Street lamp power supply monitoring control system based on Internet of Things

Abstract

The invention provides a street lamp power supply monitoring control system based on the Internet of Things, and the system comprises the steps: training a time difference change trend through employing a neural network model according to a historical data set, and generating a prediction model which comprises a prediction time difference offset and a confidence interval; smoothing the predicted time difference offset by adopting a filtering algorithm according to the output of the prediction model to obtain a smooth time difference change sequence; according to the smooth time difference change sequence and the real-time environment data, fusing the long-term time difference trend and the short-term illumination fluctuation to generate a comprehensive control signal; according to the comprehensive control signal, adopting a weight fusion method to generate a comprehensive control factor, and determining switching time and a brightness adjustment factor; according to the comprehensive control factor, a dynamic programming algorithm is adopted to generate an optimization control instruction set containing a time point and a brightness level; and the optimization control instruction set is distributed to the street lamp terminal through the synchronous control module, and a real-time synchronous control signal is generated.

Description

Technical Field

[0001] The present invention relates to the technical field of street lamp power supply monitoring, and in particular to a street lamp power supply monitoring and control system based on the Internet of Things. Background Art

[0002] Smart city development is a crucial area for promoting sustainable urban development, and IoT technology plays a key role. This is particularly true for streetlight power supply monitoring and control systems, which are directly related to urban energy efficiency and the quality of public services. Efficient streetlight control not only optimizes energy use but also enhances the intelligence of urban operations. However, existing solutions have significant limitations when adapting to complex and changing environments. Many systems rely on fixed schedules or simple sensor triggers, making them difficult to adapt to dynamic demands. This is especially true in scenarios with time zone adjustments, where control strategies often lag or mismatch, resulting in energy waste or insufficient lighting.

[0003] Time zone fluctuations, a core challenge for smart street lighting systems, present multiple technical challenges. The unpredictable nature of time zone fluctuations requires a high degree of dynamic adaptability, but current systems often lack the ability to analyze historical data and future trends, making it impossible to plan control strategies in advance. This lack of predictive capabilities makes it difficult for the system to achieve smooth transitions when faced with seasonal time zone adjustments or policy-driven time zone changes, leading to frequent adjustments to control strategies. Frequent adjustments not only increase the system's computational burden but can also affect the real-time synchronization of street lights due to response delays. This lack of real-time synchronization further exacerbates the system's difficulty in adapting to special events, such as temporary time zone adjustments during international events, making it difficult to quickly adjust strategies based on pre-set rules.

[0004] Therefore, how to design an IoT street lamp power supply monitoring and control system that can predict time difference changes based on historical data, achieve smooth transitions and quickly adapt to special scenarios has become a key issue in smart city lighting management. Summary of the Invention

[0005] The present invention provides a street lamp power supply monitoring and control system based on the Internet of Things, which mainly includes: The street lamp power supply monitoring data and ambient lighting data collected by IoT sensors are obtained, and the city time zone adjustment records and seasonal time difference change data are integrated to construct a historical data set containing timestamps, energy consumption and light intensity; based on the historical data set, a neural network model is used to train the time difference change trend to generate a prediction model containing predicted time difference offset and confidence interval; based on the output of the prediction model, a filtering algorithm is used to smooth the predicted time difference offset to obtain a smoothed time difference change sequence; based on the smoothed time difference change sequence and real-time environmental data, long-term time difference trends and short-term light fluctuations are integrated to generate a comprehensive control signal; based on the comprehensive control signal, a weight fusion method is used to generate a comprehensive control factor to determine the switching time and brightness adjustment factor; based on the comprehensive control factor, a dynamic programming algorithm is used to generate an optimized control instruction set containing time points and brightness levels; the optimized control instruction set is distributed to the street lamp terminal through the synchronous control module to generate a real-time synchronous control signal.

