A collaborative timing control method for four-way switching of a four-way shuttle car for a material bin

Abstract

This application relates to the field of warehousing and logistics equipment scheduling technology, specifically to a collaborative timing control method for four-way switching of a four-way shuttle car for material bins. The method includes: acquiring the measured duration of each reversing sub-state of the four-way shuttle car for material bins, and collecting the current operating state variables affecting the complete reversing time; storing the measured durations and current operating state variables accordingly, and fitting the predicted mean and predicted residual sequence of reversing time based on historical reversing data; deriving the confidence expansion coefficient based on the predicted mean and predicted residual sequence of reversing time, and constructing an upper bound for the individualized reversing time prediction interval; determining the expected occupancy time window based on the upper bound of the individualized reversing time prediction interval, and adjusting the uniform speed of arriving vehicles to mitigate potential intersection conflicts. This method effectively improves the accuracy of the scheduling system's control over vehicle reversing capabilities.

Description

Technical Field

[0001] This application relates to the field of warehousing and logistics equipment scheduling technology, specifically to a collaborative timing control method for four-way switching of a four-way shuttle car for material bins. Background Technology

[0002] With the development of modern automated warehousing systems, four-way shuttle carts, with their ability to flexibly switch travel directions on horizontal and vertical tracks, are widely used in high-density automated warehouses. Four-way switching, as the core action in the shuttle cart's operation, directly determines the intersection traffic efficiency in environments with dense multi-vehicle traffic. Accurately reserving intersection occupancy time during the coordinated scheduling of multiple shuttle carts is a crucial prerequisite for preventing vehicle collisions.

[0003] Existing four-way shuttle dispatch control mechanisms typically rely on fixed nominal values ​​to reserve time for intersection occupancy during reversing actions. However, in actual operation, the demand for lifting mechanism motors varies greatly depending on the load of the cargo bins. At the same time, fluctuations in battery charge and the long-term gradual wear of equipment guide components will continuously and unevenly affect the actual reversing time. The existing global fixed value evaluation method seriously ignores the individual characteristics of vehicles and the dynamic operating condition disturbance patterns.

[0004] Because it is impossible to accurately capture the microscopic effects of various operating status variables and time loss variables on the actual time of a single reversal, it is extremely easy to cause potential intersection deadlocks due to inaccurate reserved windows. This forces the dispatching system to frequently use emergency stop and start intervention, which greatly consumes the vehicle's kinetic energy and causes rigid rhythmic impacts on the system's mechanical components, reducing the overall throughput efficiency of the warehouse system and the smoothness of multi-vehicle collaborative operation. Summary of the Invention

[0005] To address the existing technical problems of individual shuttle car reversal time fluctuations leading to inaccurate reserved time windows, resulting in intersection deadlocks and reduced operational smoothness due to equipment sudden stop impacts, this application provides a collaborative timing control method for four-way switching of a four-way shuttle car.

[0006] In a first aspect, this application provides a collaborative timing control method for four-way switching of a four-way shuttle car, comprising: acquiring the measured duration of each switching sub-state corresponding to a single complete switching process of the four-way shuttle car, and collecting the current operating state quantity that affects the complete switching time; storing the measured duration and the current operating state quantity to form historical switching data, and using piecewise linear regression fitting based on the historical switching data to obtain the predicted mean of switching time and the predicted residual sequence; deriving the confidence expansion coefficient based on the predicted mean of switching time and the predicted residual sequence, and constructing an upper bound of the individualized switching time prediction interval for the current four-way shuttle car based on the confidence expansion coefficient; determining the expected occupancy time window based on the upper bound of the individualized switching time prediction interval, and when it is detected that the intersection occupancy time windows of multiple vehicles overlap, resolving potential intersection conflicts by adjusting the uniform driving speed of the arriving vehicles.

[0007] By directly extracting sub-state time variables with physical boundaries and combining them with corresponding sensitive state quantities to construct a prediction model, the specific response pattern of a single vehicle is accurately analyzed directly from the micro-loss level. Combined with the introduction of adaptive extended intervals, the vehicle's intersection conflict avoidance is adjusted from passive stopping to pre-driving adjustment.

