Train approaching identification method and system based on laser detection

Abstract

This invention relates to the field of rail transit equipment technology, and discloses a train approach recognition method and system based on laser detection. The method includes: acquiring initial data for train approach recognition and outputting an initial parameter set; calculating an adaptive detection parameter set based on the current battery status; generating a predicted wake-up window start threshold; outputting the current power consumption status based on hierarchical state switching; and performing laser detection acquisition, train approach discrimination, and alarm transmission. It solves the problems of battery depletion, alarm delay, or missed alarms caused by fixed high-frequency ranging and long wireless connections. Through the coordination of historical no-train intervals, battery status, wake-up window, power consumption status, and distance change criteria, the device maintains the timeliness of train approach recognition and alarm triggering under low-power operation. After alarm transmission, it writes back the current no-train interval, providing an updated basis for the next cycle calculation and wake-up prediction, forming a closed loop of detection and energy saving.

Description

Technical Field

[0001] This invention relates to the field of rail transit equipment technology, and in particular to a train approach identification method and system based on laser detection. Background Technology

[0002] Currently, railway workers performing maintenance work on operational railway lines need to implement train approach safety protection measures in the work area. To promptly detect train approach, laser detection equipment is typically deployed on both sides of the work area along the track direction. This equipment detects trains entering the monitoring range and sends alarm information via 5G wireless relay devices, enabling automatic early warning of train approach and reminding workers to take timely safety precautions to prevent injury. In some field work areas or where AC power is unavailable, the equipment must be powered by lithium-ion polymer batteries. The conventional implementation of such devices involves the laser detection module continuously emitting pulses at a fixed frequency and receiving echoes, while the 5G communication module maintains a constant connection with the base station and transmits data back in real time, aiming to maintain continuous detection and immediate alarms for several hours supported by battery capacity. To extend operating time, some solutions introduce simple timed sleep or voltage threshold-based frequency reduction, but these do not integrate historical train data with real-time power consumption for dynamic control. This often results in low-power strategies sacrificing detection density, making it difficult to balance battery life and real-time performance.

[0003] For example, when the system requires continuous operation in the field for 8 hours and is powered by only a single lithium-ion polymer battery, if a fixed high-frequency continuous ranging and 5G long-connection strategy is implemented, the battery will be consumed rapidly in the early stage of operation, and the battery voltage will continue to drop in the middle and late stages. This will not only cause a decrease in laser emission power and a deterioration in echo signal-to-noise ratio, but may also lead to abnormal reset of the main control circuit. At the same time, the network signaling interaction of the 5G module and the maintenance state of the RF power amplifier will further accelerate the depletion of power. When the actual train approaches the detection range, the equipment may be in a protective intermittent shutdown triggered by voltage drop or a 5G module disconnection and reconnection state, causing alarm response delays or even missed alarms. The faults are concentrated in the last stage of the mission, which seriously weakens the timeliness of train approach alarms in passive field environments, creating a prominent contradiction between high endurance and high real-time performance.

[0004] Therefore, there is an urgent need for a method that can adaptively adjust the laser detection cycle and 5G module sleep-wake strategy based on the statistical characteristics of real-time battery charge status and historical periods without train passage. Under conditions of gradually decreasing battery power and long periods without train passage, it can still preheat intensively in advance and instantly switch to high-density detection and sudden alarm mode when a train enters the detection range. This ensures that the alarm response delay does not exceed the limit while maintaining a stable effective working time of more than 8 hours, thereby improving the reliability and continuity of train approach warning during track maintenance windows on operating lines. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, the purpose of this invention is to propose a train approach recognition method based on laser detection. This method aims to solve the technical problem that existing technologies, such as laser detection modules using fixed high-frequency continuous ranging and 5G modules maintaining long-term connections, suffer from excessive power consumption, leading to insufficient power, alarm delays, or even missed alarms. In particular, under conditions of lithium-ion polymer battery power supply, continuous operation in the field for 8 hours, and ensuring high timeliness of 5G forwarding, it is impossible to achieve both low power consumption and real-time performance.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides a train approach recognition method based on laser detection.

[0007] The laser-based train proximity recognition method includes:

[0008] Step S10: Obtain train approach recognition initialization data, and perform initialization tasks based on the train approach recognition initialization data using a parameter grouping loading mechanism, and output the initialization parameter set;

[0009] Step S20: Obtain the current battery state, and based on the initialization parameter set and the current battery state, perform the laser detection period calculation task using the state of charge correction and nonlinear periodic mapping mechanism, and output the adaptive detection parameter set;

[0010] Step S30: Based on the adaptive detection parameter set, perform statistical preprocessing tasks for predicting the wake-up window using an exponentially weighted statistical prediction mechanism, and output the starting threshold of the predicting wake-up window;

[0011] Step S40: Based on the predicted wake-up window start threshold and the adaptive probe parameter set, a hierarchical state switching mechanism is used to execute a power scheduling task, and the current power consumption status is output;

[0012] Step S50: Based on the current power consumption status, perform laser detection data acquisition, train approach event judgment, and alarm data transmission, and output an alarm trigger signal;

[0013] Preferably, step S10, which involves acquiring train approach recognition initialization data and constructing the initialization parameter set based on the train approach recognition initialization data using a parameter grouping and loading mechanism, specifically includes:

[0014] Step S101: Extract dynamic power management parameters from the train approach recognition initialization data. The dynamic power management parameters include minimum laser detection period, maximum laser detection period, reference state of charge, S-shaped mapping function steepness factor, exponential weighted attenuation factor, correction amplitude factor, confidence interval coefficient, basic pre-wake-up advance coefficient, preset standard deviation lower limit, deep sleep switching ratio, preset continuous reduction quantity, high-density sampling frequency, sliding queue length, monitoring window length, preset approach ratio threshold, and preset second-order difference threshold. The deep sleep switching ratio is greater than 0 and less than 1.

[0015] Step S102: Extract historical no-train-passage time period data arranged by occurrence time from the train approach recognition initialization data. The duration of each no-train-passage time period in the historical no-train-passage time period data is the duration between when a train completely leaves the laser detection range and when the next train enters the laser detection range.

[0016] Step S103: Write the dynamic power consumption management parameters into the parameter field according to the parameter identifier, write the historical no-train-passage time period data into the historical duration field according to the occurrence time, and associate and encapsulate the parameter field and the historical duration field to generate the initialization parameter set.