[0006] Furthermore, a historical data set containing timestamps, energy consumption and light intensity is constructed, including: real-time collection of street lamp power supply monitoring data and ambient light data through Internet of Things sensors, integration of city time zone adjustment records and seasonal time difference change data, and construction of an initial data set; if the timestamp of the initial data set is missing or deviated, a time series interpolation algorithm is used to calibrate the timestamp to obtain a calibrated data set; for the energy consumption data of the calibration data set, a sliding window algorithm is used to detect outliers, and energy consumption exceeding a preset threshold is marked as an anomaly point to obtain an anomaly marked data set; based on the light intensity data of the anomaly marked data set, a clustering algorithm is used to divide the light intensity intervals, and the light intensity level corresponding to each timestamp is determined to obtain a hierarchical data set; timestamps, energy consumption and light intensity levels are extracted from the hierarchical data set, and a time series feature matrix is ​​constructed to obtain a feature data set.

[0007] Furthermore, a neural network model is used to train the time difference change trend, including: obtaining historical time difference data from a storage system, using data cleaning technology to remove outliers, and obtaining a standardized time difference data set; training the standardized time difference data set through a long short-term memory neural network, adjusting network parameters, and generating an initial time difference change trend model; using cross-validation technology to evaluate the initial time difference change trend model, calculating the prediction offset and confidence interval, and determining the prediction accuracy; if the prediction accuracy is lower than a preset threshold, adjusting the hyperparameters of the long short-term memory neural network, retraining the standardized time difference data set, and obtaining an optimized time difference change trend model; based on the optimized time difference change trend model, generating a prediction result including the time difference offset and confidence interval.

[0008] Furthermore, a filtering algorithm is used to smooth the predicted time difference offset, including: if the confidence interval of the prediction model output exceeds a preset threshold, the time difference offset is obtained from the prediction output to determine an initial offset sequence; the initial offset sequence is smoothed by an adaptive Kalman filtering algorithm to obtain a smoothed offset sequence; a time series analysis method is used to extract the trend of the smoothed offset sequence to obtain a time difference change trend; if the fluctuation amplitude of the time difference change trend exceeds a preset fluctuation threshold, the time difference change trend is secondary smoothed by a moving average algorithm to generate a stable change sequence; the autocorrelation coefficient is calculated based on the stable change sequence to determine the periodic time difference change pattern.

[0009] Furthermore, the long-term time difference trend and the short-term light fluctuation are integrated to generate a comprehensive control signal, including: by smoothing the time difference change sequence and the timestamp sequence, using the sliding window method to extract the trend characteristics of the time difference change sequence to obtain the long-term time difference trend; based on the long-term time difference trend, integrating the real-time environmental data, using the weighted average algorithm to calculate the adjustment coefficient of the environmental dynamics to the trend, and obtaining the adjusted time difference trend; if the fluctuation amplitude of the adjusted time difference trend exceeds the preset threshold, using the low-pass filtering method to smooth the trend data to obtain the smoothed time difference trend; through the light intensity data, using the fast Fourier transform algorithm to extract the frequency characteristics of the short-term light fluctuation to obtain the short-term light fluctuation; based on the smoothed time difference trend and the short-term light fluctuation, using the linear interpolation method to generate a smooth control signal.

[0010] Furthermore, a weight fusion method is used to generate a comprehensive control factor, including: obtaining a smooth time difference change sequence and real-time light intensity data, using data sequence analysis and environmental data processing to obtain a time difference offset and a light processing result; using a weight fusion method to perform weighted calculation on the time difference offset and the light processing result to generate a comprehensive control factor; if the comprehensive control factor is greater than a preset threshold, the switching time is determined; based on the comprehensive control factor, the brightness adjustment factor is calculated using the formula L=w1F+w2I, where L is the brightness adjustment factor, F is the comprehensive control factor, I is the light processing result, and w1 and w2 are preset weights to obtain the brightness adjustment factor; if the brightness adjustment factor exceeds the preset range, the light processing result is adjusted and the comprehensive control factor is recalculated.