[0008] Preferably, the step of acquiring the measured duration of each commutation sub-state corresponding to a single complete commutation process of the four-way shuttle car in the hopper, and collecting the current operating state quantities that affect the complete commutation time, includes: dividing the complete commutation process into four serially triggered, independently monitorable sub-states, and recording the time interval between the satisfaction of the precondition and the triggering of the current judgment condition for each sub-state as the measured duration to form a time-series feature vector; synchronously collecting the hopper load, battery state of charge, and cumulative number of commutations as the current operating state quantities, and merging the current operating state quantities and the time-series feature vector into the historical commutation time-series database for subsequent modeling.

[0009] By breaking down the previously unobservable and ambiguous macroscopic time consumption of the entire vehicle, the source of time data for different loss stages is clarified, which greatly improves the quality and reliability of the feature input matrix for model training.

[0010] Preferably, the step of dividing the complete commutation process into four serially triggered, independently monitorable sub-states, and recording the time interval between the satisfaction of the precondition and the triggering of the current judgment condition for each sub-state as the measured duration to form a time-series feature vector, includes: setting a single sub-state timeout protection threshold for each sub-state, adding a fixed margin to the maximum duration of the same type of historical sub-state as the single sub-state timeout protection threshold; when a sub-state does not trigger the judgment condition within the corresponding single sub-state timeout protection threshold, reporting a commutation anomaly event, and marking this commutation record as an abnormal sample for exclusion in subsequent modeling training.

[0011] Preferably, the step of deriving the confidence expansion coefficient based on the predicted mean of the reversing time and the predicted residual sequence, and constructing the upper bound of the individualized reversing time prediction interval for the current four-way shuttle car in combination with the confidence expansion coefficient, includes: calculating the standard deviation of the predicted residual sequence, and arranging all historical predicted residual sequences in ascending order of absolute value to extract the absolute value of the residual corresponding to the percentile distribution point; dividing the extracted absolute value of the residual corresponding to the percentile distribution point by the standard deviation of the predicted residual sequence as the confidence expansion coefficient, and adding the predicted mean of the reversing time to the product of the confidence expansion coefficient and the standard deviation as the upper bound of the individualized reversing time prediction interval.

[0012] By objectively relying on the inherent heavy-tailed distribution characteristics of the sample residuals, the redundancy value of the convergence interval is dynamically reduced, so that the time window expansion action effectively matches the mechanical wear and tear of the vehicle itself, avoiding the boundary underreporting problem caused by manual static experience pre-setting.

[0013] Preferably, the step of determining the expected occupancy time window based on the upper bound of the individualized reversal time prediction interval, and mitigating potential intersection conflicts by adjusting the uniform speed of arriving vehicles when overlapping intersection occupancy time windows of multiple vehicles are detected, includes: obtaining the mean and standard deviation of the historical switching interval sequence of the intersection and adding them together to generate a safety interval margin, and adding the safety interval margin to the overlapping duration of the occupancy time window as the required delay duration; calculating the ratio of the remaining travel distance of the arriving vehicle from the target intersection to the current travel speed, and dividing the remaining travel distance by the sum of the ratio and the required delay duration as the reduced uniform speed, which is then sent to the arriving vehicles.

[0014] By combining the upper limit of natural fluctuations to generate closed-loop deceleration tracking parameters, the risk of intersection nodes that would otherwise lead to rigid waiting is smoothly resolved within the remaining allowable space, and the energy reduction impact caused by frequent braking is significantly reduced.

[0015] Preferably, the step of calculating the ratio of the remaining travel distance of the arriving vehicle to the target intersection to its current travel speed, and then dividing the remaining travel distance by the sum of the ratio and the required delay time to obtain the reduced constant speed, is sent to the arriving vehicles later. This includes: determining whether the calculated reduced constant speed is lower than the minimum allowed constant speed of the vehicle; if the reduced constant speed is not lower than the minimum allowed constant speed, the arriving vehicles smoothly reduce their speed within the current road segment; if the reduced constant speed is lower than the minimum allowed constant speed, the arriving vehicles are controlled to briefly stop at the safe stopping node closest to the target intersection until the window occupied by the arriving vehicles ends.

[0016] Preferably, the step of storing the measured duration and the current operating state quantity to form historical commutation data, and using piecewise linear regression to fit the historical commutation data to obtain the predicted mean of commutation time and the predicted residual sequence, includes: calculating the autocorrelation coefficient sequence corresponding to the complete commutation time series, and deriving the white noise confidence band of the autocorrelation coefficient in combination with the length of the complete commutation time series; obtaining the lag step corresponding to the first time the autocorrelation coefficient enters the white noise confidence band as the sliding window length, and retaining the most recent commutation records within the sliding window length to refit and update the piecewise linear regression model.