[0017] Preferably, step S20, which involves obtaining the current battery state, performing laser detection period calculation based on the initialization parameter set and the current battery state using a state-of-charge correction and nonlinear periodic mapping mechanism, and outputting the adaptive detection parameter set, specifically includes:

[0018] Step S201: Extract the current state of charge from the current battery state. The ratio of the difference between the reference state of charge and the current state of charge to the reference state of charge is calculated, and the ratio is restricted to between 0 and 1 to obtain the normalized charge reduction factor. ;

[0019] Step S202: Read the historical no-train-passage time period data from the initialization parameter set, and calculate the weighted average no-train-passage interval based on the exponential weighted attenuation factor. and dynamic standard deviation It also records the duration from the moment the train completely left the laser detection range until the current moment, thus obtaining the duration of the current train-free event. ;

[0020] Step S203: Calculate the safety margin correction factor according to the following formula:

[0021]

[0022] The corrected state of charge is calculated using the following formula:

[0023]

[0024] The adaptive laser detection period is calculated using the following formula:

[0025]

[0026] in, This is the safety margin correction factor. The correction amplitude factor, The confidence interval coefficient is... To preset the lower limit of standard deviation, For the corrected state of charge, The adaptive laser detection period is... The minimum laser detection period is... The maximum laser detection period is... The steepness factor of the S-shaped mapping function. The reference state of charge, It is the hyperbolic tangent function. It is a function with maximum value. It is a minimum value function. It is a natural exponential function;

[0027] The adaptive laser detection period, the weighted average of no-train intervals, the dynamic standard deviation, the duration of the current no-train event, the normalized power reduction factor, the basic pre-wake-up advance coefficient, the minimum laser detection period, the deep sleep switching ratio, the preset continuous reduction quantity, the high-density sampling frequency, the sliding queue length, the monitoring window length, the preset proximity ratio threshold, and the preset second-order difference threshold are associated and encapsulated to generate the adaptive detection parameter set.

[0028] Preferably, in step S202, the step of calculating the weighted mean of the train interval and the dynamic standard deviation includes:

[0029] The most recent complete no-train interval duration is obtained from the historical no-train-passage time period data. The difference between the most recent complete no-train interval duration and the previous weighted no-train interval mean is calculated to obtain the interval deviation.

[0030] Based on the exponentially weighted decay factor, the most recent complete no-train interval duration and the previous weighted no-train interval mean are combined exponentially to obtain the updated weighted no-train interval mean.

[0031] Based on the exponentially weighted decay factor and the square of the interval deviation, the previous no-train interval variance is incrementally corrected to obtain the updated no-train interval variance.

[0032] The updated variance of trainless intervals is compared with the preset minimum variance. The larger of the updated variance of trainless intervals and the preset minimum variance is squared to obtain the dynamic standard deviation. The preset minimum variance is the square of the preset lower limit of standard deviation.

[0033] Preferably, step S30, which involves performing a statistical preprocessing task for predicting the wake-up window based on the adaptive detection parameter set using an exponentially weighted statistical prediction mechanism, and outputting the starting threshold of the predictive wake-up window, specifically includes:

[0034] Step S301: Extract the weighted mean of no-train interval, the dynamic standard deviation, the normalized power reduction factor, the basic pre-wake-up advance coefficient, and the minimum laser detection cycle from the adaptive detection parameter set;

[0035] Step S302: Calculate the dynamic pre-wake-up advance coefficient according to the following formula:

[0036]

[0037] Step S303: Calculate the predicted wake-up window start threshold according to the following formula:

[0038]

[0039] in, The dynamic pre-wake-up advance coefficient is... The basic pre-wake-up advance coefficient is... The threshold value for the predicted wake-up window is [value].

[0040] Preferably, step S40, which involves executing a power scheduling task based on the predicted wake-up window start threshold and the adaptive probe parameter set using a hierarchical state switching mechanism, and outputting the current power state, specifically includes:

[0041] Step S401: Extract the duration of the current no-train event, the adaptive laser detection cycle, the deep sleep switching ratio, the preset continuous reduction quantity, and the high-density sampling frequency from the adaptive detection parameter set, and determine the pre-wake boundary based on the product of the deep sleep switching ratio and the predicted wake-up window start threshold;

[0042] Step S402: When the duration of the current no-train event is less than the pre-wake-up boundary, the current power consumption state is set to deep sleep state, the laser detection module and the wireless forwarding module are turned off, and the low-power clock is kept running; when the duration of the current no-train event is not less than the pre-wake-up boundary and is less than the predicted wake-up window start threshold, the current power consumption state is set to pre-wake-up state, and the power supply preparation process of the laser detection module and the baseband initialization process of the wireless forwarding module are started.

[0043] Step S403: When the duration of the current no-train event is not less than the predicted wake-up window start threshold, the current power consumption state is set to low-frequency survey state, and the laser detection module is controlled to collect laser detection data according to the adaptive laser detection cycle; when a preset number of successively decreasing effective distance values ​​are continuously obtained, the current power consumption state is switched to high-density early warning state, the laser detection module is controlled to continuously collect the laser detection data according to the high-density sampling frequency, and the wireless forwarding module is controlled to perform network attachment.

[0044] Preferably, step S50, which involves collecting laser detection data, determining train approach events, and sending alarm data based on the current power consumption state, and then outputting the alarm trigger signal, specifically includes:

[0045] Step S501: When the current power consumption state is the high-density warning state, extract the sliding queue length, the monitoring window length, the preset proximity ratio threshold, and the preset second-order difference threshold from the adaptive detection parameter set. Write the continuously collected laser detection data into the sliding queue with the sliding queue length according to the sampling time order. Perform median filtering on the distance values ​​in the sliding queue. Subtract the distance value of the previous sampling time from the distance value of the later sampling time in every two adjacent sampling times to obtain the first-order distance difference. Subtract the previous first-order distance difference from the later first-order distance difference in every two adjacent first-order distance differences to obtain the second-order distance difference. Record the sampling time interval corresponding to each of the first-order distance differences.

[0046] Step S502: Within the monitoring window with the specified monitoring window length, the proportion of distance data points corresponding to the first-order distance difference values ​​less than 0 is calculated out of all distance data points to obtain the proportion of approaching data points. The proportion of approaching data points is compared with a preset approach proportion threshold, and the absolute value of each second-order distance difference is compared with a preset second-order difference threshold. When the proportion of approaching data points is not lower than the preset approach proportion threshold, and the absolute value of each second-order distance difference is less than the preset second-order difference threshold, a train approach event confirmation result is generated.

[0047] Step S503: In response to the confirmation result of the train approach event, calculate the ratio of the absolute value of each of the first-order distance differences to the corresponding sampling time interval, and determine the median of each ratio as the estimated train speed. Encapsulate the current distance value, the estimated train speed, and the device identifier to generate the alarm trigger signal, and control the wireless forwarding module to send the alarm trigger signal to the alarm receiving terminal. After the transmission is completed, determine the duration of the current no-train event as the current no-train interval duration, write the current no-train interval duration into the historical no-train passage time period data, and restore the current power consumption state to the deep sleep state.

[0048] The present invention also provides a train proximity recognition system based on laser detection, the system comprising:

[0049] The initialization module is used to acquire train approach recognition initialization data, perform initialization tasks based on the train approach recognition initialization data using a parameter grouping loading mechanism, and output an initialization parameter set.

[0050] The cycle calculation module is used to obtain the current battery state, and based on the initialization parameter set and the current battery state, it performs the laser detection cycle calculation task using the state of charge correction and nonlinear cycle mapping mechanism, and outputs an adaptive detection parameter set.

[0051] The wake-up prediction module is used to perform statistical preprocessing tasks for predicting the wake-up window based on the adaptive detection parameter set using an exponentially weighted statistical prediction mechanism, and output the starting threshold of the predicted wake-up window.

[0052] The power scheduling module is used to perform power scheduling tasks based on the predicted wake-up window start threshold and the adaptive probe parameter set using a hierarchical state switching mechanism, and output the current power status.

[0053] The alarm processing module is used to collect laser detection data, determine train approach events, and send alarm data based on the current power consumption status, and output an alarm trigger signal.