[0011] Furthermore, a dynamic programming algorithm is used to generate an optimization control instruction set containing time points and brightness levels, including: obtaining ambient light intensity and traffic flow data through sensors, and denoising the data using a preprocessing method to obtain a first data set; if the light intensity of the first data set is lower than a preset threshold, a dynamic programming algorithm is used to generate a control strategy containing time points and brightness levels for the light intensity and traffic flow, and determine a first control strategy; according to the first control strategy, the time point and brightness level are extracted, a first optimization instruction is generated, and a first instruction set is output; if the time point of the first instruction set overlaps with a preset power-saving time period, the brightness level is adjusted through factor analysis, a second optimization instruction is generated, and a second instruction set is output.

[0012] Furthermore, the optimized control instruction set is distributed to the street lamp terminal through the synchronous control module, including: obtaining the optimized control instruction set through the synchronous control module, collecting feedback data from the street lamp terminal, and determining the instruction distribution status; if the feedback delay is greater than the preset threshold, processing the terminal feedback data through the edge computing node to generate a local control signal; adjusting the optimized control instruction set according to the local control signal to obtain an updated control instruction; distributing the updated control instruction to the street lamp terminal through the synchronous control module to obtain new terminal feedback data; using the K-means algorithm to perform cluster analysis on the terminal feedback data, determine the abnormal status of the feedback data, and generate a synchronous control signal.

[0013] The technical solution provided by the embodiment of the present invention may have the following beneficial effects: The present invention discloses an intelligent street lamp power supply monitoring and control system based on the Internet of Things. The data on street lamp power supply and ambient lighting are collected by Internet of Things sensors, and a historical data set is constructed in combination with information on city time zone adjustment and seasonal time difference changes. The time difference change trend is predicted using a long short-term memory neural network, and smoothing is performed using an adaptive Kalman filter algorithm. Based on the smoothed time difference change sequence and real-time environmental data, a comprehensive control signal is generated, and a weight fusion method is used to determine the control factor. A dynamic programming algorithm is used to generate an optimized control instruction set for street lamp switching time and brightness adjustment, which is distributed to the street lamp terminal through a real-time synchronous control module. For special scenarios, a reinforcement learning algorithm is used to generate a temporary control instruction set. The present invention also includes historical data analysis and fuzzy logic control, which realizes the intelligent, precise and dynamic adaptation of street lamp control, effectively improving energy utilization efficiency and lighting quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a flow chart of a street lamp power supply monitoring and control system based on the Internet of Things of the present invention.

[0015] Figure 2 This is a schematic diagram of a street lamp power supply monitoring and control system based on the Internet of Things of the present invention.

[0016] Figure 3 This is another schematic diagram of a street lamp power supply monitoring and control system based on the Internet of Things of the present invention.

[0017] Figure 4 This is another schematic diagram of a street lamp power supply monitoring and control system based on the Internet of Things of the present invention.

[0018] Figure 5 This is another schematic diagram of a street lamp power supply monitoring and control system based on the Internet of Things of the present invention.

[0019] Figure 6 This is another schematic diagram of a street lamp power supply monitoring and control system based on the Internet of Things of the present invention.

[0020] Figure 7 This is another schematic diagram of a street lamp power supply monitoring and control system based on the Internet of Things of the present invention. DETAILED DESCRIPTION

[0021] To further understand the content of the present invention, the present invention is described in detail with reference to the accompanying drawings and examples. The present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are intended only to illustrate the relevant invention and are not intended to limit the invention. It should also be noted that, for ease of description, only portions relevant to the invention are shown in the accompanying drawings.

[0022] See also Figures 1 to 7 As shown, in this embodiment, a street lamp power supply monitoring and control system based on the Internet of Things may specifically include: Step S101: Obtain street lamp power supply monitoring data and ambient light data collected by IoT sensors, integrate city time zone adjustment records and seasonal time difference change data, and construct a historical data set containing timestamps, energy consumption, and light intensity; By collecting street light power supply monitoring data through IoT sensors, we can obtain street light parameters such as current, voltage, and power in real time. These data can reflect the energy consumption of street lights. For example, the power of street lights on a certain street is 50 watts at 10 pm, but drops to 30 watts at 2 am. This change may be related to traffic flow or ambient light intensity. Collecting this data through sensors can provide a basis for subsequent energy consumption analysis. The collection of ambient light data depends on light sensors, which can monitor the light intensity of the surrounding environment in real time.