[0017] Preferably, the step of deriving the confidence expansion coefficient based on the predicted mean and predicted residual sequence of the reversing time, and constructing the upper bound of the individualized reversing time prediction interval for the current four-way shuttle car in combination with the confidence expansion coefficient, includes: obtaining the quartiles and interquartile ranges of the historical distribution of the upper bound of the individualized reversing time prediction interval for vehicles in the same batch, and constructing an outlier monitoring and judgment criterion; summing the quartiles with the interquartile ranges at fixed multiples, and if the upper bound of the prediction interval of the individualized reversing time of the current four-way shuttle car exceeds the summation result, determining that the current vehicle is at a statistically significant abnormal high point and triggering a mechanism maintenance warning.

[0018] Preferably, the complete reversing process is divided into four serially triggered, independently monitorable sub-states. The four sub-states include the sequentially triggered deceleration and braking completion sub-state, lifting mechanism operation completion sub-state, new direction wheel set and track groove alignment confirmation sub-state, and brake release and completion of the first acceleration start of the new direction sub-state.

[0019] Preferably, the process of refitting and updating the piecewise linear regression model by retaining the most recent few reversing records within the length of the sliding window includes: when the vehicle is brought back online after a long period of inactivity, the current vehicle is instructed to perform several reversing actions in an unloaded state to obtain a preheating calibration reversing timing feature vector; the preheating calibration reversing timing feature vector is appended to the corresponding historical reversing data to trigger an incremental model update to correct the state deviation accumulated during the inactivity period.

[0020] This application can adaptively follow the performance changes of mechanical components throughout the life cycle of the four-way shuttle car and automatically suppress data noise introduced by external sudden working conditions. Through a continuous sliding online fitting process, the upper limit of the prediction maintains a strong correlation with the actual physical reversal time, ensuring a high-security judgment benchmark for the allocation of multiple vehicle nodes from the underlying information flow.

[0021] The intersection management model has been transformed from traditional emergency stop defense to dynamic time compensation feedforward logic. While maintaining the original site track network density, it has improved the driving continuity of individual vehicles. Combined with anomaly detection actions based on distribution characteristics, it can shift the system congestion risk from delayed fault response to early maintenance intervention, thereby improving the long-term operation and maintenance stability of the overall automated warehousing equipment. Attached Figure Description

[0022] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein: Figure 1 This is a flowchart illustrating a collaborative timing control method for four-way switching of a four-way shuttle car in the present invention.

[0023] Figure 2 This is a schematic diagram illustrating the deployment principle of a four-way shuttle car running track and intersection environment of a material bin in this invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0026] This invention discloses a collaborative timing control method for four-way switching of a four-way shuttle car for a material bin, referring to... Figure 1 This includes steps S1-S4: S1. Acquisition and recording of commutation timing sub-states.

[0027] In an optional embodiment, such as Figure 2The diagram illustrates the deployment principle of a four-way shuttle car's running track and intersection environment according to the present invention. The reversing action of the four-way shuttle car is composed of several sub-states with clear physical boundaries. An internal reversing timing state machine is established within the onboard controller, dividing the complete reversing process into four independently monitorable sub-states: deceleration and braking completion (based on the vehicle speed sensor output returning to zero), lifting mechanism operation completion (based on the lifting position switch signal triggering), new direction wheel assembly alignment confirmation with the track groove (based on the wheel assembly pressure sensor reaching rated contact force), and brake release and initial acceleration start in the new direction (based on the establishment of drive motor current). These four sub-states are triggered sequentially. The next sub-state can only proceed after the condition of the previous sub-state is met. If the condition of any sub-state is not met, the state machine remains in the current state and waits, without proceeding.

[0028] Specifically, the vehicle controller records the actual completion time of each sub-state with millisecond-level precision and calculates the measured time interval between adjacent sub-states to form a timing feature vector for this reversal. Its four components correspond to the measured duration of each of the four sub-states, in milliseconds. This timing feature vector is reported in real-time to the warehouse control system via the vehicle's wireless communication module. The warehouse control system uses the vehicle ID as an index to append it to the corresponding vehicle's historical reversal timing database. After each reversal, the number of historical records for that vehicle in the database increases accordingly, continuously accumulating samples for subsequent individualized modeling.

