Parking state monitoring system based on RISC-V

Abstract

The invention provides a parking state monitoring system based on RISC-V, and relates to the technical field of automatic control. The parking state monitoring system based on the RISC-V comprises the following modules: a data acquisition module, a signal preprocessing module, a fusion judgment module, a state recognition module, a prediction control module, an energy consumption optimization module, a communication transmission module, a cloud analysis module and a feedback regulation module. Through a multi-source sensing synchronous sampling and time sequence correction method, multiple types of sensing signals are subjected to consistent sampling under a unified time reference, and the space-time matching degree of data is remarkably improved; an adaptive filtering and dynamic noise estimation mechanism is introduced in a signal processing link, so that an input signal keeps a high signal-to-noise ratio and stability in a complex environment; in the feature fusion and state recognition stage, probability weighted fusion and correlation matrix analysis are utilized to realize information multi-layer complementation, and the accuracy and reliability of parking space state recognition are ensured.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of automatic control, in particular to a parking state monitoring system based on RISC-V. BACKGROUND

[0002] RISC-V is a reduced instruction set architecture, and the technical field of automatic control refers to a comprehensive technical field of using computers, sensors, actuators and control algorithms to monitor, analyze and regulate the running state of a controlled object in real time, so that the system performance is optimal and the process is automated. This field covers signal acquisition and processing, data fusion, predictive control, energy optimization, communication transmission and feedback regulation, and is widely used in industrial production, intelligent transportation, energy management, unmanned systems and other fields. Among them, the parking state monitoring system based on RISC-V refers to using RISC-V open instruction set architecture as the core processing platform, combining multi-source sensing technology and intelligent control algorithm, realizing real-time monitoring, identification and prediction of the state of the parking space in the parking environment, and improving the intelligent and energy-saving level of the parking lot operation through automatic scheduling and feedback regulation mechanism.

[0003] The existing automatic control technology has many levels of deficiencies in multi-source perception and dynamic control. The traditional parking space monitoring system relies on a single sensing signal, and the data is significantly disturbed by environmental factors, resulting in unstable state identification and easy false positives and false negatives. Signal acquisition often lacks a unified time reference, and different sensor outputs have phase misalignment, which reduces the fusion calculation accuracy. In the signal preprocessing process, the threshold setting is mostly a fixed value, which cannot be adaptively corrected when facing environmental noise changes, causing excessive or insufficient filtering. SUMMARY

[0004] In view of the deficiencies of the prior art, the application provides a parking state monitoring system based on RISC-V, which solves the problems of the traditional parking space monitoring system relying on a single sensing signal, the data being significantly disturbed by environmental factors, leading to unstable state identification and easy false positives and false negatives, and signal acquisition often lacking a unified time reference, different sensor outputs having phase misalignment, and reducing the fusion calculation accuracy.

[0005] To achieve the above purpose, the application is implemented by the following technical scheme: a parking state monitoring system based on RISC-V, comprising the following modules: a data acquisition module, a signal preprocessing module, a fusion judgment module, a state identification module, a predictive control module, an energy optimization module, a communication transmission module, a cloud analysis module and a feedback regulation module. The data acquisition module based on a RISC-V embedded control core adopts a multi-source sensor synchronous sampling method to acquire real-time data of parking space states. The module detects in combination with ultrasonic, infrared and magnetic resistance sensors, uses a multi-sensor synchronous timing correction algorithm to correct the consistency of the data sampling clock, ensures that multi-source signals are collected under the same time reference, and generates an original monitoring data set; The signal preprocessing module based on the original monitoring data set uses an adaptive filtering and noise suppression algorithm to denoise, smooth and normalize the collected signals, dynamically filters different types of signals through a sliding window mean method and a Kalman filtering method, uses an environmental noise estimation subroutine to realize adaptive threshold adjustment, and generates a high-confidence signal set; The fusion judgment module based on the high-confidence signal set uses a multi-sensor fusion decision algorithm to fuse signal features. The system uses a Bayesian weighted fusion method to integrate the confidence of each sensor signal, introduces a correlation matrix analysis model to dynamically adjust the weight coefficient, thereby obtaining a comprehensive judgment of the single parking space state, and generates parking space state fusion data. The state recognition module based on the parking space state fusion data uses an adaptive threshold judgment algorithm to intelligently recognize the parking space occupancy. The system first filters suspicious states through a threshold dynamic adjustment mechanism, and then uses a pattern matching recognition method to classify and judge different occupancy features, thereby generating parking state recognition results.

[0006] Preferably, the data acquisition module includes an ultrasonic acquisition submodule, an infrared detection submodule, a magnetic resistance sensing submodule, and a timing correction submodule. The ultrasonic acquisition submodule actively detects parking space distance data and calculates the time delay of reflected waves to generate initial distance signals. The infrared detection submodule detects target occlusion and temperature difference to generate infrared perception signals. The magnetic resistance sensing submodule senses geomagnetic changes and judges the strength of vehicle metal interference to generate geomagnetic response signals. The timing correction submodule uses a synchronous clock correction algorithm to align the time reference based on the aforementioned three signals to generate an original monitoring data set.

[0007] Preferably, the signal preprocessing module includes a filtering processing submodule, a noise suppression submodule, a threshold adjustment submodule, and a normalization correction submodule. The filtering processing submodule uses a sliding window mean algorithm to smooth the original monitoring data set to generate preliminary denoised signals. The noise suppression submodule uses a Kalman filtering algorithm to estimate and compensate for environmental interference in real time to generate stable signals. The threshold adaptive submodule uses a dynamic threshold adjustment algorithm to correct the upper and lower limit fluctuations of the signals based on the stable signals to generate threshold optimized signals. The normalization correction submodule performs scale standardization and linear mapping based on the threshold optimization signal to generate a high-confidence signal set.

[0008] Preferably, the fusion judgment module includes a feature extraction submodule, a weighted fusion submodule, a confidence evaluation submodule, and a correlation analysis submodule. The feature extraction submodule extracts the main feature vector of the parking space state based on the high-confidence signal set to generate a feature matrix. The weighted fusion submodule integrates multi-source features based on the feature matrix using a Bayesian weighted fusion algorithm to generate a fusion weighted result. The confidence evaluation submodule calculates the confidence score of each sensing source based on the fusion weighted result to generate a confidence index set. The correlation analysis submodule calculates the final comprehensive state determination based on the confidence index set using a correlation matrix analysis method to generate parking space state fusion data.

[0009] Preferably, the state recognition module includes a threshold determination submodule, a pattern matching submodule, an anomaly detection submodule, and a result classification submodule. The threshold determination submodule performs preliminary occupancy recognition based on the parking space state fusion data through an adaptive threshold determination algorithm to generate candidate state data. The pattern matching submodule identifies typical state features such as empty, occupied, and abnormal based on the candidate state data using a pattern matching classification method to generate a pattern classification result. The anomaly detection submodule identifies signal anomalies or false positives based on the pattern classification result using a residual detection algorithm to generate anomaly identification information. The result classification submodule integrates the pattern classification result and the anomaly identification information to output a unified recognition conclusion to generate a parking state recognition result.

[0010] Preferably, the prediction control module uses a time series prediction control algorithm to predict future parking space usage trends based on the parking state recognition result. The system analyzes historical occupancy changes through a sliding time window modeling to generate a future time period parking space utilization prediction value, providing a basis for subsequent scheduling, and generates parking space utilization prediction data. The prediction control module includes a time modeling submodule, a trend analysis submodule, a short-term prediction submodule, and a result evaluation submodule. The time modeling submodule constructs a historical time series model based on the parking state recognition result to generate a time series data set. The trend analysis submodule analyzes the occupancy change trend based on the time series data set using a sliding window trend extraction algorithm to generate trend feature data. The short-term prediction submodule predicts future parking space states based on the trend feature data using a short-term memory prediction model to generate a short-term prediction result. The result evaluation submodule performs error evaluation and correction based on the short-term prediction result and historical real data to generate parking space utilization prediction data.

[0011] Preferably, the energy consumption optimization module, based on the parking space utilization prediction data, adopts an event-triggered energy-saving control method to realize dynamic power consumption adjustment of the sensor and communication module. When the predicted traffic flow is low, the system automatically reduces the sampling frequency; when the traffic flow rises, the high-frequency sampling mode is triggered. The core algorithm is a dynamic threshold power consumption adjustment mechanism to ensure the balance between energy consumption and response speed, and to generate an energy-saving control instruction set. The energy consumption optimization module includes an event-triggering submodule, a power consumption evaluation submodule, a threshold adjustment submodule, and a task scheduling submodule. The event-triggering submodule uses an event-triggering detection method to determine whether the system state change exceeds the threshold based on the parking space utilization prediction data, and generates an event-triggering signal. The power consumption evaluation submodule uses a power dynamic evaluation algorithm to evaluate the real-time energy consumption level of the system based on the event-triggering signal, and generates power consumption state data. The threshold adaptive submodule uses a dynamic sensitivity adjustment algorithm to correct the trigger threshold range based on the power consumption state data, and generates energy consumption threshold parameters. The task scheduling submodule uses a priority task scheduling algorithm to adjust the sampling and communication task rhythm based on the energy consumption threshold parameters, and generates an energy-saving control instruction set.

