A low-power standby and human body induction wake-up bus electronic stop board energy-saving method

Abstract

This invention relates to the field of intelligent transportation technology, and in particular to an energy-saving method for electronic bus stop signs with low-power standby and human body induction wake-up. The method includes the following steps: initializing and reading basic operating data, performing multi-sensor cross-calibration and generating a calibration parameter set; applying the calibration parameter set, using a passive infrared sensor for initial screening, and collecting multimodal sensing data after triggering, outputting classification results through feature-level fusion; recognizing passenger waiting intentions, combining bus arrival and passenger flow density predictions to generate operating status control commands; sending control commands to the display control and energy module to execute dynamic control of display zones and intelligent energy scheduling based on weather predictions. This invention significantly reduces the energy consumption of electronic bus stop signs through multi-sensor data fusion, gradient wake-up, and predictive scheduling technologies, balancing user experience and lifespan, and has significant potential for large-scale promotion.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, and in particular to an energy-saving method for electronic bus stop signs with low power consumption standby and human body induction wake-up. Background Technology

[0002] As a core terminal device in the urban intelligent transportation system, electronic bus stop signs play a crucial role in providing real-time information on bus arrivals, route changes, and public services. With the advancement of smart city construction in my country, electronic bus stop signs have been deployed on a large scale in cities across the country. Traditional electronic bus stop signs generally operate 24 / 7 at full power, with display, communication, and main control modules constantly in operation, resulting in high daily energy consumption per stop. This not only increases the operating costs of urban public facilities but also contradicts the concept of green and low-carbon development. To address the energy consumption issue, existing technologies are gradually incorporating human body detection wake-up technology. This technology uses passive infrared sensors or millimeter-wave radar to detect the presence of humans, achieving basic energy-saving control by turning the screen on when a person arrives and off when they leave. Some solutions also incorporate solar-assisted power supply and simple ambient light adjustment, further reducing the operating energy consumption of electronic bus stop signs to some extent.

[0003] However, existing energy-saving solutions for electronic bus stop signs still have significant technical shortcomings. First, single-sensor detection has inherent limitations. Passive infrared sensors are easily affected by ambient temperature, with sensitivity decreasing significantly at low temperatures and severe thermal noise interference at high temperatures. Millimeter-wave radar is prone to target confusion in densely populated areas, and current technologies cannot effectively distinguish between waiting passengers and passersby, resulting in a high invalid wake-up rate. Second, existing solutions mostly adopt a two-level wake-up mode of "deep standby - normal operation," which presents an inherent contradiction between wake-up delay and energy-saving effect. Wake-up too quickly increases standby power consumption, while wake-up too slowly affects user experience. Third, energy management is relatively crude, lacking predictive scheduling capabilities based on weather forecasts, resulting in low utilization of solar power generation. Lithium batteries lack refined health management, have short lifespans, and high maintenance costs. Finally, existing technologies are all single-site independent control modes, failing to achieve coordinated energy saving for multiple bus stops within a region. They cannot cope with special scenarios such as power grid shortages and cannot meet the needs of large-scale, high-density deployment. Summary of the Invention

[0004] To address the technical deficiencies in the background technology, this invention proposes a low-power standby and human body induction wake-up method for energy-saving electronic bus stop signs, solving the aforementioned technical problems and meeting practical needs. The specific technical solution is as follows: A low-power standby and human body induction wake-up method for energy-saving electronic bus stop signs includes the following steps: After the system is powered on, it executes the initialization process, reads basic operating data, completes the cross-parameter calibration of the multi-sensor array, and generates a calibration parameter set. The target is initially screened by a passive infrared sensor using a set of calibration parameters. After the triggering conditions are met, multimodal sensing data is collected and the target classification result is output after feature-level fusion. Based on the target classification results, the waiting intention information of targets classified as passengers is generated by identifying the waiting intention of the target. Combined with bus arrival prediction and passenger flow density prediction data, the system operation status control instructions are generated. Control commands are sent to the display control module and the energy management module to execute dynamic control of display zones and intelligent energy dispatching based on weather forecasts, respectively.

[0005] Furthermore, the basic operating data includes real-time clock data, ambient temperature data, light intensity data, and historical calibration data. The multi-sensor array includes a passive infrared sensor, a millimeter-wave radar sensor, an ultrasonic ranging sensor, and an ambient light sensor. The cross-parameter calibration is to perform distance calibration on the millimeter-wave radar sensor based on the ultrasonic ranging sensor, and to perform temperature compensation calibration on the passive infrared sensor based on the ambient temperature. The specific steps for temperature compensation calibration of the passive infrared sensor based on ambient temperature are as follows: The temperature compensation coefficient is calculated based on the current ambient temperature. The original output voltage of the passive infrared sensor is calibrated using the temperature compensation coefficient to obtain the calibrated output voltage. The detection threshold is then dynamically adjusted based on the current ambient temperature. The specific details of dynamically adjusting the detection threshold are as follows: When the ambient temperature is lower than the preset low temperature threshold, the detection threshold is set to the first preset multiple of the benchmark threshold; When the ambient temperature is higher than the preset high temperature threshold, the detection threshold is set to the second preset multiple of the benchmark threshold; When the ambient temperature is between the preset low temperature threshold and the preset high temperature threshold, the detection threshold is set as the baseline threshold.

[0006] Furthermore, the specific steps for calibrating the millimeter-wave radar sensor using the ultrasonic ranging sensor as a reference are as follows: The system automatically triggers a millimeter-wave radar sensor distance calibration process every preset time interval; Detect whether there is a stationary target within a preset effective range. If so, continuously collect a preset number of raw distance data from the millimeter-wave radar sensor and the corresponding ultrasonic ranging sensor ranging data. The least squares method was used to perform linear regression calculations on the raw distance data from the millimeter-wave radar sensor and the ranging data from the ultrasonic ranging sensor to obtain the linear calibration coefficients. The calculated linear calibration coefficients are stored in the system memory to correct the original distance measurements of all subsequent millimeter-wave radar sensors.

[0007] Furthermore, the specific steps for outputting the target classification result through feature-level fusion are as follows: The calibration parameter set is applied to the passive infrared sensor and the millimeter-wave radar sensor respectively. The passive infrared sensor is used for initial target screening. When the detected change in infrared radiation exceeds the detection threshold, the millimeter-wave radar sensor and the ultrasonic ranging sensor are activated to collect multimodal sensing data. A target recognition model based on an improved evidence theory algorithm is constructed, which includes five preset categories: passengers, pets, vehicles, fallen leaves, and interference. Based on multimodal sensing data, infrared radiation variation features of passive infrared sensors, Doppler frequency shift features, micro-motion features, distance features and angle features of millimeter-wave radar sensors, precise distance features of ultrasonic ranging sensors, and illumination intensity features of ambient light sensors are extracted and input into the target recognition model for weighted fusion to calculate the comprehensive confidence level of each target category. When the highest overall confidence level is greater than the first preset confidence level threshold and the difference between the highest overall confidence level and the second highest overall confidence level is greater than the preset difference threshold, the target is determined to be the category corresponding to the highest overall confidence level; otherwise, it is determined to be an uncertain target and waits for the next frame of multimodal perception data for re-determination. If the target is determined to be an uncertain target for a consecutive preset number of frames, the target detection process is terminated and the system returns to the initial target screening state of the passive infrared sensor. During the weighted fusion process, instantaneous interference is eliminated by a time window filtering algorithm and cross-interference of multiple targets is eliminated by a spatial clustering algorithm.

[0008] Furthermore, the specific steps for generating intent information by recognizing the waiting intent of targets classified as passengers are as follows: Multimodal perception data of targets classified as passengers are continuously collected for a preset number of frames. Position data, velocity data, acceleration data and orientation angle data are extracted, and after normalization, the data is recombined using a sliding window method. The reorganized data is input into a waiting intention recognition model pre-deployed on a low-power microcontroller for inference, and the confidence values ​​corresponding to the three intentions of passengers waiting for a bus, passing by, and resting are output. Based on the confidence value, a corresponding intention instruction is generated. When the confidence of the waiting intention is greater than the second preset confidence threshold, a waiting passenger intention instruction is generated. When the confidence of the passing intention is greater than the third preset confidence threshold and the target is getting farther and farther away from the bus stop, a passing pedestrian intention instruction is generated. When neither of the above two conditions is met, a resting person intention instruction is generated. The intent confidence value and intent command are integrated into waiting intent information.

[0009] Furthermore, the specific steps for obtaining the bus arrival prediction and passenger flow density prediction data are as follows: The system receives vehicle GPS data from the bus dispatch center in real time, combines the vehicle GPS data with historical traffic data and average station dwell time data, and calculates the predicted bus arrival time data of the vehicle to the station through a pre-trained gradient boosting tree bus arrival prediction model. Retrieve historical passenger flow data for a preset historical period at this station, input it into a pre-trained passenger flow time series prediction model, and calculate the passenger flow trend and passenger flow density prediction data for this station within the preset passenger flow prediction period.

[0010] Furthermore, the specific steps for fusing and generating system operation status control instructions are as follows: Based on waiting intention information, bus arrival prediction time data, and passenger flow density prediction data, fuzzy reasoning is performed through a preset fuzzy rule base to output system operation status level instructions. The system operation status includes four levels: deep standby state, shallow standby state, pre-wake-up state, and normal operation state.

