem of assistance adaptation for unknown users, so that the assistance is more in line with the actual needs of different users.

[0025] (3) Improve operational fluency and safety: Monitor the operating status in real time. When abnormal pressure changes are detected, dynamically correct the boost force output curve to ensure that the power assist direction is synchronized with the window movement trajectory. When the visitor pauses in operation, accurately distinguish between temporary pause and operation termination, maintain the boost force output or stop the power assist in time to avoid operational difficulties caused by power assist interruption and mechanical damage to the window caused by continuous force applied beyond the expected angle. At the same time, for the elderly or children, the second-order derivative of the operating force is monitored in real time. When an abnormality is detected, the reverse braking mechanism is activated to prevent the window from opening rapidly or being damaged by impact, thereby ensuring user operation safety.

[0026] (4) Extending the service life of the equipment: Collaborative analysis of multimodal data ensures the dynamic matching of the power output and the window motion trajectory, reduces abnormal wear of mechanical components, significantly extends the service life of the window slide system, and reduces equipment maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a flow chart of an alloy window adaptive control method based on multimodal data of the present invention;

[0028] Figure 2 This is a flow chart of the visitor assistance control steps of the present invention;

[0029] Figure 3 It is a flow chart of the adaptive power-assistance calculation strategy of the present invention;

[0030] Figure 4 It is a flow chart of the human-computer interaction control steps of the present invention;

[0031] Figure 5 It is a block diagram of an alloy window adaptive control system based on multimodal data of the present invention. DETAILED DESCRIPTION

[0032] The present invention will be described in further detail below with reference to the accompanying drawings and embodiments. Identical components are denoted by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "upper," and "lower" used in the following description refer to directions in the accompanying drawings, and the terms "bottom," "top," "inner," and "outer" refer to directions toward or away from the geometric center of a particular component, respectively.

[0033] Conventional window power-assist systems often rely on a single mechanical sensor for thrust compensation, lacking the ability to identify and analyze user behavior. Fixed-parameter power-assist systems struggle to adapt to individual differences among operators of varying heights and body types, leading to power-assistance lag or overcompensation. This is particularly true in multi-family settings, where the window thrust requirements of young and elderly individuals differ significantly. Existing technologies are unable to dynamically adapt, resulting in poor operational fluidity and accelerated wear of mechanical components.

[0034] In order to solve the above problems, this application proposes the following solutions: Figure 1 As shown, it includes a data collection step, in which the facial feature data and body feature data of the window opener are collected in real time through a camera, and the axial pressure value and pressure change rate of the connecting rod are collected in real time through a pressure sensor, and the operation timestamp, window opening frequency and real-time window opening angle data are recorded; a window opening user identification step, in which when a human body approaching signal is detected by an infrared sensor, the facial feature data of the current window opener is matched with the user facial feature data pre-stored in the database through facial feature comparison. If the match is successful, the user power assistance control step is entered; otherwise, the visitor power assistance control step is entered; a user power assistance control step, in which the axial pressure value, pressure change rate and window opening angle data are obtained by calling the historical window opening behavior model of the user, the target assist force parameter is output through an adaptive assist calculation strategy, and the connecting rod is controlled to provide assist force; a visitor power assistance control step, in which the window opener is classified into a feature type according to the body feature data through a feature classification algorithm, the corresponding type of window opening behavior model is called to obtain the pressure value and calculate the assist force, and the connecting rod is controlled to provide assist force.

[0035] Among them, facial feature data refers to biometric parameters such as facial contours and the distance between facial features collected by the camera. Specifically, a convolutional neural network can be used to extract feature vectors to distinguish different user identities. Body feature data includes height estimation and limb proportion parameters, which can be obtained through a skeletal key point detection algorithm to assist in determining the user's strength level. The axial pressure value refers to the mechanical sensor measurement value in the direction of the connecting rod axis, which can be implemented using a strain gauge sensor to reflect the actual force applied by the user. The pressure change rate is obtained by differential calculation of continuous sampling data from the pressure sensor to identify the user's force trend. The timestamp record uses the real-time clock module of the embedded system to associate historical operation period characteristics.

[0036] Specifically, when the infrared sensor detects a person approaching, the system activates a multimodal data collection process. The facial image captured by the camera is then compared with a registration database after feature extraction to confirm the operator's identity. For users who successfully match, the system retrieves the force application pattern matching the current time from their historical operation database and dynamically adjusts the power assist output based on real-time pressure change trends. When a visitor is identified, the system uses the body posture analysis results to match the user to a demographic database. For example, users below a height threshold are classified as children, and the average force application model for this group is used to generate baseline power assist parameters. A pressure sensor monitors the force state of the linkage rod in real time. If an abnormal pressure change rate is detected, the power assist output curve is dynamically modified to ensure that the power assist direction is synchronized with the window's motion trajectory. Compared to existing solutions, traditional solutions only perform linear compensation based on real-time mechanical parameters and cannot distinguish between operator identity and operating habits. This solution integrates biometric recognition with historical behavioral data to achieve personalized power assist output. While existing technologies require relearning user characteristics when a visitor operates, this solution, by establishing a demographic classification model, can provide appropriate power assist support during initial contact.

[0037] Through the above technical solutions, this application can effectively reduce the force applied by users when opening windows, avoiding mechanical jamming caused by deviation in the direction of force. The temporal feature matching for registered users reduces the learning cost of repeated operations. The group feature classification mechanism for guest users solves the problem of power assistance adaptation for unknown users. The collaborative analysis of multimodal data ensures the dynamic matching of power output and window motion trajectory, significantly extending the service life of the slide rail system.

