Elevator operation state real-time recognition method and system based on cross-scale direction consistency

Abstract

The application discloses an elevator operation state real-time identification method and system based on cross-scale direction consistency, collects three-axis acceleration signals, extracts shaft vertical dynamic acceleration; constructs three causal analysis scales of short scale, medium scale and long scale, respectively calculates direction dominance and direction persistence; calculates cross-scale overall direction evidence and cross-scale direction consistency; combines uplink starting score and downlink starting score, locks the current trip direction when in the unlocked direction state; establishes an elevator legal state diagram, and determines a candidate state set according to the current confirmed state and the locked trip direction; calculates the matching degree of each candidate state according to the state observation vector of the current analysis period; calculates the retention item and the exit item according to the current state duration; respectively calculates the retention score and the conversion score, and accumulates the conversion advantage in a limited time window. The elevator operation state can be accurately identified in real time, and the elevator operation state monitoring efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of elevator inspection technology, and in particular to a method and system for real-time identification of elevator operating status based on cross-scale directional consistency. Background Technology

[0002] In existing elevator IoT monitoring, inspection, and safety supervision, accelerometers are typically installed on the elevator car, car top, car frame, or detection terminals that move with the car to acquire elevator operation information in a non-invasive manner. However, the signals measured by accelerometers are highly complex; they simultaneously include low-frequency components generated by the overall acceleration and deceleration of the elevator car, frictional vibration between the guide shoes and guide rails, structural vibration excited by the traction system, car structural resonance, door opening and closing impacts, passenger movement within the car, luggage collisions, and the superposition of multiple components such as changes in sensor installation posture, sensor zero bias, and temperature drift. In this complex signal environment, a single sample value or the statistical value of a single time window exceeding a threshold does not directly indicate that the elevator has started, decelerated, or stopped.

[0003] Currently, there are some technical solutions to address the above problems, but all of them have varying degrees of shortcomings.

[0004] Some solutions continuously integrate the acceleration signal to obtain velocity, and then drive a state machine to determine the state based on the velocity, acceleration, and duration. For example, velocity is obtained by integrating acceleration, and then the state machine is driven by a velocity threshold and a duration threshold to complete the state transition. However, in practical applications, accelerometers inevitably have zero bias and low-frequency noise. These errors accumulate over time during the integration process, causing the velocity baseline to drift over time. In high-rise elevators, the integration drift problem is more severe, and the velocity drift may exceed 0.3 m / s, rendering it unreliable.

[0005] Other approaches employ sliding window statistics, zero-crossing rate analysis, positive / negative sample ratios, or fixed consecutive counts to determine the state. While these approaches can suppress the influence of local noise to some extent, they struggle to handle continuous disturbance signals. When passengers continuously shake their feet inside the car or experience continuous impacts from carrying heavy objects, the sliding window statistical characteristics may be dominated by the disturbance signal, leading to misjudging a stationary state as an operating state. Furthermore, for short-stroke elevators without a clear constant-speed phase, these approaches often fail to promptly identify the start of the deceleration phase after the acceleration phase, resulting in a significant lag in state transition.

[0006] Some solutions incorporate barometric pressure sensors to aid in determining the elevator's position and direction of motion. For example, by fusing data from accelerometers and barometric pressure sensors, Kalman filtering is used to estimate the elevator's displacement and velocity, thereby helping to determine its operating status. However, barometric pressure sensors are significantly affected by changes in ambient air pressure, temperature, and airflow disturbances. In indoor elevator shafts, pressure changes are not only related to elevator height but are also affected by various factors such as airflow within the shaft and the start / stop of the air conditioning system, making it difficult to guarantee measurement accuracy and reliability. Furthermore, introducing additional sensors increases system cost and power consumption, hindering deployment on low-power edge computing hosts.

[0007] In addition, deep learning models are used for elevator status recognition or fault diagnosis. While these deep learning-based methods can achieve good classification results on specific datasets, they have the following inherent drawbacks: First, deep learning models require a large amount of labeled training data, while elevator operation status labeling requires manual intervention or relies on elevator controller signals, which is costly; second, the model has poor interpretability, making it difficult to explain the physical basis of a judgment result to the user; third, model inference requires significant computing resources, making it difficult to run in real time on resource-constrained edge computing hosts; and fourth, the model's generalization ability is difficult to guarantee across different elevators, installation locations, and operating environments.

[0008] The existing technology has the following problems: 1. Acceleration signals contain zero bias and random noise. Directly integrating the acceleration over a long period of time can easily lead to accumulated errors, causing the obtained speed baseline to drift and resulting in misjudgment of the elevator's operating status.

[0009] 2. The actual starting and stopping motions of the elevator car, as well as local disturbances such as passenger foot shaking, collisions with the car, local sensor shaking, and mechanical vibrations, may produce similar instantaneous fluctuations in the acceleration signal, leading to the problem of incorrect identification of the elevator's starting, stopping, and direction of travel.

[0010] 3. Under different conditions such as different elevators, different loads, different floor spacing, and different operating speeds, the duration of each operating stage varies significantly. Using a uniform fixed number of confirmations or a fixed duration threshold makes it difficult to simultaneously address the issues of stability and real-time performance in state recognition.

[0011] 4. Judging the state independently based solely on a local time window can easily lead to problems such as mismatch between the start-up and shutdown phases, contradictory running directions, incomplete state sequence, and offset state transition boundaries.

[0012] 5. During long-term online operation of elevators, abnormal impacts, strong disturbances, or incorrect identification results may be used to update the discrimination criteria, causing the identification parameters to gradually deviate from the normal state and reducing the accuracy of subsequent identification. Summary of the Invention

[0013] This invention aims to address at least one of the technical problems existing in the prior art. To this end, this invention proposes a real-time elevator operating status identification method and system based on cross-scale directional consistency. By constructing three causal scales—short, medium, and long—the directional dominance and directional persistence of each scale are calculated respectively, and then cross-scale overall directional evidence and directional consistency are synthesized to distinguish between overall car motion and local disturbances. This enables real-time and accurate identification of elevator operating status, improving the efficiency of elevator operating status monitoring.

[0014] On one hand, embodiments of the present invention provide a real-time elevator operating status identification method based on cross-scale directional consistency, including: Collect the triaxial acceleration signals output by the acceleration sensor installed on the elevator car, establish the vertical coordinates of the shaft, and extract the vertical dynamic acceleration of the shaft; Three causal analysis scales—short-scale, medium-scale, and long-scale—are constructed. Causal smoothing is performed on the vertical dynamic acceleration at each scale, and the directional dominance and directional persistence at each scale are calculated. Based on the directional dominance and directional persistence at these three scales, calculate the overall directional evidence across scales and the directional consistency across scales. Based on the cross-scale overall direction evidence and the cross-scale direction consistency, combined with the uplink initiation score and the downlink initiation score, the direction of this trip is locked when the direction is not locked. Establish a legal state diagram for the elevator, and determine a set of candidate states based on the current confirmed state and the locked travel direction. The legal state diagram for the elevator includes stationary, upward acceleration, upward constant speed, upward deceleration, downward acceleration, downward constant speed, downward deceleration, and abnormal state. A state template is established for each candidate state in the candidate state set, and the matching degree of each candidate state is calculated based on the state observation vector of the current analysis period. Establish the condition duration distribution, survival probability, and exit risk rate for each state, and calculate the hold-up and exit terms based on the duration of the current state; For the current confirmed state and each legal target state, calculate the maintenance score and transition score respectively, and accumulate the transition advantage within a finite time window. When the transition evidence meets the preset threshold, confirm the state transition.

[0015] According to some embodiments of the present invention, the calculation of directional dominance and directional persistence at these three scales respectively includes: For each scale, obtain the causal smoothed acceleration sequence within the current analysis window, calculate the positive and negative signed areas within the current analysis window respectively, and calculate the directional dominance using the following formula: In the formula, To analyze the directional dominance of scale r, the range is... A r + (k) represents the positively signed area; A r (k) represents the negative signed area; To prevent positive numbers with a denominator of zero; r is the analysis scale; k is the current analysis cycle number; The current analysis window is divided into multiple consecutive overlapping sub-windows. The directional dominance is calculated for each sub-window. Sub-windows whose absolute values ​​reach a threshold are considered valid sub-windows. Sub-windows whose signs align with the current scale's principal direction are considered co-directional sub-windows. Directional persistence is calculated using the following formula: In the formula, For scale directional continuity, normalized to ; For the mean scale The number of child windows that are in the same direction as the current main window; This represents the number of valid child windows. To prevent positive numbers with a denominator of zero.

[0016] According to some embodiments of the present invention, the calculation of cross-scale overall directional evidence and cross-scale directional consistency includes: Calculate the overall directional evidence using the following formula: In the formula, This serves as overall directional evidence, ranging from [-1, 1]. The weighting coefficients for the analysis scale r; Directional dominance; For directional continuity; These represent short-scale, medium-scale, and long-scale, respectively. Cross-scale directional consistency is calculated using the following formula: In the formula, For cross-scale directional consistency, the range is [0,1]; The weighting coefficients for the analysis scale r; The degree of directional dominance; For directional continuity; To prevent positive numbers with a denominator of zero.

[0017] According to some embodiments of the present invention, locking the direction of the current journey includes: The uplink initiation score and downlink initiation score are calculated separately. The uplink initiation score rewards positive global directional evidence, cross-scale directional consistency and long-scale low-frequency energy, and penalizes the perturbation index and lateral energy ratio. Accumulate the initiation score advantage in each direction within a limited evidence window. When the accumulated evidence in a certain direction reaches the locking threshold and the initiation score, cross-scale directional consistency, and long-scale low-frequency energy in that direction are all effective, the travel direction is locked as that direction. Once the direction is locked, a single reverse acceleration spike does not change the locked direction. The direction lock is only released when it is confirmed that the object has returned to rest, during the re-initialization process after entering an abnormal state, or when a persistent and serious directional contradiction is detected.