[0012] Preferably, the communication transmission module uses an asynchronous packet transmission algorithm based on the energy-saving control instruction set to realize efficient wireless communication of parking space state data and control instructions. The system dynamically selects the optimal communication path through a LoRa / NB-IoT dual-channel switching mechanism, and reduces data transmission volume by combining a differential compression encoding method to generate encrypted transmission data packets. The communication transmission module includes a channel management submodule, an encryption and encoding submodule, a data encapsulation submodule, and a path selection submodule. The channel management submodule uses an adaptive channel allocation algorithm to allocate data transmission paths based on the energy-saving control instruction set, and generates channel configuration information. The encryption and encoding submodule uses a differential compression encryption algorithm to encode data based on the channel configuration information, and generates an encrypted signal stream. The data encapsulation submodule uses a packet frame encapsulation method to generate a data packet structure based on the encrypted signal stream, and generates encapsulated transmission data. The path selection submodule selects the optimal transmission path through a dual-channel dynamic switching algorithm based on the encapsulated transmission data, and generates encrypted transmission data packets.

[0013] Preferably, the cloud analysis module, based on the encrypted transmission data packet, adopts a distributed parallel analysis algorithm in the cloud server to aggregate and count large-scale parking data, the system groups and manages the parking areas through clustering analysis and pattern recognition methods, and dynamically adjusts the prediction parameters by using a self-learning optimization model to realize iterative optimization of the intelligent scheduling strategy, and generates a global parking optimization scheme; The cloud analysis module comprises a data decryption submodule, a clustering analysis submodule, a parameter learning submodule and a scheduling optimization submodule. The data decryption submodule recovers the original data stream by using a symmetric key decryption algorithm based on the encrypted transmission data packet to generate decrypted data. The clustering analysis submodule groups the parking area distribution by using a density clustering algorithm based on the decrypted data to generate a clustering result set. The parameter learning submodule updates the prediction and scheduling model parameters by using a self-learning parameter optimization algorithm based on the clustering result set to generate an optimized parameter set. The scheduling optimization submodule generates a final system control scheme by using a multi-objective scheduling optimization algorithm based on the optimized parameter set to generate a global parking optimization scheme.

[0014] Preferably, the feedback adjustment module, based on the global parking optimization scheme, adopts a closed-loop feedback control algorithm to dynamically adjust the front-end system, the module sends the optimization results to each RISC-V node to control the real-time adjustment of the sampling frequency, communication mode and display logic, realizes system-level closed-loop adaptive control, and generates a system adaptive adjustment signal. The feedback adjustment module comprises a scheme analysis submodule, a parameter updating submodule, a node control submodule and a system synchronization submodule. The scheme analysis submodule analyzes the required adjustment parameters of each node based on the global parking optimization scheme to generate a control instruction set. The parameter updating submodule updates the RISC-V end control variables by using an online parameter refreshing algorithm based on the control instruction set to generate updated parameter data. The node control submodule adjusts the local sampling and communication strategy by using a node-level task control method based on the updated parameter data to generate a node execution state. The system synchronization submodule unifies the system running rhythm by using a closed-loop feedback synchronization algorithm based on the node execution state to generate a system adaptive adjustment signal.

[0015] The application provides a parking state monitoring system based on RISC-V, which has the following advantages: The application is in the continuous execution process of data acquisition, signal processing, feature fusion, state recognition, prediction analysis and energy consumption control, through multi-source sensor synchronous sampling and time sequence correction method, so that multiple types of sensing signals are consistently sampled under a unified time reference, significantly improving the spatio-temporal matching degree of data; in the signal processing link, adaptive filtering and dynamic noise estimation mechanism are introduced, so that the input signal maintains high signal-to-noise ratio and stability in complex environment; in the feature fusion and state recognition stage, probability weighted fusion and correlation matrix analysis are used to realize multi-layer complementary information, ensure the accuracy and reliability of the parking space state recognition; through time series modeling and trend prediction, the static recognition result is extended to dynamic trend information, forming a forward-looking occupancy prediction ability; in the energy consumption control and communication transmission process, through event triggering and dynamic threshold adjustment, the system resources are allocated on demand, combined with asynchronous grouping and double-channel switching mechanism to complete efficient and low-consumption communication; in the cloud processing and feedback link, through distributed clustering and self-learning model, global optimization control is formed, realizing closed-loop adaptive adjustment of front and rear ends. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a system flowchart of the application; Figure 2 is a data acquisition module schematic diagram of the application; Figure 3 is a signal preprocessing module schematic diagram of the application; Figure 4 is a fusion judgment module schematic diagram of the application; Figure 5 is a state recognition module schematic diagram of the application; Figure 6 is a prediction control module schematic diagram of the application; Figure 7 is an energy consumption optimization module schematic diagram of the application; Figure 8 is a communication transmission module schematic diagram of the application; Figure 9 is a cloud analysis module schematic diagram of the application; Figure 10 is a feedback adjustment module schematic diagram of the application. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0018] Embodiment: As Figures 1-10 shown, the embodiment of the application provides a parking state monitoring system based on RISC-V, which comprises the following modules: data acquisition module, signal preprocessing module, fusion judgment module, state recognition module, prediction control module, energy optimization module, communication transmission module, cloud analysis module, feedback regulation module; The data acquisition module is based on a RISC-V embedded control core and adopts a multi-source sensor synchronous sampling method to acquire real-time data of parking lot space state. The module detects by combining ultrasonic, infrared and magnetic resistance sensors, uses a multi-sensor synchronous timing correction algorithm to correct the consistency of data sampling clock, ensures that multi-source signals are collected under the same time reference, and generates an original monitoring data set. The signal preprocessing module is based on the original monitoring data set and uses an adaptive filtering and noise suppression algorithm to denoise, smooth and normalize the collected signals. It uses a sliding window mean method and a Kalman filter method to dynamically filter different types of signals, uses an environmental noise estimation subroutine to realize adaptive threshold adjustment, and generates a high-credibility signal set. The fusion judgment module is based on the high-credibility signal set and uses a multi-sensor fusion decision algorithm to fuse signal features. The system uses a Bayesian weighted fusion method to integrate the confidence of each sensor signal and introduces a correlation matrix analysis model to dynamically adjust the weight coefficient, thereby obtaining a comprehensive judgment of the single parking space state and generating parking space state fusion data. The state recognition module is based on the parking space state fusion data and uses an adaptive threshold judgment algorithm to intelligently recognize the parking space occupancy. The system first filters suspicious states through a threshold dynamic adjustment mechanism, and then uses a pattern matching recognition method to classify and judge different occupancy features, generating a parking state recognition result.

[0019] The data acquisition module includes an ultrasonic acquisition submodule, an infrared detection submodule, a magnetic resistance sensing submodule, and a timing correction submodule. The ultrasonic acquisition submodule performs active detection of parking space distance data and reflection wave time delay calculation to generate initial distance signals. The ultrasonic acquisition submodule performs active detection of parking space distance data. It uses an ultrasonic sensor to emit ultrasonic signals of a certain frequency, calculates the time delay of the reflected wave to obtain the distance information of the parking space. After the ultrasonic signal is emitted, it will be reflected when encountering an obstacle. The time difference between the reflected signal and the emitted signal is calculated to obtain the distance of the object. Assuming the transmission speed is 343 meters per second, if the reflection wave time delay is 0.01 seconds, the distance can be calculated by the formula: distance meters. This process can be applied to vehicle detection and parking space judgment to form parking space distance data as the basis for subsequent processing, generate initial distance signals, and accurately judge the occupancy state of the parking space.

[0020] The infrared detection sub-module detects target occlusion and temperature difference, and generates an infrared perception signal. The infrared detection sub-module is used for detecting target occlusion and temperature difference. It mainly detects the temperature change of the surrounding environment and the occlusion of objects through infrared sensors to generate an infrared perception signal. Specifically, the sensor measures the intensity of infrared radiation and compares the data with the set reference temperature value to determine whether there is object occlusion. In actual application, when a parking space is occupied by a vehicle, the temperature difference between the vehicle and the surrounding environment is large, and the infrared sensor can detect this change. For example, if the infrared sensor is set to a temperature difference threshold of 2°C, and the temperature difference of the parking space is detected to exceed this value, it is determined that the parking space has been occupied. By inversely calculating the state of the parking space, important basis is provided for subsequent parking space state judgment, forming an infrared perception signal.