[0011] Furthermore, the specific steps for performing dynamic control of the display partition are as follows: The content to be displayed is divided into three priority categories: core information, important information, and secondary information. The display module is divided into grid-like partitions, and corresponding display content is assigned to each display partition; The brightness value of each display zone is calculated based on the priority of the displayed content, ambient light intensity, viewing distance, and remaining battery power. The display status of each display zone is controlled according to the calculated brightness value. Zones with no information are completely powered off, while zones with information are lit according to the corresponding brightness value. The formula for calculating the brightness value of each display zone is as follows: , in, Indicates the first Line 1 The columns display the brightness values ​​of the zones. Indicates the base brightness value. This indicates the priority of the content displayed in this partition: a P value of 3 corresponds to core information, a P value of 2 corresponds to important information, and a P value of 1 corresponds to secondary information. This indicates the average viewing distance between the passenger and the display module, in meters. This represents the battery remaining capacity correction factor. This means that the calculation result will be limited to between 0 and 255; The base brightness value is calculated using the following formula: , in, Ambient light intensity is expressed in lux. and Indicates the pre-calibration coefficient; The specific value of the battery remaining power correction coefficient is as follows: When the remaining battery power is lower than the first preset power threshold but not lower than the second preset power threshold The value is set to the first correction coefficient; when the remaining battery power is lower than the second preset power threshold... The value is set to the second correction coefficient; when the remaining battery power is not lower than the first preset power threshold... The value is the third correction coefficient, and the first preset power threshold is greater than the second preset power threshold.

[0012] Furthermore, the specific steps for implementing the weather forecast-based intelligent energy dispatch are as follows: Connect to a third-party weather forecast application interface to obtain weather forecast data for a set time period in the future; By inputting weather forecast data into a pre-trained network energy prediction model, the solar power generation and system energy consumption for a set future time period are calculated, and a dynamic charging and discharging plan is generated. The system controls the switching between solar power, mains power, and lithium battery energy storage according to a dynamic charging and discharging plan. When there is sufficient sunshine, solar power is used first, and excess power is stored in the lithium battery pack. When the cumulative sunshine time within the predicted set period is less than the preset cumulative sunshine threshold, the system power consumption level is reduced in advance, the voice broadcast module is turned off, and the maximum brightness of the display module is reduced. The system monitors the voltage, current, and temperature data of the lithium battery pack in real time. It accurately estimates the remaining capacity and health status of the lithium battery pack using an extended Kalman filter battery state estimation algorithm and employs dynamic equalization charging technology to balance the voltage of each battery cell.

[0013] Furthermore, it also includes regional collaborative energy-saving control steps, as detailed below: Adjacent electronic bus stop signs on the same road segment form a self-organizing network through low-power wide area network communication technology, establishing a point-to-point data transmission channel between the stops; Upstream stations collect real-time data on the number of waiting passengers and the departure of vehicles, and send this data to all downstream stations on the same road segment. When a downstream station receives data from an upstream station indicating that the number of waiting passengers is greater than or equal to a preset threshold, it enters a pre-wake state with a pre-set pre-wake duration. When a downstream station receives data from an upstream station indicating that a vehicle has departed, it prepares in advance to display the real-time arrival information of that vehicle. When a power grid shortage instruction is received from the power grid dispatch center or cloud operation and maintenance platform, all electronic bus stop signs in the area will automatically adjust their wake-up strategies according to their respective passenger flow priorities, giving priority to ensuring the normal operation of preset core stops; The cloud-based operation and maintenance platform collects energy consumption data, sensor data, and battery data from each electronic bus stop sign in the area in real time. Through machine learning fault prediction algorithms, it analyzes the data and predicts problems such as dust accumulation on solar panels, sensor aging, and battery failure in advance, generating maintenance work orders and sending them to operation and maintenance personnel.

[0014] Compared with existing technologies, the energy-saving method for electronic bus stop signs with low power consumption standby and human body induction wake-up provided by the present invention has the following beneficial effects: This invention addresses the issues of low detection accuracy and poor scene adaptability of single sensors by employing multi-sensor cross-parameter calibration and multi-modal feature-level fusion technology, significantly reducing the false trigger rate. It utilizes a multi-dimensional fusion decision-making mechanism combining waiting intention recognition with bus arrival prediction and passenger flow density prediction, coupled with four-level gradient wake-up control, achieving a seamless user experience of "screen lit upon arrival" while significantly reducing invalid wake-up times. Through display content priority division and grid-based dynamic control technology, it finely manages the power consumption of the display module. Combined with weather-forecast-based predictive energy scheduling and lithium battery lifecycle health management, it effectively improves energy utilization, extends lithium battery life, and extends battery life during continuous rainy weather. Simultaneously, it introduces regional collaborative energy-saving control, achieving regional energy consumption optimization and grid emergency dispatch, significantly improving the overall energy saving rate and reducing the long-term operation and maintenance costs of electronic bus stop signs. This makes it a valuable project for large-scale promotion in the national public transportation system. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating a low-power standby and human body induction wake-up method for energy-saving electronic bus stop signs according to the present invention. Detailed Implementation

[0016] In the description of this invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "middle," and "inner," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, it should be noted that unless otherwise explicitly specified and limited, the terms "installed," "connected," and "joined" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention through specific circumstances.

[0017] The embodiments of the present invention will be described below with reference to the accompanying drawings and related examples. The embodiments of the present invention are not limited to the following examples, and the present invention relates to the relevant necessary components in this technical field, which should be regarded as well-known technology in this technical field and can be known and mastered by those skilled in this technical field.

[0018] See Figure 1 This invention provides a low-power standby and human body induction wake-up method for energy-saving electronic bus stop signs, characterized by the following steps: Step S100: After the system is powered on, the initialization process is executed, basic operating data is read, cross-parameter calibration of the multi-sensor array is completed, and a calibration parameter set is generated. Specifically, after the system powers on and resets, it first completes the hardware initialization of the low-power microcontroller, memory, communication interface, and various sensors. Then, it reads the current time data from the real-time clock module, the ambient temperature data from the temperature sensor, the light intensity data from the ambient light sensor, and the historical calibration data stored in the system memory. Based on the read environmental parameters, it performs two core calibration operations: ① It performs distance calibration on the millimeter-wave radar sensor using the ultrasonic ranging sensor as a reference to eliminate distance deviations caused by non-metallic obstacles penetrating; ② It performs temperature compensation calibration on the passive infrared sensor according to the current ambient temperature to solve the sensitivity abnormality problem under extreme temperatures. Finally, it integrates the calibrated temperature compensation coefficient and millimeter-wave radar linear calibration coefficient into a calibration parameter set and stores it in the system's non-volatile memory for subsequent steps to call.

[0019] Step S200: Apply the calibration parameter set, perform initial target screening using a passive infrared sensor, collect multimodal sensing data after the triggering conditions are met, and output the target classification result after feature-level fusion. Specifically, the calibration parameter set is loaded into the drivers of the passive infrared sensor and the millimeter-wave radar sensor respectively to complete the sensor parameter configuration; the system enters a low-power standby state by default, with only the passive infrared sensor continuously sampling at a frequency of 1Hz; when the change in infrared radiation detected by the passive infrared sensor exceeds the dynamically adjusted detection threshold, it is determined as a preliminary trigger, and the millimeter-wave radar sensor and ultrasonic ranging sensor are immediately activated; the three sensors synchronously collect target data, which, combined with the illumination data from the ambient light sensor, constitute multimodal perception data; the corresponding feature vectors are extracted from the data of each sensor, and input into the target recognition model based on the improved evidence theory algorithm for feature-level fusion; the model calculates the comprehensive confidence of each target category, and outputs the final target classification result according to the preset confidence judgment rules. The classification categories include passengers, pets, vehicles, fallen leaves, and interference.

[0020] Step S300: Based on the target classification results, identify the waiting intention of the target classified as a passenger to generate intention information, and combine the bus arrival prediction and passenger flow density prediction data to generate system operation status control instructions. Specifically, when the output target classification result is "passenger", the waiting intention recognition process is initiated; a preset number of frames of multimodal perception data of passenger targets are continuously collected, and their position, speed, acceleration, and orientation angle features are extracted. After preprocessing, the data is input into a pre-deployed quantized unidirectional long short-term memory network waiting intention recognition model; the model outputs confidence values ​​for three intentions: waiting, passing by, and resting. The corresponding intention commands are generated by combining the confidence values, and finally, the confidence values ​​and intention commands are integrated into waiting intention information; at the same time, the system receives vehicle GPS data from the bus dispatch center in real time, calculates vehicle arrival time through a pre-trained gradient boosting tree bus arrival prediction model; retrieves historical passenger flow data of the station, calculates passenger flow density within a preset time period through a passenger flow time series prediction model; inputs the waiting intention information, bus arrival prediction time data, and passenger flow density prediction data into a fuzzy comprehensive decision-maker, performs inference through 27 preset fuzzy rules, and outputs system operation status level commands. The system operation status includes four levels: deep standby, shallow standby, pre-wake-up, and normal operation.

[0021] Step S400: Send control commands to the display control module and the energy management module to execute dynamic control of display zones and intelligent energy scheduling based on weather forecasts, respectively.

[0022] Specifically, the system operation status control commands output by the fuzzy comprehensive decision-maker are sent to the display control module and the energy management module via the internal bus. The display control module switches the display mode according to the commands, divides the content to be displayed into three priorities: core information, important information, and secondary information, and performs grid-based partitioning of the display module. The brightness value of each partition is calculated based on the content priority, ambient light intensity, viewing distance, and remaining battery power. Partitions without information are completely powered off. The energy management module adjusts the overall power consumption level of the system according to the commands. At the same time, it accesses a third-party weather forecast API to obtain future weather forecast data. It predicts solar power generation and system energy consumption through a pre-trained bidirectional long short-term memory network energy prediction model, generates a dynamic charging and discharging plan, controls the power supply switching between solar energy, mains power, and lithium battery, and monitors the lithium battery status in real time to achieve full life cycle health management. In addition, the energy management module is also responsible for coordinating the collaborative energy-saving control of adjacent bus stops within the area.