[0029] Furthermore, while recording the timing feature vector, the on-board controller also simultaneously collects and reports three types of state variables that affect the commutation time: the load of the hopper at the time of the commutation, the current state of battery charge, and the cumulative number of commutations since the vehicle left the factory. These three types of state variables, together with the timing feature vector, constitute a complete feature label data record, which is stored in the historical database. Among them, the duration of the sub-state of the lifting mechanism completing its action is most sensitive to the load. When the load increases, the lifting motor needs to overcome greater gravity to do work, and the time of this sub-state is correspondingly extended. The duration of the sub-state of brake release and completion of the first acceleration start in the new direction is most sensitive to the state of battery charge. When the charge is low, the peak current of the drive motor is limited, and the acceleration establishment process is slower. The duration of the sub-state of confirmation of alignment between the new direction wheel set and the track groove shows a slow increasing trend with the increase of the cumulative number of commutations, reflecting the gradual wear of the lifting mechanism's guide components. The three types of influencing factors are independent of each other and jointly determine the actual value of a single complete commutation time.

[0030] Next, for the duration of abnormal sub-states under extreme operating conditions, the vehicle controller sets a timeout protection threshold for each sub-state. Since the natural fluctuation range of the duration of each sub-state can be characterized by the standard deviation of the duration of similar historical sub-states, the timeout threshold for each sub-state is taken as the maximum value of the duration of similar historical sub-states of the vehicle plus a fixed margin. The fixed margin is taken as an integer multiple of the standard deviation of the historical duration of the sub-state, for example, 2 times the standard deviation, so that the timeout threshold covers the upper limit of normal fluctuations while retaining sufficient sensitivity for anomaly identification. If a sub-state does not trigger the judgment condition within the timeout threshold, the vehicle controller immediately reports a reversing anomaly event to the warehouse control system and marks this reversing record as an anomalous sample. Samples marked as anomalous are excluded from the training set in subsequent modeling to prevent extreme outliers introduced by occasional failures such as mechanical jamming from contaminating the parameter estimation of the prediction model.

[0031] In this way, by breaking down the reversing process into four sub-states with clear physical boundaries and collecting the measured duration of each sub-state, the originally unobservable reversing time is transformed into a computable temporal feature vector, providing a real and reliable data foundation for subsequent individualized modeling and effectively improving the scheduling system's cognitive accuracy of each vehicle's reversing capability.

[0032] S2, Construction of Individualized Commutation Time Prediction Interval.

[0033] In an optional embodiment, after the warehouse control system has accumulated a sufficient number of valid records in the historical reversing time sequence database of the shuttle, it constructs an individualized reversing time prediction model for each shuttle. Using the current load, current battery state of charge, and cumulative reversing count as input features, and the complete reversing time as the prediction target, a piecewise linear regression method is used to fit the historical valid records of the shuttle to obtain the predicted mean reversing time of the shuttle under the current state combination. The piecewise linear regression divides the input feature space into several intervals, independently fitting linear coefficients within each interval to capture the piecewise influence of load, battery charge, and wear level on reversing time. For example, the segment node for the load feature can be the median value between the two endpoints of empty and fully loaded; the segment node for the battery state of charge can be the boundary threshold between sufficient and low charge; and the segment node for the cumulative reversing count can be the empirical inflection point number when the mechanism wears into an accelerated phase.

[0034] Specifically, after obtaining the predicted mean, the predicted residual sequence is further calculated. This is the sequence formed by the difference between the measured complete reversal time and the corresponding predicted mean in all valid historical records of the vehicle. The standard deviation of this residual sequence is then calculated to characterize the actual fluctuation range of the vehicle's reversal time under the current state combination. The upper bound of the reversal time prediction interval is defined as the predicted mean plus the product of the confidence expansion coefficient and the standard deviation.