[0023] For example, at noon on a sunny day, the light intensity may reach 1000 lux, while on cloudy days or at night it may drop below 10 lux. These data can help determine whether it is necessary to turn on street lights or adjust the brightness of street lights, thereby achieving energy-saving goals. Urban time zone adjustment records refer to the adjustments made by cities to time zones in different seasons or special circumstances.

[0024] For example, some cities adopt daylight saving time in the summer, advancing the time by one hour, and return to standard time in the winter. This adjustment affects the switching time of street lights, so it needs to be included in the dataset to ensure the accuracy of the timestamp. The seasonal time difference change data reflects the changes in daylight hours in different seasons.

[0025] For example, in summer, when the daylight hours are longer, the streetlights can be turned on later; in winter, when the daylight hours are shorter, the streetlights need to be turned on earlier. These data can help optimize the streetlight control strategy to better meet actual needs. When constructing a historical dataset containing timestamps, energy consumption, and light intensity, the above data needs to be integrated.

[0026] For example, the timestamp of a certain day, the energy consumption data of street lights, and the ambient light intensity data are integrated together to form a complete data record. This data set can provide comprehensive information for subsequent analysis and prediction. The accuracy of the timestamp is crucial for data analysis. If the timestamp is missing or deviated, it may lead to inaccurate analysis results.

[0027] For example, if the timestamp of a record is missing, the timestamps of the previous and next records can be interpolated to infer the missing time point. This calibration method can ensure the integrity and accuracy of the data set. Outlier detection in energy consumption data is an important step in ensuring data quality.

[0028] For example, if the energy consumption value of a certain record suddenly soars to 1000 watts, while it is normally only 50 watts, this may be an anomaly caused by sensor failure or external interference. The sliding window algorithm can detect such anomalies and mark them as abnormal points to avoid interference with the analysis results. The division of light intensity levels can help better understand the changing patterns of ambient light.

[0029] For example, dividing light intensity into three levels: high, medium, and low can more intuitively analyze the energy consumption of street lamps under different lighting conditions. This grading method can provide a basis for subsequent optimization control. The construction of the time series feature matrix is ​​to convert data such as timestamps, energy consumption, and light intensity levels into features that can be used for analysis.

[0030] For example, the data of a certain day can be arranged in chronological order to form a matrix, where each row represents a time point and each column represents a feature. This matrix can provide a basis for time series analysis. Through the time series decomposition algorithm, the cyclical change trend of energy consumption can be separated.

[0031] For example, the energy consumption of street lights on a certain street fluctuates periodically every night. This fluctuation may be related to the periodic changes in traffic flow or ambient light intensity. Through the decomposition algorithm, this periodic trend can be extracted to provide a basis for subsequent predictions. The linear interpolation algorithm can be used to predict energy consumption changes at future timestamps.

[0032] For example, based on the energy consumption data of the past week, the energy consumption value for a certain day in the future can be predicted. This prediction method can provide a reference for the optimized control of street lights, thereby achieving energy saving goals.

[0033] Step S102: Based on the historical data set, a neural network model is used to train the time difference change trend to generate a prediction model including a predicted time difference offset and a confidence interval.

[0034] Based on historical data sets, a neural network model is used to train the time difference change trend and generate a prediction model that includes the predicted time difference offset and confidence interval. First, the historical data set is the basis for time difference change trend analysis. It usually includes timestamps and corresponding time difference values. This data may come from multiple sensors or systems and records the time difference changes at different time points. To ensure data quality, historical data needs to be cleaned to remove outliers and noise.

[0035] For example, if the time difference value at a certain point in time deviates significantly from the normal range, it may be due to sensor failure or external interference. Such data needs to be eliminated, and the cleaned data needs to be standardized to convert the time difference value to a unified scale to facilitate the training of the neural network model. Next, the standardized time difference dataset is trained using a long short-term memory neural network (LSTM). LSTM is a special recurrent neural network that can capture long-term dependencies in time series data. During the training process, the LSTM model learns the changing patterns of historical time difference data, gradually adjusts the network parameters, and generates an initial time difference change trend model.