[0018] According to some embodiments of the present invention, calculating the matching degree of each candidate state based on the state observation vector of the current analysis period includes: Form the state observation vector for the current analysis period: In the formula, This is the state observation vector for the kth analysis period; Evidence is provided for the overall direction, within the range [-1,1]. For cross-scale directional consistency, the range is [0,1]. and These represent the zero-acceleration stability at mesoscale and longscale, respectively. This represents the normalization result for long-scale low-frequency energy. The disturbance index; This refers to the horizontal energy ratio; Signal quality, range [0,1]; For each candidate state s, the matching value of each feature is calculated using the membership function in the pre-established state template, and the weighted matching degree is calculated using the following formula: In the formula, The degree of matching between the current observation and the candidate state s, ranging from [0,1]; Let f(j) represent the membership function of the j-th feature in state s; For the corresponding normalized features; For template weights, when s belongs to the candidate set Only when this occurs will it participate in the normal conversion comparison; if If the signal quality falls below the threshold, normal matching will be paused or reduced. The validity of the data is verified and the case enters the abnormal pending confirmation process.

[0019] According to some embodiments of the present invention, the extraction of vertical dynamic acceleration of the wellbore includes: Once it is confirmed that the elevator is stable and stationary, the direction is not locked, there is no continuous disturbance, and the signal quality is normal, the most recent reliable stationary sample is selected, and the median of the three axes is taken as the three-axis reliable stationary baseline vector. Based on the installation calibration or reliable static gravity direction combined with positive and negative sign calibration, establish the vertical unit vector of the shaft, and uniformly define the upward direction along the shaft as positive; The triaxial dynamic acceleration vector is obtained by subtracting the triaxial reliable static baseline vector from the triaxial acceleration measurement vector. The triaxial dynamic acceleration vector is then projected onto the vertical unit vector of the shaft to obtain the vertical dynamic acceleration of the shaft.

[0020] According to some embodiments of the present invention, the method further includes the steps of establishing the conditional duration distribution, survival probability, and exit hazard rate of the elevator operating state: For each state s, establish a duration probability distribution under given condition c; Calculate the survival probability using the following formula: In the formula, Let Pr be the survival probability, representing the probability that state s will persist to the d-th period under condition c; Pr represents the probability; D s Let be a random variable representing the duration of state s; u is the number of possible duration periods; The maximum duration period; This represents the probability that the running state s continues for exactly u analysis cycles under condition c; Calculate the exit risk rate using the following formula: In the formula, The discrete exit risk rate represents the conditional probability that state s will exit at the end of the current period when it continues to the d-th period under condition c, and its range is limited to [0,1]. This represents the probability that state s continues for exactly d analysis periods under condition c; Probability of survival; To prevent positive numbers with a denominator of zero.

[0021] According to some embodiments of the present invention, the method further includes the step of calculating the impulse closure residual for the same stroke start-up and braking segments: For the candidate acceleration segments of this journey, the area of ​​signed acceleration is calculated using the following formula: In the formula, The area of ​​signed acceleration for initiating accelerated candidate segments, in m / s². This is the set of sampled indices from the start candidate boundary to the acceleration end candidate boundary; For long-scale smooth vertical dynamic acceleration; It is a long-scale reliable static residual center; The sampling period; For the candidate braking and deceleration segments of this journey, the area of ​​signed acceleration is calculated using the following formula: In the formula, The signed acceleration area for the candidate segments of braking and deceleration, in m / s². The set of sample indices between the deceleration start candidate boundary and the stop candidate boundary; Calculate the impulse closure residual using the following formula: In the formula, The range of the start / stop impulse closure residual is limited to [0,1]. The area of ​​signed acceleration for initiating accelerated candidate segments; The area of ​​signed acceleration for candidate segments of braking and deceleration; Represents absolute value; To prevent positive numbers with a denominator of zero.

[0022] According to some embodiments of the present invention, the method further includes the step of jointly correcting the state transition boundary within a finite historical neighborhood: Using the initial state transition boundary as the center, a finite search interval is set for each boundary to be corrected, and a set of candidate boundary combinations is constructed. For each candidate boundary combination, the joint evaluation value is calculated using the following formula: In the formula, This is the joint evaluation value of candidate boundary combination B; For a finite set of analysis periods affected by candidate boundary adjustments; The degree of state matching; This represents the state corresponding to the k-th period after being divided according to B; This represents the number of state fragments. and These are the nth state segment and its duration; Let c be the duration probability under condition c; Consistency reward weight; Weighting for impulse residual penalties; This represents the closed residual obtained after re-dividing the segments according to candidate boundary combination B; For perturbation penalty weights; Q k The comprehensive disturbance index; The candidate boundary combination that maximizes the joint evaluation value and satisfies the constraints of a valid state diagram and the shortest residence time is selected as the final modified boundary.

[0023] On the other hand, embodiments of the present invention provide a real-time elevator operating status identification system based on cross-scale directional consistency, used to implement the above-mentioned real-time elevator operating status identification method based on cross-scale directional consistency, including: An accelerometer sensor, which is installed on the elevator car, is used to collect triaxial acceleration signals; An edge computing host, which is used for real-time signal processing and elevator operation status identification; A cloud service platform is used to receive data from multiple edge computing hosts, perform data analysis, and centrally manage the data.

[0024] The embodiments of the present invention have at least the following beneficial effects: First, by fusing cross-scale directional evidence, the system effectively distinguishes between overall motion and local disturbances. Directional dominance and directional persistence are calculated at short, medium, and long time scales, and then cross-scale overall directional evidence and directional consistency are synthesized. Continuous overall car motion can generate mutually supporting directional evidence at all three scales, while local collisions, passenger foot shaking, and short-term mechanical impacts typically only produce responses at the short scale, lacking sustained directional evidence at medium and long scales. This allows for accurate identification of the elevator's true operating state under strong disturbance environments, significantly improving the ability to distinguish between state recognition and local disturbances.

[0025] Secondly, by locking the travel direction, the technical bias of misjudging the elevator as stationary during the uniform speed phase is resolved. Through the travel direction locking mechanism, when the elevator enters the uniform speed phase and the vertical dynamic acceleration approaches zero, it is accurately identified as moving at a uniform upward or downward speed based on the locked travel direction, rather than being misjudged as stationary. This overcomes the long-standing technical bias in the field that "zero acceleration means stationary".

[0026] Third, establish a state-related duration probability model to adaptively adjust the timing of transitions. Establish conditional duration distributions, survival probabilities, and exit risk rates for each operating state, so that different states such as acceleration, constant speed, deceleration, and short-stroke have differentiated hold / exit judgment criteria; this can adaptively adjust the timing of state transitions to avoid premature switching or excessive delay.

[0027] Fourth, integral drift is avoided through start-stop impulse closure verification. The area of ​​signed acceleration is calculated only for a limited segment of the same stroke during start-up acceleration and braking deceleration, and the physical consistency is evaluated using the impulse closure residual; this provides a basis for verifying the integrity of the state sequence while avoiding long-term integral drift.

[0028] Fifth, a finite historical boundary correction is adopted to balance real-time performance and accuracy. A dual-track mode of "real-time output of transient labels + post-event finite neighborhood correction" is employed. The real-time phase ensures the timeliness of state output; after the journey ends, the boundary position is optimized only within a finite candidate neighborhood near the initial boundary by jointly considering the degree of state matching, duration probability, cross-scale consistency, impulse closure residual, and perturbation exponent, effectively reducing the boundary delay caused by evidence accumulation.

[0029] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0030] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a flowchart of a real-time elevator operation status identification method based on cross-scale directional consistency according to an embodiment of the present invention; Figure 2 This is a functional block diagram of an elevator operation status real-time identification system based on cross-scale directional consistency according to an embodiment of the present invention. Figure 3 This is a schematic diagram of elevator operation state transition in the real-time elevator operation state identification method based on cross-scale directional consistency according to an embodiment of the present invention. Figure 4 This is a schematic diagram of the complete downward shaft vertical dynamic acceleration of the elevator operation status real-time identification method based on cross-scale directional consistency according to an embodiment of the present invention. Figure 5 This is a complete schematic diagram of the downward low-frequency operation trend of the elevator operation status real-time identification method based on cross-scale directional consistency according to an embodiment of the present invention; Figure 6 This is a complete downward high-frequency residual diagram of the real-time elevator operation status identification method based on cross-scale directional consistency according to an embodiment of the present invention; Figure 7 This is a complete downward high-frequency intensity diagram of the elevator operation status real-time identification method based on cross-scale directional consistency according to an embodiment of the present invention. Detailed Implementation

[0031] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0032] In the description of this invention, it should be understood that the orientation descriptions, such as up, down, front, back, left, right, etc., are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0033] In the description of this invention, "several" means one or more, "multiple" means two or more, "greater than," "less than," "exceeding," etc. are understood to exclude the stated number, and "above," "below," "within," etc. are understood to include the stated number. If "first," "second," etc. are used in the description, they are only for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance or implicitly indicating the number of indicated technical features or the order of the indicated technical features.

[0034] In the description of this invention, unless otherwise explicitly defined, the terms "setting", "installing", "connecting" and "linking" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0035] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0036] Please see Figures 1 to 3 This embodiment provides a real-time elevator operation status identification method based on cross-scale directional consistency, mainly including steps S101~S108: S101. Collect the three-axis acceleration signal output by the acceleration sensor installed on the elevator car, establish the vertical coordinates of the shaft, and extract the vertical dynamic acceleration of the shaft.

[0037] S102. Construct three causal analysis scales: short-scale, medium-scale, and long-scale. Perform causal smoothing on the vertical dynamic acceleration respectively, and calculate the directional dominance and directional persistence at each of the three scales.

[0038] S103. Based on the directional dominance and directional persistence at these three scales, calculate the overall directional evidence across scales and the directional consistency across scales.

[0039] S104. Based on cross-scale overall direction evidence and cross-scale direction consistency, combined with uplink initiation score and downlink initiation score, lock the direction of this trip when the direction is not locked.

[0040] S105. Establish a valid elevator state diagram and determine the candidate state set based on the current confirmed state and the locked travel direction. The valid elevator state diagram includes stationary, upward acceleration, upward constant speed, upward deceleration, downward acceleration, downward constant speed, downward deceleration, and abnormal state.

[0041] S106. Establish a state template for each candidate state in the candidate state set, and calculate the matching degree of each candidate state based on the state observation vector of the current analysis period.

[0042] S107. Establish the condition duration distribution, survival probability, and exit risk rate for each state, and calculate the hold-up and exit terms based on the duration of the current state.