[0035] The confidence expansion coefficient is derived by the system from the actual distribution of the vehicle's historical reversal time residual sequence. All historical residual sequences of the vehicle are arranged in ascending order of absolute value, and the absolute value of the residual corresponding to the 99.7th quantile is denoted as... ,but:

[0036] Furthermore, the logic behind the above derivation is as follows: if the residual sequence strictly follows a normal distribution, then according to the 3σ rule, the confidence expansion coefficient should be exactly equal to 3; however, the actual commutation time residuals often exhibit a heavy-tailed distribution due to operating condition disturbances such as sudden changes in load and sudden drops in battery power. The absolute value of the residuals corresponding to the 99.7% quantile will be greater than 3 times the standard deviation. At this time, the confidence expansion coefficient is automatically amplified to a measured value greater than 3, objectively reflecting the true heavy-tailed thickness of the vehicle's residual distribution; the value of the confidence expansion coefficient is completely anchored to the actual distribution pattern of the vehicle's residual samples and dynamically converges with the continuous accumulation of historical data.

[0037] This explanation clarifies the basis for the 99.7% setting. In industrial safety control engineering, for random disturbance errors that strictly follow a normal distribution, according to classical... Criteria, data falls The cumulative probability within the interval is 99.7%, meaning that the interval can encompass the natural time fluctuations under all working conditions with a very high degree of confidence; therefore, the absolute value of the residual corresponding to its 99.7% quantile is taken.

[0038] Subsequently, as battery power decreases, the predicted mean of the braking release and initial acceleration in the new direction sub-state increases, thereby increasing the predicted mean of the complete reversal time and correspondingly increasing the upper bound of the prediction interval. Conversely, as mechanism wear intensifies, the historical measured values ​​of the alignment confirmation sub-state between the new direction wheelset and the track groove become more dispersed, resulting in a larger residual standard deviation and a correspondingly larger upper bound of the prediction interval. Both types of factors are transmitted to the upper bound of the prediction interval through their respective sub-state durations, causing the intersection reservation window to expand adaptively and thus avoiding insufficient window reservation. Compared to a globally uniform fixed nominal value, the upper bound of the prediction interval varies for each vehicle due to individual differences and is continuously and dynamically updated as the vehicle state changes, realizing the transformation of the intersection occupancy time window reservation from a static global value to a dynamic individual value.

[0039] Thus, by constructing an individualized prediction interval upper bound based on the actual distribution of each vehicle's residual sequence, the probability of window reservation failure due to ignoring individual differences by fixed nominal values ​​is effectively reduced, providing an accurate and reliable time window input for subsequent multi-vehicle intersection conflict detection.

[0040] S3, Multi-vehicle intersection time window conflict detection and speed feedforward adjustment.

[0041] In an optional embodiment, when planning a route for each shuttle, the scheduling system calculates the estimated arrival time of each shuttle at each intersection based on its current location, speed, and route node sequence, and determines its estimated occupancy time window for that intersection by combining the upper bound of the shuttle's prediction interval. For all intersections in the same track network, the scheduling system maintains an intersection time window occupancy table, recording the occupancy windows allocated to each shuttle. Whenever a new route planning request arrives, or when a real-time location update of a shuttle already en route causes a shift in its estimated arrival time, the system performs conflict detection on the relevant intersections. If two shuttles are detected to have overlapping occupancy time windows at the same intersection, it is determined to be a potential conflict, triggering a speed feedforward adjustment process.

[0042] Specifically, the core approach to conflict resolution is to reduce the speed of vehicles arriving later, rather than forcing them to stop and wait. Let the remaining distance between the arriving vehicles and the target intersection be... The current driving speed is The time required to delay its arrival at the intersection The overlap time between the two vehicles' occupancy windows is added to a safety margin. The safety margin is derived by the system from the historical switching interval sequence of the intersection. Any measured sample in the historical switching interval sequence refers to the time difference between the moment when the rear of the preceding vehicle completely leaves the physical boundary of the intersection and the moment when the front of the following vehicle triggers its entry into the intersection boundary during a continuous passage event between two vehicles at the target intersection. The measured sample collection method is as follows: the system automatically records the time difference of each safe intersection using photoelectric sensors or track pressure detectors deployed at the four-way boundaries of the intersection. For sample selection, a sliding time window mechanism is used to dynamically capture the most recent samples from the intersection. Next, the intersection records constitute the current historical switch interval sequence. The value range can be 50-200 times. This sampling principle based on a finite-length sliding window can promptly eliminate outdated and invalid data, ensuring that the sequence characteristics accurately reflect the actual physical condition response of the intersection under the influence of track wear, temperature, and microscopic deformation. The mean of the historical switching interval sequence of the intersection is added to its standard deviation.