[0038] This application further proposes to perform timing matching by combining the current real-time system time with the historical window opening behavior model in the user assistance control step to obtain the window opening angle data, axial pressure value and pressure change rate of the current window opener at the corresponding time, wherein the timing matching locks the relevant data in the historical window opening behavior model by matching the operation timestamp with the real-time system time.

[0039] Among them, timing matching refers to a processing mechanism that establishes an association between real-time time and historical operation time. It can be implemented by a timestamp matching algorithm. Its function is to eliminate the impact of differences in user behavior in different time periods on the assistance strategy; the operation timestamp refers to the precise time mark that records each window opening operation of the user. It can be generated by the system clock module and used to construct a data index in the time dimension; the real-time system time refers to the actual moment when the current window opening operation occurs. It can be obtained by the network timing protocol and used to compare the time window with historical data; the axial pressure value refers to the pressure value detected along the axis of the connecting rod. It can be measured by a strain sensor. Its function is to reflect the user's force intensity characteristics; the pressure change rate refers to the change in the pressure value per unit time. It can be calculated by the differential algorithm to calculate the slope of the pressure sensor signal, which is used to characterize the dynamic characteristics of the user's force.

[0040] Specifically, when the system detects a match in user identity, it extracts the current time information through the time calibration module and retrieves the user's operation records in the same time period from the historical database. For example, if the current time is 9 a.m., the user's window angle average, pressure peak and its changing pattern in the 8-10 a.m. period are preferentially retrieved. Through the time window comparison algorithm, the real-time time and the historical operation timestamp are similarly calculated to filter out the subset of historical data with the closest time characteristics. In this way, the system can accurately obtain the user's typical force application pattern in a specific period of time, and generate time-adaptive thrust parameters based on the real-time feedback data of the current pressure sensor. When there are obvious behavioral differences between users in different time periods, this mechanism can avoid the mistaken application of the midday force application mode to the early morning scene, ensuring that the power output is synchronized with the user's current biorhythm.

[0041] Compared with existing technologies, traditional power-assistance systems usually only match data based on user identity, ignoring the impact of time factors on force application habits. For example, the window opening angle requirements of the same user in the morning and at noon may be significantly different, and existing technologies cannot respond to this dynamically. This solution introduces a matching mechanism between timestamps and real-time time, and superimposes time trajectory analysis on identity recognition, so that the generation process of power-assistance parameters not only associates individual user characteristics, but also integrates time context information. This dual-dimensional data mapping relationship can effectively solve the power-assistance mismatch problem caused by traditional methods ignoring time patterns.

[0042] Through the above technical solution, this application can dynamically adjust the power assist parameters based on the user's behavioral characteristics in different time periods, avoiding misjudgments caused by missing time dimension data. For example, during the period when the user habitually opens the door at a specific angle, the system can preload the corresponding pressure change curve in advance, thereby quickly generating a matching power assist output when a similar force application pattern is detected. This mechanism can significantly reduce abnormal wear of mechanical components caused by power assist lag or deviation, while improving user operating comfort in different time scenarios.

[0043] This application further proposes that after each window opening operation is completed, the time characteristic data, pressure change curve and actual window opening angle data of this operation are input into the corresponding user behavior model, and the model parameters are updated through an incremental learning algorithm. The registered user model adopts an individualized parameter update strategy, and the guest user model adopts a group feature optimization strategy.

[0044] Temporal feature data refers to time series information including operation timestamps and operation time periods. Time series analysis methods can be used to extract periodic features to capture the temporal regularity of user window-opening behavior. The pressure change curve refers to the dynamic trajectory of the lever pressure during operation. This can be generated using continuous sampling data from a pressure sensor and reflects the force and rhythm of the user's operation. Actual window opening angle data refers to the final window opening angle achieved. This can be obtained using an angle sensor or image recognition algorithm and used to verify the consistency of the operation results with the intended target. Incremental learning algorithms are machine learning algorithms that support online updates. These can employ online gradient descent or sliding window techniques to optimize model parameters in real time. Individualized parameter update strategies are model parameter adjustment methods for registered users. These can employ a weighted average algorithm to fuse new and historical data to maintain the uniqueness of user behavior. Group feature optimization strategies are model optimization methods for guest users. Cluster analysis can be used to extract common group features to overcome poor model generalization due to insufficient individual data.

[0045] Specifically, the model optimization step forms a feature vector set by collecting multi-dimensional dynamic data during the operation process, and automatically triggers the model update process after each window opening operation. For the registered user model, the time characteristics of the current operation are aligned with the historical operation records, and the effective data fragments are screened through the sliding window mechanism. The model parameters are gradually updated using the exponential decay weighted algorithm to retain long-term behavioral characteristics while incorporating the latest operation mode. For the guest user model, the current operation feature vector is first matched with the group feature library for similarity. After identifying the user category, the mean shift clustering algorithm is used to update the position parameters of the category center point, and the category boundary threshold is adjusted to optimize the classification accuracy. An online verification mechanism is used in the incremental learning process to perform error analysis on the model prediction results and the actual operation data, and the learning rate parameters are dynamically adjusted to balance the convergence speed and stability.