[0021] The magnetoresistance sensing sub-module senses geomagnetic changes and judges the strength of vehicle metal interference to generate a geomagnetic response signal. The magnetoresistance sensing sub-module judges whether there is metal object (such as vehicle) interference by sensing geomagnetic changes. The geomagnetic sensor can detect the change of the geomagnetic field above or near the parking space. When the vehicle enters or exits, the influence of the metal object on the geomagnetic field will fluctuate, and the sensor generates a geomagnetic response signal according to these fluctuation changes. For example, the geomagnetic sensor detects the geomagnetic change in the vertical direction to obtain an amplitude value. Assuming that the initial geomagnetic value is 0.5 Gauss, when a vehicle is parked in the parking space, the geomagnetic value fluctuates to 0.6 Gauss, indicating that the parking space may be occupied. At this time, the sensor triggers the detection result and outputs the geomagnetic response signal to provide supplementary information for the occupancy state of the parking space.

[0022] The time sequence correction sub-module aligns the time base based on the aforementioned three signals using a synchronous clock correction algorithm to generate an original monitoring data set.

[0023] The time sequence correction sub-module ensures that the collection times of ultrasonic, infrared, and magnetoresistance signals are consistent through a synchronous clock correction algorithm, solving the time delay difference problem between signals. The time sequences of the three types of sensing signals are different, which may cause time errors during data fusion, thereby affecting the final parking space judgment. Through the time sequence correction algorithm, the data of different sensors can be time sequence corrected to ensure that they are compared under the same time base. For example, assuming that the collection time of the ultrasonic signal is T1, the collection time of the infrared signal is T2, and the collection time of the magnetoresistance signal is T3, the time stamps of these signals are adjusted to be synchronized using the time sequence correction algorithm, generating a unified time base. On this basis, the original monitoring data set is generated through time sequence correction, ensuring accurate and error-free subsequent data processing.

[0024] The signal preprocessing module comprises a filtering processing submodule, a noise suppression submodule, a threshold adjustment submodule, and a normalization correction submodule. The filtering processing submodule performs smoothing processing on the original monitoring data set based on a sliding window mean algorithm to generate a preliminary denoising signal. The filtering processing submodule performs smoothing processing on the original monitoring data set based on a sliding window mean algorithm to generate a preliminary denoising signal. The sliding window mean algorithm applies a fixed window size sliding window to each group of data in the original data to calculate the mean value of the data in the window for smoothing processing. In actual application, a window size can be specified, such as 5 data points, and the mean value of the 5 data points before and after the current data point is calculated to remove short-term fluctuations or noise in the data. For example, if the data set is [3.1, 3.3, 3.0, 2.8, 3.2, 3.4, 3.3], a 5-point window is selected, and the first data point is processed. The window is [3.1, 3.3, 3.0, 2.8, 3.2], and the mean value is 3.08. The window of the second data point is [3.3, 3.0, 2.8, 3.2, 3.4], and the mean value is 3.14. Similarly, the smoothed signal is obtained to effectively remove high-frequency noise and generate a preliminary denoising signal.

[0025] The noise suppression submodule performs real-time estimation and compensation of environmental interference based on a Kalman filter algorithm to generate a steady-state signal. The noise suppression submodule performs real-time estimation and compensation of environmental interference based on a Kalman filter algorithm to generate a steady-state signal. The Kalman filter algorithm recursively estimates the system state and continuously corrects the estimation error, which is suitable for real-time processing of dynamic signals. In actual application, the Kalman filter reduces the impact of noise by weighting the difference between the predicted value and the measured value. For example, assuming that the true state of the system is X, the predicted value is , the measured value is Z, the Kalman gain is K, and the new estimated value is , where K is a weighting coefficient calculated from the error covariance matrix. If K is large, the system relies more on the measured value, and if K is small, the system relies more on the predicted value. By continuously iterating and updating, the impact of noise is gradually reduced, and a steady-state signal is generated.

[0026] The threshold adaptive submodule generates a threshold optimized signal by using a dynamic threshold adjustment algorithm to correct the upper and lower limit fluctuations of the steady-state signal. The threshold adaptive sub-module generates a threshold optimization signal by correcting the upper and lower limit fluctuations of the signal based on the steady-state signal using a dynamic threshold adjustment algorithm. The dynamic threshold adjustment algorithm adjusts the upper and lower limit of the signal in real time by setting certain dynamic adjustment rules and combining the fluctuation range of the steady-state signal. For example, the upper and lower limits of the signal fluctuation range are set as a dynamic range to avoid the influence of extreme values of the signal on normal data processing. In the adjustment process, the standard deviation of the current signal fluctuation range can be calculated to determine the change range of the threshold. Assuming that the standard deviation of the steady-state signal is 0.5, if the upper limit of the threshold is set as the mean value of the steady-state signal plus a multiple of the standard deviation, and the lower limit is the mean value minus a multiple of the standard deviation, through such dynamic adjustment, the signal fluctuation range is ensured to be within a reasonable range, avoiding signal distortion caused by excessive fluctuation, and generating a threshold optimization signal.

[0027] The normalization correction sub-module performs scale standardization and linear mapping based on the threshold optimization signal to generate a high-confidence signal set.

[0028] The normalization correction sub-module performs scale standardization and linear mapping based on the threshold optimization signal to generate a high-confidence signal set. Scale standardization and linear mapping are performed by adjusting the amplitude and range of the signal to adapt to specific application requirements and algorithm requirements. For example, by mapping the signal to the range of [0, 1], the maximum value of the signal can be mapped to 1 and the minimum value can be mapped to 0 using a linear mapping formula. Assuming that the minimum value of the threshold optimization signal is 2 and the maximum value is 8, the signal can be normalized to a new value through the linear mapping formula: to ensure that the signal is within the adaptive range, thereby forming a high-confidence signal set for subsequent processing or analysis.

[0029] The fusion judgment module includes a feature extraction sub-module, a weighted fusion sub-module, a confidence evaluation sub-module, and a correlation analysis sub-module. The feature extraction sub-module extracts the main feature vector of the parking space state based on the high-confidence signal set to generate a feature matrix. The feature extraction submodule extracts the main feature vector of the parking space state based on a high-confidence signal set, generates a feature matrix, and the high-confidence signal set is derived from the normalized output of ultrasonic, infrared, and geomagnetic signals. First, the signals are time-aligned and the stable section with a signal fluctuation range within ±10% is selected as the input sample. The feature components are calculated through the signal change rate, duration, and fluctuation amplitude. In specific operation, assuming that the sampling period is 0.1 seconds, the infrared signal temperature difference change is 1.8°C, the geomagnetic change is 0.09 Gauss, and the ultrasonic distance change is 0.25 meters, the three are recorded as temperature features, magnetic field features, and distance features, respectively. Each feature is composed of 10 consecutive sampling points to form a window, and the average, maximum, and variance of each window are calculated. Then, through standardization processing, they are mapped to the [0, 1] interval. Subsequently, according to the time sequence, the feature vector F=[f1, f2, f3,..., fn] is spliced, and the feature matrix is arranged in matrix form, where the rows represent different sampling times, and the columns represent different signal types. This matrix serves as the input basis for subsequent fusion calculation and is used to generate the feature matrix.

[0030] The weighted fusion submodule integrates multi-source features based on the feature matrix using the Bayesian weighted fusion algorithm to generate a fusion weighted result. The weighted fusion submodule integrates multi-source features based on the feature matrix using the Bayesian weighted fusion algorithm to generate a fusion weighted result. In specific operation, the conditional probability of the features from different sensors is estimated, and the weight of each sensor output corresponds to a feature vector group, which is set to where the infrared signal stability weight is 0.35, the geomagnetic signal weight is 0.30, and the ultrasonic signal weight is 0.35. When each sensor signal fluctuates, a weighted average calculation is performed, and the likelihood probability distribution of each type of feature is judged for fusion. For example, when the confidence probability of the ultrasonic detection distance feature is 0.85, the confidence probability of the infrared detection temperature difference feature is 0.78, and the confidence probability of the geomagnetic signal is 0.81, the weighted calculation method is used to obtain the fused confidence value P=0.35×0.85+0.30×0.81+0.35×0.78=0.8145. This fusion value represents the joint judgment strength of multi-source features in a specific time period. By dynamically adjusting the weight value range between [0.2, 0.5], the noise proportion and historical error of different sensor sources are corrected to ensure that the fusion weighted result can reflect the comprehensive state of multi-source features at the numerical level, generating a fusion weighted result.

[0031] The confidence evaluation submodule calculates the confidence score of each sensor source based on the fusion weighted result to generate a confidence index set. The confidence evaluation submodule calculates the confidence score of each sensing source based on the fusion weighting result, generates a confidence index set, and needs to quantitatively evaluate the signal stability and consistency of each sensing source during the calculation process. First, set the confidence calculation reference value C0=0.8 as the intermediate reference level. When the fluctuation amplitude of the sensing source signal is within ±5%, it is recorded as high stability, with a score interval of [0.85, 1.0]. When the fluctuation is within ±10%, it is recorded as medium stability, with a score interval of [0.65, 0.85]. When the fluctuation exceeds ±10%, it is recorded as low stability, with a score interval of [0.5, 0.65]. In practical applications, for example, the standard deviation of the infrared sensing signal is 0.07°C for five consecutive samplings, which belongs to the medium stability interval, and the confidence score is 0.74. The geomagnetic sensing signal changes by 0.03 Gauss, which belongs to the high stability interval, and the score is 0.90. The ultrasonic signal changes by 0.22 meters, which belongs to the low stability interval, and the score is 0.60. The scores of the three are arranged as an index vector C=[0.74, 0.90, 0.60], which is used to construct the confidence index set. The index set provides the basis for fusion in the subsequent steps and is used to generate the confidence index set.