[0023] In one embodiment of the present invention, the basic operating data includes real-time clock data, ambient temperature data, light intensity data, and historical calibration data; the multi-sensor array includes a passive infrared sensor, a millimeter-wave radar sensor, an ultrasonic ranging sensor, and an ambient light sensor; and the cross-parameter calibration is to perform distance calibration on the millimeter-wave radar sensor based on the ultrasonic ranging sensor, and to perform temperature compensation calibration on the passive infrared sensor according to the ambient temperature. Specifically, during system initialization, the system reads the year, month, day, hour, minute, and second data from the real-time clock module, the ambient temperature data from the digital temperature sensor (accuracy ±0.5℃), and the light intensity data from the ambient light sensor (range 0-100000 lux) via the I2C bus. It also reads historical calibration data stored during the previous system run from the Flash non-volatile memory. The multi-sensor array adopts a "master-slave complementary" architecture: the passive infrared sensor acts as the master sensor for low-power initial screening, the millimeter-wave radar sensor acts as the auxiliary sensor for high-precision target identification, the ultrasonic ranging sensor acts as the reference sensor for distance calibration, and the ambient light sensor acts as the auxiliary sensor for interference suppression and display brightness adjustment.

[0024] The specific steps for temperature compensation calibration of the passive infrared sensor based on ambient temperature are as follows: The temperature compensation coefficient is calculated based on the current ambient temperature. The original output voltage of the passive infrared sensor is calibrated using the temperature compensation coefficient to obtain the calibrated output voltage. The detection threshold is then dynamically adjusted based on the current ambient temperature. Specifically, the temperature compensation coefficient is calculated using the following formula: , in, Indicates temperature as Temperature compensation coefficient at that time This indicates the current ambient temperature, in degrees Celsius. , , , The coefficients of the pre-trained third-order polynomial are represented; the calibrated output voltage is calculated using the following formula: in, This indicates the output voltage of the calibrated passive infrared sensor. This indicates the raw output voltage of the passive infrared sensor, measured in millivolts. These are zero-order polynomial coefficients, room temperature reference coefficients, representing the compensation reference value at 25℃, with a typical value of 1.023; These are first-order polynomial coefficients, linear temperature correction coefficients, used to compensate for the linear change in sensor output with temperature; a typical value is -0.0087. These are the coefficients of a second-order polynomial, a second-order nonlinear correction coefficient, used to compensate for the second-order temperature drift in the sensor output; a typical value is 0.00012. These are the coefficients of a third-order polynomial, representing cubic nonlinearity correction coefficients, used to compensate for higher-order nonlinear errors under extreme temperatures; a typical value is -5.2 × 10⁻⁶. -7 The above typical coefficients were obtained through batch temperature calibration experiments, as detailed below: A passive infrared sensor was placed in a high and low temperature test chamber, with temperature points set in 5℃ increments within the range of -40℃ to 85℃. After stabilizing at each temperature point for 30 minutes, the sensor's output voltage to a standard human body heat source was collected. Using the output voltage at 25℃ as a benchmark, the compensation coefficient for each temperature point was calculated. The compensation coefficient for all temperature points was fitted with a third-order polynomial using the least squares method to obtain... , , , The difference in sensor coefficients between different batches is less than 5%, and the adaptation coefficient can be obtained through the same calibration method to ensure consistency in mass production.

[0025] The specific details of dynamically adjusting the detection threshold are as follows: When the ambient temperature is lower than the preset low temperature threshold, the detection threshold is set to the first preset multiple of the benchmark threshold; When the ambient temperature is higher than the preset high temperature threshold, the detection threshold is set to the second preset multiple of the benchmark threshold; When the ambient temperature is between the preset low temperature threshold and the preset high temperature threshold, the detection threshold is set as the baseline threshold.

[0026] Specifically, when the ambient temperature is below the preset low-temperature threshold (typically -10℃), the detection threshold is reduced to 0.7 times the baseline threshold to improve detection sensitivity at low temperatures; when the ambient temperature is above the preset high-temperature threshold (typically 40℃), the detection threshold is increased to 1.5 times the baseline threshold to suppress thermal noise interference at high temperatures; the baseline threshold (typically 150mV) remains unchanged in the normal temperature range. The detection threshold refers to the trigger voltage threshold of the passive infrared sensor, which is a voltage threshold preset by the system. When the calibrated voltage value output by the passive infrared sensor is greater than or equal to this threshold, the system determines that "a suspected human target has been detected," triggering the subsequent millimeter-wave radar and ultrasonic sensor activation process; when the output voltage is less than this threshold, the system maintains a low-power standby state, with only the passive infrared sensor continuously sampling. The baseline threshold is the factory-preset baseline detection threshold for the passive infrared sensor.

[0027] In one embodiment of the present invention, the specific steps for calibrating the millimeter-wave radar sensor using an ultrasonic ranging sensor as a reference are as follows: Step S101: The system automatically triggers the millimeter-wave radar sensor distance calibration process once every preset time interval; Specifically, the system has a built-in calibration timer that automatically triggers a millimeter-wave radar distance calibration process at preset time intervals (typically 24 hours). Calibration can also be triggered remotely through a cloud-based operations and maintenance platform. The calibration process is prioritized to be executed when the system is in deep standby mode and there is no personnel activity, to avoid affecting normal services.

[0028] Step S102: Detect whether there is a stationary target within the preset effective range. If there is, continuously collect a preset number of raw distance data from the millimeter-wave radar sensor and the corresponding ultrasonic ranging sensor ranging data. Specifically, the system first controls a millimeter-wave radar sensor and an ultrasonic ranging sensor to simultaneously scan the area in front, detecting whether there are stationary targets within a preset effective range (0.5-5 meters). The method for determining a stationary target is: the target's distance change is less than ±5cm in 10 consecutive frames of data. If a stationary target is detected, 50 sets of raw millimeter-wave radar distance data and corresponding ultrasonic ranging data are continuously acquired, with each set of data acquired at a 100ms interval to ensure data stability and representativeness.

[0029] Step S103: Use the least squares method to perform linear regression calculation on the raw distance data of the millimeter-wave radar sensor and the ranging data of the ultrasonic ranging sensor to obtain the linear calibration coefficient; Specifically, the least squares method was used to perform linear regression analysis on the 50 sets of collected data to solve for the linear calibration coefficients α and β. The linear regression model is Rcal = α × Radar + β, where Radar is the original range of the millimeter-wave radar and Rcal is the calibrated range. The least squares method finds the best function match for the data by minimizing the sum of squared errors, which can effectively reduce the influence of random noise and ensure calibration accuracy.

[0030] Step S104: Store the calculated linear calibration coefficients in the system memory to correct the original distance measurements of all subsequent millimeter-wave radar sensors.

[0031] Specifically, the calculated linear calibration coefficients α and β are stored in the system's Flash memory, overwriting the previous calibration coefficients. All subsequent raw range measurements from the millimeter-wave radar are corrected using this linear model to obtain calibrated range values, which are then used for target identification and wake-up decisions.

[0032] In one embodiment of the present invention, the specific steps for outputting the target classification result through feature-level fusion are as follows: Step S201: Apply the calibration parameter set to the passive infrared sensor and the millimeter-wave radar sensor respectively. Perform initial target screening through the passive infrared sensor. When the detected change in infrared radiation exceeds the detection threshold, activate the millimeter-wave radar sensor and the ultrasonic ranging sensor to collect multimodal sensing data. Specifically, the generated calibration parameter sets (passive infrared temperature compensation coefficient and millimeter-wave radar linear calibration coefficient) are loaded into the corresponding sensor's drive register to complete the dynamic configuration of sensor parameters. The system defaults to a low-power standby state, with only the passive infrared sensor continuously sampling at a frequency of 1Hz, and the MCU in stop mode, resulting in overall power consumption of <0.15W. When the output voltage of the passive infrared sensor after two consecutive calibration samples is greater than or equal to the dynamic detection threshold, it is considered a preliminary trigger. The MCU immediately exits stop mode and synchronously activates the millimeter-wave radar sensor and ultrasonic ranging sensor via GPIO pins. The three sensors synchronously acquire data at a frequency of 20Hz, and feature extraction is performed after acquiring three consecutive frames of data to ensure data stability and consistency.

[0033] Step S202: Construct a target recognition model based on an improved evidence theory algorithm. The target recognition model includes five preset categories: passengers, pets, vehicles, fallen leaves, and interference. Specifically, a target recognition model is pre-constructed, comprising five preset categories: passengers, pets, vehicles, fallen leaves, and interference. The core of the model is an improved DS evidence theory algorithm, specifically optimized to address the counterintuitive results that traditional DS evidence theory produces when there is significant evidence conflict. First, a basic probability allocation function is defined, assigning initial confidence levels to each sensor based on its characteristics and advantages. For example, the initial confidence level for the passive infrared sensor is 0.6 for passengers, 0.2 for pets, 0.1 for vehicles, 0.05 for fallen leaves, and 0.05 for interference; the initial confidence level for the millimeter-wave radar sensor is 0.7 for passengers, 0.1 for pets, 0.15 for vehicles, 0 for fallen leaves, and 0.05 for interference; the initial confidence level for the ultrasonic ranging sensor is 0.5 for passengers, 0.1 for pets, 0.2 for vehicles, 0.1 for fallen leaves, and 0.1 for interference; the initial confidence level for the ambient light sensor is 0.9 for interference and 0 for all other categories. The uncertainty for all sensors is 0. Based on this, an improved evidence combination rule was designed. When the calculated evidence conflict coefficient K is less than 0.6, the traditional DS combination rule is used for fusion. When the evidence conflict coefficient K is greater than or equal to 0.6, the weighted average combination rule is used, in which the weight of the millimeter-wave radar sensor is 0.4, the weight of the passive infrared sensor is 0.3, the weight of the ultrasonic ranging sensor is 0.2, and the weight of the ambient light sensor is 0.1. This effectively solves the problem of identification errors caused by multi-sensor evidence conflict.