[0043] This value objectively anchors the safety margin to the upper limit of the natural fluctuation of the intersection's travel time, which is a conservative value and conforms to the basic common sense in safety engineering that the margin should be greater than the natural fluctuation of the system. After determining the required delay time, the vehicles arriving later are instructed to travel at the reduced constant speed within the remaining road segment, so that the time taken for them to reach the intersection is exactly equal to the original travel time plus the delay time. The reduced speed is:

[0044] The speed reduction is: The remaining travel distance is physically the length of the track segment from the current position of the arriving vehicle to the target intersection, which is maintained in real time by the path planning module. When a conflict detection is triggered, the arriving vehicle has not yet reached the intersection, and this distance is greater than zero. The speed adjustment command is sent to the on-board controller of the arriving vehicle through the wireless communication module. The vehicle smoothly reduces its speed within the current road segment and drives towards the target intersection at a constant speed after the speed reduction.

[0045] Next, during the speed adjustment process, the system continuously monitors the real-time position of arriving vehicles. If other disturbances cause the arrival time prediction to deviate again, the speed reduction is updated and reissued, forming a closed-loop tracking. In extreme cases, where the remaining travel distance is extremely short but the required delay time is large, resulting in the calculated reduced speed being lower than the minimum allowable constant speed of the vehicle, the system determines that the speed feedforward reduction method cannot complete the required delay within the current road segment. Instead, it orders the arriving vehicle to briefly stop and wait at the safe stopping node closest to the target intersection until the occupancy window of the first arriving vehicle ends before resuming travel. This method ensures that even in extreme conditions not covered by the speed feedforward reduction method, the system can still reliably resolve intersection conflicts and prevent intersection occupancy conflicts caused by window overlap.

[0046] In this way, speed feedforward reduction replaces the traditional stopping and waiting method to resolve intersection conflicts. Vehicles arriving later can smoothly reduce their speed within the remaining road segment to avoid intersection conflicts, eliminating the kinetic energy loss and impact caused by sudden stops and starts. The overall throughput efficiency and smoothness of the system are significantly improved.

[0047] S4. Reversal completion feedback and online correction of the prediction model.

[0048] In an optional embodiment, after each commutation operation, the onboard controller reports the complete commutation time-series feature vector, along with the current load, current battery state of charge, and cumulative commutation count, to the warehouse control system. The system appends the new data to the vehicle's historical commutation time-series database and triggers an incremental update of the vehicle's prediction model. The incremental update uses a sliding window approach, retaining only the most recent commutation records for model refitting. The window length is adaptively determined by the autocorrelation characteristics of the vehicle's commutation time series. Specifically, let the length of the vehicle's complete commutation time series be... Calculate its autocorrelation coefficient sequence, with a lag number of steps. Based on existing white noise confidence band criteria in time series analysis, the upper bound of the 95% white noise confidence band for the autocorrelation coefficient is... The lower bound is The window length is taken from the lag number corresponding to the first time the correlation coefficient enters the confidence band.

[0049] The historical window length retained by this method precisely covers all historical samples in the vehicle's reversal time series that still have statistical significance. Historical samples beyond this lag step are statistically degraded into white noise and no longer provide information gain for the current prediction. The value of the window length is entirely determined by the autocorrelation structure of the vehicle's measured sequence. Based on the most recent reversal records within the sliding window, a piecewise linear regression model is refitted, updating the predicted mean and residual standard deviation, and then updating the upper bound of the prediction interval. The updated upper bound of the prediction interval takes effect immediately and is used for reserving intersection occupancy windows in the vehicle's subsequent path planning. As the vehicle's reversal time gradually increases due to mechanical wear, the upper bound of the prediction interval adaptively increases, and the intersection reservation window of the scheduling system expands synchronously, always matching the vehicle's actual reversal capability, effectively avoiding the problem of window reservation failure caused by equipment aging.

[0050] Next, the system continuously monitors the upper bound of the prediction interval for each vehicle to identify vehicles with abnormally high turn-off times. The monitoring and judgment criteria adopt the existing outlier identification method of box plots: when the upper bound of a vehicle's prediction interval exceeds the upper quartile of the historical distribution of the upper bound of the prediction interval for the same batch of vehicles plus 1.5 times the interquartile range, the upper bound of that vehicle's prediction interval is determined to be a statistically abnormally high point. After this judgment is triggered, the system automatically generates a maintenance warning for that vehicle's mechanism, prompting maintenance personnel to check the lifting mechanism. This warning is triggered earlier than the time when the turn-off time exceeds the reserved window at the intersection, leading to an actual conflict. It is a preventive maintenance intervention, rather than a reactive fault response.