[0032] The correlation analysis submodule calculates the final comprehensive state judgment based on the confidence index set using the correlation matrix analysis method to generate parking space state fusion data.

[0033] The correlation analysis submodule calculates the final comprehensive state judgment based on the confidence index set using the correlation matrix analysis method to generate parking space state fusion data. In the specific process, first, construct a three-dimensional confidence matrix R, input the confidence scores of the infrared, geomagnetic, and ultrasonic signals as column vectors, and determine the correlation strength between signals by calculating the correlation coefficient pair by pair. The correlation coefficient has a value range of [-1, 1]. In the vehicle detection scene, when the change trends of two signals are consistent, the correlation coefficient is close to 1, and when the change trends are opposite, the correlation coefficient is close to -1. For example, the correlation coefficient of infrared and geomagnetic is 0.82, the correlation coefficient of geomagnetic and ultrasonic is 0.68, and the correlation coefficient of infrared and ultrasonic is 0.74. The overall correlation strength is determined according to the weighted average result of each correlation coefficient When is greater than 0.7, it is judged that the parking space state is "occupied", less than 0.5 is judged as "idle", and between the two is judged as "transition state". In the example , the output parking space state fusion data is "occupied". This process generates the final parking space state fusion data through the calculation and quantitative comparison of the correlation matrix.

[0034] The state recognition module includes a threshold judgment submodule, a pattern matching submodule, an anomaly detection submodule, and a result classification submodule The threshold judgment submodule performs preliminary occupancy recognition based on the parking space state fusion data through an adaptive threshold judgment algorithm to generate candidate state data. The threshold determination sub-module generates candidate state data based on the preliminary occupation recognition through an adaptive threshold determination algorithm based on the parking space state fusion data. In specific operation, first, infrared temperature features, geomagnetic signal intensity, and ultrasonic distance features are extracted from the fusion data as three types of input signals. The signals are time-synchronized and value-normalized, and all inputs are mapped to the 0 to 1 interval. Then, an adaptive threshold is set The threshold is dynamically adjusted based on the mean and standard deviation of the past 10 minutes of historical data. When the geomagnetic signal value is higher than 0.65, the infrared signal temperature difference is greater than 0.6, and the ultrasonic distance feature is lower than 0.3, it is determined that there is a vehicle signal at that moment and recorded as an occupied state sample. In the period without vehicle passing, if the three are less than 0.4, 0.3, and 0.5 respectively, it is determined as an idle state sample. By comparing each sample signal, the occupation ratio in different time windows is calculated. When the ratio exceeds the set reference value of 0.6, the candidate state is generated as "occupied", otherwise as "idle". For example, in a certain detection period, the sampled geomagnetic signal is 0.72, the infrared signal is 0.65, and the ultrasonic signal is 0.25. The threshold judgment conditions all meet the occupied interval range, so the candidate state "occupied" is output. This process runs continuously to generate candidate state data.

[0035] The pattern matching sub-module generates pattern classification results based on candidate state data using pattern matching classification to identify typical state features such as idle, occupied, and abnormal. The pattern matching sub-module generates pattern classification results based on candidate state data using pattern matching classification to identify typical state features such as idle, occupied, and abnormal. In the execution process, the candidate state sequence is first divided into time windows of 60 seconds, each window containing multiple groups of signal feature data. The fluctuation feature, persistence feature, and conversion frequency are extracted as three key indicators. The fluctuation feature is defined as stable when the signal standard deviation is below 0.05, slightly fluctuating when it is between 0.05 and 0.15, and abnormal fluctuation when it exceeds 0.15, which is mapped to idle, occupied, and abnormal three preliminary state patterns. In the calculation process, the most frequent mode category in each window is counted as the main classification result of that window. For example, in a certain time period, the infrared signal average is 0.68, the fluctuation standard deviation is 0.04, the geomagnetic signal average is 0.71, and the ultrasonic signal average is 0.28. The three match the occupied mode feature set {high geomagnetic + high infrared + low ultrasonic wave}, corresponding to the mode "occupied". If there is a situation of severe infrared signal fluctuation and sudden drop in geomagnetic signal, it matches the abnormal mode. The classification result vector M = [m1, m2,..., mn] is output through pattern matching, where each m_i represents the recognition mode label of the time window, and the final pattern classification result is formed.

[0036] The anomaly detection submodule identifies signal anomalies or false positives based on the pattern classification results using a residual detection algorithm to generate anomaly identification information. The anomaly detection submodule identifies signal anomalies or false positives based on the pattern classification results using a residual detection algorithm to generate anomaly identification information. In implementation, a prediction model is first established according to the occupied and idle state samples in the pattern classification results, and the weighted average of the signal in the last period is used to construct the expected value set E = [e1, e2,..., en]. Then, the current measured signal set S = [s1, s2,..., sn] is compared, and the residual The deviation amplitude is obtained, and if the absolute value of the residual exceeds the set threshold , it is marked as an anomaly. The threshold is set according to the signal type, such as geomagnetic signals , infrared signals , and ultrasonic signals . When the sample infrared signal measured value is 0.85, and the expected value is 0.72, the residual is calculated, and the sample is determined to be abnormal. On this basis, the anomaly type number is recorded and the anomaly identification matrix A is generated, where each row represents the anomaly state flag of different sensing sources (0 for normal, 1 for abnormal). For example, A = [0, 1, 0] indicates that only the infrared signal is abnormal. This matrix provides the basis for false positive correction in subsequent processing, and generates anomaly identification information.

[0037] The result classification submodule integrates the pattern classification results and anomaly identification information to output a unified recognition conclusion and generate a parking state recognition result.

[0038] The result classification submodule integrates the pattern classification results and anomaly identification information to output a unified recognition conclusion and generate a parking state recognition result. In the execution process, the pattern classification result vector M and the anomaly identification matrix A are first read. For each time period, the number of anomaly identifications corresponding to the state pattern is determined. When the number of anomaly identifications exceeds one-third of the total number of signals, the corresponding pattern result is marked as "abnormal to be confirmed". Otherwise, the original pattern category is maintained. Then, the normal signal proportion is calculated by weighting according to the signal weight coefficients w = [0.4, 0.35, 0.25]. If the weighted result is higher than the reference judgment value 0.65, the final state is output as "occupied". If it is lower than 0.4, the final state is output as "idle". If it is between the two, the final state is output as "transitional state". For example, at a certain time, the infrared signal is normal, the geomagnetic signal is normal, and the ultrasonic signal is abnormal. The weighted sum of normal signals P = 0.4 * 1 + 0.35 * 1 + 0.25 * 0 = 0.75 > 0.65, and the final recognition result is output as "occupied". When the same result appears continuously for three times within the detection period, it is recorded as a stable recognition state, and a parking state recognition result is generated.

[0039] The prediction control module adopts a time series prediction control algorithm to predict the future parking space usage trend based on the parking state recognition result, the system analyzes the historical occupancy change through sliding time window modeling, generates the future period parking space utilization prediction value, provides basis for subsequent scheduling, and generates parking space utilization prediction data; The prediction control module includes a time modeling submodule, a trend analysis submodule, a short-term prediction submodule, and a result evaluation submodule. The time modeling submodule constructs a historical time series model based on the parking state recognition result, and generates a time series data set. The time modeling submodule constructs a historical time series model based on the parking state recognition result, and in the execution process, first, the occupancy identifier of each parking space in a unit time interval is extracted from the state recognition result of the parking system, and is arranged in time sequence to form an initial data stream. For example, for the parking space numbered P01, it is sampled every 5 minutes from 09:00 to 11:00 on November 7, 2025, a total of 24 data points are collected, if the occupancy state is recorded as 1 and the idle state is recorded as 0, the state sequence [1, 1, 1, 0, 0, 1,...] is obtained, secondly, the data is time-aligned and missing is completed, when sampling loss occurs, the adjacent time point interpolation method is used to complete, if 09:45 has no record, the average state value 0.5 of 09:40 and 09:50 is taken and rounded up to 1, and then the state sequences of all parking spaces are aggregated according to the time stamp to form a multi-dimensional time matrix T(t, p Where t represents the time index, and p is the parking space number. The columns in the matrix are smoothed, when the state changes frequently for 3 consecutive time points, the smoothing window width w=3 is used for smoothing, so that the data fluctuation amplitude is controlled in the interval ±0.2, then the daily average occupancy period and the change rate parameter of each parking space are extracted, for example, the P01 parking space is occupied for 15 hours in 24 hours, the occupancy rate is 15 / 24=0.625, and the change rate is the number of state changes / total sampling points=8 / 288=0.028, which belongs to the "moderate fluctuation interval" (0.02~0.05), finally, the time series and fluctuation characteristic values of each parking space are stored in the database to form a historical time series model, and a time series data set is generated.