[0054] The present invention also provides a train approach recognition device based on laser detection, the train approach recognition device based on laser detection includes: a memory, a processor, and a train approach recognition program based on laser detection stored in the memory and executable on the processor, the train approach recognition program based on laser detection implementing the above method when executed by the processor.

[0055] The present invention also provides a computer program product, the computer program product including a laser detection-based train approach recognition program, which implements the above-described method when executed by a processor.

[0056] The beneficial effects of this invention are as follows:

[0057] This invention encapsulates dynamic power management parameters and historical no-train-passage time period data into an initialization parameter set through a parameter grouping and loading mechanism, so that subsequent ranging cycles, wake-up windows and power state switching have a unified parameter source and historical statistical basis.

[0058] By using a state-of-charge correction and nonlinear periodic mapping mechanism, the current battery state, historical no-train interval statistics, and periodic boundaries are encapsulated into an adaptive detection parameter set, enabling the laser detection period to change with the battery level and the train arrival patterns on the line.

[0059] The predicted wake-up window threshold is obtained by using an exponentially weighted statistical prediction mechanism, so that the equipment does not need to measure distance at a fixed frequency for a long time, but gradually enters a detectable state before the train may approach.

[0060] The current power consumption status is output through a hierarchical state switching mechanism, enabling the laser detection module and wireless relay module to operate in stages: deep sleep state, pre-wake-up state, low-frequency survey state, and high-density early warning state. When the current power consumption status is high-density early warning state, continuous ranging, median filtering, distance differential, and proximity ratio discrimination are performed. After confirming the train's approach, an alarm trigger signal is sent, ensuring that the alarm trigger is based on the continuous distance change trend rather than a single-point threshold. Attached Figure Description

[0061] Figure 1 This is a schematic flowchart of the first embodiment of a train approach recognition method based on laser detection according to the present invention. Detailed Implementation

[0062] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0063] 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.

[0064] Example 1: As Figure 1 The diagram shown is a flowchart of a first embodiment of a train approach recognition method based on laser detection according to the present invention.

[0065] In the first embodiment, the laser detection-based train proximity recognition method includes:

[0066] Step S10: Obtain train approach recognition initialization data, and perform initialization tasks based on the train approach recognition initialization data using a parameter grouping loading mechanism, and output the initialization parameter set;

[0067] Train approach identification initialization data is a set of data configured or accumulated by the equipment before entering the current train approach identification process. It consists of dynamic power management parameters and historical no-train-passage time data arranged by occurrence time. Dynamic power management parameters are used to specify the adjustable range of the ranging cycle, the state of charge correction magnitude, the statistical weight of historical intervals, the pre-wake-up advance degree, the switching boundaries of each power state, and the train approach discrimination conditions. Historical no-train-passage time data records the duration from when a train completely leaves the laser detection range until the next train enters the range. The two types of data have different meanings: the former defines how the equipment is allowed to operate, while the latter reflects when trains usually reappear on the current line in actual operation.

[0068] Dynamic power consumption management parameters are written into parameter fields according to parameter identifiers. The minimum and maximum laser detection cycles define the upper and lower boundaries of the cycle during low-frequency surveys; the reference state of charge and the steepness factor of the S-shaped mapping function are used to determine the degree of influence of power changes on the ranging cycle; the exponentially weighted attenuation factor, correction amplitude factor, confidence interval coefficient, and preset standard deviation lower limit are used to handle the center position, fluctuation range, and safety margin of historical train intervals; the basic pre-wake-up advance coefficient and deep sleep switching ratio are used to determine the time position of the transition from deep sleep to pre-wake-up; the preset continuous reduction quantity, high-density sampling frequency, sliding queue length, monitoring window length, preset proximity ratio threshold, and preset second-order difference threshold are used to constrain the judgment of subsequent train proximity events.

[0069] Historical train-free interval data is written to the historical duration field according to the event occurrence time. Each record starts when the train completely leaves the laser detection range and ends when the subsequent train first enters the range. Using this unified start and end point avoids including time spent within the monitoring range, alarm transmission time, or equipment re-sleep time in the train-free interval data. The train-free interval duration written back in step S50 after alarm transmission also uses the same metric; therefore, new samples can directly participate in the next round of index-weighted statistics without needing to re-convert the data meaning.

[0070] After the parameter fields and historical duration fields are written, the parameter grouping and loading mechanism associates and encapsulates them according to the same device identifier. The resulting initial parameter set retains both the independent identifier of each control parameter and the chronological order of historical intervals. When reading this parameter set in step S20, the cycle boundary and correction coefficient can be obtained by parameter identifier, while the most recent complete train-free interval and previous statistical status can be obtained in chronological order, avoiding mixing configuration parameters and observation samples into the same numerical processing. Therefore, the initial parameter set is not a simple data splicing, but a common entry point for the engineering boundaries and line timing statistics required for subsequent cycle calculations.

[0071] The associated encapsulation also ensures that the fate of each parameter in subsequent steps remains clear: the cycle boundary, reference state of charge, steepness factor, and correction parameters are used in S20 to calculate the adaptive laser detection cycle; the basic pre-wake-up advance coefficient and statistical results are used in S30 to determine the prediction wake-up window start threshold; the deep sleep switching ratio, continuous reduction quantity, and high-density sampling frequency are used in S40 for state switching; and the queue length, monitoring window length, and two discrimination thresholds are used in S50 for event confirmation. Each input can find a specific use in subsequent processing chains, avoiding the collection of data unrelated to train approach recognition during the initialization phase.

[0072] Fixed-cycle equipment typically loads only a ranging frequency and an alarm distance upon startup. Regardless of whether the line is in peak passenger flow, off-peak nighttime conditions, or long periods without trains, the laser detection module and wireless relay module operate at the same rhythm. This approach cannot distinguish between the states of "not yet approaching the usual arrival time" and "approaching or exceeding the usual arrival time." This embodiment retains the historical time sequence of no-train intervals, allowing recent operating intervals to have a more direct impact on subsequent statistical results. At the same time, it limits the adjustment range with minimum cycle, maximum cycle, and various discrimination thresholds, preventing the equipment from changing its operating rhythm without boundaries based on a single interval change.

[0073] At a line protection point without mains power supply, the minimum laser detection period can be set to 2 seconds, the maximum laser detection period to 30 seconds, the baseline state of charge to 0.80, the steepness factor of the S-shaped mapping function to 6, the exponentially weighted attenuation factor to 0.25, the correction amplitude factor to 0.20, the confidence interval coefficient to 1.5, the basic pre-wake-up advance coefficient to 1.8, the preset standard deviation lower limit to 15 seconds, and the deep sleep switching ratio to 0.70. The preset continuous reduction quantity can be set to 3, the high-density sampling frequency can be set to 10Hz, the sliding queue length can be set to 9, the monitoring window length can be set to 8, the preset proximity percentage threshold can be set to 0.75, and the preset second-order differential threshold can be set to 0.60 meters.