[0051] Specifically, the logic for generating maintenance warnings and the logic for reserving intersection occupancy windows are independent and do not interfere with each other. Even if a vehicle has triggered a maintenance warning, its upper bound of the prediction interval will continue to be dynamically updated. The dispatching system will still reserve an intersection occupancy window for it based on the latest upper bound of the prediction interval, ensuring that the system can still dispatch the vehicle normally before the maintenance personnel complete the maintenance, and will not force a shutdown due to the warning. After the maintenance personnel complete the maintenance of the lifting mechanism, they can manually clear the maintenance warning status of the vehicle through the maintenance management interface of the warehouse control system. The system will then reset the cumulative reversing frequency influencing factor of the vehicle and re-establish the baseline with the first batch of reversing records after the maintenance, so that the prediction model reconverges from the initial state after the maintenance.

[0052] Furthermore, for scenarios where vehicles are brought back online after a long period of downtime, the records in the historical commutation timing database may deviate from the current actual state due to deep battery discharge or changes in ambient temperature during the downtime. Therefore, the system performs a pre-warm-up calibration process when the vehicle is brought back online: before officially accepting scheduling tasks, the vehicle performs several consecutive commutation actions in an unloaded state, the obtained commutation timing feature vectors are appended to the historical database, and an incremental model update is immediately triggered to correct the state deviations accumulated during downtime, ensuring that the upper bound of the prediction interval used by the first batch of scheduling tasks after the vehicle is brought back online reflects the vehicle's current true commutation capability. The specific number of repetitions is determined by a comprehensive trade-off between the minimum number of effective samples required for model refitting and the time cost of warehouse preheating, with an optimal range of 3 to 5 repetitions. It should be noted that at least 3 consecutive samples are required to initially calculate the mean and standard deviation, thereby effectively capturing the initial state fluctuations after shutdown and restart. At the engineering level, if there are more than 5 consecutive no-load reversals, it will excessively prolong the preparation time for vehicle launch. Therefore, selecting 3 to 5 repetitions can achieve the best balance between quickly correcting state deviations and ensuring equipment launch efficiency.

[0053] In this way, by combining online updates via sliding windows with maintenance warnings based on box plot criteria, the system maintains adaptive convergence of prediction accuracy throughout the vehicle's lifecycle and can trigger maintenance intervention in the early stages of structural deterioration, effectively suppressing the risk of intersection deadlock caused by equipment aging. The long-term stability and maintainability of the overall system are significantly enhanced.

Claims

1. A collaborative timing control method for four-way switching of a four-way shuttle car for a material bin, characterized in that, include: The measured duration of each commutation sub-state corresponding to a single complete commutation process of the four-way shuttle car in the hopper is obtained, and the current running state variables affecting the complete commutation time are collected, including: dividing the complete commutation process into four serially triggered, independently monitorable sub-states, and recording the time interval between the satisfaction of the precondition and the triggering of the current judgment condition for each sub-state as the measured duration to form a time series feature vector; synchronously collecting the hopper load, battery state of charge, and cumulative number of commutations as current running state variables, and merging the current running state variables and time series feature vectors into the historical commutation time series database for subsequent modeling use; The four sub-states include the deceleration and braking completion sub-state triggered sequentially, the lifting mechanism action completion sub-state, the new direction wheel set and track groove alignment confirmation sub-state, and the brake release and completion of the first acceleration start in the new direction sub-state. The measured duration and the current running status quantity are stored to form historical commutation data. Based on the historical commutation data, piecewise linear regression is used to fit and obtain the predicted mean of commutation time and the predicted residual sequence. The confidence expansion coefficient is derived based on the predicted mean of reversing time and the predicted residual sequence. Then, the upper bound of the individualized reversing time prediction interval for the current four-way shuttle in the material bin is constructed using the confidence expansion coefficient. This includes: calculating the standard deviation of the predicted residual sequence and arranging all historical predicted residual sequences in ascending order of absolute value to extract the absolute value of the residual corresponding to the percentile distribution point; dividing the extracted absolute value of the residual corresponding to the percentile distribution point by the standard deviation of the predicted residual sequence to obtain the confidence expansion coefficient; and adding the predicted mean of reversing time to the product of the confidence expansion coefficient and the standard deviation to obtain the upper bound of the individualized reversing time prediction interval. The estimated occupancy time window is determined based on the upper bound of the individualized reversal time prediction interval. When overlapping occupancy time windows of multiple vehicles are detected, potential intersection conflicts are mitigated by adjusting the uniform speed of arriving vehicles. This includes: obtaining the mean and standard deviation of the historical switching interval sequence of the intersection and adding them together to generate a safety margin; adding the safety margin to the overlapping duration of the occupancy time window as the required delay duration; calculating the ratio of the remaining travel distance of the arriving vehicle from the target intersection to its current travel speed; and dividing the remaining travel distance by the sum of the ratio and the required delay duration as the reduced uniform speed, which is then sent to the arriving vehicles.