[0046] The individualized parameter update strategy includes, first, performing time series alignment and data screening, nonlinearly aligning the current operation timestamp with the time axis of the historical operation record, identifying the operation mode of similar time periods (such as 7-8 am), eliminating the deviation caused by the fluctuation of the operation time, and dynamically adjusting the window size according to the user's average daily operation frequency (for example, if the user opens the window 3 times a day on average, the window width is set to 9 operation records), and only retaining the stable operation fragments with the pressure change rate fluctuation within the window less than the preset threshold (such as ±15%); then dynamically assigning feature weights, calculating the attenuation coefficient according to the interval between the current operation time and the historical record, and the longer the interval, the lower the weight of the historical data (for example, 24 hours ago). The data weight decays to 30% of the original value), and by comparing the difference between the actual window opening angle and the model predicted angle, the credibility weight is calculated (for example, when the angle deviation is less than 5 degrees, a 90% weight is given, and when the deviation exceeds 15 degrees, only a 50% weight is retained). Finally, the parameters are progressively fused, and the historical model parameters and the current operation data are weighted averaged according to the time decay factor and the credibility weight to ensure that long-term behavioral characteristics (such as habitual window opening angles) are not covered by short-term abnormal operations. The gradient direction of the model prediction error is calculated based on the current operation data, and an adaptive learning rate (the larger the error, the higher the learning rate) is used to fine-tune the parameters along the negative direction of the gradient, but the single update amplitude is limited to no more than 20% to prevent overfitting. The group feature optimization strategy involves constructing a multidimensional feature space, integrating temporal features (e.g., encoding operation time periods as morning / noon / evening), pressure dynamics (mean, variance, peak rate of change), and operational behavior (number of pauses, total duration) into a seven-dimensional feature vector. Z-score normalization is performed on features of different dimensions to eliminate the impact of numerical range differences on clustering. Dynamic clustering and group updates are then performed, and the KL divergence between the current operational features and the existing group feature library is calculated to measure distribution similarity. When the similarity exceeds 70%, the user is considered to belong to the group category. For successfully matched group categories, the category center position is dynamically adjusted based on the current operation duration ratio. For example, a time-consuming windowing operation will cause the category center to shift toward the "slow operation" direction. When the similarity is lower than the threshold, a new group category is created using the current feature as the seed, with the initial confidence set to 50%. It requires five subsequent similar operations to be verified before it can be confirmed as a stable category. The classification boundary is then optimized, and the statistical distance threshold is calculated based on the sample distribution within the group. When new samples increase the dispersion of group characteristics (such as a 20% increase in the standard deviation), the classification boundary is automatically tightened, and abnormal samples are filtered. Operation data that deviates from the group center by more than 3σ for three consecutive times is marked as an outlier and an independent analysis process is initiated. When a group data accumulates more than 100 items, its principal component features are extracted (if it is found that the group prefers fast windowing), and common features are injected into similar group models (such as the elderly user group) to achieve model construction.

[0047] Compared with existing technologies, traditional user behavior model updates typically rely on periodic batch training, which suffers from poor data timeliness and high computational resource consumption, and fails to differentiate between the optimization needs of individual and group users. This solution achieves real-time online updates through incremental learning, effectively improving the model's responsiveness to changes in user behavior. Furthermore, through differentiated strategies for individual updates for registered users and optimization for guest user groups, it improves the precision of window-opening assistance control for unknown users while maintaining personalized service levels.

[0048] Through this technical solution, the user behavior model can continuously adapt to changes in user operation patterns, avoiding the accumulation of control errors caused by long-term data drift. The personalized update mechanism of the registered user model ensures that the power-assistance control parameters are closely aligned with user habits, while the group optimization strategy of the guest user model enhances the system's adaptability to unfamiliar users, ultimately improving the overall intelligence and robustness of the window power-assistance control system.

[0049] This application further proposes visitor assistance control steps, such as Figure 2 As shown, it also includes establishing a window opening action continuity model through time series analysis of the pressure change rate. When it is detected that the pressure value fluctuates intermittently within a predetermined time and the fluctuation amplitude exceeds a preset threshold, it is determined to be an operation pause state, and an operation continuity evaluation parameter is generated and the window opening instruction is completed according to the operation continuity evaluation parameter.

[0050] Time series analysis of the pressure change rate refers to analyzing the dynamic trend of continuous pressure data collected by the pressure sensor along the time dimension. Specifically, a sliding window algorithm is used to segment the pressure change rate and extract fluctuation characteristics in combination with discrete wavelet transforms. This analysis can capture abnormal fluctuation signals that indicate discontinuous movements during force application. Intermittent fluctuations refer to discontinuous changes in pressure values along the time axis. Specifically, this can be detected by determining whether the difference in the pressure change rate between adjacent sampling points exceeds a fluctuation amplitude threshold. For example, a fluctuation event is identified when the difference in the pressure change rate between two adjacent sampling periods exceeds 30%. The operation continuity assessment parameter is a quantitative indicator generated by statistically analyzing the frequency, duration, and amplitude of fluctuation events within a predetermined time window. Specifically, a weighted summation algorithm is used to map multiple fluctuation characteristics into a single assessment value, which indicates whether the current operation is temporarily paused or terminated. The preset threshold is a boundary value that is dynamically adjusted based on the statistical distribution of operating force among different user groups. For example, the preset fluctuation amplitude threshold for elderly users is 25%, and for children it is 20%.