[0074] The historical duration field of the same device can sequentially store complete no-train intervals of 820 seconds, 780 seconds, 760 seconds, 740 seconds, and 720 seconds. During loading, the aforementioned power management parameters are entered into the parameter field, and the five interval records are entered into the historical duration field according to their occurrence time, and then encapsulated together with the device identifier. In this way, step S20 can confirm that the laser detection cycle is always within the range of 2 to 30 seconds, and can also identify that the recent no-train intervals are gradually shortening from 820 seconds to 720 seconds. This trend will not directly trigger an alarm in S10, but will be passed to the next step as statistical input, thereby maintaining the boundary of responsibilities between initialization, cycle calculation, and event discrimination.

[0075] Step S20: Obtain the current battery state, and based on the initialization parameter set and the current battery state, perform the laser detection period calculation task using the state of charge correction and nonlinear periodic mapping mechanism, and output the adaptive detection parameter set;

[0076] The current battery state is the remaining battery status given by the device's power management unit at the current calculation moment, where the current state of charge (SOC) characterizes the ratio of currently available power to rated capacity. The adaptive detection parameter set is a collection of associated data after the cycle calculation is completed. In addition to the adaptive laser detection cycle, it also retains the weighted mean of no-train intervals, dynamic standard deviation, duration of the current no-train event, normalized power decay factor, and control parameters required for subsequent wake-up prediction, state switching, and proximity detection. Retaining these intermediate values ​​allows S30 to S50 to directly use the results of the same cycle calculation, eliminating the need to repeatedly interpret power and historical intervals in different modules.

[0077] The normalized charge reduction factor is calculated as follows: first, subtract the current state of charge from the reference state of charge; then, divide the difference by the reference state of charge; finally, limit the result to between 0 and 1. When the current state of charge reaches or exceeds the reference state of charge, the factor value will not be lower than 0; when the current state of charge is abnormally low, the factor will not exceed 1. This limiting process converts the remaining state of charge of batteries with different capacities into a uniform proportion, and subsequent calculations focus on the relative degree of charge reduction, rather than the absolute ampere-hours of a particular battery model.

[0078] The weighted mean of train-free intervals is calculated by exponentially weighting the most recent complete train-free interval with the previous weighted mean. The exponential weighting decay factor determines the proportion of new samples included in the statistical results, with the remaining proportion carried over by the previous statistical results. Therefore, recent changes in train intervals are gradually reflected in the mean, without causing sudden jumps in the central value due to a single, accidental long or short interval. The dynamic standard deviation is updated using the interval deviation of the most recent complete train-free interval relative to the previous weighted mean, and the square of this deviation is used to incrementally correct the previous variance. Before taking the square root, it is compared with a preset minimum variance, retaining at least the volatility corresponding to the preset lower limit of the standard deviation.

[0079] The duration of the current train-free event is accumulated from the moment the previous train completely left the laser detection range until the current calculation time. This duration, along with the weighted mean of train-free intervals and the dynamic standard deviation, is used for safety margin correction: when the duration is still significantly lower than the historically common interval, the correction result tends to retain a longer ranging period; when the duration is close to or exceeds the historically common interval, the correction result increases the corrected state of charge, resulting in a relatively shorter ranging period output by the nonlinear periodic mapping. Hyperbolic tangent processing limits the correction amplitude to a controllable range, and the corrected state of charge is subsequently limited again to between 0 and 1 to prevent the period from exceeding the preset boundary due to prolonged train-free periods.

[0080] Nonlinear periodic mapping is performed between the minimum and maximum laser detection periods. When the state of charge (SOC) is high, the mapping result shifts towards the minimum period, increasing the ranging density. When the SOC is low, the mapping result shifts towards the maximum period to reduce invalid sampling under low charge conditions. The steepness factor of the S-shaped mapping function controls the rate of change near the reference SOC, preventing the period from frequently and significantly changing with minor fluctuations in SOC. With the sequential application of charge constraints and time margin corrections, the device avoids maintaining sparse ranging when a train might approach due to low charge, and also avoids ignoring battery endurance limits due to longer waiting times.

[0081] The co-encapsulation of statistical and periodic results within the same calculation is also of practical significance. The weighted mean, dynamic standard deviation, and normalized power degradation factor used in S30 are derived from the same current battery state and the same set of historical intervals as the current adaptive laser detection cycle, thus avoiding the time misalignment where the ranging cycle has been updated according to the new samples while the wake-up threshold still references the old statistics. The state switching parameters and discrimination thresholds required by S40 and S50 are transmitted along with this parameter set, ensuring that the power consumption scheduling boundaries and event confirmation conditions remain consistent within a single identification cycle.

[0082] If the cycle is set solely based on the current state of charge, low-charge devices will typically reduce the ranging frequency indiscriminately, potentially failing to obtain new distance values ​​for an extended period even when the train is approaching a historical arrival interval. Conversely, if the cycle is set solely based on historical intervals, the devices may prematurely engage in intensive detection when the battery is insufficient. Here, we describe the endurance pressure using a normalized power decline factor and the time pressure using the position of the current duration of no-train events relative to the weighted mean and dynamic standard deviation, with the cycle derived from the same nonlinear mapping. By incorporating both pressures into a single bounded cycle, the trade-off between energy saving and proximity detection can be continuously adjusted according to track conditions.

[0083] Using the aforementioned line data, assuming the previous weighted average no-train interval was 750 seconds, the most recent complete no-train interval was 720 seconds, and the exponential weighted decay factor was 0.25, the updated weighted average no-train interval is approximately 742.5 seconds. After incremental correction based on the squared interval deviation and the previous variance, a dynamic standard deviation of approximately 37 seconds is obtained. This value is higher than the preset lower limit of 15 seconds, therefore 37 seconds is used in subsequent calculations. When the current state of charge is 0.56 and the reference state of charge is 0.80, the normalized charge decay factor is 0.30.

[0084] When the duration of the current no-train event is approximately 700 seconds, the system uses the weighted average of 700 seconds and 742.5 seconds, along with the dynamic standard deviation of 37 seconds, for safety margin correction. With a correction amplitude factor of 0.20 and a confidence interval coefficient of 1.5, the resulting corrected state of charge remains within the range of 0 to 1. Combining this with period boundaries of 2 seconds and 30 seconds, and a steepness factor of 6 for nonlinear mapping, an adaptive laser detection period of approximately 22 to 24 seconds can be obtained. If the duration continues to increase and exceeds the historically common range, the period will gradually shorten within the same boundary, rather than jumping directly to 2 seconds, thus providing smooth and interpretable detection parameters for the S30 wake-up prediction.

[0085] Step S30: Based on the adaptive detection parameter set, perform statistical preprocessing tasks for predicting the wake-up window using an exponentially weighted statistical prediction mechanism, and output the starting threshold of the predicting wake-up window;

[0086] The predicted wake-up window start threshold is the timeframe at which the device should transition from the pre-wake-up state to the low-frequency monitoring state after the current train-free event duration reaches a certain point. This threshold represents the time threshold for power consumption scheduling, is not equivalent to the train approach event confirmation time, and does not directly generate an alarm. The exponentially weighted statistical prediction mechanism reads the weighted mean of the train-free interval, dynamic standard deviation, normalized power decrease factor, basic pre-wake-up advance coefficient, and minimum laser detection cycle from the adaptive detection parameter set, ensuring that the threshold calculation and step S20 use the same round of historical statistics and power status.