[0036] For example, suppose historical data shows that the time difference tends to gradually increase between 10 a.m. and 12 p.m. every day. The LSTM model will learn this pattern and reflect it in the prediction. To evaluate the model's prediction accuracy, cross-validation technology is used to verify the initial model. Cross-validation divides the dataset into multiple subsets, and one of the subsets is used as the validation set in turn, and the remaining subsets are used as the training set to comprehensively evaluate the model's performance.

[0037] For example, the dataset is divided into 5 subsets, 4 subsets are used to train the model each time, and the remaining 1 subset is used for verification. This is repeated 5 times, and the average prediction error is calculated. If the prediction accuracy is lower than the preset threshold, for example, the average error exceeds 0.5 seconds, it is necessary to adjust the hyperparameters of the LSTM model, such as the learning rate and the number of hidden layer nodes, and retrain the model until the accuracy requirements are met. After the model training is completed, a prediction result containing the predicted time difference offset and the confidence interval is generated. The time difference offset refers to the difference between the predicted time difference and the actual time difference, and the confidence interval reflects the credible range of the prediction result.

[0038] For example, the model predicts that the time difference at a certain point in time is 1.2 seconds, and the confidence interval is 1.0 to 1.4 seconds, which means that there is a 95% probability that the actual time difference will fall within this range. The calculation of the confidence interval is usually based on the performance of the model on the validation set, combined with statistical methods. Finally, the prediction results are verified through time series analysis technology to determine whether they conform to the fluctuation pattern of historical data.

[0039] For example, if historical data shows that the time difference fluctuates periodically within a specific time period, and the prediction results also reflect this pattern, it means that the model has a high reliability. If the verification results show that the prediction results are consistent with the historical data, the final time difference change trend model is output for subsequent analysis.

[0040] Step S103: Based on the output of the prediction model, a filtering algorithm is used to smooth the predicted time difference offset to obtain a smoothed time difference variation sequence.

[0041] According to the output of the prediction model, a filtering algorithm is used to smooth the predicted time difference offset to obtain a smoothed time difference change sequence. First, the core idea of ​​the filtering algorithm is to reduce the influence of noise and outliers by weighted averaging the data, thereby obtaining a smoother data sequence. In the processing of time difference offset, the prediction model may produce unstable prediction results due to data fluctuations or noise. The filtering algorithm can effectively suppress these fluctuations and improve data stability.

[0042] For example, assuming that the time difference offset output by the prediction model is [1.2, 1.5, 1.8, 2.0, 1.7], which contains a certain amount of fluctuation, after processing by the filtering algorithm, a smoothed sequence [1.3, 1.4, 1.6, 1.8, 1.7] may be obtained, and the fluctuation amplitude is significantly reduced. The selection of the filtering algorithm is usually based on the data characteristics. For example, the Kalman filter is suitable for dynamic systems and can dynamically adjust the filtering parameters according to the prediction error to adapt to the changes in the data. In the smoothing processing of the time difference offset, the Kalman filter can dynamically adjust the filtering weight according to the confidence interval of the prediction model and the actual observation value to ensure the accuracy of the smoothed sequence.

[0043] For example, when the confidence interval of the prediction model is large, the Kalman filter will reduce its reliance on the predicted value and refer more to historical data, thereby avoiding smoothing errors caused by inaccurate predictions. The goal of smoothing is to reduce high-frequency noise in the data while retaining the trend information of the data. In the processing of time difference offset, the smoothed sequence can not only reflect the overall change trend of the time difference, but also provide a more reliable basis for subsequent trend extraction and periodic analysis.

[0044] For example, the smoothed time difference change sequence [1.3, 1.4, 1.6, 1.8, 1.7] can clearly show the gradual increase trend of the time difference without interfering with the analysis results due to local fluctuations. The parameter setting of the filtering algorithm has an important influence on the smoothing effect.