[0034] Step S203: Based on multimodal sensing data, extract the infrared radiation change features of the passive infrared sensor, the Doppler frequency shift features, micro-motion features, distance features and angle features of the millimeter-wave radar sensor, the precise distance features of the ultrasonic ranging sensor, and the illumination intensity features of the ambient light sensor, respectively, input them into the target recognition model for weighted fusion, and calculate the comprehensive confidence level of each target category; Specifically, feature vectors are extracted from the raw data of each sensor. Specifically, the passive infrared sensor data yields two features: the amplitude and frequency of infrared radiation variation. The millimeter-wave radar sensor data yields four features: Doppler frequency shift, micro-motion characteristics, target distance, and horizontal angle. The ultrasonic ranging sensor data yields the target's precise distance feature. The ambient light sensor data yields the rate of change of light intensity feature, ultimately forming a 10-dimensional multi-source feature vector. These feature vectors are then input into the corresponding basic probability allocation functions to calculate the preliminary confidence level of each sensor for the five target categories. Then, the improved evidence combination rule defined in step 2 is used to perform weighted fusion processing on the preliminary confidence levels output by the four sensors, finally calculating the comprehensive confidence level for each target category. The comprehensive confidence level ranges from 0 to 1, with a higher value indicating a greater probability that the target belongs to that category.

[0035] Step S204: When the highest overall confidence level is greater than the first preset confidence level threshold and the difference between the highest overall confidence level and the second highest overall confidence level is greater than the preset difference threshold, the target is determined to be the category corresponding to the highest overall confidence level; otherwise, it is determined to be an uncertain target, and the next frame of multimodal perception data is waited for re-determination. Specifically, target classification is performed based on the calculated overall confidence level. When the highest overall confidence level is greater than the first preset confidence threshold (typically 0.7), and the difference between the highest and second highest overall confidence levels is greater than the preset difference threshold (typically 0.2), the system classifies the target as belonging to the category corresponding to the highest overall confidence level. If the above conditions are not met, the target is classified as uncertain, and the system will wait for the next frame of multimodal sensing data for re-judgment to avoid classification errors caused by single data errors. If three consecutive frames of data are classified as uncertain targets, the system will automatically terminate the target detection process, return to the passive infrared sensor target screening state, and the microcontroller will re-enter the low-power stop mode to prevent system resource waste and program deadlock caused by continuous uncertainty, ensuring system stability and low-power characteristics.

[0036] Step S205: If the target is determined to be an uncertain target for a consecutive preset number of frames, the target detection process is terminated and the passive infrared sensor target screening state is returned. During the weighted fusion process, instantaneous interference is eliminated by time window filtering algorithm and cross interference of multiple targets is eliminated by spatial clustering algorithm.

[0037] Throughout the feature fusion calculation process, the system simultaneously executes two interference suppression algorithms to further improve recognition accuracy. The first is a time window filtering algorithm, which sets a 300-millisecond time window to retain only target features that persist within that time window, effectively eliminating transient interference signals such as lightning, flashing car lights, and falling leaves. The second is a spatial clustering algorithm, which uses the DBSCAN algorithm to cluster the point cloud data output by the millimeter-wave radar, merging point clouds within a distance of less than 0.5 meters into a single target. This eliminates feature confusion caused by the overlapping movement of multiple targets, ensuring the accuracy of target feature extraction.

[0038] In one embodiment of the present invention, the specific steps for generating intent information by recognizing the waiting intent of targets classified as passengers are as follows: Step S301: Continuously collect multimodal perception data of targets classified as passengers for a preset number of frames, extract position data, velocity data, acceleration data and orientation angle data, and reassemble the data using a sliding window method after normalization processing; Specifically, when the output target classification result is "passenger", the system immediately initiates a time-series data acquisition process, continuously acquiring multimodal perception data of targets classified as passengers for a preset number of frames. The data acquisition source consists of a calibrated millimeter-wave radar sensor and an ultrasonic ranging sensor, with an acquisition frequency of 20 Hz and a typical preset frame count of 10 frames, corresponding to an observation duration of 500 milliseconds. This duration ensures sufficient motion features are captured without causing wake-up delay due to excessive observation time. From the acquired multimodal perception data, the system extracts the target's position, velocity, acceleration, and orientation angle data. The position data is the target's two-dimensional coordinates relative to the bus stop sign; the velocity data is the target's instantaneous velocity; the acceleration data is the target's instantaneous acceleration; and the orientation angle data is the angle between the target's motion direction and the normal to the front of the bus stop sign. All extracted data undergoes min-max normalization to map data of different dimensions to the 0-1 range, eliminating the influence of dimensional differences on the model's inference results. After normalization, a sliding window method is used to reorganize the data. The sliding window size is 10 frames and the step size is 2 frames. The overlapping window processing enhances the temporal continuity of the data and improves the stability of subsequent model inference.

[0039] Step S302: Input the reconstructed data into the waiting intention recognition model pre-deployed on the low-power microcontroller for inference, and output the confidence values ​​corresponding to the three intentions of passengers waiting for a bus, passing by, and resting. Specifically, the reconstructed time-series data is input into a quantized unidirectional long short-term memory network (Uni-LSTM) waiting intention recognition model pre-deployed on a low-power microcontroller for inference. This model uses INT8 quantization technology for lightweight processing, with a model size of no more than 300 kilobytes and a single-frame inference time of no more than 50 milliseconds, fully meeting the real-time operation requirements of the low-power microcontroller. The model architecture consists of five layers. The first layer is the input layer, with an input dimension of 10×4, corresponding to 10 frames of temporal data and 4 feature dimensions. The second layer is a unidirectional long short-term memory network layer, containing 64 hidden units. It extracts temporal features based only on current and historical frame data, accurately identifying the target's motion trend without relying on future data. The third layer is a global average pooling layer, which reduces the dimensionality of the output of the unidirectional long short-term memory network layer, reducing the number of model parameters and computational cost. The fourth layer is a fully connected layer, containing 32 neurons, which uses the ReLU activation function for non-linear transformation. The fifth layer is the output layer, which uses the Softmax activation function to output the confidence values ​​corresponding to three intentions: waiting for a bus, passing by, and resting. The sum of the three confidence values ​​is 1. The reasoning process is as follows: the recombined time series data is first input into the input layer for format conversion, then enters the unidirectional long short-term memory network layer for forward reasoning to extract time series features, the global average pooling layer compresses the extracted features, the fully connected layer performs non-linear mapping on the compressed features, and finally the output layer converts the mapping result into the probability distribution of three intentions, i.e. the corresponding confidence values, through the Softmax function.

[0040] Step S303: Generate corresponding intent instructions based on confidence values. When the confidence of the waiting intent is greater than the second preset confidence threshold, generate a waiting passenger intent instruction. When the confidence of the passing intent is greater than the third preset confidence threshold and the target is getting farther and farther away from the bus stop, generate a passing pedestrian intent instruction. When neither of the above two conditions is met, generate a resting person intent instruction. Specifically, the system generates corresponding intent commands based on three intent confidence values ​​output by the model. When the confidence of the waiting intent is greater than the second preset confidence threshold (typically 0.8), it indicates that the target has a clear waiting behavior, and a waiting passenger intent command is generated. This command will trigger the system to execute subsequent pre-wake-up or normal operation procedures. When the confidence of the passing intent is greater than the third preset confidence threshold (typically 0.7) and the target is getting farther and farther away from the bus stop, it indicates that the target is just passing through the bus stop area without a waiting intent, and a passing pedestrian intent command is generated. The system will maintain its current low-power state and will not perform a wake-up operation. When neither of the above two conditions is met, it indicates that the target is lingering near the bus stop but has no clear waiting behavior, and is judged to be a resting person. A resting person intent command is generated, and the system only wakes up the low-power basic display module to display basic route information, without illuminating the high-power LCD display module.

[0041] Step S304: Integrate the intent confidence value and intent command into waiting intent information.

[0042] Specifically, the intent confidence value output by the model and the generated intent commands are integrated into structured waiting intent information, which includes three parts: intent type field, confidence value field, and timestamp field. The intent type field identifies the passenger's specific intent, with values ​​for waiting passenger, passerby, or resting person; the confidence value field stores the confidence value of the corresponding intent; and the timestamp field records the time when intent recognition was completed, used for time synchronization during subsequent fusion decision-making. The integrated waiting intent information is sent to the fuzzy comprehensive decision-maker via an internal bus, serving as one of the core inputs for generating system operation status control commands.