2. The collaborative timing control method for four-way switching of a four-way shuttle car in a material bin according to claim 1, characterized in that, The complete commutation process is divided into four independently monitorable sub-states triggered sequentially, and the time interval between the satisfaction of the precondition and the triggering of the current judgment condition for each sub-state is recorded as the measured duration to form a timing feature vector, including: Set a single substate timeout protection threshold for each substate, and use the maximum duration of the same type of substate in history plus a fixed margin as the single state timeout protection threshold. When a substate fails to trigger the judgment condition within the corresponding single substate timeout protection threshold, a commutation anomaly event is reported, and this commutation record is marked as an abnormal sample to be excluded in subsequent modeling training.

3. The collaborative timing control method for four-way switching of a four-way shuttle car in a material bin according to claim 1, characterized in that, The calculation involves dividing the remaining distance of the vehicle from the target intersection by the ratio of its current speed to the calculated speed, and then using the sum of the remaining distance, the ratio, and the required delay time as the adjusted constant speed, which is then sent to the arriving vehicles. This includes: Determine whether the calculated reduced constant speed is lower than the minimum allowable constant speed of the vehicle. If the reduced constant speed is not lower than the minimum constant speed, the vehicle arriving later will smoothly reduce its speed within the current road segment. When the reduced constant speed is lower than the minimum allowable constant speed, the vehicle that arrives later will briefly stop at the safe stopping point closest to the target intersection and wait until the window occupied by the vehicle that arrived first ends.

4. The collaborative timing control method for four-way switching of a four-way shuttle car in a material bin according to claim 1, characterized in that, The process of storing the measured duration and the current operating state quantity to form historical reversal data, and using piecewise linear regression to fit the historical reversal data to obtain the predicted mean of reversal time and the predicted residual sequence, includes: Calculate the autocorrelation coefficient sequence corresponding to the complete commutation time series, and derive the white noise confidence band of the autocorrelation coefficients based on the length of the complete commutation time series; The lag step number corresponding to the first time the autocorrelation coefficient enters the white noise confidence band is obtained as the sliding window length, and the most recent commutation records within the sliding window length are retained to refit and update the piecewise linear regression model.

5. The collaborative timing control method for four-way switching of a four-way shuttle car in a material bin according to claim 1, characterized in that, The method involves deriving the confidence expansion coefficient based on the predicted mean and predicted residual sequence of the reversing time, and then constructing an upper bound for the individualized reversing time prediction interval for the current four-way shuttle car in combination with the confidence expansion coefficient, including: Obtain the quartiles and interquartile ranges of the historical distribution of the upper bound of the individualized reversal time prediction interval for vehicles in the same batch, and construct outlier monitoring and judgment criteria. The interquartiles are summed with the interquartile ranges at fixed multiples. If the upper bound of the predicted interval for the individualized reversal time of the current four-way shuttle car exceeds the summation result, the current vehicle is determined to be at a statistically abnormal high point and a mechanism maintenance warning is triggered.

6. The collaborative timing control method for four-way switching of a four-way shuttle car in a material bin according to claim 4, characterized in that, The process of refitting and updating the piecewise linear regression model by retaining the most recent commutation records within the length of the sliding window includes: When a vehicle is brought back online after a long period of downtime, the current vehicle is made to perform several reversing actions in an unloaded state to obtain the preheating calibration reversing timing feature vector. The preheating calibration commutation timing feature vector is appended to the corresponding historical commutation data to trigger an incremental model update to correct the state deviation accumulated during downtime.

Images