[0045] For example, in moving average filtering, the choice of window size requires a balance between smoothing effect and response speed. A window that is too large may lead to over-smoothing and loss of data details; a window that is too small may not effectively suppress noise. In processing time difference offsets, it is usually necessary to select an appropriate window size based on the fluctuation characteristics of the data. For example, a window size of 3 can effectively smooth the data without losing trend information. The smoothed time difference change series can provide a higher-quality data foundation for subsequent analysis.

[0046] For example, in time series analysis, smoothed sequences can more accurately reflect the long-term trends and cyclical characteristics of time differences, providing strong support for optimizing forecasting models. Smoothing the predicted time difference offset through filtering algorithms not only improves data stability but also provides a more reliable basis for subsequent analysis and decision-making.

[0047] Step S104 , based on the smoothed time difference variation sequence and the real-time environmental data, the long-term time difference trend and the short-term illumination fluctuation are integrated to generate a comprehensive control signal.

[0048] The process of generating a comprehensive control signal by integrating the long-term time difference trend and short-term illumination fluctuation based on the smoothed time difference change sequence and real-time environmental data can be discussed from multiple aspects. First, the smoothed time difference change sequence is extracted through the sliding window method. The size of the sliding window can be adjusted according to actual needs. For example, the window size is selected as 10 time points, which can capture the overall trend of the time difference change while avoiding the influence of local fluctuations. The long-term time difference trend extracted by the sliding window can reflect the overall direction of the time difference change. For example, the time difference change in a day may show a trend of gradually increasing or decreasing. Next, the environmental dynamics in the real-time environmental data have an impact on the long-term The adjustment of time difference trends plays an important role. The environmental dynamics can be calculated through a weighted average algorithm. For example, the weights of factors such as ambient temperature and humidity are set to 0.6 and 0.4 respectively. This can comprehensively consider the impact of different environmental factors on the time difference trend. If the fluctuation amplitude of the adjusted time difference trend is too large, for example, exceeding the preset threshold of 0.5, a low-pass filtering method is used for smoothing. Low-pass filtering can remove high-frequency noise and retain low-frequency trends, making the time difference trend more stable. The extraction of short-term light fluctuations depends on the fast Fourier transform algorithm, which can convert light intensity data from the time domain to the frequency domain, thereby extracting the frequency characteristics of light fluctuations.

[0049] For example, light intensity data may contain fluctuations with a period of 5 minutes. Fast Fourier transforms can accurately identify this frequency feature. Extracting short-term light fluctuations can reflect rapid changes in light intensity. For example, when clouds move, light intensity can change significantly in a short period of time. When fusing long-term time difference trends and short-term light fluctuations, linear interpolation can effectively combine the two.

[0050] For example, the long-term time difference trend may show that the time difference is gradually increasing, while the short-term light fluctuation shows that the light intensity has decreased in a short period of time. A preliminary control signal can be generated through linear interpolation. This signal contains both the long-term trend of the time difference change and the short-term fluctuation of the light intensity. If the number of mutation points in the preliminary control signal exceeds the preset threshold, for example, more than 3, the median filtering method is used to eliminate these mutation points. The median filtering can effectively remove outliers and make the control signal smoother. Finally, the signal amplitude is adjusted through normalization processing to generate a comprehensive control signal. The normalization processing can adjust the amplitude of the control signal to between 0 and 1, for example, adjusting the maximum value of the control signal to 1 and the minimum value to 0. This can ensure that the control signal has consistent performance under different environmental conditions. The generation of the comprehensive control signal can comprehensively consider the influence of time difference changes and light intensity, thereby providing a more accurate control factor for switching time and brightness adjustment.

[0051] Step S105 , based on the integrated control signal, a weight fusion method is used to generate an integrated control factor to determine the switching time and the brightness adjustment factor; based on the integrated control factor, a dynamic programming algorithm is used to generate an optimized control instruction set including time points and brightness levels.