[0043] In one embodiment of the present invention, the specific steps for obtaining the bus arrival prediction and passenger flow density prediction data are as follows: Step S305: Receive vehicle GPS data sent by the bus dispatch center in real time, combine the vehicle GPS data with historical road condition data and average station dwell time data, and calculate the predicted bus arrival time data of the vehicle at the station through the pre-trained gradient boosting tree bus arrival prediction model. Specifically, the system receives real-time GPS data from the bus dispatch center via a 4G / 5G communication module. This data includes the vehicle's unique identifier, real-time latitude and longitude coordinates, instantaneous speed, data collection timestamp, and the vehicle's route number. Upon receiving the raw data, preprocessing is performed. A Kalman filter algorithm is used to correct GPS drift errors, and abnormal data exceeding the route's speed limit or deviating from the bus route's position are removed. Then, the vehicle's latitude and longitude coordinates are matched onto the electronic map of the bus route, and the remaining physical distance from the stop is calculated. The calculated remaining distance, current vehicle speed, historical traffic data for the corresponding road segment, average dwell time at the stop, current weekday, and whether it is a holiday are integrated into a 12-dimensional input feature vector, which is then input into a pre-trained gradient boosting tree bus arrival prediction model for calculation. This model employs an ensemble learning architecture, consisting of 80 regression trees of depth 5, with a learning rate of 0.1 and squared error as the loss function. The calculation process is as follows: The first regression tree, based on the input feature vector, initially predicts the vehicle's arrival time. Each subsequent regression tree learns the residual between the previous tree's prediction and the actual arrival time, iterating to reduce the prediction error. Finally, the predictions from all regression trees are weighted and summed to obtain the final predicted bus arrival time for the current station, with a prediction error not exceeding 30 seconds. The model employs a daily offline update mechanism, incrementally training the model every morning using the previous day's actual arrival data to ensure that prediction accuracy continuously improves with the accumulation of operational data.

[0044] Step S306: Retrieve historical passenger flow data for a preset historical period at this station, input it into the pre-trained passenger flow time series prediction model, and calculate the passenger flow change trend and passenger flow density prediction data for this station within the preset passenger flow prediction period.

[0045] Specifically, the system retrieves historical passenger flow data for a preset historical period from local non-volatile memory. The typical preset historical period is 30 days, with data granularity at 15-minute intervals. Each data entry includes information such as the number of waiting passengers, date, day of the week, and whether it is a holiday. After retrieval, the historical passenger flow data is preprocessed. Linear interpolation is used to fill in missing data, and then min-max normalization is used to map passenger flow numbers to the 0-1 range, eliminating the impact of numerical range differences on the model's prediction results. The preprocessed historical passenger flow data is then input into a pre-trained passenger flow time series prediction model. This model employs a bidirectional long short-term memory network architecture, capable of simultaneously capturing both long-term trends and short-term fluctuations in passenger flow data. The model input dimension is 96×1, corresponding to 96 15-minute granular passenger flow data entries from the past 24 hours; the output dimension is 4×1, corresponding to passenger flow density prediction data for four 15-minute intervals within the next hour. The calculation process is as follows: the input historical passenger flow sequence first enters a bidirectional long short-term memory network layer, extracting temporal features from both the forward and backward directions; then, a fully connected layer performs a nonlinear transformation on the extracted features; finally, it outputs the passenger flow trend and passenger flow density prediction data for the station within a preset passenger flow prediction period. The passenger flow density prediction data is displayed at 15-minute intervals, reflecting the number of waiting passengers at different times. The model adopts a weekly offline update mechanism, retraining the model every Sunday using the actual passenger flow data from the previous week to adapt to the seasonal changes in passenger flow patterns.

[0046] In one embodiment of the present invention, the specific steps of the fusion generation system operation state control command are as follows: Step S307: Based on waiting intention information, bus arrival prediction time data, and passenger flow density prediction data, perform fuzzy inference through a preset fuzzy rule base, and output system operation status level instructions. The system operation status includes four levels: deep standby state, shallow standby state, pre-wake-up state, and normal operation state.

[0047] Specifically, the waiting intention information, bus arrival prediction time data, and passenger flow density prediction data are all fuzzified, converting precise numerical values ​​into corresponding fuzzy linguistic values. First, the fuzzy set partitioning and membership function for each variable are defined: the waiting intention confidence is divided into three fuzzy sets: low, medium, and high, using a triangular membership function, where low corresponds to the interval 0 to 0.4, medium to 0.3 to 0.7, and high to 0.6 to 1; the bus arrival prediction time is divided into three fuzzy sets: far, medium, and near, using a trapezoidal membership function, where far corresponds to the interval 15 to 30 minutes, medium to 5 to 15 minutes, and near to 0 to 5 minutes; the passenger flow density prediction data is divided into three fuzzy sets: low, medium, and high, using a triangular membership function, where low corresponds to the interval 0 to 0.5 people / square meter, medium to 0.4 to 1.2 people / square meter, and high to 1.0 to 2 people / square meter. For each precise input value, the degree to which the value belongs to each fuzzy language value is obtained by calculating its membership degree on the corresponding fuzzy set, thus completing the fuzzification conversion.

[0048] The fuzzy comprehensive decision-maker infers from the fuzzified input data using a pre-set fuzzy rule base. This rule base contains 27 complete rules, covering all possible combinations of the three input variables. The rule design follows the principle of "waiting intention first, arrival time second, and passenger flow density as an auxiliary factor." Specific rules are as follows: When the confidence level of the waiting intention is high, if the predicted bus arrival time is near, regardless of passenger flow density, a normal operation status command is output; if the predicted bus arrival time is medium, and the passenger flow density is low or medium, a pre-wake-up status command is output; if the passenger flow density is high, a normal operation status command is output; if the predicted bus arrival time is far, and the passenger flow density is low or medium, a shallow standby status command is output; if the passenger flow density is high, a pre-wake-up status command is output. When the confidence level of the waiting intention is medium, if the predicted bus arrival time is near and the passenger flow density is low or medium, a pre-wake-up state level command is output; if the passenger flow density is high, a normal operation state level command is output. If the predicted bus arrival time is medium, a shallow standby state level command is output regardless of passenger flow density. If the predicted bus arrival time is far, a deep standby state level command is output; if the passenger flow density is high, a shallow standby state level command is output. When the confidence level of the waiting intention is low, if the predicted bus arrival time is near and the passenger flow density is low, a shallow standby state level command is output; if the passenger flow density is medium or high, a pre-wake-up state level command is output. If the predicted bus arrival time is medium and the passenger flow density is low, a deep standby state level command is output; if the passenger flow density is medium or high, a shallow standby state level command is output. If the predicted bus arrival time is far, a deep standby state level command is output regardless of passenger flow density. Each rule is represented in the form of "IF-THEN", for example, "IF waiting intention confidence is high AND bus arrival prediction time is near AND passenger flow density is low THEN output normal operation status level instruction". The reasoning process adopts the Mamdani reasoning method, which merges the conclusions of rules that meet the conditions to obtain the fuzzy reasoning result.

[0049] The centroid method is used to defuzzify the fuzzy results obtained from fuzzy inference, converting the fuzzy sets into precise numerical outputs. First, the centroid coordinates of the region enclosed by the membership function curve of the fuzzy inference result and the horizontal axis are calculated; the horizontal coordinate value corresponding to this centroid is the precise defuzzified value. This precise value is then mapped to four system operating state levels: values ​​less than 0.5 correspond to deep standby, 0.5 to 1.5 to shallow standby, 1.5 to 2.5 to pre-wake-up, and greater than 2.5 to normal operation. Based on the mapping results, corresponding system operating state level commands are generated and sent to the energy management module and display control module via the internal bus. These two modules then execute the corresponding power control and display control operations, respectively.

[0050] In one embodiment of the present invention, the specific steps for performing dynamic control of the display partition are as follows: Step S401: Divide the content to be displayed into three priority categories: core information, important information, and secondary information; Specifically, the system first prioritizes all content to be displayed, categorizing it into three levels based on its importance and timeliness: core information, important information, and secondary information. Core information consists of real-time information that passengers are most concerned about, including countdown timers for bus arrivals on the current route, real-time vehicle locations, and temporary route change notices. This type of information must be clearly visible in any environment. Important information includes basic service information that passengers need to know, such as route and station lists, first and last bus times, and fare information. The brightness of this type of information can be appropriately reduced in low-power mode. Secondary information is auxiliary information, including public service advertisements, weather forecasts, and convenience service notices. This type of information can be completely turned off when the battery is low or there are no passengers. Content priority is automatically assigned by the system based on information type and can also be adjusted remotely through a cloud-based operations and maintenance platform.

[0051] Step S402: Divide the display module into grid-like partitions and assign corresponding display content to each display partition; Specifically, the display module is divided into grid-like partitions, dividing the entire display screen into several independent and controllable rectangular grid units. The size of each grid unit is determined based on the resolution of the display module and the minimum font size of the displayed content, typically 32×32 pixels. Each grid unit has an independent power control circuit and brightness adjustment circuit, which can be turned on, off, or adjusted independently without affecting each other. After partitioning, the system allocates the content to the corresponding grid units according to the position and size of the content to be displayed. One display content may cover multiple adjacent grid units, and only one priority content can be displayed in the same grid unit at a time. When multiple contents overlap, the content with higher priority is displayed first. The display module is a liquid crystal display panel, an active light-emitting diode (OLED) panel, or an LED dot matrix screen with full array local dimming (FALD) function.

[0052] Step S403: Calculate the brightness value of each display zone based on the display content priority, ambient light intensity, viewing distance, and remaining battery power; Specifically, the system collects ambient light intensity, average passenger viewing distance, and remaining battery power data in real time. Combining this with the content priority of each display zone, it calculates the target brightness value for each grid cell using a preset brightness calculation formula. The brightness calculation follows the principle of "higher priority, higher brightness; brighter environment, higher brightness; farther distance, higher brightness; lower battery, lower brightness," achieving optimal energy consumption while ensuring information readability. The calculated brightness value is limited to a range of 0 to 255, where 0 represents completely off and 255 represents maximum brightness, conforming to the brightness adjustment standards of mainstream display modules.