[0051] Specifically, when the pressure sensor detects changes in the axial pressure of the linkage rod, it collects real-time pressure change rate data and inputs it into the time series analysis module. The pressure change rate is segmented using a sliding time window, and the discrete wavelet transform is used to extract fluctuation characteristics within each time window, including amplitude, frequency, and duration. If multiple fluctuation amplitude events exceeding a preset threshold are detected within a set time window, the current operation is deemed to be at risk of stalling. An operation continuity assessment parameter is then generated by summing the cumulative number of fluctuation events and the sum of the amplitudes. This parameter is mapped to a continuous range from 0 to 1, where 0 indicates complete termination of the operation and 1 indicates continuous operation. When the assessment parameter is below 0.5, the operation is deemed terminated and the assist force output is stopped. When the assessment parameter is between 0.5 and 0.8, the operation is deemed to be temporarily paused and the current assist force is maintained. When the assessment parameter is above 0.8, the operation is deemed to be continuous and the original assist strategy is executed. This system dynamically identifies the user's operational intent, avoiding misjudgments due to insufficient physical strength.

[0052] The window opening action continuity model is configured as follows:

[0053]

[0054] Where C is the operational continuity assessment parameter, with a value range of [0,1], where 1 indicates complete continuity and 0 indicates complete interruption; t is the current time; Δt is the time span of the sliding window; DWT(·) is the discrete wavelet transform function; P'(τ) is the time derivative of the pressure change rate; σ is the standard deviation of the pressure signal; Γ(·) is the characteristic scale normalization function; Φ(·) is the multi-factor coupling function; N is the number of sampling points in the detection window; H(·) is the step function; ΔP'k is the pressure change rate difference at the kth sampling point; λk is the dynamic fluctuation amplitude threshold; κ is the time attenuation coefficient; τk is the duration of the kth fluctuation; M is the number of significant fluctuation events; β is the amplitude nonlinear adjustment factor; δm is the amplitude of the mth fluctuation event; ξm is the time interval between adjacent fluctuation events; arctan(·) is the inverse tangent constraint function; and log(·) is the natural logarithmic damping term. The wavelet transform coefficients in the window are scale-normalized by using the Gamma function, where: ψ is the Morlet wavelet basis function, a=2 is the scale parameter, and b=Δt is the denominator of the translation parameter Used to eliminate signal noise interference.

[0055] The first summation uses the Heaviside function to filter out the value exceeding the dynamic threshold λ k Volatility events: Basic threshold λ base =0.3, T = 24-hour period, exponential term A weighted mechanism (k=0.05) that decays over time is implemented, the arctan function is introduced into the product term to constrain the nonlinear growth of the fluctuation amplitude, the Sigmoid function is used for dynamic adjustment, and the logarithmic term suppresses the influence of short-interval high-frequency fluctuations.

[0056] Compared with existing technologies, existing technologies usually only judge the operation status through a single dimension of the absolute value or change rate of the pressure value, and cannot effectively distinguish between normal force fluctuations and real pause events. For example, some methods use a fixed time threshold to judge the interruption of the operation, and determine that it is terminated when the pressure value is lower than the threshold for more than 5 seconds, but this method cannot handle scenarios with frequent intermittent force application. This solution combines time series analysis with multi-dimensional fluctuation feature extraction to dynamically capture the differences in continuity and discontinuity of operating behaviors, and uses operation continuity evaluation parameters to achieve refined state discrimination. This method avoids the misjudgment problem caused by static thresholds and improves the accuracy of pause recognition.

[0057] Through the above technical solution, this application can accurately distinguish between temporary pauses and operation termination states when the force applied by uncommon window openers is discontinuous due to lack of physical strength. By dynamically analyzing the fluctuation characteristics of the pressure change rate, the system can maintain the thrust output during the user's short rest, avoiding operational difficulties caused by the interruption of the assist force. At the same time, when the real termination intention is detected, the assist force is stopped in time to prevent the window from being mechanically damaged due to continuous force applied beyond the expected angle. This technical solution significantly improves the smoothness and safety of window opening operations for non-registered user groups.

[0058] The present application further proposes that when the operation is determined to be in a pause state, the angle difference between the initial pause position and the final stop position is recorded, the angle compensation coefficient is calculated based on the angle difference, and the visitor assist force value is corrected by the angle compensation coefficient correction.

[0059] Among them, the operation pause state refers to the state judgment generated when the pressure sensor detects that the pressure value fluctuates intermittently within a predetermined time and the fluctuation amplitude exceeds the preset threshold. Specifically, it can be achieved by using a time series analysis algorithm to establish a window opening action continuity model. The model judges the operation continuity by analyzing the fluctuation characteristics of the pressure change rate. The angle difference refers to the difference between the window opening angle at the initial pause and the window opening angle at the final stop position. Specifically, it can be achieved by using an angle sensor to collect window opening angle data in real time and through a difference calculation module. The difference calculation module obtains a numerical value by comparing the absolute difference of the two angle measurements. The angle compensation coefficient refers to a proportional correction factor generated according to the size of the angle difference. Specifically, the linear regression algorithm or the piecewise function mapping relationship can be used to convert the angle difference into a compensation coefficient. The compensation coefficient acts on the correction term in the thrust calculation formula.

[0060] Specifically, after detecting that the pressure fluctuation triggers the operation pause judgment, the system continues to collect window opening angle data until the operation is completely terminated. The angle data of the initial pause position and the angle difference of the final stop position are input into the compensation calculation module, which generates a corresponding compensation coefficient based on the absolute value of the difference. For example, when the angle difference is 5 degrees, the compensation coefficient can be set to 1.1. This coefficient will be superimposed on the baseline thrust parameter of the subsequent power assist calculation link, so that the next time the same user operates, the thrust output by the system will increase by 10%. In the specific implementation, the mapping relationship between the compensation coefficient and the angle difference is realized through a preset compensation coefficient table, which is established based on the operation pause test data of different user groups.