[0040] The trend analysis submodule analyzes the occupancy change trend based on the time series data set, and generates trend characteristic data by using a sliding window trend extraction algorithm. The trend analysis submodule analyzes the occupancy change trend based on the time series data set, and in the execution process, first, the sliding window size w=6 (corresponding to 30 minutes) is set, and the occupancy rate change When is defined as an upward trend, is defined as a stable trend, Defined as a downward trend, in actual calculations, taking parking space P01 as an example, if the average occupancy rate over the past 30 minutes... Current window average occupancy ,but If the area falls within an upward trend range, the window result is marked as "Increasing Occupancy," and the trend continuity coefficient is subsequently recorded in multiple consecutive windows. ,definition The proportion of windows with the same trend to the total number of windows, when Once a trend is identified as a continuing trend, time periods of stable upward or downward movement are selected, and then the trend strength parameter is calculated. This is used to quantify the magnitude of change and form a trend characteristic data table, such as during a certain period. This indicates a change in occupancy rate of 0.03 every 5 minutes, categorized as a "slight increase". The data is categorized as "significantly rising" and the final output is a set of trend characteristic data for each time period.

[0041] The short-term prediction submodule uses a short-term memory prediction model based on trend feature data to predict the future parking space status and generate short-term prediction results. The short-term prediction submodule uses a short-term memory prediction model based on trend feature data to predict future parking space status. During execution, the trend features of the past hour are first used as the input sequence. ,in For occupancy rate, For trend strength, As a continuity coefficient, a prediction step size h=3 is set to correspond to the next 15 minutes, a short-term prediction sample set is established, and then the predicted value is generated by weighted averaging. The weighting coefficients were set to α=0.6, β=0.25, and γ=0.15, and the calculations were performed using actual data from the parking lot, such as the current... ,but It belongs to the medium occupancy range. The predicted parking space status is "occupied". When three consecutive prediction results fall within the same interval, the result is output as a short-term prediction result matrix P=[p1,p2,...,pn], where each... This indicates the parking space status at a future time step.

[0042] The results evaluation submodule performs error assessment and correction based on short-term forecast results and historical real data, and generates parking space utilization forecast data.

[0043] The results evaluation submodule performs error assessment and correction based on short-term forecast results and historical real data. During execution, it first extracts the short-term forecast result sequence P and the real occupancy rate sequence R for the same time period, and calculates the error vector. and with average error As an evaluation benchmark, if If the prediction deviation is deemed too large, a correction process is executed, during which an adjustment coefficient is defined. ,when When the ratio is below 0.9, the adjustment ratio will be adjusted accordingly. For example, the predicted average occupancy rate during the period from 10:00 to 11:00 on November 7, 2025. Actual occupancy rate ,but No correction is needed. However, if the prediction is 0.55 and the actual value is 0.75, then... Correction results After correction, the error was reduced to 0.16. Finally, the corrected sequence was bound with the timestamp to form a parking space utilization prediction dataset for subsequent periodic updates and model retraining.

[0044] The energy consumption optimization module, based on parking space utilization prediction data, adopts an event-triggered energy-saving control method to achieve dynamic power consumption adjustment of the sensor and communication modules. When the predicted traffic flow is low, the system automatically reduces the sampling frequency; when the traffic flow increases, a high-frequency sampling mode is triggered. The core algorithm is a dynamic threshold power consumption adjustment mechanism to ensure a balance between energy consumption and response speed, and to generate an energy-saving control instruction set. The energy consumption optimization module includes an event triggering submodule, a power consumption evaluation submodule, a threshold adjustment submodule, and a task scheduling submodule; The event triggering submodule uses parking space utilization prediction data and an event triggering detection method to determine whether the system state change exceeds the threshold and generates an event triggering signal. The event triggering submodule constructs an input sequence based on parking space utilization prediction data. During execution, it first reads the parking space utilization prediction dataset output from the previous stage, where the occupancy rate of each parking space in the next 15 minutes, 30 minutes, and 60 minutes is stored in numerical form. For example, the predicted occupancy rate of parking space P01 is [0.68, 0.74, 0.79]. Next, it sets the system state change threshold. The initial threshold is set with reference to the average fluctuation range of the past week. and take The rate of change is then calculated for each forecast period. ,when If the time frame is determined to be a state change, then in instance P01, if the occupancy rate in the previous time period was 0.55 and the current prediction is 0.74, then... The event detection flag is set to 1 if the event is triggered, and 0 otherwise if the event is triggered. The trigger sensitivity coefficient is then set according to the different parking space zones. ground area Underground area ,Will and Weighted comparison, when Event trigger signal is generated , for example, underground parking lot When , ground area , then trigger, all event signals are stored in the event queue table , forming the final event trigger signal output.

[0045] The power consumption evaluation submodule evaluates the real-time energy consumption level of the system based on the event trigger signal using the power dynamic evaluation algorithm, generating energy consumption state data; The power consumption evaluation submodule performs real-time energy consumption calculation based on the event trigger signal. During execution, it first collects current power , voltage U, current I and other parameters from system sensors and control units, and calculates the power consumption difference corresponding to each event trigger , if the energy consumption changes significantly, the average power consumption per unit time is calculated again according to the time weighting , in a specific example, assuming that the initial power of the parking lot lighting system is 15.5W, and after the event trigger, it rises to 17.3W, then ΔP=1.8W>0.2W is recorded as an energy consumption increase event, and then the dynamic power consumption evaluation coefficient is set combining with environmental temperature, illumination and other auxiliary parameters , take for night, for day, the corrected real-time power consumption , calculate and output the energy consumption state data table in hours, if all three consecutive samples are greater than 0.2W, then the energy consumption state of the corresponding time period is marked as "high", when between 10-15W is defined as "medium", and less than 10W is defined as "low", through this process, the energy consumption state data sequence is generated.

[0046] The threshold adaptive submodule modifies the trigger threshold range using a dynamic sensitivity adjustment algorithm based on the energy consumption state data, generating energy consumption threshold parameters; The threshold adaptive submodule performs dynamic sensitivity adjustment based on the energy consumption state data. During execution, it first imports the previous stage energy consumption state sequence, sets the initial energy consumption threshold , and records the power consumption value sampled every 5 minutes , for example [12.8, 14.6, 15.9, 17.2, 14.1], calculates the average power consumption , fluctuation amplitude , when the energy consumption fluctuation is high, the threshold interval is expanded, the dynamic correction coefficient is modified , the threshold is updated , the sensitivity adjustment parameter is calculated again , according to the number of triggers in the past 1 hour ratio of the actual number of times to the expected number of times , when indicates that the sensitivity is too high and needs to be adjusted down by 5%, when indicates that the sensitivity is too low and needs to be adjusted up by 5%, for example, the number of times of triggering in the last 1 hour is 18, and the expected number of times is 15, , after adjustment , an updated energy consumption threshold parameter table is generated, recording the time of each adjustment, the old threshold, the new threshold and the adjustment direction, for subsequent task calling.

[0047] The task scheduling submodule adjusts the rhythm of sampling and communication tasks based on the energy consumption threshold parameter, using a priority task scheduling algorithm to generate a set of energy-saving control instructions.

[0048] The task scheduling submodule adjusts the rhythm of sampling and communication tasks based on the energy consumption threshold parameter, in the execution process, first match the updated energy consumption threshold parameter with the power consumption demand of each task , establish a power consumption priority table, when defined as a high-power task, between is a medium-power task, less than 0.5 is a low-power task, which is used to determine the priority weight of the task , take the high-power task , medium-power task , low-power task , and then adjust the task period according to the priority , where is the initial period, assuming that the initial period of the sampling task , the communication task , if the current power consumption is medium, then when the power consumption rises to the high interval, the system automatically redistributes the task rhythm and outputs the instruction set C={T_1',T_2',...} to form the energy-saving control instruction set for dynamic scheduling in the system execution period.