[0053] Step S404: Control the display status of each display zone according to the calculated brightness value. The zone with no information is completely powered off, and the zone with information is lit according to the corresponding brightness value. Specifically, based on the calculated target brightness value for each grid cell, the display state of each zone is controlled by the PWM (Pulse Width Modulation) circuit of the display control module. For grid cells with a target brightness value of 0, the system directly cuts off their power supply, completely shutting down the zone, reducing its energy consumption to zero. For grid cells with a target brightness value greater than 0, the system adjusts the duty cycle of the PWM signal according to the target brightness value to control the backlight brightness of the zone, achieving the target brightness. The system updates the brightness values ​​of all zones every 500 milliseconds, enabling dynamic adjustment of the display state and ensuring timely response to environmental changes or content updates.

[0054] The formula for calculating the brightness value of each display zone is as follows: , This formula is a multi-factor product model. It uses a base brightness value as a baseline, multiplying it by a content priority correction factor, a viewing distance correction factor, and a remaining battery power correction factor to obtain the final zone brightness value. The `clamp` function is used to limit the result to between 0 and 255, avoiding invalid values ​​that exceed the display module's brightness adjustment range.

[0055] in, Indicates the first Line 1 The column displays the brightness value of the partition, ranging from 0 to 255 integers. This parameter is the output of the formula and is directly used to control the backlight brightness of the corresponding grid cell.

[0056] This represents the base brightness value, which is the display brightness at a standard viewing distance (3 meters), under standard content priority (secondary information), and when fully charged. The unit is nits. This parameter is determined by the ambient light intensity and is calculated using the base brightness value formula.

[0057] This indicates the priority of the content displayed in this partition. The P value is 3 for core information, 2 for important information, and 1 for minor information. This indicates the average viewing distance between passengers and the display module, measured in meters. This parameter is measured in real time by a millimeter-wave radar sensor. The system counts all currently detected distances between passengers and the bus stop sign and calculates the average as... The value is updated at a frequency of 1 Hz. When no passenger is detected, The default value is 3 meters.

[0058] This represents the battery remaining capacity correction factor. This means that the calculation result will be limited to between 0 and 255; The base brightness value is calculated using the following formula: , This formula uses a logarithmic model to establish the relationship between ambient light intensity and baseline brightness, which aligns with the human eye's perception of brightness. Since the human eye perceives brightness in a logarithmic rather than a linear manner, using a logarithmic model ensures comfortable viewing and readability of displayed content under varying ambient light intensities, preventing poor visibility in strong light or excessive glare in low light.

[0059] in, Ambient light intensity is expressed in lux. and The parameters represent the pre-calibration coefficients; k is the brightness gain coefficient, a pre-calibrated dimensionless constant with a typical value of 20. This parameter adjusts the slope of the base brightness value as a function of ambient light intensity, obtained through experimental calibration to ensure clear visibility of content under strong light. b is the brightness offset coefficient, a pre-calibrated dimensionless constant with a typical value of 50. This parameter sets the minimum base brightness value to ensure visibility of content in complete darkness while avoiding excessive brightness that could cause light pollution.

[0060] The specific value of the battery remaining power correction coefficient is as follows: When the remaining battery power is lower than the first preset power threshold but not lower than the second preset power threshold The value is set to the first correction coefficient; when the remaining battery power is lower than the second preset power threshold... The value is set to the second correction coefficient; when the remaining battery power is not lower than the first preset power threshold... The value is set to the third correction coefficient, where the first preset battery level threshold is greater than the second preset battery level threshold. The typical values ​​for the first preset battery level threshold are 20%, the second preset battery level threshold is 10%, the first correction coefficient is typically 0.5, the second correction coefficient is typically 0.3, and the third correction coefficient is typically 1.0. The remaining battery power is estimated in real-time using an extended Kalman filter battery state estimation algorithm.

[0061] In one embodiment of the present invention, the specific steps for performing the weather forecast-based intelligent energy dispatch are as follows: Step S405: Connect to the third-party weather forecast application interface to obtain weather forecast data for a set future duration; Specifically, the system connects to a third-party publicly available weather forecast application interface via a 4G / 5G communication module to obtain weather forecast data for a set future duration, typically 72 hours, with data updates every 6 hours. The acquired weather forecast data includes hourly light intensity data, ambient temperature data, and precipitation probability data. Light intensity data is in lux, temperature data is in degrees Celsius, and precipitation probability data is a decimal between 0 and 1. Upon receiving the raw weather forecast data, the system first performs data validity verification, removing abnormal data exceeding reasonable ranges. For temporarily missing data, linear interpolation is used to fill in the gaps, ensuring data continuity and completeness. After verification, the data is organized into a time-series data format according to chronological order and stored in the system's local memory for subsequent use by energy prediction models.

[0062] Step S406: Input the weather forecast data into the pre-trained network energy prediction model to calculate the solar power generation and system energy consumption for the future set time period, and generate a dynamic charging and discharging plan; Specifically, the pre-processed weather forecast data is input into a pre-trained network energy prediction model to calculate hourly solar power generation and system energy consumption within a set future timeframe. This model employs a bidirectional long short-term memory network architecture and is a deep learning model specifically designed for energy time series forecasting tasks. It can simultaneously capture both long-term trends and short-term fluctuations in energy data, achieving significantly higher prediction accuracy than traditional statistical forecasting methods.

[0063] The network energy prediction model architecture consists of six layers. The first layer is the input layer, with an input dimension of 72×3, corresponding to three features: light intensity, temperature, and precipitation probability for the next 72 hours. The second layer is a bidirectional long short-term memory network layer, containing 128 hidden units, capable of processing time-series data from both forward and backward directions simultaneously to extract temporal dependencies. The third layer is an attention mechanism layer, which can automatically learn the importance weights of features at different time steps, focusing on key time points that have a significant impact on energy prediction. The fourth layer is a fully connected layer, containing 64 neurons, using the ReLU activation function for nonlinear transformation to map the extracted time-series features to a high-dimensional space. The fifth layer is a dropout layer, randomly discarding 20% ​​of the neurons to prevent the model from overfitting. The sixth layer is the output layer, containing two neurons, outputting the solar power generation and system energy consumption at the corresponding time step, respectively, in watt-hours.

[0064] The network energy prediction model training process is completed on a cloud server. The training dataset includes historical solar power generation data, historical system energy consumption data, and historical weather forecast data for corresponding periods over a preset historical period, typically 12 months. During training, the Adam optimizer is used with a learning rate of 0.001 and a mean squared error loss function. The model parameters are continuously adjusted using backpropagation until the model's loss on the validation set no longer decreases. After training, the model is converted to INT8 quantization format and deployed to the local low-power microcontroller of the bus stop electronic display, enabling local real-time inference without relying on cloud computing, thus reducing communication power consumption and latency.

[0065] The reasoning process is as follows: the time series data of the weather forecast for the future set duration is input into the input layer, the time series features are extracted through the bidirectional long short-term memory network layer, the attention mechanism layer performs weighted processing on the features, the fully connected layer performs nonlinear mapping, and finally the output layer outputs the predicted values ​​of solar power generation and system energy consumption for the future set duration hour by hour.

[0066] Based on the solar power generation and system energy consumption forecast data output by the network energy prediction model, and combined with the current remaining capacity of the lithium battery pack, a dynamic charge-discharge plan for a predetermined duration is generated. The charge-discharge plan is calculated in hourly units, with each time unit including four parameters: solar power output, lithium battery charging power, lithium battery discharging power, and mains power output. The plan generation follows the principle of "solar priority, lithium battery buffer, and mains backup." When there is sufficient sunshine and solar power generation exceeds system energy consumption, the excess power is used to charge the lithium battery pack; when solar power generation is less than system energy consumption, the lithium battery pack is used to discharge and supplement the shortfall; when the remaining capacity of the lithium battery pack falls below a preset lower threshold, the system switches to mains power. Simultaneously, the plan considers the charge-discharge efficiency and cycle life of the lithium battery to avoid overcharging and over-discharging, thus extending the battery's lifespan. The system updates the charge-discharge plan every 6 hours based on the latest weather forecast data and lithium battery status to ensure the plan's accuracy and adaptability.

[0067] Step S407: Control the switching between solar power supply, mains power supply and lithium battery energy storage according to the dynamic charging and discharging plan. When there is sufficient sunshine, solar power supply is used first, and excess power is stored in the lithium battery pack. When the cumulative sunshine time within the predicted set time period is less than the preset cumulative sunshine threshold, reduce the system power consumption level in advance, turn off the voice broadcast module and reduce the maximum brightness of the display module. Specifically, according to the generated dynamic charging and discharging plan, the energy management module controls the seamless switching between solar power supply, mains power supply, and lithium battery energy storage through relays and power conversion circuits. When there is sufficient sunlight, the energy management module prioritizes connecting the solar power supply circuit to power the system load, while simultaneously connecting the lithium battery charging circuit to store excess solar power in the lithium battery pack. When solar power generation is insufficient, the energy management module connects the lithium battery discharging circuit, with the lithium battery pack supplementing the power supply. When the remaining charge of the lithium battery pack falls below a preset lower threshold, the energy management module automatically switches to the mains power supply circuit to ensure uninterrupted system operation. When the predicted cumulative sunshine duration within a set period is less than the preset cumulative sunshine threshold, the system reduces its overall power consumption level in advance, shuts down the voice broadcast module, and reduces the maximum brightness of the display module to 50% of its normal brightness, further extending the system's battery life during continuous cloudy and rainy weather.