[0043] S108. For the current confirmed state and each legal target state, calculate the maintenance score and the transition score respectively, and accumulate the transition advantage within a limited time window. When the transition evidence meets the preset threshold, confirm the state transition.

[0044] Please see Figure 2 This embodiment discloses a real-time elevator operating status identification system based on cross-scale directional consistency, used to implement the above-mentioned real-time elevator operating status identification method based on cross-scale directional consistency, including: Accelerometer 100 is installed on the elevator car to collect triaxial acceleration signals. Edge computing host 200 is used for real-time signal processing and elevator operation status recognition; The cloud service platform 300 is used to receive data from multiple edge computing hosts 200 for data analysis and centralized management.

[0045] The following is a detailed description of the elevator operation status real-time identification method and system based on cross-scale directional consistency provided in this embodiment.

[0046] 1. Acquire triaxial acceleration signals, establish vertical coordinates of the shaft, and extract dynamic acceleration. The triaxial measurements from the accelerometer's own coordinate system are converted into vertical dynamic acceleration aligned with the elevator shaft direction, while lateral dynamic acceleration is also obtained. Vertical dynamic acceleration serves as the primary input for subsequent status recognition, while lateral dynamic acceleration assists in identifying collisions, foot shaking, guide rail vibration, and loose installation.

[0047] In the formula, k is the discrete sampling number; For the first The triaxial acceleration measurement vector at each sampling time; These are the outputs of the accelerometer's x, y, and z measurement axes, respectively, in units of... ; superscript Represents vector transpose; sampling rate 1000 Hz, sampling period .

[0048] Once it is confirmed that the elevator is stable and stationary, its direction is not locked, there is no continuous disturbance, the sensors are not saturated, there are no consecutive data loss points, and the installation attitude is stable, the most recent reliable static sample is selected to form a set. The reliable settling time is 3 seconds, but it can be adjusted according to the sampling rate and the level of vibration at the site.

[0049] In the formula, This is a three-axis reliable static baseline vector; Let be the triaxial acceleration measurement vector at the i-th sampling time. A set of credible static samples The sample number in the array; median is the median operator, which means taking the median of the samples on each of the three axes. It includes the gravity measurement component at rest, the sensor's static zero bias under fixed installation conditions, and the stable structural bias; therefore, the scalar gravity g is not subtracted again subsequently. Using the median instead of a single static sample value reduces the impact of accidental shocks on the baseline.

[0050] Establish the vertical unit vector of the shaft. And it is uniformly stipulated that the upward direction along the shaft is the positive direction. It can be obtained from installation calibration, or from a reliable static gravity direction combined with a sign calibration; if the calculated direction is opposite to the upward specification, then the vector is multiplied by -1.

[0051] In the formula, Let be the vertical unit vector of the wellbore in the sensor coordinate system; This represents the L2 norm. This constraint ensures that the projection results retain the acceleration dimension.

[0052] Once the direction is calibrated, upward acceleration generates positive vertical dynamic acceleration, while downward acceleration generates negative vertical dynamic acceleration.

[0053] In the formula, The three-axis dynamic acceleration vector after removing the reliable static baseline, in m / s². This is the triaxial acceleration measurement vector, representing the original measurement value at the current moment; This is a three-axis reliable static baseline vector.

[0054] When the elevator remains stationary and the sensor attitude remains unchanged, It should fluctuate slightly around the zero vector.

[0055] The formula for vertical projection is: In the formula, For the first The vertical dynamic acceleration of the shaft at each sampling time, in m / s²; according to the rule that upward is positive, during upward acceleration... Normally positive, usually negative during upward deceleration; usually negative during downward acceleration, usually positive during downward deceleration; usually close to zero during stationary and near-uniform motion phases. This symbol only indicates the direction of acceleration and cannot represent the direction of car motion alone.

[0056] The formula for lateral projection is: In the formula, I is the lateral dynamic acceleration vector; I is the third-order identity matrix; The projection matrix is ​​in the vertical direction; This is the planar projection matrix orthogonal to the vertical direction of the shaft. The horizontal composite amplitude is expressed in m / s².

[0057] When there is a significant increase in elevation but insufficient evidence in the vertical direction at medium and long scales, it indicates a localized collision, passenger movement, or loose installation. Updates at a slower rate only when the reliable static condition is continuously met; freezes when the elevator is moving, colliding, shaking, experiencing sudden attitude changes, sensor saturation, or data loss. If a continuous attitude change is detected, the condition is redefined first. With a new reliable static baseline, normal recognition is restored. If the linear acceleration with gravity removed is directly output, gravity removal is not repeated; only coordinate transformation, static residual correction, and positive direction unification are performed.

[0058] 2. Construct three causal analysis scales: short, medium, and long. The same vertical dynamic acceleration is observed using historical windows of varying lengths. Short-scale windows are used to detect start-up, braking, impact, and directional reversal; medium-scale windows are used to determine whether the direction continues; and long-scale windows are used to determine whether the car is forming an overall motion trend. All windows use only data from the current moment and earlier, avoiding the use of future samples, which facilitates real-time implementation.

[0059] In the formula, To analyze the causal smoothing results at scale r, the unit is m / s²; To represent the analytical scale, use These represent short-scale, medium-scale, and long-scale, respectively. The number of sampling points included in the analysis scale r; l is the historical sample index relative to the current time. Let be the weight coefficient of the l-th historical sample, and satisfy . ; For the kth Vertical dynamic acceleration of the shaft at l sampling times.

[0060] This formula is implemented using moving average, exponential smoothing, finite impulse response low-pass filtering, or equivalent methods with the same causal smoothing effect.

[0061] For example in During sampling, short-scale, medium-scale, and long-scale measurements covered approximately 0.2 s, 1.0 s, and 3.0 s, respectively, corresponding to... Short-scale collisions should be sufficient to capture the initiation spike, medium-scale collisions should cover the directional continuous process, and long-scale collisions should be significantly longer than a single local collision.

[0062] In the formula, For the first The high-frequency residual at each sampling time, in units of ; For the first Vertical dynamic acceleration of the shaft at each sampling time; This is a long-scale smoothing result.

[0063] Localized collisions, foot shaking, and high-frequency mechanical vibrations typically cause The energy increases, and the overall start-up or braking of the car will leave low-frequency components with the same direction in the medium and long dimensions.

[0064] 3. Establish a static characteristic baseline and perform robust normalization. The numerical ranges of energy, jerk, and lateral vibration may differ depending on the elevator, installation location, and sampling rate. This step utilizes reliable static data to establish the center and scale of each feature, enabling features with different dimensions to be included in the same scoring system.

[0065] In the formula, For the first Item features percentile; Indicates the first In the 1st reliable static analysis cycle, the 1st Original features; Assign a feature number; Percentile represents percentiles, with values ​​of 10, 50, and 90; Percentile indicates quantile statistics. It is a set of reliable static samples.

[0066] The formula for truncation normalization is: In the formula, For the first The first analysis cycle Normalization results of the features; This represents the original value of the j-th feature in the k-th analysis period; It is the 50th percentile of the j-th feature, i.e., the center of rest; It is the 90th percentile of the j-th feature, i.e., the high percentile; It is the 10th percentile of the j-th feature, i.e., the lowest percentile; Indicates the scale of stationary fluctuations; This is the minimum scale allowed for this feature, to prevent static samples from becoming too uniform and causing the denominator to approach zero; To prevent positive numbers with a denominator of zero; This represents the truncation function; This is the maximum value function, representing the value q(j, 90). q(j,10) and The larger value in the denominator is used as the fluctuation scale.

[0067] Directional features truncated to Energy, stability, perturbation, and mass characteristics are truncated to For features that are already within a fixed range, this range can be used directly for linear normalization.

[0068] When the number of stationary samples is insufficient, the equipment's factory-standard scale parameters should be used preferentially, and unconfirmed runtime segments should not be used to reverse-engineer a stationary baseline. Freeze the system when foot jitter, collisions, pose changes, saturation, or dropped points are detected. and Updates are made to prevent abnormal data from expanding into the normal range.

[0069] 4. Calculate directional evidence, stability, and perturbation characteristics at each scale. For each scale Take the end time of the current analysis period. Previous A window composed of smooth samples First, the residual centers at this scale are obtained from reliable static data. If both the baseline and scaling filter are stable, near ,reserve It can absorb small residuals caused by filters and finite windows.

[0070] In the formula, For scale The positively signed area within the window, in units of ; The sampling sequence number within the window; This means only positive values ​​are retained; To analyze the smoothed acceleration at scale r; To determine the credible static residual center at the analysis scale r; The sampling period. The larger the value, the more pronounced the continuous positive acceleration within the window.

[0071] In the formula, To analyze the area of ​​the absolute value of the negative portion within the scale r window, the unit is... . The larger the value, the more pronounced the continuous negative acceleration within the window. and All are non-negative quantities, used to compare the cumulative effects in positive and negative directions.

[0072] In the formula, To analyze the directional dominance of scale r, the range is... ; To prevent positive numbers with a denominator of zero. When near The positive direction is dominant, close to The negative direction is dominant, close to When the positive and negative effects cancel each other out or both are very small. When the direction is below the effective threshold, this scale is recorded as having no effective direction, and motion cannot be determined solely by the occasional positive or negative deviation of the numerical ratio.

[0073] To calculate directional persistence, The area is then divided into several consecutive overlapping sub-windows. The directional dominance is calculated for each sub-window; sub-windows whose absolute value reaches the sub-window directional threshold are recorded as valid sub-windows, and those whose signs are consistent with the current scale's main direction are recorded as co-directional sub-windows.

[0074] In the formula, For scale directional continuity, normalized to ; This represents the number of valid child windows. For the mean scale The number of child windows that are in the same direction as the current main window; To prevent positive numbers with a denominator of zero. When for hour, . The height indicates that the direction exists consecutively across multiple sub-windows. A low value indicates that the directional evidence is only a transient spike or frequent reversals.

[0075] In the formula, To analyze the zero-acceleration stability of scale r, the range is... ; Number of samples in the window; For indicator functions, take the condition if it is true. Otherwise take ; To analyze the smoothed acceleration at scale r; b r The credible static residual center is at scale r; For analytical scale The zero-band threshold, in units of It is determined by the high quantile range of reliable static data. The higher the value, the longer the proportion of samples at that scale lies within the zero-acceleration band. Both stationary and uniform motion may exhibit higher values. ,therefore The state cannot be determined independently without direction locking and stage memory.