[0049] The communication transmission module, based on the energy-saving control instruction set, uses an asynchronous packet transmission algorithm to realize efficient wireless communication of parking space state data and control instructions, the system dynamically selects the optimal communication path through the LoRa / NB-IoT dual-channel switching mechanism, and combines the differential compression encoding method to reduce the data transmission amount, and generates encrypted transmission data packets; The communication transmission module includes a channel management submodule, an encryption and encoding submodule, a data encapsulation submodule, and a path selection submodule; The channel management submodule, based on the energy-saving control instruction set, uses an adaptive channel allocation algorithm to allocate data transmission paths and generates channel configuration information; ​The channel management submodule extracts the task power consumption weight information based on the energy saving control instruction set and reads the current available wireless channel set. In the parking lot environment, it is assumed that the available channels are A and B, which correspond to different frequency bands and bandwidths. The bandwidth values are set according to the device specifications, with A bandwidth being 2 MHz and B bandwidth being 1 MHz. According to the previous energy consumption scheduling results, the priority weights of the sampling task and the communication task are given , and then the business data rate requirement is converted according to the weight . The calculation refers to the task data size and transmission period, for example, the sampling task transmission data packet size is 500 Byte, and the period is 15 s, then , the communication task transmission data packet size is 1500 Byte, and the period is 30 s, then , R_i and the channel bandwidth call proportion coefficient λ are set to 0.8 to ensure business continuity. The demand rate and bandwidth correlation are compared, and when , it is determined that the channel supports the task. In the A channel , all task rate requirements are supported, so it can be added to the candidate list. In the B channel, 0.8×1MHz=0.8MHz also supports all tasks. Based on the priority comparison item included in the energy consumption control instruction set, the candidate channels are sorted and the channel congestion factor is calculated , where , the current number of connected devices and the occupancy percentage are referred to, for example, the channel A current occupancy rate is 40%, which is judged as the medium level interval (30%~60%), and the channel B occupancy rate is 70%, which is judged as the high level interval (60%). Select the low-congestion channel A as the sampling task allocation path. The communication task is compared again due to low delay tolerance, and channel A is selected as the main path and channel B as the backup path. The channel allocation relationship corresponding to each task is sorted to generate channel configuration information.

[0050] The encryption coding submodule generates an encrypted signal stream based on the channel configuration information using a differential compression encryption algorithm; The encryption coding submodule based on channel configuration information first reads the original data packet byte stream and sets the differential reference value d_ref to obtain the difference between consecutive data, such as sampling data [102, 108, 111, 113] Byte, and the differential sequence is obtained by differentiating each item , and then the differential sequence is divided into coding levels according to the data change amplitude. When the differential absolute value , it is judged as the low change interval , it is judged as the medium change interval , For high variation interval, different encoding lengths are set according to the interval, such as low variation encoding 2 bits, medium variation encoding 3 bits, and high variation encoding 4 bits. Taking the parking lot temperature and humidity sensor as an example, the differential value corresponds to the encoding length of 2 bits, 3 bits, and 2 bits, respectively. Then, the XOR encryption key k is introduced to the differential value. The key is 8 bits long, which is referred to the security level of the system, such as k = 10101100b. The bit-by-bit XOR processing is performed on each encoding field to form an encrypted bit stream If the differential value is 6, the binary 000110 is XORed with the key to obtain the encrypted result 10110110. Then, the encoding index and the differential encrypted data are spliced in a fixed order to form an encrypted signal stream. At the same time, the time stamp and the device number appearing in the data are processed by bit shift masking using an independent encryption coefficient α set as the hexadecimal constant 0xF. All encrypted segments are arranged in the order of channel configuration for subsequent packaging.

[0051] The data packaging submodule generates a data packet structure based on the encrypted signal stream using a grouping frame packaging method to generate packaged transmission data; The data packaging submodule based on the encrypted signal stream first sets the grouping frame structure parameters, wherein the frame header length H is set to 4 bytes including synchronization identifier and frame sequence number, the payload area length L is set to be dynamically divided according to the length of the encrypted signal stream, such as a maximum payload of 48 bytes per frame, and the frame tail length T is set to 2 bytes for error checking. The encrypted signal stream is segmented by 48 bytes, such as a total length of 100 bytes, which is divided into 3 frames, two frames of 48 bytes, and one frame of 4 bytes. Then, a frame sequence number segment Seq is assigned to each frame, which is incremented from 0, for example, Seq = [0, 1, 2]. The frame header synchronization identifier is set to a fixed binary 0101b converted to 1 byte, and the remaining frame header contains a time stamp of 2 bytes and a device ID of 1 byte, which are combined into 4 bytes and added to the front of each encrypted payload. Then, a check value c generation process is performed on each frame. The check value is obtained by byte-by-byte accumulation and taking the last 2 bytes, such as If the sum of all bytes in a certain frame is 123456, then c = 123456 mod 65536 = 57920, which is converted to 2 bytes and added to the tail. After frame-by-frame packaging, a complete data packet structure is formed, and the data packets are arranged in the order of task priority for transmission.

[0052] The path selection submodule selects the optimal transmission path based on the packaged transmission data through a dual-channel dynamic switching algorithm to generate encrypted transmission data packets.

[0053] The path selection submodule reads the available transmission path set based on the packaged transmission data, which is set as the main channel A and the backup channel B. At the same time, the real-time channel quality indicators including error rate and delay are extracted The interval is divided into defined as low error level, medium level, high level, d is divided into low delay when d < 50 ms, medium delay when 50 ~ 100 ms, high delay when 100 ms, in the example Channel A measures medium level, medium level, and then the channel quality comprehensive judgment basis is set according to the weight coefficient β for error rate, for delay, the index value difference is weighted and added to form a quality score , wherein The lower the comparison score, the higher the priority, so the main path is set to channel A, and channel B is reserved for switching conditions after event triggering, forming an encrypted transmission data packet output.

[0054] The cloud analysis module is based on the encrypted transmission data packet, and a distributed parallel analysis algorithm is used in the cloud server to aggregate and count large-scale parking data. The system groups and manages the parking areas through clustering analysis and pattern recognition methods, and dynamically adjusts the prediction parameters using a self-learning optimization model to realize iterative optimization of intelligent scheduling strategies and generate a global parking optimization scheme. The cloud analysis module includes a data decryption submodule, a clustering analysis submodule, a parameter learning submodule, and a scheduling optimization submodule. The data decryption submodule is based on the encrypted transmission data packet and uses a symmetric key decryption algorithm to restore the original data stream to generate decrypted data. The data decryption submodule based on the encrypted transmission data packet first reads the frame header, frame body, and frame tail information, parses the encrypted bit stream E_b and key index field, loads the symmetric key k from the system security storage unit, sets the key length to 8 bits k = 10101100b, then reorders the received encrypted bit stream according to the frame sequence number. In the parking lot vehicle detection terminal scene, assuming that the receiver obtains the frame sequence number Seq = [0, 1, 2] corresponding to the encrypted data length of 48 bytes, 48 bytes, and 4 bytes, respectively, perform exclusive or inverse operation processing on each frame segment to restore the difference sequence If a certain encrypted byte is 10110110b and the key is XORed to recover 000110b, the original byte sequence [102, 108, 111, 113] is obtained by accumulating the consecutive difference sequence, and the data consistency is checked according to the accumulated check value field in the frame tail. Assuming that the check value of a frame is 57920, the sum of the received bytes is taken modulo 65536, and if the result is the same, it is judged as valid data, otherwise it is rejected and re-requested. Then, all frame segments are spliced to generate a continuous original data stream, and the timestamp and device number field are processed by reverse bit shift masking. The masking coefficient a used in the encryption stage is a hexadecimal constant 0xF, corresponding to a shift of 4 bits. Therefore, the opposite direction shift is performed during decryption to restore the original data.

[0055] The clustering analysis submodule groups the parking area distribution based on the decrypted data using a density clustering algorithm to generate a clustering result set; The clustering analysis submodule extracts the vehicle parking coordinates (x, y , stay duration t and occupation state s, etc. from the decrypted data. Assuming that the sample data set in the parking lot contains N=10 parking point coordinates [(5, 8), (6, 9), (7, 8), (20, 25), (21, 24), (22, 26), (40, 42), (41, 40), (60, 65), (61, 66)], the density clustering method is used to divide the cluster set. First, the Euclidean distance between any two points is calculated According to experience, the distance threshold is set to 5m, and is defined as a neighborhood point, and the minimum core point quantity MinPts is set to 3 as the clustering reference value. The number of neighborhood points for each point is counted When , it is judged as a core point, otherwise it is a boundary or noise point. For example, the neighborhood of point (6, 9 contains (5, 8) and (7, 8 , a total of 2 points, which is judged as a non-core point, while the neighborhood of (21, 24 contains (20, 25) and (22, 26 , a total of 2 points, which is also less than 3 but is connected to another core point and is classified into the same cluster. In the Python environment, the distance matrix can be calculated in batches through the numpy array, and the neighborhood relationship can be determined according to the threshold. Finally, the point set connected by density forms a clustering cluster, such as three clustering clusters C1={(5, 8), (6, 9), (7, 8)}, C2={(20, 25), (21, 24), (22, 26)}, C3={(40, 42), (41, 40)}, and the remaining points are defined as independent clusters. The clustering result set is output.