[0087] The dynamic pre-wake-up advance factor is amplified from the basic pre-wake-up advance factor according to the normalized power decline factor. When the power decline factor is 0, the basic advance level is used; as the power decline factor gradually increases, the advance factor increases accordingly. The reason for this is that under low power conditions, the laser detection module's power supply stability and the wireless relay module's baseband initialization may require more preparation time. The threshold should be shifted forward relative to the historical average vehicle-to-vehicle interval, rather than waiting until it is close to the historical average before the module is activated.

[0088] When determining the threshold, the weighted average of no-train intervals is first used as the historical center, and then the threshold is shifted forward based on the advance amount represented by the dynamic pre-wake-up advance coefficient and the dynamic standard deviation. The larger the dynamic standard deviation, the more dispersed the recent distribution of no-train intervals. Setting a fixed wake-up time only around the average is likely to be later than the actual train arrival time, so a wider advance interval needs to be reserved. The calculation result is also compared with the minimum laser detection period and a value not less than that period is taken to prevent negative values ​​or thresholds that cannot be used for state scheduling when the historical average is too small or fluctuates too much.

[0089] The exponentially weighted mean gives recent train intervals a significant impact on the threshold, while the dynamic standard deviation provides a margin for uncertainty. When these two factors work together, as recent train intervals continuously shorten, the historical center gradually shifts forward, and the wake-up window is brought forward accordingly. Even when intervals don't shorten significantly but fluctuations suddenly increase, the standard deviation term still amplifies the advance. Conversely, when recent intervals steadily lengthen, the threshold gradually shifts backward. Therefore, the equipment schedules wake-up based on both the "recent center position" and the "recent dispersion," rather than simply averaging all historical data and using a fixed time for an extended period.

[0090] The predicted wake-up window start threshold and the pre-wake-up boundary serve different functions. The former determines when low-frequency surveying begins, while the latter, determined by S40 in conjunction with the deep sleep switching ratio, is further subdivided forward to arrange module power supply and baseband initialization. Decomposing a statistical prediction result into two state boundaries allows for the preservation of independent time periods for hardware preparation without altering the meaning of historical vehicle-to-device predictions. Even if a decrease in battery power causes the predicted wake-up window to advance, a definite sequential relationship is maintained between deep sleep, pre-wake-up, and low-frequency surveying.

[0091] Fixed-time wake-up is typically set based on timetables or empirical constants. However, when trains are delayed, additional services are added, intervals change, or the equipment has just completed an alarm, the fixed wake-up time can easily misalign with the actual no-train interval. This can cause the equipment to repeatedly sample far in advance, or to start only after the target has already entered the ranging range. In this embodiment, after each alarm is completed, S50 writes back the new no-train interval, S20 updates the statistics, and S30 calculates the next round of thresholds, allowing power consumption scheduling to be continuously corrected using the actual operating records of the measurement point where the equipment is located.

[0092] Taking the aforementioned weighted average train interval of 742.5 seconds, dynamic standard deviation of 37 seconds, normalized energy reduction factor of 0.30, and basic pre-wake-up advance coefficient of 1.8 as an example, the dynamic pre-wake-up advance coefficient is approximately 2.34. After shifting the statistical advance amount corresponding to this advance coefficient forward from 742.5 seconds, the predicted wake-up window start threshold is approximately 656 seconds, significantly greater than the minimum laser detection period of 2 seconds. Therefore, approximately 656 seconds is used as the threshold for this round. This calculation process does not interpret "approximately 656 seconds" as the inevitable arrival time of the train, but rather as the starting boundary for low-frequency surveys.

[0093] When the interval between trains is further shortened from 720 seconds in subsequent rounds, the weighted average will continue to decrease, and the equipment will enter low-frequency surveying earlier in the next round. If the interval is extended to more than 900 seconds multiple times at night and the fluctuations converge, the threshold will be shifted later, thus extending the deep sleep and pre-wake-up phases accordingly. The single threshold output by S30 and the adaptive detection parameter set are sent to S40 together. The former provides the time boundary, and the latter provides the current duration, ranging cycle, deep sleep switching ratio, and sampling parameters. The two together support the hierarchical state switching.

[0094] Step S40: Based on the predicted wake-up window start threshold and the adaptive probe parameter set, a hierarchical state switching mechanism is used to execute a power scheduling task and output the current power status;

[0095] The current power consumption state is a unified identifier for the operational status of the laser detection module and the wireless relay module at the current moment, including deep sleep state, pre-wake-up state, low-frequency patrol state, and high-density early warning state. The hierarchical state switching mechanism first extracts the duration of the current train-free event, the adaptive laser detection cycle, the deep sleep switching ratio, the preset continuous reduction quantity, and the high-density sampling frequency from the adaptive detection parameter set. Then, it determines the pre-wake-up boundary by multiplying the deep sleep switching ratio by the predicted wake-up window start threshold. Thus, a train-free waiting process is divided into four stages: sleep preparation, module preparation, trend patrol, and event confirmation.

[0096] When the duration of no train events is less than the pre-wake-up boundary, the device enters a deep sleep state. At this time, the laser detection module and wireless forwarding module are turned off, retaining only the low-power clock that can accumulate the duration and trigger subsequent state judgments. Since the S30 has already determined the wake-up time position based on recent train intervals and current battery level, deep sleep is not unconditionally continuous, but rather reduces the standby power consumption of the two types of high-power modules before approaching the statistical wake-up interval.

[0097] When the duration reaches the pre-wake-up boundary but has not yet reached the predicted wake-up window start threshold, the state switches to the pre-wake-up state. The laser detection module enters the power supply preparation process, and the wireless relay module performs baseband initialization. However, during this stage, continuous ranging is not performed at the high-density sampling frequency, nor is alarm data sent. By advancing the module preparation from the proximity event confirmation stage, the time required for power supply stabilization and communication initialization can be scheduled before the train enters the high-risk period, while avoiding maintaining a wireless network connection for an extended period during the deep sleep phase.

[0098] After the duration reaches the predicted wake-up window start threshold, the system enters a low-frequency survey state. The laser detection module collects distance data according to the adaptive laser detection cycle obtained in step S20. The task in this state is not to immediately confirm the train, but to observe the direction of distance change with a sampling rhythm constrained by both power and historical intervals. When a preset number of successively decreasing effective distance values ​​are continuously obtained, the distance sequence shows a continuous approaching trend. Only then does the system switch to a high-density warning state and continuously collect data at a high-density sampling frequency, while simultaneously controlling the wireless forwarding module to complete network attachment.

[0099] Switching directly from deep sleep to full-speed ranging and wireless online operation would result in a complete module startup and communication power consumption for each statistical wake-up, even if the ranging result indicates no train is present, incurring a high-power consumption process. Conversely, waiting until the distance value meets the alarm conditions before initiating wireless forwarding would add network attachment time to the alarm transmission phase. The four-level state separates time accumulation, module preparation, trend discovery, and intensive confirmation, ensuring that high power consumption increases progressively with train approach evidence, and network preparation is completed before the final alarm.

[0100] The criteria for determining the four states progress from time statistics to distance trends. In deep sleep and pre-wake states, only the duration of the current train-free event is compared to two time boundaries; continuous distance data acquisition is not required. In low-frequency survey state, distance values ​​are acquired, but only it is determined whether a specified number of consecutive decreasing sequences have formed. Only in high-density early warning state is continuous distance measurement data handed over to the S50 for window analysis. The processing load increases with the sufficiency of evidence, avoiding premature median filtering, differential calculations, and wireless alarm transmission before a distance trend is established.