[0076] In addition to direction and stability, high-frequency energy is also calculated. Periodicity Horizontal energy ratio and accelerometer energy Each component is normalized to a certain value before being included in the perturbation score. .

[0077] In the formula, High-frequency residuals within a short-scale window The root mean square of , with the original unit being m / s²; W represents the number of samples in the short-scale window. S (k) represents the short-scale window. An increase indicates a strong, rapid vibration or shock relative to the overall trend over a long timescale.

[0078] Calculate the root mean square of low-frequency acceleration within a long-scale window: In the formula, is the long-scale low-frequency energy of the kth analysis period, in m / s². This represents the number of sampling points contained in the long-scale window; For a long-scale window; i is the sampling index within the window; It represents long-scale causal smooth acceleration.

[0079] The temperature rises significantly during the elevator's acceleration and deceleration phases, then drops to near zero during constant speed operation and when stationary. After normalization, As the state observation vector X k One of the feature components is used for subsequent state template matching and start score calculation.

[0080] In the formula, The proportion of transverse dynamic energy to total dynamic energy within a mesoscale window, ranging from... ; Mesoscale window; This represents the horizontal composite amplitude. It is vertical dynamic acceleration; To prevent positive numbers with a denominator of zero. The height indicates that the motion mainly comes from lateral disturbances rather than the normal vertical movement of the car.

[0081] In the formula, This is the root mean square of the vertical acceleration within a short-scale window, with the original unit being m / s³. This represents the acceleration change between adjacent samples; Δt is the sampling period. Collisions and mechanical spikes typically produce large [acceleration variations]. .

[0082] In the formula, Let be the comprehensive disturbance index for the k-th analysis period, ranging from [0,1]. The normalized high-frequency energy; The normalized jerk energy; Indicates periodicity; The proportion of transverse dynamic energy to total dynamic energy within the mesoscale window; , , and All are weighting coefficients.

[0083] Periodic You can remove the mean. The normalized autocorrelation maximum value within the preset human body shaking or mechanical vibration cycle range is [0,1]. A value close to 1 indicates the presence of significant repetitive vibrations; if periodic features are not used, their weights can be set to 0 and the weights of the remaining disturbances can be renormalized. The weights can be determined by the ability of each disturbance feature in the general data to distinguish between normal operation and shaking / collision samples, or they can be provided by engineering calibration. The higher the value, the less suitable it is to trigger state transitions solely based on local acceleration spikes.

[0084] 5. Calculate cross-scale overall directional evidence and directional consistency. By combining results from various scales, the overall directional evidence reflects whether the current trend is more positive or negative, while cross-scale consistency reflects whether multiple scales support the same direction.

[0085] The formula for calculating overall directional evidence is: In the formula, This serves as overall directional evidence, ranging from [-1, 1]. To analyze the weighting coefficients of scale r and ; The degree of directional dominance; For directional continuity; These represent short-scale, medium-scale, and long-scale, respectively. A positive value indicates that overall positive evidence dominates, while a negative value indicates that overall negative evidence dominates. The larger the absolute value, the stronger the directional evidence. When initiating confirmation, the weighting of medium and long-term scales can be increased, while the sensitivity of short-term scales is retained during impact detection.

[0086] The formula for calculating cross-scale directional consistency is: In the formula, For cross-scale directional consistency, the range is [0,1]; The degree of directional dominance; The numerator represents the directional persistence; the numerator is the absolute value of the synthesized signed evidence at each scale; the denominator is the weighted sum of the absolute values ​​of the evidence at each scale. To prevent positive numbers with a denominator of zero, when multiple scales are aligned, the numerator and denominator are close together. Approaching 1; when different scale directions cancel each other out reduce.

[0087] when Below the total effective threshold for direction When all scales lack effective directions, a direct definition is made. and put Considered neutral, this avoids interpreting static, minute noise as high consistency when the denominator is very small. Local collisions typically only make them significant at short scales; evidence of sustained homing at medium and long scales is lacking, making it difficult to simultaneously obtain high consistency. and .

[0088] 6. Form state observation vectors, establish state templates, and calculate the matching degree of candidate states. The obtained direction, stability, energy, disturbance, and mass characteristics are combined to form the state observation vector for the current analysis period. The state observation vector is not a raw waveform, but rather a summary of the overall motion direction, stability, and disturbance level within the current finite window.

[0089] In the formula, This is the state observation vector for the kth analysis period; Evidence is provided for the overall direction, within the range [-1,1]. For cross-scale directional consistency, the range is [0,1]. and These represent the zero-acceleration stability at mesoscale and longscale, respectively. This represents the normalization result for long-scale low-frequency energy. The disturbance index; This refers to the horizontal energy ratio; The signal quality is defined in the range [0,1], where 1 indicates good quality. The analysis period is updated every ΔT, where ΔT can be equal to several sampling periods Δt.

[0090] Before calculating state matching, the candidate range is limited by the previous confirmed state, the locked direction, and the legal state diagram to avoid directly comparing states with the same local acceleration sign but different physical stages.

[0091] In the formula, This is the set of candidate states for the k-th analysis period; This is a confirmation of the status from the previous analysis period; Lock the direction of travel; the value can be NONE, UP, or DOWN. A predefined legal state diagram; Candidate represents the state from which the current state can be maintained or transitioned.

[0092] When the direction is not locked, the main comparisons are static, static with disturbance, upward acceleration, and downward acceleration; locking UP excludes the normal downward state; locking DOWN excludes the normal upward state.

[0093] A state template is established for each specific state s. The template is not a fixed acceleration curve for a certain journey, but a set of statistical parameters and matching rules for each normalized feature in that state.

[0094] In the formula, is the template for state s; j is the feature number; This indicates that feature j uses high value, low value, near zero or interval matching in state s; The template center or turning point threshold; To allow for fluctuations or the transition width of membership functions; These are non-negative feature weights. Features that are irrelevant to a certain state can have their weights set to 0, and the remaining weights are renormalized.

[0095] In the formula, The template center or turning point threshold; The true state label of the k-th sample is provided for reference system or manual review; The corresponding normalized feature; median represents the median. The reference label is obtained before deployment from the elevator controller's operating signal, speed reference signal, floor arrival signal, or through manual verification; access to these reference signals is not required for online identification.

[0096] In the formula, and These are the 90th and 10th percentiles of feature j in state s, respectively. This is the minimum positive transition width. This setting prevents a small number of overly consistent samples from making the template too narrow. When there is no individual data, a general template established by multiple similar elevators can be used first, or an initial template can be set based on physical relationships such as positive and negative directions, near-zero stability, and low disturbance.

[0097] In the formula, is the high-value membership function; x is the current normalized feature; T is the turning point where high values ​​begin to dominate; The transition width is given by exp, which is the natural exponential function. When x is significantly greater than T... When it is close to 1, it is close to 0 when it is significantly less than T.

[0098] In the formula, For low-value membership functions, the smaller the feature value, the higher the degree of matching; It is a near-zero membership function; The near-zero scale is positive. For the expectation to fall within the interval... The features within the range are used to form interval matching using the product of an ascending membership function and a descending membership function. The outputs of all membership functions are restricted to [0,1].

[0099] In the formula, The degree of matching between the current observation and the candidate state s, ranging from [0,1]; Template weights; Let f(j) represent the membership function of the j-th feature in state s; These are the corresponding normalized features. Only s belongs to the candidate set. It only participates in the normal conversion comparison at that time. If If the signal quality falls below the threshold, normal matching will be paused or reduced. The validity of the data is verified and the case enters the abnormal pending confirmation process.

[0100] It only indicates whether the current window resembles state s in terms of local features, and cannot determine the final state on its own. Both stationary and uniform motion may have near-zero dynamic acceleration, and both upward acceleration and downward deceleration may have positive acceleration; the final state must continue to be combined with direction lock, current stage, legal path, duration, and limited evidence accumulation.

[0101] 7. Determine the direction of this journey based on evidence of cross-scale overall motion. When in IDLE state or IDLE_DIST state and When =NONE, upward and downward initiation scores are generated respectively. The score rewards directional evidence, consistency, and long-scale motion energy, and penalizes perturbations and lateral energy.

[0102] In the formula, The score for the upward start, ranging from [0,1]; This indicates that the line is truncated to [0,1]. Evidence for the overall direction; For cross-scale directional consistency; The long-scale low-frequency energy of the k-th analysis period; and These represent the turning point and width of positive directional evidence, respectively. , , , and All are weighting coefficients. The first three reflect the positive overall startup, while the last two are used to reject strong disturbances and lateral shocks. Each weighting coefficient is determined by general calibration data, or it can be initially set as equal in importance and then adjusted using reliable complete journey.

[0103] In the formula, The score for initiating the downlink, ranging from [0,1]; Transform negative overall evidence into positive matching input.

[0104] Using the same set of parameters can maintain symmetry in the up and down scoring, or can be calibrated separately when there are significant differences in the vibration of the elevator going up and down in reality.

[0105] In the formula, Accumulate evidence for the direction; u indicates that the direction to be locked is UP or DOWN; Score for that direction; Score for the opposite direction; The length of the finite evidence window for directional locking, calculated over the analysis period; The forgetting coefficient is between 0 and 1; This requires a minimum advantage for this direction over the opposite direction. Only... Reach the minimum score required to start and Simultaneously effective and Only when the locking threshold is reached will it be locked. Set it to u and enter the corresponding acceleration state.

[0106] Once direction is locked, a single reverse spike will not change the locked direction. Direction locking is only released when the reinitialization process after confirming recovery from the IDLE state and entry into the ABNORMAL state is complete, or when a persistent severe direction discrepancy is detected and a reliable static baseline is re-established.

[0107] 8. Establish valid state diagrams and state paths The legal state diagram specifies which states each confirmed state can remain in or transition to. Normal operation does not allow a direct switch from an uplink state to a downlink state, nor does it allow a direct switch from a downlink state to an uplink state; a change of direction must be followed by a stop, release of direction lock, and a new start confirmation.