[0056] The parameter learning sub-module updates the prediction and scheduling model parameters based on the clustering result set using a self-learning parameter optimization algorithm to generate an optimized parameter set; The parameter learning sub-module first calculates the average occupancy rate of each cluster based on the clustering result set and the average stay duration . For example, the average occupancy rate of C1 is 0.82, the average stay duration is 45 minutes, the average occupancy rate of C2 is 0.65, the average stay duration is 30 minutes, and the average occupancy rate of C3 is 0.40, the average stay duration is 20 minutes. These statistics are input into the self-learning parameter optimization algorithm as an input vector. The prediction model parameter θ is adjusted through error correction iteration. The initial parameter is taken as , where w represents the occupancy rate weight and t represents the time weight. The error term δ is set as the average value of the difference between the actual and predicted occupancy rates. For example, if the current predicted occupancy rate is 0.7 and the actual occupancy rate is 0.82, then δ = 0.12. The learning rate η is taken as 0.1. The update formula is described as follows: the new parameter is equal to the old parameter minus the learning rate multiplied by the error weighting value. The updated parameter is calculated for each clustering cluster, and an optimized parameter set is formed. For example, after 5 iterations, θ = [0.388, 0.612]. The weight range is adjusted according to the density level of different parking areas. When , it is defined as a high-density area, , it is a medium-density area, , it is a low-density area. The corresponding parameter adjustment interval is set as follows: the occupancy weight of the high-density area is increased to 0.7, the occupancy weight of the medium-density area is kept at 0.6, and the occupancy weight of the low-density area is reduced to 0.5. The final output parameter adjustment result is compiled into an optimized parameter set.

[0057] The scheduling optimization sub-module generates a final system control scheme based on the optimized parameter set using a multi-objective scheduling optimization algorithm to generate a global parking optimization scheme.

[0058] The scheduling optimization sub-module loads the θ parameter vector based on the optimized parameter set and reads the parking demand prediction value and the number of available parking spaces M. The multi-objective scheduling optimization algorithm is expanded into a weight summation model. The predicted number of arriving vehicles is matched with the remaining parking spaces M to calculate the matching degree. The weight corresponds to the occupancy rate, corresponds to the time factor. In the parking lot instance , the resource allocation ratio is calculated. The comprehensive scheduling score is calculated for each cluster , where is the maximum stay duration of the entire area, which is assumed to be 60 minutes. The values are substituted into to obtain . The scores are sorted and divided into priority scheduling levels according to the score intervals. The high-level interval is defined as , the medium-level is , and the low-level is C1 is high, C2 is medium, C3 is low, and then generate a scheduling instruction set for each cluster according to the level, where high level allocates priority control ratio 0.5, medium level 0.3, and low level 0.2, and the calculation result is that the number of allocated parking spaces is 20, 12 and 8 respectively, and the final global parking optimization scheme is formed by combining all partitions.

[0059] The feedback adjustment module uses a closed-loop feedback control algorithm to perform dynamic adjustment on the front-end system based on the global parking optimization scheme. This module sends the optimization results to each RISC-V node to control the real-time adjustment of sampling frequency, communication mode and display logic, realizes system-level closed-loop adaptive control, and generates system adaptive adjustment signals. The feedback adjustment module includes a scheme analysis submodule, a parameter update submodule, a node control submodule, and a system synchronization submodule. The scheme analysis submodule analyzes the required adjustment parameters of each node based on the global parking optimization scheme and generates a control instruction set. The scheme analysis submodule based on the global parking optimization scheme first extracts the region number, vehicle allocation ratio, and parking space utilization threshold from the optimization scheme file. Let the region number set be A = {A1, A2, A3}, and each region corresponds to a parking space utilization threshold The range is set to [0.6, 0.9], and when It is considered as a saturation zone. During the analysis process, the control parameters of each node are read, including sampling interval , adjustment coefficient and signal power and other indicators. Assuming that A1 node , compared with A2 node , by comparing the difference between the current and the reference parameter table to determine whether the adjustment amount needs to be reset. When threshold , the update command is triggered, where Take 2 as the integer limit. Further group analysis of control parameters is performed, and the adjustment parameters of different nodes are mapped to the instruction field. The instruction frame is output in hexadecimal format, for example, node A1 generates control frame 0x01A50F corresponding to the adjustment sampling and power instructions. After mapping and encoding all nodes, a unified control instruction set is generated.

[0060] The parameter update submodule updates the RISC-V end control variables using an online parameter refresh algorithm based on the control instruction set and generates updated parameter data. The parameter update submodule executes the online refresh algorithm on the RISC-V end based on the control instruction set. First, parse each field in the control instruction set into a control variable table. Let the control parameters include Three categories, for each category of parameters, read the current register value. and instructions Perform difference comparison ,when If register data needs to be updated, a step-refresh method is used, with each step incrementing the value. To reduce the number of mutations, it is assumed The update step is 0.4 seconds, and the process is repeated 5 times to complete the refresh. Simultaneously, a normalized weight calculation method is used to set the update magnitude for the power parameter P, and the power weight is defined. The range is [0.5, 1.0], and the value is taken when the node is under high load. Medium load Low load Assuming node A1's current load rate λ = 0.75, which is between 0.7 and 0.9, it is considered to have a high load and is assigned a value accordingly. and new power Write to the register, compare the refreshed register data with the cache table. If they match, record the update flag flag=1; otherwise, flag=0. Finally, integrate and output the updated parameter data.

[0061] Based on the updated parameter data, the node control submodule executes node-level task control methods to adjust local sampling and communication strategies and generate node execution status. The node control submodule performs sampling and communication strategy adjustment processes at each node based on the updated parameter data. First, it extracts the node identifier ID, sampling period Δt, transmission power P, and adjustment coefficient β from the updated data. These parameters are then loaded into the configuration buffer of the node's local controller. Assuming node N3's current sampling period is 5s, and after the update it becomes 6s, the communication power is adjusted from 12dBm to 10dBm, and β is increased from 0.6 to 0.8, the number of sampling windows is recalculated based on the change in sampling period. ,in Assuming the scheduling period length is 60 seconds, then The sampling rate is changed from 12 to 10. The difference in the sampled data before and after the change is compared to determine whether the sampling rhythm is stable. The rhythm is considered stable when the difference between three consecutive rounds of sampling is less than the threshold ε=0.05; otherwise, it is automatically corrected. Simultaneously, it performs bandwidth allocation judgment on the communication strategy, and when the node occupancy rate... Defined as bandwidth weighting for high-load areas Between Defined as medium load Defined as low load Assuming an N3 load factor of 0.65, allocate a bandwidth weight of 0.6 and adjust the transmission rate to the baseline rate. The node records the status field {ID,Δt,P,β,R} and generates the node execution status.

[0062] The system synchronization submodule uses a closed-loop feedback synchronization algorithm based on the node execution status to unify the operating rhythm of the entire system and generate adaptive adjustment signals for the system.

[0063] The system synchronization submodule collects the timestamps of all nodes based on their execution status. Status indicators With feedback delay Calculate the global synchronization offset for each node. ,in The average timestamp of all nodes, when Synchronization threshold Determined as an out-of-synchronization node, threshold Let's set the time interval to 0.3 seconds. Assume the timestamps of the five nodes are [10.0, 10.2, 10.1, 10.5, 9.8] seconds. ,correspond If the offset between nodes 4 and 5 exceeds 0.3 seconds, a correction needs to be performed. The correction step size α is 0.5 times the offset, i.e., node 4 rolls back by 0.19 seconds and node 5 moves forward by 0.16 seconds, while simultaneously updating the system clock table to account for feedback delays. Synchronization levels are distinguished by their size. For high synchronization region, For the central synchronization zone, Assuming a low synchronization zone, node 1 is judged to be in the high synchronization zone with a feedback delay of 45ms, and node 3 is judged to be in the low synchronization zone with a delay of 110ms. The system records the synchronization level and correction parameters. After all node offsets are corrected, the average offset and delay level are aggregated to generate the system adaptive adjustment signal.

[0064] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A RISC-V based parking condition monitoring system, characterized in that, Comprise the following modules: data acquisition module, signal preprocessing module, fusion judgment module, state recognition module, predictive control module, energy consumption optimization module, communication transmission module, cloud analysis module, feedback regulation module; The data acquisition module is based on a RISC-V embedded control core and adopts a multi-source sensor synchronous sampling method to acquire real-time data of the parking space state. The module detects by combining ultrasonic, infrared and magnetic resistance sensors, uses a multi-sensor synchronous timing correction algorithm to correct the consistency of the data sampling clock, and ensures that the multi-source signals are collected under the same time reference to generate an original monitoring data set. The signal preprocessing module is based on the original monitoring data set and uses an adaptive filtering and noise suppression algorithm to denoise, smooth and normalize the collected signals. It dynamically filters different types of signals by using the sliding window mean method and Kalman filtering method, and uses an environmental noise estimation subroutine to realize adaptive threshold adjustment to generate a high-confidence signal set. The fusion judgment module is based on the high-confidence signal set and uses a multi-sensor fusion decision algorithm to fuse signal features. The system uses a Bayesian weighted fusion method to integrate the confidence of each sensor signal and introduces a correlation matrix analysis model to dynamically adjust the weight coefficient, thereby obtaining a comprehensive judgment of the single parking space state and generating parking space state fusion data. The state recognition module is based on the parking space state fusion data and uses an adaptive threshold decision algorithm to intelligently recognize the parking space occupancy. The system first filters suspicious states through a threshold dynamic adjustment mechanism, and then classifies and judges different occupancy features using a pattern matching recognition method to generate parking state recognition results.