[0068] Step S408: Monitor the voltage, current and temperature data of the lithium battery pack in real time, accurately estimate the remaining capacity and health status of the lithium battery pack using the extended Kalman filter battery state estimation algorithm, and use dynamic equalization charging technology to balance the voltage of each battery cell.

[0069] Specifically, the system monitors the voltage, current, and temperature data of the lithium battery pack in real time at a frequency of 1 Hz. The collected data is processed using an extended Kalman filter (EPF) battery state estimation algorithm to accurately estimate the remaining capacity and health status of the lithium battery pack. The EPF algorithm effectively overcomes the nonlinear characteristics of lithium batteries and the influence of measurement noise, controlling the remaining capacity estimation error to within 3%. Simultaneously, the system employs dynamic equalization charging technology, monitoring the voltage of each individual battery cell in real time. When the voltage difference between battery cells exceeds a preset threshold, the equalization circuit is automatically activated to supplement the charging of cells with lower voltage and discharge the charging of cells with higher voltage, balancing the voltage of each battery cell and preventing overcharging or over-discharging of individual cells, thus extending the overall lifespan of the lithium battery pack.

[0070] In one embodiment of the present invention, a regional collaborative energy-saving control step is further included, as follows: Step S501: Adjacent electronic bus stop signs on the same road segment form an ad hoc network through low-power wide area network communication technology to establish a point-to-point data transmission channel between the stops; Specifically, adjacent electronic bus stop signs on the same road segment automatically form a distributed self-organizing network using low-power wide-area network (LPWAN) communication technology, eliminating the need for manual network parameter configuration. After system startup, each stop sign periodically sends a broadcast beacon containing its own stop number, geographical location, and communication address information. Upon receiving the beacon, adjacent stops automatically establish neighbor relationships, forming a network topology based on road segments. Based on these neighbor relationships, stops communicate directly through a point-to-point data transmission channel, without requiring a cloud server intermediary. The communication distance can reach up to 3 kilometers, with a communication frequency of once every 10 seconds, a single data transmission volume of less than 100 bytes, and overall communication power consumption below 0.05W. This communication channel is specifically used for transmitting real-time data related to regional collaborative control and is independent of the 4G / 5G channel used for cloud communication, ensuring the real-time performance and reliability of collaborative control.

[0071] Step S502: The upstream station collects real-time data on the number of waiting passengers and the departure data of vehicles at its station, and sends it to all downstream stations on the same road segment; Specifically, upstream stations collect two types of core data in real time and send them to all downstream stations on the same route. The first type is passenger count data, which comes from target identification results. The system counts the number of all currently detected targets classified as passengers, updating every 5 seconds. The second type is vehicle departure data. When the bus dispatch center sends GPS data showing that a vehicle has left the station and is heading towards a downstream station, the system immediately generates vehicle departure data, including the vehicle number, the route it belongs to, and the departure timestamp. When the number of waiting passengers is greater than or equal to a preset threshold (typically 3 people), the upstream station immediately sends the passenger count data to all downstream stations; when a vehicle departure is detected, the vehicle departure data is immediately sent to all downstream stations, ensuring that downstream stations can obtain upstream situational information in a timely manner.

[0072] Step S503: When the downstream station receives data from the upstream station indicating that the number of waiting passengers is greater than or equal to a preset number threshold, it enters the pre-wake state with a preset pre-wake time; when the downstream station receives data from the upstream station indicating that a vehicle has left, it prepares to display the real-time arrival information of the vehicle in advance. Specifically, after receiving data from the upstream station, the downstream station executes corresponding collaborative control operations based on the data type. When it receives data indicating that the number of waiting passengers is greater than or equal to a preset threshold, the downstream station pre-wakes up for a pre-set time (typically 30 seconds), switches the display module from deep standby to shallow standby, pre-initializes the display driver circuit, and loads basic route information to ensure that the screen can be lit up immediately to display complete information when passengers arrive, eliminating wake-up delay. When it receives vehicle departure data from the upstream station, the downstream station immediately calculates the estimated arrival time of the vehicle based on its speed and distance, prepares to display the vehicle's real-time arrival information, and automatically switches to normal operation when the vehicle is about to arrive, ensuring the real-time nature and accuracy of the arrival information.

[0073] Step S504: When a power grid power shortage instruction is received from the power grid dispatch center or cloud operation and maintenance platform, all electronic bus stop signs in the area will automatically adjust their wake-up strategies according to their respective passenger flow priorities, giving priority to ensuring the normal operation of the preset core stations. Specifically, when a power grid shortage instruction is received from the power grid dispatch center or cloud-based operation and maintenance platform, all electronic bus stop signs in the area automatically enter a collaborative energy-saving mode. The system classifies stops within the area according to preset passenger flow priority rules. Passenger flow priority is determined based on the stop's average daily passenger volume, geographical importance, and route coverage, and is divided into three levels: core stops, important stops, and ordinary stops. Each stop automatically adjusts its wake-up strategy based on its own passenger flow priority: core stops maintain normal operation and prioritize power supply; important stops reduce wake-up sensitivity and extend the minimum wake-up interval to 2 minutes; ordinary stops are only woken up when a bus is about to arrive, remaining in deep standby mode at other times. Through this tiered dispatching method, while ensuring service quality in the core area, the overall energy consumption of the area is minimized, responding to the power grid's peak-shaving needs.

[0074] Step S505: The cloud-based operation and maintenance platform collects energy consumption data, sensor data, and battery data of each bus electronic stop sign in the area in real time. Through machine learning fault prediction algorithms, it analyzes the data and predicts problems such as dust accumulation on solar panels, sensor aging, and battery failure in advance, generates maintenance work orders, and sends them to the operation and maintenance personnel.

[0075] Specifically, the cloud-based operation and maintenance platform collects real-time operational data from various electronic bus stop signs within the area via 4G / 5G communication channels. This includes hourly energy consumption data for each stop, operational status data for each sensor, and voltage, current, and temperature data for the lithium battery packs. The platform inputs the collected historical data into a pre-trained machine learning fault prediction model. By analyzing abnormal trends in the data, it predicts potential equipment failures in advance. For example, when the power generation of the solar panel is consistently more than 30% lower than the average for the same period, a dust accumulation fault on the solar panel is predicted; when the detection accuracy of the sensor continues to decline, a sensor aging fault is predicted; and when the charging and discharging efficiency of the lithium battery continues to decrease, a battery performance degradation fault is predicted. After a fault is predicted, the platform automatically generates a maintenance work order containing the location of the faulty site, the fault type, and a suggested repair plan. This order is then sent to the operation and maintenance personnel via SMS or mobile application, realizing a shift from reactive maintenance to proactive predictive maintenance.

[0076] This invention fundamentally solves the core pain points of traditional electronic bus stop signs, such as high energy consumption, high false trigger rate, incompatibility between wake-up delay and energy saving effect, poor adaptability to extreme environments, and high operation and maintenance costs, by constructing a full-link closed-loop energy control system that includes "multi-sensor cross-calibration, multi-modal feature fusion target recognition, accurate recognition of waiting intention, dual-dimensional prediction of bus arrival and passenger flow, fuzzy comprehensive gradient decision-making, refined control of display partitions, predictive energy scheduling, regional collaborative energy saving, and cloud-based intelligent operation and maintenance". Multi-sensor cross-parameter calibration technology significantly improves detection accuracy under extreme temperatures. Combined with improved evidence theory-based multimodal fusion target recognition and lightweight waiting intention recognition technology, the overall false trigger rate of the system is greatly reduced. The adoption of a four-level gradient wake-up mechanism and gridded display partition dynamic control technology effectively reduces the energy consumption of the display module while achieving a seamless user experience. Through weather forecast-based predictive energy scheduling and lithium battery life cycle health management, the utilization rate of solar energy is significantly improved, effectively extending the lifespan of lithium batteries and the driving range during continuous cloudy and rainy weather. Furthermore, the introduction of regional self-organizing network collaborative energy saving and cloud-based predictive operation and maintenance technology realizes the upgrade from single-site refined control to regional-level energy consumption optimization, significantly reducing equipment failure rate and long-term operation and maintenance costs. It has engineering value for large-scale promotion in urban public transportation systems nationwide and significant socio-economic benefits.

[0077] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A low-power standby and human body induction wake-up method for energy-saving electronic bus stop signs, characterized in that, Includes the following steps: After the system is powered on, it executes the initialization process, reads basic operating data, completes the cross-parameter calibration of the multi-sensor array, and generates a calibration parameter set. The target is initially screened by a passive infrared sensor using a set of calibration parameters. After the triggering conditions are met, multimodal sensing data is collected and the target classification result is output after feature-level fusion. Based on the target classification results, the waiting intention information of targets classified as passengers is generated by identifying the waiting intention of the target. Combined with bus arrival prediction and passenger flow density prediction data, the system operation status control instructions are generated. Control commands are sent to the display control module and the energy management module to execute dynamic control of display zones and intelligent energy dispatching based on weather forecasts, respectively.