[0108] The normal uplink path is IDLE→ACC_UP→UNIFORM_UP→DEC_UP→IDLE; short paths without a clear constant-speed phase allow IDLE→ACC_UP→DEC_UP→IDLE. The normal downlink path is symmetrical to this. The IDLE and IDLE_DIST states can switch bidirectionally; either state can transition to the ABNORMAL state when there are severe abnormalities in signal quality or sequence, and after recovery, it first enters the INIT state or the trusted IDLE state.

[0109] In addition to satisfying the graph structure, candidate transitions must also simultaneously satisfy the corresponding directional characteristics, zero-acceleration stability, shortest dwell time, perturbation rejection, and limited evidence requirements. For example, the ACC_UP→UNIFORM_UP state requires positive evidence decay and Z... M Z L The ACC_UP→DEC_UP state requires cross-scale evidence to continuously change from positive to negative and the current process to be locked in the UP state; only one accompanying high value is required. The DEC_UP state is not allowed when there is a negative spike.

[0110] 9. Establish the probability of state duration, survival probability, and exit risk rate. Different elevator states have different reasonable durations. For example, it should not immediately disengage when acceleration begins, while a constant speed state can last for a longer period. To avoid using the same fixed number of confirmations for all states, duration statistics are established for each specific state.

[0111] In the formula, Let be the number of cycles that state s has lasted up to the k-th analysis cycle; This is the analysis cycle number for the current state s; ΔT represents the actual duration in seconds; ΔT is the state analysis update cycle, in seconds. This occurs when the state is first confirmed to have entered the state. After maintaining this for one cycle, add 1.

[0112] set up Let be a random variable representing the duration of state s, and allow different probability distributions under condition c. Condition c must at least indicate short-stroke, normal-stroke, or different elevator numbers, and may also include load range.

[0113] In the formula, Pr represents the survival probability, indicating the probability that state s has persisted for d cycles or longer; u represents the possible number of cycles. The maximum duration period; This represents the probability that the running state s continues for exactly u analysis cycles under condition c. When there are insufficient reliable samples... A general segmented model is constructed based on the shortest, typical, and longest durations of the project.

[0114] In the formula, The discrete exit risk rate, limited to [0,1], represents the conditional probability of exiting the state after the current period ends when state s has continued for the dth period. This represents the probability that state s continues for exactly d analysis periods under condition c; To avoid positive numbers with a denominator of zero, when the state has not reached the physical minimum dwell time, it can be... The limit is close to 0; after entering the normal exit range, it gradually increases.

[0115] In the formula, The logarithm of the duration for which state s is maintained is denoted by ; log is the natural logarithm. When the risk rate is low, The larger the duration, the smaller the penalty for the hold term; when the duration of the state enters the high exit probability range, the hold term decreases.

[0116] In the formula, This is a logarithmic term representing the duration allowed to exit the current state. When the danger rate is low, this term is small, thus suppressing premature transitions; as the danger rate increases, this term increases. However, the duration only regulates the hold and exit tendencies and cannot trigger a transition independently in the absence of target state signal characteristics.

[0117] During a cold start, a piecewise linear hazard rate can be established based on the shortest physical duration, typical interval, and longest abnormal duration for each state, or a truncated log-normal distribution can be used. After accumulating sufficient complete strokes with low disturbances, legal paths, and eventual recovery to a stationary state with impulse closure, the hazard rate can be updated based on short strokes, normal strokes, or specific elevator update experience.

[0118] 10. Calculate the retention score, transform the score, and accumulate limited evidence of transformation. Let the current confirmed state be r, and the legal target state be s, where s belongs to the current candidate set. The system calculates the scores for maintaining r and transitioning from r to s, respectively. All observed matches, rewards, and penalties should be normalized or weighted to a comparable scale.

[0119] In the formula, To maintain the score of the current state r; The degree of matching between the current observation and the state template r; The prior score for maintaining r in the state graph; The retention term obtained above; This represents the number of cycles the current state has lasted. The non-negative perturbation penalty weight; This is the perturbation index. If the current state itself allows it to persist under perturbation, then... Set a smaller value.

[0120] In the formula, The score for transitioning from r to the legal target state s; The degree of matching between the current observation and the target state template s; For valid transformations, illegal transformations are either excluded from the candidate set or set to negative infinity. This is the exit item for the current state r; The consistency reward weight is for non-negative directions. For cross-scale consistency; The non-negative perturbation penalty weight; The disturbance index. For transitions that primarily rely on homing to zero stability rather than direction formation, it can be reduced. Alternatively, a corresponding stability reward can be used as an alternative.

[0121] For example, The value is 0.1; illegal transitions are not included in the candidate set. The value of the preserved term Γ(r,r) is 0. Different weights are assigned based on the physical rationality of the state transition, for example, for IDLE→ACC_UP. The value can be slightly higher than other conversions to reflect the importance of initiation detection.

[0122] In the formula, The transition advantage of the target state s relative to the current state r is determined. A Δ greater than 0 indicates that the current window is more supportive of the transition, but a single window advantage is still insufficient to confirm a state change. A minimum effective advantage is set for each valid transition. This is to eliminate small fluctuations in the ratings.

[0123] In the formula, This represents the cumulative evidence of the transition from r to s within the most recent finite interval; ρ is the length of the evidence window, calculated over the analysis period; i is the period number within the window; ρ is the forgetting coefficient between 0 and 1, with the closer to 1 indicating that the old evidence decays more slowly. The minimum conversion advantage is achieved; maximum only accumulates positive advantages exceeding δ. Since we no longer participate in history, we will not accumulate outdated evidence indefinitely.

[0124] Appear Below the quality threshold, direction lock contradicts target state, or continuous When the basic observation conditions for the target state are no longer met in a given cycle, the corresponding [condition] should be reset. Only the following criteria are considered: legal path, target state matching, shortest dwell time in the current state, exit risk rate, cross-scale or zero-return stability condition, and perturbation veto. A state transition is confirmed only when all thresholds are met simultaneously.

[0125] 11. Entry, maintenance, and exit judgments for each specific state (1) INIT state and IDLE state When the device is first started and lacks a reliable static baseline or travel direction history, it outputs the INIT state, without forcibly determining whether it is stationary or moving at a constant speed. Continuous fulfillment of the condition... Approaching zero and higher Lower Established after good condition and no orientation lock It then enters the IDLE state. The conditions for maintaining the IDLE state are insufficient evidence of overall direction and no effective cumulative evidence for initiation; when there is high perturbation but no overall direction at medium and long scales, it can temporarily enter the IDLE_DIST state.

[0126] (2) From a standstill to upward acceleration or downward acceleration In IDLE or IDLE_DIST state, if consistently higher The positive evidence at both the mesoscale and longscale scales is valid. and The startup requirements have been met. and If not rejected, and the cumulative upward direction evidence reaches the threshold, then UP is locked and ACC_UP is confirmed. Correspondingly, negative cross-scale evidence and When dominant, lock DOWN and confirm ACC_DOWN. Only short-scale spikes, rapid direction reversals, or... When the temperature is high, maintain the original state.

[0127] (3) Acceleration of upward movement and acceleration of downward movement The condition for maintaining the ACC_UP state is The positive or decaying startup evidence still conforms to the acceleration template and has not yet met the legal exit conditions; ACC_DOWN is symmetric to its sign. After the accelerated state reaches the minimum dwell time, if... Back to near zero and If the velocity continues to rise without evidence of deceleration in the opposite direction, it can transition to the corresponding uniform velocity state. If the cross-scale directional evidence directly and stably reverses, it is permissible to directly enter the corresponding deceleration state along the shortest path.

[0128] (4) Uniform speed going up and uniform speed going down UNIFORM_UP status requirements It has been previously confirmed that the upward acceleration has occurred and the current vertical dynamic acceleration has returned to zero on the medium to long scale. and The deceleration is relatively high and there is no evidence of sustained negative deceleration. Since the UP state is still locked, it is not considered IDLE at this time. The UNIFORM_DOWN state is symmetrical to it. A single reverse spike appears during the constant velocity phase, accompanied by a high [speed / deceleration]. or At the same time, maintain a constant speed and output a disturbance flag during operation.

[0129] (5) Upward deceleration and downward deceleration When UP is locked, it remains negative. ,high Evidence of a legitimate exit supports entry into the DEC_UP state; when the DOWN state is locked, it continues to move positively. Supports entering the DEC_DOWN state. The deceleration state persists until evidence of deceleration in the opposite direction disappears, and the long-scale component returns to the stationary baseline. rise, The trend is downward and there is no new valid directional evidence. After reaching the threshold for stopping the accumulation of evidence, it enters the IDLE state and the directional lock is released.

[0130] (6) IDLE_DIST and ABNORMAL states The overall direction lacks sufficient evidence, but or When the value is high, the IDLE_DIST state is output, and the static baseline and template update are frozen. When there is sensor saturation, continuous data loss, abrupt attitude changes, loose installation, persistent severe conflict across scale directions, extremely abnormal state duration, or significant mismatch in start / stop impulses, an abnormal pending confirmation or ABNORMAL state is output. After the abnormality disappears, the IDLE state and normal judgment will only resume after continuous and reliable static conditions are regained.

[0131] 12. Calculate the impulse closure residual for the start-up and braking segments of the same stroke. A normal stroke begins at rest and eventually returns to rest; therefore, the velocity change during the start-up phase should be offset by the opposite velocity change during the braking phase. This step calculates the signed acceleration area only for the segmented, finite segments of start-up and braking, for physical consistency verification.

[0132] In the formula, The area of ​​signed acceleration for initiating accelerated candidate segments, in m / s². This is the set of sampled indices from the start candidate boundary to the acceleration end candidate boundary; For long-scale smooth vertical dynamic acceleration; It is a long-scale reliable static residual center; The sampling period. It is only used to represent the effect of velocity change on this finite segment.

[0133] In the formula, The signed acceleration area for the candidate segments of braking and deceleration, in m / s². This is the set of sampled indices between the candidate boundary for deceleration and the candidate boundary for stopping. During normal uplink... Usually positive, It is usually negative; the sign is reversed when it is in a normal downward direction.