2. The RISC-V based parking condition monitoring system according to claim 1, characterized in that, The data acquisition module includes an ultrasonic acquisition submodule, an infrared detection submodule, a magnetic resistance sensing submodule, and a timing correction submodule. The ultrasonic acquisition submodule performs active detection of parking space distance data and reflection wave time delay calculation to generate initial distance signals. The infrared detection submodule detects target occlusion and temperature difference to generate infrared perception signals. The magnetic resistance sensing submodule senses geomagnetic changes and judges the strength of vehicle metal interference to generate geomagnetic response signals. The timing correction submodule uses a synchronous clock correction algorithm to align the time reference based on the aforementioned three signals to generate an original monitoring data set.

3. The RISC-V based parking condition monitoring system according to claim 1, wherein, The signal preprocessing module includes a filtering processing submodule, a noise suppression submodule, a threshold adjustment submodule, and a normalization correction submodule. The filtering processing submodule uses a sliding window mean algorithm to smooth the original monitoring data set to generate preliminary denoised signals. The noise suppression submodule uses a Kalman filtering algorithm to estimate and compensate for environmental interference in real time to generate steady-state signals. The threshold adaptive submodule uses a dynamic threshold adjustment algorithm to correct the upper and lower limit fluctuations of the signals based on the steady-state signals to generate threshold optimized signals. The normalization correction submodule performs scale standardization and linear mapping based on the threshold optimized signals to generate a high-confidence signal set.

4. The RISC-V based parking condition monitoring system according to claim 1, wherein, The fusion judgment module includes a feature extraction submodule, a weighted fusion submodule, a confidence evaluation submodule, and a correlation analysis submodule. The feature extraction submodule extracts the main feature vector of the parking space state based on the high-confidence signal set to generate a feature matrix. The weighted fusion sub-module integrates multi-source features based on the feature matrix using a Bayesian weighted fusion algorithm to generate a fusion weighting result; The confidence evaluation sub-module calculates the credibility score of each sensing source based on the fusion weighting result to generate a confidence index set; The correlation analysis sub-module analyzes the final comprehensive state judgment using a correlation matrix analysis method based on the confidence index set to generate parking space state fusion data.

5. The RISC-V based parking condition monitoring system according to claim 1, wherein, The state recognition module includes a threshold judgment sub-module, a pattern matching sub-module, an anomaly detection sub-module, and a result classification sub-module; The threshold judgment sub-module performs preliminary occupancy recognition based on the parking space state fusion data using an adaptive threshold judgment algorithm to generate candidate state data; The pattern matching sub-module identifies typical state characteristics of idle, occupied, and abnormal based on the candidate state data using a pattern matching classification method to generate a pattern classification result; The anomaly detection sub-module identifies signal anomalies or false positives using a residual detection algorithm based on the pattern classification result to generate anomaly identification information; The result classification sub-module outputs a unified recognition conclusion by combining the pattern classification result and the anomaly identification information to generate a parking state recognition result.

6. The RISC-V based parking condition monitoring system according to claim 1, wherein, The prediction control module predicts future parking space usage trends using a time series prediction control algorithm based on the parking state recognition result. The system analyzes historical occupancy changes through sliding time window modeling to generate future parking space utilization prediction values, providing a basis for subsequent scheduling, and generating parking space utilization prediction data; The prediction control module includes a time modeling sub-module, a trend analysis sub-module, a short-term prediction sub-module, and a result evaluation sub-module; The time modeling sub-module constructs a historical time series model based on the parking state recognition result to generate a time series data set; The trend analysis sub-module analyzes the occupancy change trend using a sliding window trend extraction algorithm based on the time series data set to generate trend feature data; The short-term prediction sub-module predicts future parking space states using a short-term memory prediction model based on the trend feature data to generate a short-term prediction result; The result evaluation sub-module performs error evaluation and correction based on the short-term prediction result and historical true data to generate parking space utilization prediction data.

7. The RISC-V based parking condition monitoring system according to claim 1, wherein, The energy optimization module adjusts the dynamic power consumption of sensors and communication modules using an event-triggered energy-saving control method based on parking space utilization prediction data. When predicted traffic volume is low, the system automatically reduces the sampling frequency. When traffic volume rises, a high-frequency sampling mode is triggered. The core algorithm is a dynamic threshold power consumption adjustment mechanism to ensure the balance between energy consumption and response speed, generating energy-saving control instruction sets; The energy optimization module includes an event trigger sub-module, a power consumption evaluation sub-module, a threshold adjustment sub-module, and a task scheduling sub-module; The event trigger sub-module determines whether the system state change exceeds the threshold value using an event trigger detection method based on parking space utilization prediction data to generate an event trigger signal; The power consumption evaluation sub-module evaluates the real-time energy consumption level of the system using a power dynamic evaluation algorithm based on the event trigger signal to generate power consumption state data; The threshold adaptive sub-module adjusts the trigger threshold range using a dynamic sensitivity adjustment algorithm based on the power consumption state data to generate energy consumption threshold parameters; The task scheduling submodule adjusts the sampling and communication task rhythm based on the energy consumption threshold parameter, adopts a priority task scheduling algorithm, and generates an energy-saving control instruction set.

8. The RISC-V based parking condition monitoring system of claim 1, wherein, The communication transmission module adopts an asynchronous packet transmission algorithm based on the energy-saving control instruction set, realizes efficient wireless communication of the parking space state data and control instructions, dynamically selects the optimal communication path through the LoRa / NB-IoT dual-channel switching mechanism, and reduces the data transmission amount by combining the differential compression encoding method, and generates an encrypted transmission data packet. The communication transmission module includes a channel management submodule, an encryption and encoding submodule, a data encapsulation submodule, and a path selection submodule. The channel management submodule adopts an adaptive channel allocation algorithm based on the energy-saving control instruction set to allocate data transmission paths and generates channel configuration information. The encryption and encoding submodule adopts a differential compression encryption algorithm based on the channel configuration information to encode data and generates an encrypted signal stream. The data encapsulation submodule adopts a packet frame encapsulation method based on the encrypted signal stream to generate a data packet structure and generates encapsulated transmission data. The path selection submodule selects the optimal transmission path through a dual-channel dynamic switching algorithm based on the encapsulated transmission data and generates an encrypted transmission data packet.

9. The RISC-V based parking condition monitoring system according to claim 1, wherein, The cloud analysis module adopts a distributed parallel analysis algorithm in the cloud server to aggregate and count large-scale parking data based on the encrypted transmission data packet, groups the parking areas through clustering analysis and pattern recognition methods, dynamically adjusts the prediction parameters by using a self-learning optimization model, iteratively optimizes the intelligent scheduling strategy, and generates a global parking optimization scheme. The cloud analysis module includes a data decryption submodule, a clustering analysis submodule, a parameter learning submodule, and a scheduling optimization submodule. The data decryption submodule restores the original data stream by using a symmetric key decryption algorithm based on the encrypted transmission data packet and generates decrypted data. The clustering analysis submodule groups the parking area distribution by using a density clustering algorithm based on the decrypted data and generates a clustering result set. The parameter learning submodule updates the prediction and scheduling model parameters by using a self-learning parameter optimization algorithm based on the clustering result set and generates an optimized parameter set. The scheduling optimization submodule generates the final system control scheme by using a multi-objective scheduling optimization algorithm based on the optimized parameter set and generates a global parking optimization scheme.

10. The RISC-V based parking condition monitoring system according to claim 1, wherein, The feedback adjustment module dynamically adjusts the front-end system by using a closed-loop feedback control algorithm based on the global parking optimization scheme, sends the optimization results to each RISC-V node, controls the real-time adjustment of the sampling frequency, communication mode, and display logic, realizes system-level closed-loop adaptive control, and generates a system adaptive adjustment signal. The feedback adjustment module includes a scheme analysis submodule, a parameter update submodule, a node control submodule, and a system synchronization submodule. The scheme analysis submodule analyzes the required adjustment parameters of each node based on the global parking optimization scheme and generates a control instruction set. The parameter update submodule updates the RISC-V end control variables by using an online parameter refresh algorithm based on the control instruction set and generates updated parameter data. The node control submodule adjusts the local sampling and communication strategy by using a node-level task control method based on the updated parameter data and generates a node execution state. The system synchronization submodule adopts a closed-loop feedback synchronization algorithm to unify the operation rhythm of the whole system based on the node execution state, and generates a system adaptive adjustment signal.

Images