[0061] Compared to existing technologies, traditional power-assisted control systems only perform linear compensation based on real-time pressure data, failing to account for deviations in actual window opening amplitude caused by operational interruptions. In existing technologies, when a user interrupts an operation due to lack of strength, the system fails to recognize the angle difference before and after the interruption, resulting in a mismatch between subsequent power-assisted parameters and the user's actual needs. This solution dynamically captures the angle change during operational pauses and establishes an interruption compensation mechanism, enabling the power-assisted control system to proactively correct errors caused by inconsistent operation.

[0062] Through the above technical solution, this application effectively solves the problem of inaccurate power assist parameter adjustment during intermittent operation by unregistered users. By quantitatively analyzing the angular deviation before and after the pause in operation, dynamic compensation correction of the power assist parameters is achieved. This solution enables the power assist system to adjust output parameters based on the user's actual operation trajectory, avoiding incomplete window opening due to insufficient power assist or excessive wear of the mechanical structure due to power assist overload.

[0063] This application further proposes an adaptive power-assisted computing strategy, such as Figure 3 As shown, the historical average thrust of the current window opener is obtained by calculation, and the current thrust value is calculated based on the average thrust combined with the real-time pressure change rate and the current axial pressure value.

[0064] Among them, the historical average thrust refers to the thrust baseline value obtained based on the user's historical operation data statistics. Specifically, the sliding window averaging method or the weighted average method can be used to calculate the average thrust in several past window opening operations to reflect the user's long-term force application habits.

[0065] The real-time pressure change rate refers to the rate at which the axial pressure of the connecting rod changes with time during the current operation cycle. Specifically, a differential calculation method can be used to process the continuous sampling data of the pressure sensor to capture the dynamic characteristics of the user's operation.

[0066] The current axial pressure value refers to the real-time collected axial pressure measurement data of the connecting rod, which can be obtained through a strain gauge pressure sensor installed on the connecting rod and is used to represent the user's current actual force application state.

[0067] Specifically, when calculating the thrust, the user's historical operation records are first extracted from the database. After eliminating abnormal data through time series screening, the average thrust is calculated to generate personalized benchmark parameters. The real-time axial pressure value is then input into the differential calculation module to obtain the pressure change per second as a dynamic adjustment factor. Finally, the historical average thrust, the real-time pressure change rate and the current axial pressure value are weighted and fused through a linear superposition algorithm to output the thrust command value at the current moment. In this process, the synergy between historical data and real-time parameters enables the system to maintain the consistency of user operating habits while compensating for sudden force changes.

[0068] The thrust calculation configuration is:

[0069]

[0070] Where Fc is the boost force value after compensation, with a range of [0.2Fb, 1.8Fb], Fb is the baseline force value, >1 indicates enhanced compensation, and <1 indicates reduced force correction; Ψ(·) is the normalization function of the historical data spectrum; θ is the adaptive time window parameter; H(ν) is the Hermite polynomial expansion of the historical boost force; ζ is the time series attenuation factor; π is the circumference of the circle; is the feature fusion operator; erfi(·) is the imaginary error function; ιι is the pressure change rate sensitivity coefficient; is the pressure change rate; ∈ is the pressure baseline calibration quantity; sech(·) is the hyperbolic secant function; is the nonlinear compression factor; Pτ is the observed value of the current axial pressure processed by time delay τ.

[0071] Compared to existing technologies, traditional power-assistance control methods typically perform single-dimensional calculations based on real-time pressure data or fixed historical thresholds, making it impossible to distinguish individual user differences or respond to changes in operating rhythm. This solution, by building a multi-source data fusion model, for the first time combines long-term behavioral characteristics with instantaneous operating status, effectively addressing the problem of power-assistance hysteresis caused by differences in user force application habits.

[0072] Through the above technical solution, this application can dynamically adjust the baseline power assist value according to the user's historical operating rules, and at the same time perform fine-tuning based on the real-time force application trend, so that the power assist output is consistent with the user's inherent operating mode and can adapt to the actual needs in different scenarios, significantly reducing the window jamming caused by insufficient power assist or overload.

[0073] This application further proposes setting up a categorized assistance calculation strategy in the visitor assistance control step. This strategy calculates the visitor's assistance value based on the average assistance force baseline value of the population after feature classification, combined with the real-time pressure change rate and the historical average value of the window opening angle.

[0074] Among them, the categorized power assist calculation strategy refers to the power assist calculation rules established for groups of uncommon window openers with different characteristic types. Specifically, it can be implemented by using a statistical modeling method based on group behavior data. By dividing window openers into multiple group categories according to age and body attributes, a corresponding benchmark parameter library is established for each category. The average power assist benchmark value refers to the average power assist level required by the same group of people in historical operations. Specifically, the time-weighted average algorithm can be used to calculate the same historical data to eliminate data fluctuations caused by individual differences. The real-time pressure change rate refers to the instantaneous change rate of the pressure on the connecting rod during the operation. Specifically, it can be obtained by using differential calculation after collecting time series data through a pressure sensor to reflect the instantaneous force characteristics of the operator. The historical average value of the window opening angle refers to the average value of the window opening angle achieved by the same group of people in historical operations. Specifically, the sliding window averaging method can be used to statistically analyze the same operation records in the database to establish an angle change benchmark.