[0134] In the formula, The range of the start / stop impulse closure residual is limited to [0,1]. Represents absolute value; To prevent positive numbers with a denominator of zero. The closer it is to 0, the more the starting and stopping effects cancel each other out; A larger value indicates that there may be problems with the state segment, candidate boundary, baseline, or signal quality.

[0135] The deceleration is not yet over; only the observed portion can be used. A transient closed lower bound is formed, but normal deceleration is not rejected prematurely based on this; the complete calculation is performed only after confirmation. If the complete residual exceeds the update threshold, the system reduces the confidence of the current travel sequence, prohibits individual template updates, and allows relocation of the start or deceleration boundary within a limited candidate boundary neighborhood.

[0136] It should be noted that Ω acc and Ω dec During the real-time processing phase, the confirmed state transition time is used as the temporary boundary; after a complete cycle is completed and the finite historical boundary is corrected, the boundary can be based on the corrected boundary B. Recalculate I acc I dec and R I This is used for subsequent template update decisions.

[0137] 13. Jointly modify the state transition boundary within a finite historical neighborhood. Real-time state transitions require accumulating evidence, so confirmation typically occurs later than the actual physical boundary. This step does not backtrack through the entire infinite history; instead, it sets up a finite number of candidate intervals near each initial boundary and re-compares a small number of candidate boundary combinations.

[0138] In the formula, B is the candidate boundary vector for one stroke; The sampling or analysis cycle number is the nth state transition boundary; N is the number of boundaries to be corrected in this process. The search interval is a finite range near the initial boundary; Let be the Cartesian product of each finite interval. Each Ideally, only a limited time before and after the confirmation time should be covered, without traversing the entire original data.

[0139] In the formula, This is the joint evaluation value of candidate boundary combination B; For a finite set of analysis periods affected by candidate boundary adjustments; This represents the state corresponding to the k-th period after being divided according to B; This indicates the degree of matching for this state; This represents the number of state fragments. and These are the nth state segment and its duration; For duration probability; Consistency reward weight; Weighting for impulse residual penalties; For perturbation penalty weights; This represents the closed residual obtained after re-dividing the segment according to B.

[0140] In the formula, This is the final modified boundary combination used; Indicates the choice to make The largest candidate; the candidate must also satisfy a valid state diagram, state sequence, and shortest dwell time for each state. During real-time verification, only the most recent boundary can be corrected first, and after stopping verification, the start, constant speed, and deceleration boundaries can be put into a finite combination for stroke-level verification.

[0141] 14. Output transient and confirmation tags, handle exceptions, and selectively update individual templates. In each analysis cycle, the system first outputs a transient real-time label, transient confidence level, and accompanying disturbance flag for current display and alarm purposes. Upon meeting the confirmation criteria, a confirmation label is output; after completing finite boundary correction, the corrected transition timestamp is also output. The confirmation label is used for travel statistics, state reconstruction, and storage of reliable training data.

[0142] Abnormal and ABNORMAL states are not considered normal template update samples. After the sensor recovers, the system must regain a reliable stationary baseline, good signal quality, and direction unlocked conditions before resuming normal judgment from the INIT or ABNORMAL state.

[0143] Maintain template centers for each state s and each feature j. Transition width and duration distribution. The robust centers of the newly acquired samples that meet the update criteria are denoted as... .

[0144] In the formula, and These are the template centers of the j-th item in states s before and after the update, respectively. This represents the median or truncated mean of the current reliable batch. The update rate is between 0 and 1; a smaller value indicates a slower update. Transition width. The batch quantile span can be updated using the same forgetting method, but a minimum lower limit and a single change upper limit must be set.

[0145] For the elevator operating status set and main transition rules, please refer to Table 1 and Table 2 respectively.

[0146] Table 1 Elevator Operating Status Set Table Table 2 Elevator Operation Status Transfer Rules The following explains the specific identification and transition process of the states INIT, IDLE, IDLE_DIST, ACC_UP, UNIFORM_UP, DEC_UP, ACC_DOWN, UNIFORM_DOWN, DEC_DOWN, and ABNORMAL.

[0147] The following is a detailed explanation using the complete descent as an example. This descent consists of three phases: acceleration, constant speed operation, and deceleration.

[0148] 1. Obtain the vertical dynamic acceleration of the shaft from data from a triaxial sensor. First, a reliable set of stationary data Ω is selected from the stable stationary data from 0.5s to 1.8s before the start of the journey. idle Substituting the sampled values ​​of each axis into the median formula described above, we obtain the stationary baseline. m / s². For the current triaxial acceleration Subtracting the stationary baseline yields the dynamic acceleration vector. After completing the installation orientation calibration, the vertical unit vector e of the shaft is... z =[0,0,1] T Therefore, Substitute into the vertical projection formula You can get Figure 4 The vertical dynamic acceleration of the shaft is shown; simultaneously, a is obtained according to the lateral projection formula. h (k) is used for subsequent judgment of lateral disturbances.

[0149] For example, at 4.5s, a d (k)=[ 0.010320, 0.007453, 0.776875] T m / s², substituting into the vertical projection formula, we get a v (k)= 0.776875 m / s². Since upward is defined as positive, this negative value indicates that the car generated downward dynamic acceleration during the downward start-up phase, but at this time the operating state is not directly determined based on a single sampling point.

[0150] Table 3 Examples of Triaxial Dynamic Acceleration 2. Obtain low-frequency operating components and high-frequency residuals The short-scale, medium-scale, and long-scale windows are set to 0.2s, 1.0s, and 3.0s, respectively, with corresponding sampling points of N. S =200、N M =1000、N L =3000. Let a v (k) Substituting into the above causal weighted smoothing formula, we obtain the low-frequency components at each scale. .

[0151] Figure 5 Give long-scale low-frequency components This is used to characterize the overall acceleration, constant speed, and deceleration trends of the car. Then, from... ,get Figure 6 The high-frequency residuals are shown, and calculated using the root mean square formula over the most recent 200 samples. Figure 7 The high frequency intensity shown ...

[0152] Taking point 4.5 as an example, the smoothing calculation yields... =-0.804120 m / s², and thus obtain =0.027245 m / s²; calculated within a short-scale window. =0.023005 m / s². Low-frequency components and Both are clearly negative, but the high-frequency residuals are smaller, indicating that the change mainly comes from the overall downward start of the car, rather than from collisions or passenger disturbances that occur only in a short time.

[0153] Table 4 Examples of Low-Frequency, High-Frequency, and High-Frequency Intensity Calculations 3. Forming normalized observations and cross-scale directional evidence. The system first calculates the quantile centers and scales of each feature using reliable static data, and then calculates the results using the truncated normalization formula. This was done to reduce the impact of differences in vibration amplitude between elevators of different dimensions and individual elevators. Subsequently, the positive and negative areas were accumulated within each scale window. and Calculate the dominance of the direction , directional continuity and zero acceleration stability and combined Horizontal energy ratio and accelerometer energy Formation of disturbance index .

[0154] At 4.5 s, ,and Substituting the above results into the formulas for calculating overall directional evidence and cross-scale directional consistency, we obtain... =-0.999377、 =1. This means that the short, medium, and long scales continuously support the same negative direction; simultaneously... =0.000307、 =0.164777, which does not constitute a disturbance rejection. Therefore, this signal segment provides credible evidence of a downward initiation direction.

[0155] Table 5 Examples of cross-scale directional evidence calculation Table 6 Examples of Stability, Energy, and Disturbance Characteristic Calculations 4. Calculate the degree of state matching and lock the travel direction. , , , , , , , and Equal components form the state observation vector and the candidate state template obtained from the trusted complete journey. Compare and obtain the matching score for each candidate state. In the IDLE stage, before the direction is locked, the cross-scale directional evidence, consistency, long-scale energy, perturbation exponent, and lateral energy ratio are substituted into the above-mentioned initiation score formula. At 4.5 s, the result is obtained... =0.750967、 =0.300988; After the downlink score continues to dominate and the limited cumulative directional evidence reaches the threshold, it will Set to DOWN to transition from IDLE state to ACC_DOWN state.

[0156] Table 7. Examples of Directional Start-up Scores and Status Outputs 5. Identify uniform and decelerating states by combining legal paths, stage memory, and duration constraints. After locking the DOWN state, compare states only within the legal candidate set; do not release the direction lock due to a single reverse spike. After 6.93 s, the low-frequency components gradually return to the zero band; at 20.0 s, =0.022140 m / s², = = =0, =0, and = =1 indicates that there is currently no new evidence of acceleration or deceleration direction. Because It is still DOWN, and the ACC_DOWN state has already met the minimum dwell time and exit risk rate conditions, so it is identified as the UNIFORM_DOWN state, rather than the IDLE state, which is also close to zero in terms of acceleration.

[0157] At 42.5 seconds, when the elevator approaches the target floor... =0.785781 m / s², =0.822690 m / s², = =1、 =0.999808, thus obtaining =0.999904、 =1. This positive evidence, if it occurs during a stationary phase without a locked direction, can be considered a candidate for upward initiation; however, at this moment... =DOWN, the previous confirmed state was UNIFORM_DOWN, and the valid state diagram only allows it to enter the downward deceleration candidate. Therefore, the system interprets this positive acceleration as the upward deceleration when the car stops downward. After the DEC_DOWN matching score, duration hazard rate and limited transition evidence all meet the conditions, it confirms the entry into the DEC_DOWN state at 40.70s, instead of misjudging it as the ACC_UP state.

[0158] 6. Verify stroke closure, correct state boundaries, and output results. Perform finite-interval integration on the initial state segment for the acceleration and deceleration phases to obtain the results. =-2.521161 m / s =2.530690 m / s; Substituting into the impulse closure residual formula, we get =0.001886. This value is close to 0, indicating that the area of ​​signed acceleration during the start and stop of the same downstroke is basically closed. It can be used to support the integrity of the state sequence, but the integral result is not used as the recursive velocity or displacement.

[0159] Table 8. State boundaries and duration of the complete downlink journey Real-time limited evidence first gives the initial boundary =[3.0, 7.1, 40.9, 45.3] s. The objective function is computed only within a finite historical neighborhood of 0.5 s before and after each initial boundary. By combining state matching degree, condition duration probability, cross-scale consistency, impulse closure residual, signal quality, and legal state diagram, the corrected boundary is obtained. =[2.79, 6.93, 40.70, 45.08] s. Therefore, the durations of the ACC_DOWN, UNIFORM_DOWN, and DEC_DOWN states are 4.14 s, 33.77 s, and 4.38 s, respectively.