2. The energy-saving method for low-power standby and human body induction wake-up of electronic bus stop signs according to claim 1, characterized in that, The basic operating data includes real-time clock data, ambient temperature data, light intensity data, and historical calibration data. The multi-sensor array includes a passive infrared sensor, a millimeter-wave radar sensor, an ultrasonic ranging sensor, and an ambient light sensor. The cross-parameter calibration is to perform distance calibration on the millimeter-wave radar sensor based on the ultrasonic ranging sensor, and to perform temperature compensation calibration on the passive infrared sensor based on the ambient temperature. The specific steps for temperature compensation calibration of the passive infrared sensor based on ambient temperature are as follows: The temperature compensation coefficient is calculated based on the current ambient temperature. The original output voltage of the passive infrared sensor is calibrated using the temperature compensation coefficient to obtain the calibrated output voltage. The detection threshold is then dynamically adjusted based on the current ambient temperature. The specific details of dynamically adjusting the detection threshold are as follows: When the ambient temperature is lower than the preset low temperature threshold, the detection threshold is set to the first preset multiple of the benchmark threshold; When the ambient temperature is higher than the preset high temperature threshold, the detection threshold is set to the second preset multiple of the benchmark threshold; When the ambient temperature is between the preset low temperature threshold and the preset high temperature threshold, the detection threshold is set as the baseline threshold.

3. The energy-saving method for low-power standby and human body induction wake-up of electronic bus stop signs according to claim 2, characterized in that, The specific steps for calibrating the millimeter-wave radar sensor using an ultrasonic ranging sensor as a reference are as follows: The system automatically triggers a millimeter-wave radar sensor distance calibration process every preset time interval; Detect whether there is a stationary target within a preset effective range. If so, continuously collect a preset number of raw distance data from the millimeter-wave radar sensor and the corresponding ultrasonic ranging sensor ranging data. The least squares method was used to perform linear regression calculations on the raw distance data from the millimeter-wave radar sensor and the ranging data from the ultrasonic ranging sensor to obtain the linear calibration coefficients. The calculated linear calibration coefficients are stored in the system memory to correct the original distance measurements of all subsequent millimeter-wave radar sensors.

4. The energy-saving method for low-power standby and human body induction wake-up of electronic bus stop signs according to claim 2, characterized in that, The specific steps for outputting the target classification result through feature-level fusion are as follows: The calibration parameter set is applied to the passive infrared sensor and the millimeter-wave radar sensor respectively. The passive infrared sensor is used for initial target screening. When the detected change in infrared radiation exceeds the detection threshold, the millimeter-wave radar sensor and the ultrasonic ranging sensor are activated to collect multimodal sensing data. A target recognition model based on an improved evidence theory algorithm is constructed, which includes five preset categories: passengers, pets, vehicles, fallen leaves, and interference. Based on multimodal sensing data, infrared radiation variation features of passive infrared sensors, Doppler frequency shift features, micro-motion features, distance features and angle features of millimeter-wave radar sensors, precise distance features of ultrasonic ranging sensors, and illumination intensity features of ambient light sensors are extracted and input into the target recognition model for weighted fusion to calculate the comprehensive confidence level of each target category. When the highest overall confidence level is greater than the first preset confidence level threshold and the difference between the highest overall confidence level and the second highest overall confidence level is greater than the preset difference threshold, the target is determined to be the category corresponding to the highest overall confidence level. Otherwise, it is determined to be an uncertain target, and the next frame of multimodal perception data is used for reassessment; If the target is determined to be an uncertain target for a consecutive preset number of frames, the target detection process is terminated and the system returns to the initial target screening state of the passive infrared sensor. During the weighted fusion process, instantaneous interference is eliminated by a time window filtering algorithm and cross-interference of multiple targets is eliminated by a spatial clustering algorithm.

5. The energy-saving method for low-power standby and human body induction wake-up of electronic bus stop signs according to claim 1, characterized in that, The specific steps for generating intent information by recognizing waiting intentions from targets classified as passengers are as follows: Multimodal perception data of targets classified as passengers are continuously collected for a preset number of frames. Position data, velocity data, acceleration data and orientation angle data are extracted, and after normalization, the data is recombined using a sliding window method. The reorganized data is input into a waiting intention recognition model pre-deployed on a low-power microcontroller for inference, and the confidence values ​​corresponding to the three intentions of passengers waiting for a bus, passing by, and resting are output. Based on the confidence value, a corresponding intention instruction is generated. When the confidence of the waiting intention is greater than the second preset confidence threshold, a waiting passenger intention instruction is generated. When the confidence of the passing intention is greater than the third preset confidence threshold and the target is getting farther and farther away from the bus stop, a passing pedestrian intention instruction is generated. When neither of the above two conditions is met, a resting person intention instruction is generated. The intent confidence value and intent command are integrated into waiting intent information.

6. The energy-saving method for low-power standby and human body induction wake-up of electronic bus stop signs according to claim 1, characterized in that, The specific steps for obtaining the bus arrival prediction and passenger flow density prediction data are as follows: The system receives vehicle GPS data from the bus dispatch center in real time, combines the vehicle GPS data with historical traffic data and average station dwell time data, and calculates the predicted bus arrival time data of the vehicle to the station through a pre-trained gradient boosting tree bus arrival prediction model. Retrieve historical passenger flow data for a preset historical period at this station, input it into a pre-trained passenger flow time series prediction model, and calculate the passenger flow trend and passenger flow density prediction data for this station within the preset passenger flow prediction period.

7. The energy-saving method for low-power standby and human body induction wake-up of electronic bus stop signs according to claim 1, characterized in that, The specific steps for the fusion generation system operation status control command are as follows: Based on waiting intention information, bus arrival prediction time data, and passenger flow density prediction data, fuzzy reasoning is performed through a preset fuzzy rule base to output system operation status level instructions. The system operation status includes four levels: deep standby state, shallow standby state, pre-wake-up state, and normal operation state.

8. The energy-saving method for low-power standby and human body induction wake-up of electronic bus stop signs according to claim 1, characterized in that, The specific steps for performing dynamic control of the display partition are as follows: The content to be displayed is divided into three priority categories: core information, important information, and secondary information. The display module is divided into grid-like partitions, and corresponding display content is assigned to each display partition; The brightness value of each display zone is calculated based on the priority of the displayed content, ambient light intensity, viewing distance, and remaining battery power. The display status of each display zone is controlled according to the calculated brightness value. Zones with no information are completely powered off, while zones with information are lit according to the corresponding brightness value. The formula for calculating the brightness value of each display zone is as follows: ， in, Indicates the first Line number The columns display the brightness values ​​of the zones. Indicates the base brightness value. This indicates the priority of the content displayed in this partition: a P value of 3 corresponds to core information, a P value of 2 corresponds to important information, and a P value of 1 corresponds to secondary information. This indicates the average viewing distance between the passenger and the display module, in meters. This represents the battery remaining capacity correction factor. This means that the calculation result will be limited to between 0 and 255; The base brightness value is calculated using the following formula: ， in, Ambient light intensity is expressed in lux. and Indicates the pre-calibration coefficient; The specific value of the battery remaining power correction coefficient is as follows: When the remaining battery power is lower than the first preset power threshold but not lower than the second preset power threshold The value is set to the first correction coefficient; when the remaining battery power is lower than the second preset power threshold... The value is set to the second correction coefficient; when the remaining battery power is not lower than the first preset power threshold... The value is the third correction coefficient, and the first preset power threshold is greater than the second preset power threshold.

9. The energy-saving method for low-power standby and human body induction wake-up of electronic bus stop signs according to claim 1, characterized in that, The specific steps for implementing the weather forecast-based intelligent energy dispatch are as follows: Connect to a third-party weather forecast application interface to obtain weather forecast data for a set time period in the future; By inputting weather forecast data into a pre-trained network energy prediction model, the solar power generation and system energy consumption for a set future time period are calculated, and a dynamic charging and discharging plan is generated. The system controls the switching between solar power, mains power, and lithium battery energy storage according to a dynamic charging and discharging plan. When there is sufficient sunshine, solar power is used first, and excess power is stored in the lithium battery pack. When the cumulative sunshine time within the predicted set period is less than the preset cumulative sunshine threshold, the system power consumption level is reduced in advance, the voice broadcast module is turned off, and the maximum brightness of the display module is reduced. The system monitors the voltage, current, and temperature data of the lithium battery pack in real time. It accurately estimates the remaining capacity and health status of the lithium battery pack using an extended Kalman filter battery state estimation algorithm and employs dynamic equalization charging technology to balance the voltage of each battery cell.

10. The energy-saving method for low-power standby and human body induction wake-up of electronic bus stop signs according to claim 1, characterized in that, It also includes regional collaborative energy-saving control steps, as detailed below: Adjacent electronic bus stop signs on the same road segment form a self-organizing network through low-power wide area network communication technology, establishing a point-to-point data transmission channel between the stops; Upstream stations collect real-time data on the number of waiting passengers and the departure of vehicles, and send this data to all downstream stations on the same road segment. When a downstream station receives data from an upstream station indicating that the number of waiting passengers is greater than or equal to a preset threshold, it enters a pre-wake state with a pre-set pre-wake duration. When a downstream station receives data from an upstream station indicating that a vehicle has departed, it prepares in advance to display the real-time arrival information of that vehicle. When a power grid shortage instruction is received from the power grid dispatch center or cloud operation and maintenance platform, all electronic bus stop signs in the area will automatically adjust their wake-up strategies according to their respective passenger flow priorities, giving priority to ensuring the normal operation of preset core stops; The cloud-based operation and maintenance platform collects energy consumption data, sensor data, and battery data from each electronic bus stop sign in the area in real time. Through machine learning fault prediction algorithms, it analyzes the data and predicts problems such as dust accumulation on solar panels, sensor aging, and battery failure in advance, generating maintenance work orders and sending them to operation and maintenance personnel.

Images