[0075] Specifically, after the guest user completes the feature classification, the pre-stored average assist force benchmark value is first called according to the group category to which they belong. This benchmark value is obtained by analyzing the pressure data distribution characteristics in the historical operations of the same group of people. Then, the pressure change rate of the linkage rod is collected in real time, and the benchmark value is corrected in real time by dynamically capturing the mutation characteristics during the force application process. At the same time, combined with the average window opening angle formed in the historical operations of the same group of people, a correlation model between angle change and assist force demand is established. For example, when it is detected that the real-time pressure change rate of the elderly user group shows a fluctuation pattern of first rising and then falling, the system will automatically increase the assist force value in the initial stage and gradually reduce the assist force output when the angle approaches the average window opening angle of this group of people. This forms a composite calculation model based on group behavior patterns, dynamic operation characteristics and historical angle data, so that the assist force output not only conforms to the common characteristics of the group, but also can adapt to real-time operation changes.

[0076] The categorized power assist calculation and categorized power assist value configuration are as follows:

[0077]

[0078] Among them, F g is the categorized thrust force value, with a range of [0.5F0, 2.0F0], where F0 is the baseline force, >1 indicates enhanced compensation, and <1 indicates reduced force correction; Ψ(·) is the group baseline normalization function; N is the number of samples in the same group; B n is the baseline boost of the nth sample; J m(·) is the mth-order Bessel function of the first kind; θ n is the window angle of the nth sample; Φ(·) is the pressure dynamic coupling function; is the pressure change acceleration; κ is the time series convolution factor; Θ(·) is the angle correlation function; Li 1 / 2 (·) is a half-order multi-logarithmic function; η is the angle sensitivity coefficient; Δθ is the deviation between the real-time angle and the historical mean; λ is the pressure gradient damping coefficient; is the pressure change rate gradient.

[0079] The model classifies visitor users according to their body shape, and then uses Bessel functions to perform angle correlation modeling on the historical baseline thrust of the group. It uses Gaussian convolution kernel integral to process the pressure change acceleration signal to capture the sudden force characteristics, and combines multiple logarithmic functions to establish a nonlinear compensation mechanism for angle deviation. It also introduces a pressure gradient damping term to suppress high-frequency jitter interference, and finally generates a dynamically adjusted categorized thrust value through the composite operation of the group baseline normalization function, the pressure dynamic coupling function and the angle correlation function. The value range is continuously adjustable between 50% and 200% of the baseline value, realizing intelligent power output that retains the common characteristics of similar user operations and can respond to pressure mutations and angle deviations in real time. When the pressure acceleration is detected to exceed 10kPa / s 2 The enhanced compensation mode is triggered when the angle deviation exceeds 15°, and the safety force reduction mechanism is activated when the angle deviation exceeds 15°, ensuring the adaptive control stability and safety of the system in different operating scenarios.

[0080] Compared with existing technologies, traditional visitor window-opening assistance solutions usually only use fixed assist force values or single-dimensional pressure data calculations, and are unable to distinguish the operational characteristics of different groups of people. For example, some solutions directly use a preset universal assist force curve, resulting in insufficient thrust for elderly users or overload for child users. This solution, by establishing a group behavior model and a dynamic parameter fusion mechanism, not only solves the problem of calculation benchmarks when there is no individual historical data, but also captures operational dynamics through real-time pressure changes, achieving differentiated adaptation for groups of people with different characteristics. Compared with the calculation method that relies solely on individual historical data, this solution significantly improves calculation accuracy and operational adaptability in scenarios where data is missing.

[0081] Through the above technical solution, this application can establish a reasonable thrust calculation benchmark based on group behavior patterns in the absence of individual operation data, and dynamically adjust the output parameters based on real-time pressure changes, effectively solving the problem of poor thrust adaptation caused by individual differences among users who rarely open windows. At the same time, by introducing the average of historical window opening angles, it avoids angle misjudgments caused by differences in operating habits and reduces the risk of mechanical damage to windows caused by improper assist.

[0082] This application further proposes human-computer interaction control steps, such as Figure 4 As shown, when it is detected that the real-time axial pressure value of the connecting rod is lower than the preset threshold and reaches the preset time, the fully automatic auxiliary mode is started, and the corresponding assist force is given to the connecting rod according to the current window opening user judgment result.

[0083] The real-time axial pressure value refers to the force along the axial direction of the linkage rod, as measured in real time by a pressure sensor. This can be achieved by measuring the force using a strain gauge pressure sensor and converting it into an electrical signal, representing the actual thrust currently applied by the user. The preset threshold refers to the pre-set critical pressure value that triggers the fully automatic assist mode. The threshold range can be determined through experimental testing or historical data analysis, for example, by setting it to 20%-30% of the average force applied by a normal user, to determine whether the user is applying insufficient force. The preset time refers to the duration of continuous insufficient force required to trigger the fully automatic mode. This can be achieved by using a timer module to record the length of time the pressure value remains below the threshold, for example, by setting it to 1.5 to 3 seconds to eliminate the effects of transient pressure fluctuations. The fully automatic assist mode refers to the state in which the system actively takes over control of the window opening power output. This can be achieved by the intelligent control module outputting a preset constant assist force or a dynamically adjusted assist force sequence to the linkage rod to maintain the continuity of the window opening action. The windowed user identification result refers to the result of determining the user identity or type through facial feature comparison or body shape classification. Specifically, it can be achieved by using a convolutional neural network for feature matching or a support vector machine for group classification, which is used to differentiate the corresponding boost parameters.