[0160] Finally, transient tags are output in real time, and confirmation tags and corresponding timestamps are output after joint correction based on limited history. In this example, the complete output is IDLE→ACC_DOWN→UNIFORM_DOWN→DEC_DOWN→IDLE. Because the travel signal quality is normal, the disturbance is low, the state path is valid, and the impulse is closed, it is allowed to be used for limited updates of the individual state template; if any of these conditions are not met, the identification result will still be output, but the template will not be updated using this travel signal.

[0161] 7. Other Statuses and Special Circumstances (1) INIT state Upon startup, since a reliable static baseline has not yet been established and the historical direction of this trip has not been obtained, INIT is output first, indicating that the status is pending confirmation.

[0162] For example, if the elevator starts during a period of constant downward speed, the vertical dynamic acceleration of the shaft may be close to zero, but it's impossible to determine whether the elevator is currently stationary or moving at a constant downward speed. Therefore, instead of directly outputting the IDLE or UNIFORM_DOWN state, the INIT state is maintained.

[0163] When the vertical dynamic acceleration of the elevator shaft is continuously detected to be stable within the zero-band range, no effective directional evidence is formed at short, medium, and long scales, the disturbance index is lower than the stationary disturbance threshold, and the signal quality is normal, a credible stationary baseline is established, and the system transitions from INIT to IDLE. If a complete deceleration and stopping process is detected first after startup, a credible stationary baseline can also be established after the elevator finally returns to a stable station, and then the system transitions from INIT to IDLE.

[0164] (2) IDLE state The elevator stops at a certain floor, and the vertical dynamic acceleration of the shaft fluctuates slightly near the stationary baseline. At this time, the directional dominance at short, medium, and long scales is low, the overall directional evidence across scales is close to zero, the zero acceleration stability at medium and long scales is relatively high, and the travel direction lock variable is NONE, so the system transitions to the IDLE state.

[0165] In IDLE state, slow updates to the trusted static baseline are allowed, but updates are only performed when signal quality is normal, disturbance index is low, and no evidence of a start-up direction has been formed.

[0166] (3) IDLE_DIST state The elevator remains stopped at the floor, but passengers inside the car are shaking their legs, walking around, or moving their luggage. At this point, multiple acceleration spikes appear in the short-scale signal, with an increase in the proportion of high-frequency energy, jerk energy, or lateral energy, but no consistent positive or negative directional evidence is formed at the mesoscale and long-scale.

[0167] For example, the short-scale directional dominance briefly becomes positive and then negative, cross-scale directional consistency is low, and the perturbation index exceeds the stationary perturbation threshold. This situation is not classified as an uplink or downlink start, but rather transitions from the IDLE state to the IDLE_DIST state.

[0168] In the IDLE_DIST state, the static baseline and individual state template updates are frozen. When passenger activity ceases, high-frequency disturbances gradually decrease, the shaft vertical dynamic acceleration stabilizes again within the static zero band, and no valid start-up evidence is formed for several consecutive analysis cycles, the system recovers from IDLE_DIST to IDLE state.

[0169] (4) Uplink status recognition The uplink state uses the same calculation process as the complete downlink implementation, but the direction sign and legal state path are changed accordingly. In the IDLE state, the uplink initiation score is determined when consistent positive initiation evidence is consistently generated across short, medium, and long scales. consistently higher than the downtrend initiation score Furthermore, when the evidence from a limited number of directions reaches a threshold, the direction of travel is locked to a variable. Set it to UP, and switch from IDLE to ACC_UP state.

[0170] After the upward acceleration ends, the low-frequency components return to the zero band, and the stability of zero acceleration at medium and long scales increases. Before sustained negative evidence is formed, the state transitions from ACC_UP to UNIFORM_UP. Approaching the target floor, the cross-scale directional evidence turns into sustained negative acceleration. Combining the UP directional lock and the legal state path, this negative acceleration is interpreted as upward braking and deceleration, and the state transitions from UNIFORM_UP to DEC_UP. After the stopping condition is met, the directional lock is released, and the state returns to IDLE, thus obtaining IDLE→ACC_UP→UNIFORM_UP→DEC_UP→IDLE.

[0171] (5) Short strokes without a distinct constant speed phase For short travel distances such as between adjacent floors, the elevator car may accelerate and then immediately enter deceleration without a stable constant speed phase. If continuous positive deceleration evidence is detected directly after the ACC_DOWN state, and the DEC_DOWN template matching degree, duration constraint, and finite transition evidence all reach their thresholds, then ACC_DOWN is allowed to directly transition to the DEC_DOWN state; for short upward travel, ACC_UP is allowed to directly transition to the DEC_UP state. Non-existent constant speed states are not forcibly filled in.

[0172] (6) Local impact and mechanical vibration When an elevator is stationary or in operation, and there are passenger foot stomping, luggage collisions, guide rail joint vibrations, or mechanical vibrations, large positive or negative spikes may appear on a short scale. However, on medium and long scales, consistent evidence in the same direction is usually not formed. or The change is treated as a local disturbance: when stationary, IDLE_DIST can be output; during operation, the original running state is retained and a disturbance flag is added, and the travel direction is not changed due to a single spike.

[0173] (7) Signal abnormality and ABNORMAL status When sensor saturation, continuous sample loss, abrupt attitude changes, signal quality consistently below the threshold, state duration significantly exceeding the allowable range, state sequence violating a valid state diagram, or impulse closure residual exceeding the threshold is detected, the system outputs an ABNORMAL status or a corresponding anomaly flag. During the anomaly period, the stationary baseline and individual template updates are frozen; after the signal returns to normal, the system needs to re-obtain a reliable stationary segment or a complete valid state change before resuming normal state confirmation.

[0174] (8) Restrictions on itinerary verification and template updates Both impulse closure and template update are performed after the current trip ends, without affecting the real-time transient tags previously output with an analysis cycle of 0.1 s. Only complete trips with normal signal quality, low disturbance, legal state path, reasonable duration, and impulse closure are used to update the corresponding state template; trips that do not meet the above conditions can still retain the identification results and anomaly records, but will not participate in template updates.

[0175] To address the issue that real-time elevator operation status identification is easily affected by factors such as sensor bias, random noise, changes in installation posture, local passenger movements, and mechanical vibration, this paper aims to reliably identify specific elevator operation states—stationary, accelerating upwards, constant speed upwards, decelerating upwards, accelerating downwards, constant speed downwards, decelerating downwards, and abnormal—without relying on external floor signals or traction machine control signals. By constructing three causal scales (short, medium, and long), the paper calculates the directional dominance and directional persistence at each scale, and then synthesizes cross-scale overall directional evidence and directional consistency to distinguish between overall car motion and local disturbances. A travel direction locking mechanism ensures that even when acceleration returns to zero during the constant speed phase, the paper correctly identifies the constant speed state instead of misjudging it as stationary. The paper establishes the duration probability, survival probability, and exit hazard rate for each operation state, adaptively adjusting the state transition timing. The paper calculates the impulse closure residual for the same travel start-stop segment to verify the physical consistency of the state sequence. Finally, it jointly corrects the state transition boundary within a finite historical neighborhood to reduce boundary delays caused by evidence accumulation. With low computational load and strong interpretability, it meets real-time requirements and is suitable for real-time deployment of edge computing hosts. It can accurately identify the elevator operating status in real time and improve the efficiency of elevator operating status monitoring.

[0176] The real-time elevator operation status identification method and system based on cross-scale directional consistency provided in this invention have the following beneficial effects: 1. Cross-scale directional evidence fusion effectively distinguishes between overall motion and local perturbations. Directional dominance and directional persistence are calculated at short, medium, and long time scales, and cross-scale overall directional evidence and directional consistency are further calculated. Continuous overall car motion can form mutually supporting directional evidence at multiple scales, while local collisions, passenger foot shaking, and short-term mechanical shocks typically only produce responses at certain scales. This mechanism enables the present invention to accurately identify the true operating state of the elevator even under strong disturbance environments, significantly improving the ability to distinguish between normal operating conditions and local disturbances.

[0177] 2. Stroke direction locking mechanism accurately identifies the constant speed phase. The current acceleration signal is interpreted by combining the travel direction lock, the previously confirmed state, and the legal state transition relationship. When the elevator enters the constant speed phase and the vertical dynamic acceleration of the shaft approaches zero, the system can still accurately identify it as either upward or downward constant speed based on the locked travel direction and the previously confirmed acceleration phase, fundamentally eliminating the possibility of misjudging constant speed movement as stationary. This mechanism overcomes the long-standing technical bias in those skilled in the art that zero acceleration equates to stationary motion.

[0178] 3. State-dependent duration probability model for adaptive adjustment of transition timing. For different operating states, separate probability of conditional duration, survival probability, and exit risk rate are established. State maintenance and exit are controlled by combining state transition evidence within a finite time window. Compared to using a uniform fixed number of confirmations for all states, this invention can adaptively adapt to the differences in state durations such as acceleration, constant speed, deceleration, and short travel. It can respond promptly in short travel and avoid premature exit in long travel, significantly reducing the possibility of premature or excessively delayed state switching.

[0179] 4. Start / stop impulse closure verification to ensure physical consistency. The signed acceleration area is calculated only for candidate acceleration and deceleration segments within the same stroke, over a finite interval. The physical consistency of the state sequence and candidate boundaries is evaluated using the start-stop impulse closure residual. This acceleration area is not used for recursive calculation of velocity or displacement across strokes, thus providing auxiliary basis for confirming the complete stroke and verifying state boundaries while avoiding long-term integral drift.

[0180] 5. Limited historical boundary correction, balancing real-time performance and accuracy. During real-time state confirmation, transient state labels are continuously output. After each cycle, the state boundary is corrected only within a limited candidate neighborhood near the initial transition boundary by jointly utilizing state matching degree, duration probability, cross-scale directional consistency, perturbation exponent, and start-stop impulse closure residual. This dual-track mode of "real-time transient output + post-event limited correction" effectively reduces the state transition time deviation caused by evidence accumulation while retaining the real-time state output capability, making the final output state transition timestamp closer to the true physical boundary.