[0101] Taking the approximately 656-second predicted wake-up window start threshold obtained from S30 and the deep sleep switching ratio of 0.70 as an example, the pre-wake-up boundary is approximately 459 seconds. When the duration of the current no-train event is 430 seconds, the device remains in deep sleep; when it reaches 470 seconds, the device enters the pre-wake-up state, begins laser module power supply preparation and wireless baseband initialization; after 656 seconds, the device switches to low-frequency surveying and obtains distance values ​​according to an adaptive cycle of approximately 22 to 24 seconds. The state boundaries are all derived from the aforementioned parameter set and thresholds, and do not rely on temporary judgments by operators.

[0102] If, during the low-frequency survey phase, three effective distance values ​​of 1000 meters, 984 meters, and 969 meters are obtained sequentially, and these three distance values ​​meet the requirement of successive decreases within a preset consecutive decrease quantity of 3, the system immediately sets the current power consumption state to a high-density warning state. The laser detection module switches to continuous sampling at 10Hz, the wireless forwarding module performs network attachment, and the current power consumption state output by S40 is subsequently used as the execution condition for S50. If the low-frequency distance values ​​do not form the prescribed consecutive decrease sequence, the device continues low-frequency surveying and will not directly trigger an alarm process due to a single small distance value.

[0103] Step S50: Based on the current power consumption status, perform laser detection data acquisition, train approach event judgment, and alarm data transmission, and output an alarm trigger signal;

[0104] The alarm trigger signal is a data packet sent to the alarm receiving terminal after a train approach event is confirmed. It encapsulates at least the current distance value, the estimated train speed, and the device identifier. S50 uses the current power consumption state as the entry condition and only performs continuous distance analysis when the state is a high-density warning state. The system reads the sliding queue length, monitoring window length, preset proximity ratio threshold, and preset second-order difference threshold from the adaptive detection parameter set, so that this round of dense discrimination uses the same set of engineering parameters loaded in S10 and associated and encapsulated by S20.

[0105] Continuously acquired laser detection data is written into a sliding queue in chronological order of sampling time. The queue retains only the most recent distance data corresponding to the preset sliding queue length. Median filtering corrects for occasional deviations based on the median position of the distance values ​​within the queue, ensuring that single reflection anomalies do not directly alter the overall proximity trend. After filtering, the system subtracts the distance value from the previous sampling time from the distance value at the next sampling time to obtain the first-order distance difference; then, it subtracts the previous first-order distance difference from the next first-order distance difference to obtain the second-order distance difference, simultaneously retaining the sampling time interval spanned by each first-order distance difference.

[0106] A first-order distance difference less than 0 indicates that the target distance measured at a later time step is less than that at a previous time step, serving as evidence of the target's approach direction. The system calculates the proportion of such data points among all distance data points within the monitoring window, obtaining the approach data point proportion. If this proportion is not lower than a preset approach proportion threshold, it indicates that most distance changes within the window follow the same direction; if the absolute value of each second-order distance difference is less than a preset second-order difference threshold, it indicates that there are no abrupt jumps in adjacent distance changes exceeding the allowable range. Only when both conditions are met simultaneously will a train approach event confirmation result be generated.

[0107] In an optional dual-path detection implementation, the laser detection module includes a first laser detection unit and a second laser detection unit spaced apart along the track extension direction. The two laser detection units respectively form a first detection position and a second detection position, and the numbers of the two detection positions and their azimuth relationship relative to the alarm point are pre-written into the STM32 main control board program. Each laser detection unit converts the detected distance signal into an analog laser voltage quantity corresponding to the ranging result. After signal conditioning, the signal is input to the analog-to-digital converter port of the STM32 main control board. The STM32 main control board continuously samples the two analog laser voltage quantities according to a unified timing benchmark and records the trigger time when either analog laser voltage quantity exceeds a trigger threshold matching the train entering the corresponding detection position. When the train sequentially triggers the first and second laser detection units, it is determined that the train is running in the first direction based on the pre-established correspondence between the detection position number and the track direction; when the train sequentially triggers the second and first laser detection units, it is determined that the train is running in the second direction, opposite to the first direction. The difference between the two trigger times is used as the time difference for the train to pass through the two detection positions, and the train speed can be calculated by combining this with the preset distance between the two detection positions. The STM32 main control board will use the direction determination result corresponding to the trigger sequence as evidence of the train approach direction only when the time difference is within an effective time range that matches the line's allowable speed and the preset spacing. It will then jointly confirm this result with the aforementioned proportion of approach data points and the second-order distance difference criterion to exclude false triggering caused by single-path instantaneous occlusion, sudden reflection changes, or non-train targets.

[0108] The proportion of data points approaching a target is used to constrain the consistency of the direction of change, while the second-order distance difference is used to constrain the continuity of the change process. They address different problems. When only checking whether the distance is continuously decreasing, multiple negative jumps caused by occasional occlusion might be mistaken for an approaching train. When only checking whether the change is smooth, targets that are slowly moving away or essentially stationary might also meet the smoothness condition. Using both the direction proportion and the second-order change as confirmation conditions allows the monitoring window to simultaneously possess the characteristics of "most distances are decreasing" and "there are no abnormal abrupt changes in the decreasing process."

[0109] The sliding queue length and monitoring window length constrain the range of recent data retention and the current discrimination range, respectively. The queue is continuously updated as new sampled data is written, ensuring that filtering always revolves around the most recent distance sequence. The monitoring window extracts data segments from recent results in the queue for proportion and second-order difference comparison. The queue length provides the data foundation required for filtering and continuous differencing, while the monitoring window length determines the time span covered by a single confirmation. Both are uniformly given by the initialization parameter set, thus preventing inconsistencies in the judgment criteria caused by temporary changes in the data range during operation.

[0110] Single-point distance thresholds are easily affected by birds, pedestrians, falling objects, and strong local reflections. For example, when an interfering object briefly crosses the optical path, the distance may suddenly drop from 900 meters to 35 meters, and then recover to 895 meters. Even though 35 meters is lower than the traditional alarm distance, the first-order change direction of this sequence is reversed, and the second-order change will be significantly increased, thus failing to simultaneously satisfy the proportion condition and the second-order difference condition of this embodiment. The distance sequence formed by continuous train approach maintains the same direction of change across multiple sampling moments, making it more suitable for entry speed estimation and alarm encapsulation.

[0111] Using the 10Hz high-density sampling scenario, a monitoring window, after median filtering, yields continuous distance values ​​of 120.0 meters, 117.8 meters, 115.7 meters, 113.5 meters, 111.4 meters, 109.2 meters, 107.1 meters, and 105.0 meters. Adjacent distance values ​​all show a decreasing trend, with the proportion of data points approaching each other exceeding the preset 0.75. The variation between adjacent first-order distance differences is small, and the absolute value of each second-order distance difference is also below 0.60 meters. Based on this, the system generates a train approach event confirmation result, instead of immediately triggering an alarm upon the first measurement of 120.0 meters.