[0084] Specifically, when the pressure sensor detects that the axial pressure value of the linkage rod is continuously lower than the preset threshold value for more than the preset time, the system determines that the user's force is insufficient to maintain normal window opening operation. At this time, the intelligent control module terminates the user's manual control mode and switches to the fully automatic assistance mode, and the system takes over the drive control of the linkage rod. While starting the fully automatic assistance mode, the system selects the corresponding boost force calculation strategy based on the judgment result of the current user: if the user is an identified registered user, the personalized boost parameters stored in the historical window opening behavior model are called; if the user is an unidentified visitor, the baseline boost parameters of the corresponding group type are called based on the body feature classification results. The boost force parameters are dynamically adjusted through the real-time pressure change rate and the target window opening angle to ensure that the auxiliary force output by the linkage rod works in conjunction with the user's remaining force to maintain the window moving smoothly along the preset trajectory.

[0085] Compared with existing technologies, traditional automatic window opening systems usually trigger fully automatic mode based on only a single pressure threshold, which is prone to false triggering or response delays due to instantaneous pressure fluctuations. This solution uses the dual judgment conditions of pressure threshold and time threshold to effectively distinguish between short-term force fluctuations and continuous insufficient force, thereby improving the accuracy of mode switching. In addition, existing technologies mostly use fixed boost force strategies, while this solution matches differentiated boost parameters based on user identity or type, realizing the combination of personalized boost and group feature adaptation, avoiding the problems of overdrive or insufficient boost that may be caused by a unified boost value.

[0086] Through the above technical solution, this application can switch to fully automatic control mode in a timely manner when the user's force is insufficient, avoiding window jamming or angle deviation caused by thrust interruption. The dual triggering conditions reduce the probability of misjudgment and ensure that the system only intervenes in control when the force is continuously insufficient. The differentiated boosting strategy based on user type matching not only ensures the continuity of registered users' operating habits, but also provides visitors with reasonable assistance that conforms to the group characteristics, ultimately achieving a smooth transition and precise control of the window opening process.

[0087] The present application further proposes to monitor the second-order derivative of the operating force in real time when the user is classified as an elderly or child group, and when it is detected that the second-order derivative exceeds a preset safety threshold, activate the reverse braking mechanism to apply reverse braking force to the window linkage rod.

[0088] Classifying users as elderly or children refers to analyzing facial or body shape data collected by cameras using image recognition algorithms to categorize the user opening the window into a specific user group. This classification is used to enable differentiated security protection mechanisms based on the force application characteristics of different user groups.

[0089] The second-order derivative of the operating force refers to the acceleration of the axial pressure applied to the linkage over time. This can be calculated using a differential calculation module after collecting pressure data from a pressure sensor. This parameter is used to capture high-frequency signals caused by sudden force changes.

[0090] The preset safety threshold is a critical acceleration value established based on the force application patterns of elderly or children. It can be obtained through experimental data statistics or machine learning model training. This threshold is used to determine whether to trigger braking action.

[0091] The reverse braking mechanism, which drives the linkage lever to generate a braking force opposite to the current window opening direction, can be achieved through an electromagnetic brake or reverse motor drive. This mechanism is used to offset uncontrolled movement caused by abnormal force.

[0092] Specifically, when an elderly user or child opens the window, the pressure sensor continuously collects axial pressure data from the linkage rod and calculates its time-varying second-order derivative through a differential operation module. When this second-order derivative value exceeds a preset safety threshold, it indicates a sudden force abnormality, such as a pressure surge caused by a slipped hand. At this point, the brake control unit immediately outputs a reverse drive signal to the linkage rod, causing the brake to generate a braking force opposite to the current window opening direction, thereby suppressing the abnormal acceleration of the window within milliseconds of the sudden force change. During this process, second-order derivative monitoring can identify dangerous conditions 50-100 milliseconds earlier than conventional pressure value or first-order derivative monitoring, providing more sufficient response time for reverse braking.

[0093] Compared with existing technologies, conventional safety protection solutions only monitor the absolute value or rate of change of pressure and are unable to effectively identify sudden force changes. For example, the anti-pinch device disclosed in the prior art triggers braking based on a pressure threshold, but its response delay makes it impossible to prevent mechanical impact caused by sudden force changes. This solution captures the sudden force acceleration characteristics through second-order derivative monitoring. Combined with dynamic safety thresholds set for specific user groups, it can initiate precise braking at the early stage of force anomalies. Compared with existing technologies, it shortens response time by more than 60%, while avoiding frequent false triggering caused by misjudging normal force fluctuations.

[0094] Through the above technical solution, the present application effectively prevents elderly or child users from opening windows rapidly or being damaged by impact due to uncontrolled force application during the window opening process. It identifies dangerous conditions by real-time monitoring of second-order derivative parameters and immediately applies reverse braking force when abnormal acceleration is detected, so that the window movement is decelerated to a stationary state within 20 milliseconds, ensuring user operation safety and equipment integrity.

[0095] An adaptive control system for alloy windows based on multimodal data, such as Figure 5 Shown, including,

[0096] The data acquisition module collects the facial and body feature data of the person opening the window in real time through a camera, collects the axial pressure value and pressure change rate of the linkage rod in real time through a pressure sensor, and records the operation timestamp, window opening frequency and real-time window opening angle data;

[0097] The window opening user identification module detects a human approach signal through the infrared sensor and matches the facial feature data of the current window opener with the facial feature data of users pre-stored in the database through facial feature comparison. If the match is successful, the module enters the user assistance control step; otherwise, the module enters the visitor assistance control step;

[0098] The user power assist control module obtains the axial pressure value, pressure change rate and window opening angle data by calling the user's historical window opening behavior model, outputs the target power assist force parameters through the adaptive power assist calculation strategy, and controls the linkage rod to provide power assist force;

[0099] The visitor assistance control module classifies the window opener into a characteristic type according to the body feature data through a feature classification algorithm, calls the corresponding window opening behavior model to obtain a pressure value and controls the linkage rod to provide an assist force.