[0181] 6. Selective template updates to prevent parameter degradation. Only trips that meet multiple conditions, including a valid state path, a credible final restoring to a static state, low disturbance, normal signal quality, and a complete impulse closure residual below the update threshold, participate in template updates. Abnormal, low-confidence, and strongly disturbed trips are excluded. This mechanism effectively prevents abnormal data from contaminating normal templates, ensuring the accuracy and stability of identification during long-term online operation.

[0182] 7. Completely non-intrusive deployment, highly adaptable. Only an accelerometer needs to be installed on the elevator car; there's no need to connect it to the elevator controller, encoder, or floor control signals. Installation is simple and has no impact on the original elevator control system. All calculations can be completed in real time on the edge computing host, eliminating the need to upload raw data to the cloud, thus reducing communication bandwidth requirements and privacy and security risks.

[0183] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for real-time identification of elevator operating status based on cross-scale directional consistency, characterized in that, include: Collect the triaxial acceleration signals output by the acceleration sensor installed on the elevator car, establish the vertical coordinates of the shaft, and extract the vertical dynamic acceleration of the shaft; Three causal analysis scales—short-scale, medium-scale, and long-scale—are constructed. Causal smoothing is performed on the vertical dynamic acceleration at each scale, and the directional dominance and directional persistence at each scale are calculated. Based on the directional dominance and directional persistence at these three scales, calculate the overall directional evidence across scales and the directional consistency across scales. Based on the cross-scale overall direction evidence and the cross-scale direction consistency, combined with the uplink initiation score and the downlink initiation score, the direction of this trip is locked when the direction is not locked. Establish a legal state diagram for the elevator, and determine a set of candidate states based on the current confirmed state and the locked travel direction. The legal state diagram for the elevator includes stationary, upward acceleration, upward constant speed, upward deceleration, downward acceleration, downward constant speed, downward deceleration, and abnormal state. A state template is established for each candidate state in the candidate state set, and the matching degree of each candidate state is calculated based on the state observation vector of the current analysis period. Establish the condition duration distribution, survival probability, and exit risk rate for each state, and calculate the hold-up and exit terms based on the duration of the current state; For the current confirmed state and each legal target state, calculate the maintenance score and transition score respectively, and accumulate the transition advantage within a finite time window. When the transition evidence meets the preset threshold, confirm the state transition.

2. The real-time elevator operation status identification method based on cross-scale directional consistency according to claim 1, characterized in that, The extraction wellbore vertical dynamic acceleration includes: Once it is confirmed that the elevator is stable and stationary, the direction is not locked, there is no continuous disturbance, and the signal quality is normal, the most recent reliable stationary sample is selected, and the median of the three axes is taken as the three-axis reliable stationary baseline vector. Based on the installation calibration or reliable static gravity direction combined with positive and negative sign calibration, establish the vertical unit vector of the shaft, and uniformly define the upward direction along the shaft as positive; The triaxial dynamic acceleration vector is obtained by subtracting the triaxial reliable static baseline vector from the triaxial acceleration measurement vector. The triaxial dynamic acceleration vector is then projected onto the vertical unit vector of the shaft to obtain the vertical dynamic acceleration of the shaft.

3. The real-time elevator operation status identification method based on cross-scale directional consistency according to claim 1, characterized in that, The calculation of directional dominance and directional persistence at these three scales includes: For each scale, obtain the causal smoothed acceleration sequence within the current analysis window, calculate the positive and negative signed areas within the current analysis window respectively, and calculate the directional dominance using the following formula: In the formula, To analyze the directional dominance of scale r, the range is... A r + (k) represents the positively signed area; A r (k) represents the negative signed area; To prevent positive numbers with a denominator of zero; r is the analysis scale; k is the current analysis cycle number; The current analysis window is divided into multiple consecutive overlapping sub-windows. The directional dominance is calculated for each sub-window. Sub-windows whose absolute values ​​reach a threshold are considered valid sub-windows. Sub-windows whose signs align with the current scale's principal direction are considered co-directional sub-windows. Directional persistence is calculated using the following formula: In the formula, For scale directional continuity, normalized to ; For the mean scale The number of child windows that are in the same direction as the current main window; This represents the number of valid child windows. To prevent positive numbers with a denominator of zero.

4. The real-time elevator operation status identification method based on cross-scale directional consistency according to claim 3, characterized in that, The calculation of cross-scale overall directional evidence and cross-scale directional consistency includes: Calculate the overall directional evidence using the following formula: In the formula, This serves as overall directional evidence, ranging from [-1, 1]. The weighting coefficients for the analysis scale r; The degree of directional dominance; For directional continuity; These represent short-scale, medium-scale, and long-scale, respectively. Cross-scale directional consistency is calculated using the following formula: In the formula, For cross-scale directional consistency, the range is [0,1]; The weighting coefficients for the analysis scale r; The degree of directional dominance; For directional continuity; To prevent positive numbers with a denominator of zero.

5. The real-time elevator operating status identification method based on cross-scale directional consistency according to claim 1, characterized in that, The process of locking the direction of this trip includes: The uplink initiation score and downlink initiation score are calculated separately. The uplink initiation score rewards positive global directional evidence, cross-scale directional consistency and long-scale low-frequency energy, and penalizes the perturbation index and lateral energy ratio. Accumulate the initiation score advantage in each direction within a limited evidence window. When the accumulated evidence in a certain direction reaches the locking threshold and the initiation score, cross-scale directional consistency, and long-scale low-frequency energy in that direction are all effective, the travel direction is locked as that direction. Once the direction is locked, a single reverse acceleration spike does not change the locked direction. The direction lock is only released when it is confirmed that the object has returned to rest, during the re-initialization process after entering an abnormal state, or when a persistent and serious directional contradiction is detected.

6. The real-time elevator operating status identification method based on cross-scale directional consistency according to claim 5, characterized in that, The calculation of the matching degree of each candidate state based on the state observation vector of the current analysis period includes: Form the state observation vector for the current analysis period: In the formula, This is the state observation vector for the kth analysis period; Evidence is provided for the overall direction, within the range [-1,1]. For cross-scale directional consistency, the range is [0,1]. and These represent the zero-acceleration stability at mesoscale and longscale, respectively. This represents the normalization result for long-scale low-frequency energy. The disturbance index; This refers to the horizontal energy ratio; Signal quality, range [0,1]; For each candidate state s, the matching value of each feature is calculated using the membership function in the pre-established state template, and the weighted matching degree is calculated using the following formula: In the formula, The degree of matching between the current observation and the candidate state s, ranging from [0,1]; Let f(j) represent the membership function of the j-th feature in state s; For the corresponding normalized features; For template weights, when s belongs to the candidate set Only then will it participate in the normal conversion comparison.

7. The real-time elevator operating status identification method based on cross-scale directional consistency according to claim 1, characterized in that, The method also includes the steps of establishing the conditional duration distribution, survival probability, and exit hazard rate of the elevator operating status: For each state s, establish a duration probability distribution under given condition c; Calculate the survival probability using the following formula: In the formula, Let Pr be the survival probability, representing the probability that state s will persist to the d-th period under condition c; Pr represents the probability; D s Let be a random variable representing the duration of state s; u is the number of possible duration periods; The maximum duration period; This represents the probability that the running state s continues for exactly u analysis cycles under condition c; Calculate the exit risk rate using the following formula: In the formula, The discrete exit risk rate represents the conditional probability that state s will exit at the end of the current period when it continues to the d-th period under condition c, and its range is limited to [0,1]. This represents the probability that state s continues for exactly d analysis periods under condition c; Probability of survival; To prevent positive numbers with a denominator of zero.

8. The real-time elevator operating status identification method based on cross-scale directional consistency according to claim 1, characterized in that, The method also includes the step of calculating the impulse closure residual for the start-up and braking segments of the same stroke: For the candidate acceleration segments of this journey, the area of ​​signed acceleration is calculated using the following formula: In the formula, The area of ​​signed acceleration for initiating accelerated candidate segments, in m / s². This is the set of sampled indices from the start candidate boundary to the acceleration end candidate boundary; For long-scale smooth vertical dynamic acceleration; It is a long-scale reliable static residual center; The sampling period; For the candidate braking and deceleration segments of this journey, the area of ​​signed acceleration is calculated using the following formula: In the formula, The signed acceleration area for the candidate segments of braking and deceleration, in m / s². The set of sample indices between the deceleration start candidate boundary and the stop candidate boundary; Calculate the impulse closure residual using the following formula: In the formula, The range of the start / stop impulse closure residual is limited to [0,1]. The area of ​​signed acceleration for initiating accelerated candidate segments; The area of ​​signed acceleration for candidate segments of braking and deceleration; Represents absolute value; To prevent positive numbers with a denominator of zero.

9. The real-time elevator operating status identification method based on cross-scale directional consistency according to claim 1, characterized in that, The method further includes the step of jointly revising the state transition boundary within a finite historical neighborhood: Using the initial state transition boundary as the center, a finite search interval is set for each boundary to be corrected, and a set of candidate boundary combinations is constructed. For each candidate boundary combination, the joint evaluation value is calculated using the following formula: In the formula, This is the joint evaluation value of candidate boundary combination B; For a finite set of analysis periods affected by candidate boundary adjustments; The degree of state matching; This represents the state corresponding to the k-th period after being divided according to B; This represents the number of state fragments. and These are the nth state segment and its duration; Let c be the duration probability under condition c; Consistency reward weight; Weighting for impulse residual penalties; This represents the closed residual obtained after re-dividing the segments according to candidate boundary combination B; For perturbation penalty weights; Q k The comprehensive disturbance index; The candidate boundary combination that maximizes the joint evaluation value and satisfies the constraints of a valid state diagram and the shortest residence time is selected as the final modified boundary.

10. A real-time elevator operating status identification system based on cross-scale directional consistency, used to implement the real-time elevator operating status identification method based on cross-scale directional consistency as described in any one of claims 1 to 9, characterized in that, include: An accelerometer sensor, which is installed on the elevator car, is used to collect triaxial acceleration signals; An edge computing host, which is used for real-time signal processing and elevator operation status identification; A cloud service platform is used to receive data from multiple edge computing hosts, perform data analysis, and centrally manage the data.

Images