[0112] Upon confirmation of the train approach event, the system divides the absolute value of each first-order distance difference by the corresponding sampling time interval to obtain the speed estimates for each adjacent sampling segment. The median of these ratios is then taken as the estimated train speed. Using the median reduces the impact of slight deviations at individual distance points on the speed results. For the aforementioned 0.1-second sampling interval and adjacent distance variations of approximately 2.1 to 2.2 meters, the speed estimates for each segment are concentrated around 21 to 22 meters per second; the resulting median can be used as the estimated train speed for this test.

[0113] The system encapsulates the current distance value of 105.0 meters, the estimated train speed, and the device identifier into an alarm trigger signal, which is then sent to the alarm receiving terminal by the wireless forwarding module that has completed network attachment. After transmission, the system determines the duration of the current no-train event accumulated since the last time a train completely left the laser detection range as the current no-train interval duration, writes it into the historical no-train passage time data, and then restores the current power consumption state to deep sleep. The newly written interval becomes the most recent complete sample in the next round and participates in the mean and variance update of S20, thus forming a closed loop from proximity recognition, alarm transmission, historical statistical update, and the next round of energy-saving scheduling.

[0114] Therefore, the alarm trigger signal is not an isolated final output. Its completion time simultaneously marks the end of this round of high-density early warning processes, and the written-back no-train interval changes the weighted mean and dynamic standard deviation for the next round. The updated statistical results further affect the adaptive laser detection cycle, the predicted wake-up window start threshold, and the duration range of each power consumption state. The equipment uses the actual operating interval to correct subsequent scheduling after each real train approach event, rather than relying on the fixed train-to-train interval set during initial deployment.

[0115] Example 2: Furthermore, the present invention provides a laser-based train approach recognition system that employs a laser-based train approach recognition method from the above embodiments, thereby solving the technical problem of laser-based train approach recognition. The beneficial effects of the laser-based train approach recognition system provided by the present invention are the same as those of the laser-based train approach recognition method provided in the above embodiments, and other technical features of the laser-based train approach recognition system are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0116] Example 3: This invention provides a laser-based train proximity recognition device. The device includes at least one processor and a memory communicatively connected to the processor. The memory stores instructions executable by the processor, which are then executed to enable the processor to perform the laser-based train proximity recognition method described in Example 1. The laser-based train proximity recognition device in this invention may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. This laser-based train proximity recognition device is merely an example and should not limit the functionality or scope of the invention. The device may also include a processing unit (e.g., a central processing unit, a graphics processing unit), which can perform various appropriate actions and processes based on a program stored in a read-only memory or a program loaded from a storage device into a random access memory. The random access memory also stores various programs and data required for the operation of a laser-based train proximity detection device. The processing unit, read-only memory, and random access memory are interconnected via a bus. The I / O interface is also connected to the bus. Typically, the following systems can be connected to the I / O interface: input devices including touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices including liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices including magnetic tapes, hard disks, etc.; and communication devices. The communication device allows the laser-based train proximity detection device to communicate wirelessly or wiredly with other devices to exchange data. While a laser-based train proximity detection device with various systems has been described, it should be understood that it is not required to implement or possess all of the described systems. Alternatively, more or fewer systems may be implemented.

[0117] Example 4: This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the laser detection-based train approach recognition method described above. The computer program product provided by this invention can solve the technical problem of laser detection-based train approach recognition. Compared with the prior art, the beneficial effects of the computer program product provided by this invention are the same as those of the laser detection-based train approach recognition method provided in the above embodiments, and will not be repeated here.

[0118] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device, or installed from a read-only memory. When the computer program is executed by a processing device, it performs the functions defined in the methods of the embodiments disclosed in this invention.

[0119] It should be understood that the various parts disclosed in this invention can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0120] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the present invention and its equivalents, the present invention also intends to include these modifications and variations.

Claims

1. A train approach recognition method based on laser detection, characterized in that, The method includes: Step S10: Obtain train approach recognition initialization data, and perform initialization tasks based on the train approach recognition initialization data using a parameter grouping loading mechanism, and output the initialization parameter set; Step S20: Obtain the current battery state, and based on the initialization parameter set and the current battery state, perform the laser detection period calculation task using the state of charge correction and nonlinear periodic mapping mechanism, and output the adaptive detection parameter set; Step S30: Based on the adaptive detection parameter set, perform statistical preprocessing tasks for predicting the wake-up window using an exponentially weighted statistical prediction mechanism, and output the starting threshold of the predicting wake-up window; Step S40: Based on the predicted wake-up window start threshold and the adaptive probe parameter set, a hierarchical state switching mechanism is used to execute a power scheduling task, and the current power consumption status is output; Step S50: Based on the current power consumption status, perform laser detection data acquisition, train approach event judgment, and alarm data transmission, and output an alarm trigger signal.

2. The train proximity recognition method based on laser detection as described in claim 1, characterized in that, Step S10, which involves acquiring train approach recognition initialization data and constructing the initialization parameter set based on the train approach recognition initialization data using a parameter grouping and loading mechanism, specifically includes: Step S101: Extract dynamic power management parameters from the train approach recognition initialization data. The dynamic power management parameters include minimum laser detection period, maximum laser detection period, reference state of charge, S-shaped mapping function steepness factor, exponential weighted attenuation factor, correction amplitude factor, confidence interval coefficient, basic pre-wake-up advance coefficient, preset standard deviation lower limit, deep sleep switching ratio, preset continuous reduction quantity, high-density sampling frequency, sliding queue length, monitoring window length, preset approach ratio threshold, and preset second-order difference threshold. The deep sleep switching ratio is greater than 0 and less than 1. Step S102: Extract historical no-train-passage time period data arranged by occurrence time from the train approach recognition initialization data. The duration of each no-train-passage time period in the historical no-train-passage time period data is the duration between when a train completely leaves the laser detection range and when the next train enters the laser detection range. Step S103: Write the dynamic power consumption management parameters into the parameter field according to the parameter identifier, write the historical no-train-passage time period data into the historical duration field according to the occurrence time, and associate and encapsulate the parameter field and the historical duration field to generate the initialization parameter set.

3. The train proximity recognition method based on laser detection as described in claim 2, characterized in that, Step S20, which involves obtaining the current battery state, performing laser detection period calculation based on the initialization parameter set and the current battery state using a state-of-charge correction and nonlinear periodic mapping mechanism, and outputting the adaptive detection parameter set, specifically includes: Step S201: Extract the current state of charge from the current battery state. The ratio of the difference between the reference state of charge and the current state of charge to the reference state of charge is calculated, and the ratio is restricted to between 0 and 1 to obtain the normalized charge reduction factor. ; Step S202: Read the historical no-train-passage time period data from the initialization parameter set, and calculate the weighted average no-train-passage interval based on the exponential weighted attenuation factor. and dynamic standard deviation It also records the duration from the moment the train completely left the laser detection range until the current moment, thus obtaining the duration of the current train-free event. ; Step S203: Calculate the safety margin correction factor according to the following formula: The corrected state of charge is calculated using the following formula: The adaptive laser detection period is calculated using the following formula: in, This is the safety margin correction factor. The correction amplitude factor, The confidence interval coefficient is... To preset the lower limit of standard deviation, For the corrected state of charge, The adaptive laser detection period is... The minimum laser detection period is... The maximum laser detection period is... The steepness factor of the S-shaped mapping function. The reference state of charge, It is the hyperbolic tangent function. It is a function with maximum value. It is a minimum value function. It is a natural exponential function; The adaptive laser detection period, the weighted average of no-train intervals, the dynamic standard deviation, the duration of the current no-train event, the normalized power reduction factor, the basic pre-wake-up advance coefficient, the minimum laser detection period, the deep sleep switching ratio, the preset continuous reduction quantity, the high-density sampling frequency, the sliding queue length, the monitoring window length, the preset proximity ratio threshold, and the preset second-order difference threshold are associated and encapsulated to generate the adaptive detection parameter set.