[0100] The above shows and describes the basic features, principles, and advantages of the present invention. It should be noted that the present invention is not limited to the above embodiments, which are only some embodiments. Without departing from the spirit and scope of the present invention, various improvements and supplements made are considered to be within the scope of protection of the present invention.

Claims

1. An alloy window adaptive control method based on multimodal data, characterized in that: The method comprises the following steps: a data collection step, in which the facial feature data and body feature data of the window opener are collected in real time by a camera, the axial pressure value and pressure change rate of the linkage rod are collected in real time by a pressure sensor, and the operation timestamp, window opening frequency and real-time window opening angle data are recorded; The window opening user identification step, when the infrared sensor detects a human approach signal, matches the facial feature data of the current window opener with the facial feature data of users pre-stored in the database through facial feature comparison. If the match is successful, the user assistance control step is entered; Otherwise, it will enter the visitor assistance control step; The user power assist control step obtains the axial pressure value, pressure change rate and window opening angle data by calling the user's historical window opening behavior model, outputs the target power assist force parameters through the adaptive power assist calculation strategy, and controls the linkage rod to provide power assist force; The visitor assistance control step classifies the window opener into a characteristic type according to the body feature data through a feature classification algorithm, calls the corresponding type of window opening behavior model to obtain the pressure value and calculate the assist force, and controls the linkage rod to provide the assist force.

2. The alloy window adaptive control method based on multimodal data according to claim 1, characterized in that: The user-assisted control step also includes performing time series matching based on the current real-time system time and the historical window opening behavior model to obtain the window opening angle data of the current window opener at the corresponding time and the axial pressure value and pressure change rate applied to the window linkage rod by the current window opener. The time series matching includes matching the operation timestamp of the window opening operation recorded by the current window opener with the real-time system time of this window opening, and locking the window opening angle data, axial pressure value and pressure change rate in the historical window opening behavior model of the current window opener at the corresponding time.

3. The alloy window adaptive control method based on multimodal data according to claim 1, characterized in that: It also includes a model optimization step. After each window opening operation is completed, the time characteristic data, pressure change curve and actual window opening angle data of this operation are input into the corresponding user behavior model. The user behavior model is a model constructed based on the historical window opening behavior of users and visitors of different feature types, and the model parameters are updated through an incremental learning algorithm.

4. The alloy window adaptive control method based on multimodal data according to claim 1, characterized in that: The visitor assistance control step also includes establishing a window opening action continuity model through time series analysis of the pressure change rate. When it is detected that the pressure value fluctuates intermittently within a predetermined time and the fluctuation amplitude exceeds a preset threshold, it is determined to be an operation pause state, and an operation continuity evaluation parameter is generated and the window opening instruction is completed according to the operation continuity evaluation parameter.

5. The alloy window adaptive control method based on multimodal data according to claim 1, characterized in that: The visitor assist control step further includes, when it is determined that the operation is in a pause state, recording the angle difference between the initial pause position and the final stop position, calculating the angle compensation coefficient based on the angle difference, and correcting the visitor assist force value by correcting the angle compensation coefficient.

6. The alloy window adaptive control method based on multimodal data according to claim 1, characterized in that: The adaptive power assist calculation strategy includes obtaining the historical average power assist force of the current window opener by calculation, and obtaining the current power assist force value by calculating the average power assist force in combination with the real-time pressure change rate and the current axial pressure value.

7. The alloy window adaptive control method based on multimodal data according to claim 1, characterized in that: The visitor assistance control step also includes a categorized assistance calculation strategy, which includes calculating the assistance value in the visitor assistance control based on the average assistance force baseline value of the population after the feature classification, combined with the real-time pressure change rate and the historical average value of the window opening angle.

8. The alloy window adaptive control method based on multimodal data according to claim 1, characterized in that: It also includes a human-computer interaction control step. When it is detected that the real-time axial pressure value of the connecting rod is lower than the preset threshold and reaches the preset time, the fully automatic auxiliary mode is started, and the corresponding assist force is given to the connecting rod according to the current window opening user judgment result.

9. The alloy window adaptive control method based on multimodal data according to claim 1, characterized in that: When the person opening the window is identified as an elderly person or a child through the user's facial feature data or feature classification, the second-order derivative of the operating force is monitored in real time. When it is detected that it exceeds the preset safety threshold, the reverse braking mechanism is activated to apply reverse braking force to the window linkage rod.

10. An alloy window adaptive control system based on multimodal data, characterized in that: include, The data acquisition module collects the facial and body feature data of the person opening the window in real time through a camera, collects the axial pressure value and pressure change rate of the linkage rod in real time through a pressure sensor, and records the operation timestamp, window opening frequency and real-time window opening angle data; The window opening user identification module detects a human approach signal through the infrared sensor and matches the facial feature data of the current window opener with the facial feature data of users pre-stored in the database through facial feature comparison. If the match is successful, the user assistance control step is entered; Otherwise, it will enter the visitor assistance control step; The user power assist control module obtains the axial pressure value, pressure change rate and window opening angle data by calling the user's historical window opening behavior model, outputs the target power assist force parameters through the adaptive power assist calculation strategy, and controls the linkage rod to provide power assist force; The visitor assist control module classifies the window opener into a characteristic type according to the body feature data through a feature classification algorithm, calls the corresponding window opening behavior model to obtain a pressure value, and controls the linkage rod to provide an assist force.

Images