4. The train proximity recognition method based on laser detection as described in claim 3, characterized in that, Step S202, the step of calculating the weighted mean of the train interval and the dynamic standard deviation includes: The most recent complete no-train interval duration is obtained from the historical no-train-passage time period data. The difference between the most recent complete no-train interval duration and the previous weighted no-train interval mean is calculated to obtain the interval deviation. Based on the exponentially weighted decay factor, the most recent complete no-train interval duration and the previous weighted no-train interval mean are combined exponentially to obtain the updated weighted no-train interval mean. Based on the exponentially weighted decay factor and the square of the interval deviation, the previous no-train interval variance is incrementally corrected to obtain the updated no-train interval variance. The updated variance of trainless intervals is compared with the preset minimum variance. The larger of the updated variance of trainless intervals and the preset minimum variance is squared to obtain the dynamic standard deviation. The preset minimum variance is the square of the preset lower limit of standard deviation.

5. The train proximity recognition method based on laser detection as described in claim 3, characterized in that, Step S30, which involves performing a statistical preprocessing task for predicting the wake-up window based on the adaptive detection parameter set using an exponentially weighted statistical prediction mechanism, and outputting the starting threshold of the predictive wake-up window, specifically includes: Step S301: Extract the weighted mean of no-train interval, the dynamic standard deviation, the normalized power reduction factor, the basic pre-wake-up advance coefficient, and the minimum laser detection cycle from the adaptive detection parameter set; Step S302: Calculate the dynamic pre-wake-up advance coefficient according to the following formula: Step S303: Calculate the predicted wake-up window start threshold according to the following formula: in, The dynamic pre-wake-up advance coefficient is... The basic pre-wake-up advance coefficient is... The threshold value for the predicted wake-up window is [value].

6. The train proximity recognition method based on laser detection as described in claim 5, characterized in that, Step S40, which involves executing a power scheduling task based on the predicted wake-up window start threshold and the adaptive probe parameter set using a hierarchical state switching mechanism, and outputting the current power consumption state, specifically includes: Step S401: Extract the duration of the current no-train event, the adaptive laser detection cycle, the deep sleep switching ratio, the preset continuous reduction quantity, and the high-density sampling frequency from the adaptive detection parameter set, and determine the pre-wake boundary based on the product of the deep sleep switching ratio and the predicted wake-up window start threshold; Step S402: When the duration of the current no-train event is less than the pre-wake-up boundary, the current power consumption state is set to deep sleep state, the laser detection module and the wireless forwarding module are turned off, and the low-power clock is kept running; when the duration of the current no-train event is not less than the pre-wake-up boundary and is less than the predicted wake-up window start threshold, the current power consumption state is set to pre-wake-up state, and the power supply preparation process of the laser detection module and the baseband initialization process of the wireless forwarding module are started. Step S403: When the duration of the current no-train event is not less than the predicted wake-up window start threshold, the current power consumption state is set to low-frequency survey state, and the laser detection module is controlled to collect laser detection data according to the adaptive laser detection cycle; when a preset number of successively decreasing effective distance values ​​are continuously obtained, the current power consumption state is switched to high-density early warning state, the laser detection module is controlled to continuously collect the laser detection data according to the high-density sampling frequency, and the wireless forwarding module is controlled to perform network attachment.

7. The train proximity recognition method based on laser detection as described in claim 6, characterized in that, Step S50, which involves collecting laser detection data, determining train approach events, and sending alarm data based on the current power consumption state, and then outputting the alarm trigger signal, specifically includes: Step S501: When the current power consumption state is the high-density warning state, extract the sliding queue length, the monitoring window length, the preset proximity ratio threshold, and the preset second-order difference threshold from the adaptive detection parameter set. Write the continuously collected laser detection data into the sliding queue with the sliding queue length according to the sampling time order. Perform median filtering on the distance values ​​in the sliding queue. Subtract the distance value of the previous sampling time from the distance value of the later sampling time in every two adjacent sampling times to obtain the first-order distance difference. Subtract the previous first-order distance difference from the later first-order distance difference in every two adjacent first-order distance differences to obtain the second-order distance difference. Record the sampling time interval corresponding to each of the first-order distance differences. Step S502: Within the monitoring window with the specified monitoring window length, the proportion of distance data points corresponding to the first-order distance difference values ​​less than 0 is calculated out of all distance data points to obtain the proportion of approaching data points. The proportion of approaching data points is compared with a preset approach proportion threshold, and the absolute value of each second-order distance difference is compared with a preset second-order difference threshold. When the proportion of approaching data points is not lower than the preset approach proportion threshold, and the absolute value of each second-order distance difference is less than the preset second-order difference threshold, a train approach event confirmation result is generated. Step S503: In response to the confirmation result of the train approach event, calculate the ratio of the absolute value of each of the first-order distance differences to the corresponding sampling time interval, and determine the median of each ratio as the estimated train speed. Encapsulate the current distance value, the estimated train speed, and the device identifier to generate the alarm trigger signal, and control the wireless forwarding module to send the alarm trigger signal to the alarm receiving terminal. After the transmission is completed, determine the duration of the current no-train event as the current no-train interval duration, write the current no-train interval duration into the historical no-train passage time period data, and restore the current power consumption state to the deep sleep state.

8. A laser-based train approach recognition system, applied to the laser-based train approach recognition method according to any one of claims 1 to 7, characterized in that, The laser-based train proximity recognition system includes: The initialization module is used to acquire train approach recognition initialization data, perform initialization tasks based on the train approach recognition initialization data using a parameter grouping loading mechanism, and output an initialization parameter set. The cycle calculation module is used to obtain the current battery state, and based on the initialization parameter set and the current battery state, it performs the laser detection cycle calculation task using the state of charge correction and nonlinear cycle mapping mechanism, and outputs an adaptive detection parameter set. The wake-up prediction module is used to perform statistical preprocessing tasks for predicting the wake-up window based on the adaptive detection parameter set using an exponentially weighted statistical prediction mechanism, and output the starting threshold of the predicted wake-up window. The power scheduling module is used to perform power scheduling tasks based on the predicted wake-up window start threshold and the adaptive probe parameter set using a hierarchical state switching mechanism, and output the current power status. The alarm processing module is used to collect laser detection data, determine train approach events, and send alarm data based on the current power consumption status, and output an alarm trigger signal.

9. A train proximity identification device based on laser detection, characterized in that, The laser-based train approach recognition device includes: a memory, a processor, and a laser-based train approach recognition program stored in the memory and executable on the processor. When the laser-based train approach recognition program is executed by the processor, it implements a laser-based train approach recognition method according to any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a laser-based train approach recognition program, which, when executed by a processor, implements a laser-based train approach recognition method according to any one of claims 1 to 7.

Images