[0052] The implementation process of generating a comprehensive control factor using the weighted fusion method can be illustrated by the following example: Assuming the time difference offset data sequence analysis result is 0.8 (normalized value), the real-time light intensity processing result is 0.6, and the preset weights w1 and w2 are 0.7 and 0.3, the comprehensive control factor F = 0.7 × 0.8 + 0.3 × 0.6 = 0.74. If the threshold is set to 0.7, since F > 0.7, the control signal is generated, and a dynamic programming algorithm is required. Dynamic programming optimizes switching times and brightness levels through staged decision-making. For example, if the night is divided into six periods, each lasting 2 hours, with an initial brightness level of 30%, the algorithm iteratively calculates the optimal brightness adjustment value for each period based on the comprehensive control factor. For example, if the brightness is increased to 50% in the first period and reduced to 20% in the third period, the final output instruction set is [(20:00, 50%), (22:00, 40%), (00:00, 20%)]. The specific application of the dynamic programming algorithm in generating the optimized control instruction set is reflected in the period division and state transition design. Taking streetlight control as an example, the state is defined as the light intensity and energy cost of the current time period, and the decision variable is the brightness adjustment range. The algorithm works backwards from the end of the time period. For example, if the light intensity drops sharply between 22:00 and 24:00, the state transitions to "high energy demand," at which point the brightness level must be increased to 70%. Conversely, during the 04:00-06:00 period, the state transitions to "low energy demand," with the brightness level reduced to 10%. By comparing the cumulative costs of each state (e.g., the weighted sum of energy consumption and comfort), the path with the lowest cost is selected, ultimately generating the instruction set [(04:00, 10%), (05:30, 30%)]. The synergy between brightness adjustment factor calculation and dynamic programming can be demonstrated through an environmental data feedback mechanism. When the brightness adjustment factor L = 0.7 × 0.74 + 0.3 × 0.6 = 0.698 exceeds the preset range [0.2, 0.6], the lighting processing results must be readjusted. For example, multiplying the original illumination data (0.6) by the correction factor (0.8) yields a new value (0.48). The comprehensive control factor (F) is recalculated as 0.7 × 0.8 + 0.3 × 0.48 = 0.704. At this point, L = 0.7 × 0.704 + 0.3 × 0.48 = 0.637, which still exceeds the limit. The dynamic programming brightness level constraint is then further adjusted, compressing the maximum brightness from 80% to 60%, ultimately generating the revised instruction set [(20:00, 45%), (22:00, 35%)]. The combined optimization effect of weight fusion and dynamic programming is reflected in the multi-objective balance. For example, during rush hour (high time difference offset) and rainy weather (low light intensity), the weight distribution focuses on the time difference data (w1=0.8), and the comprehensive control factor tends to extend the lighting time; dynamic planning increases the brightness from 21:00 to 23:00 to 60% based on this factor, while reducing the brightness during non-peak hours to 15%, ensuring safety and reducing energy consumption.Relying solely on single lighting data may result in over-dimming or response delays. However, after integrating time difference data, the system can predict the changing trend of pedestrian flow and adjust the brightness level in advance.

[0053] Step S106: Distribute the optimized control instruction set to the street lamp terminal through the synchronous control module to generate a real-time synchronous control signal.

[0054] When distributing optimized control instruction sets to streetlight terminals via the synchronization control module, a low-latency communication link must first be established. For example, the LoRaWAN protocol is used for data transmission between the terminal and the module. The channel bandwidth is set to 125kHz, and the transmission interval is 5 seconds. This ensures that the instruction set, encapsulated in JSON format, can be delivered within 200ms. During the distribution process, the module attaches a timestamp and sequence number to each instruction. Upon receipt, the terminal immediately returns an acknowledgment frame. If the synchronization control module does not receive an acknowledgment within 500ms, a command retransmission mechanism is triggered, with a maximum retransmission limit of three to ensure reliable instruction delivery. The key to generating real-time synchronization control signals lies in time alignment and state consistency.

[0055] For example, when the instruction set requires that streetlights in a certain area be uniformly turned on to 70% brightness at 6:30 PM, the synchronization control module calibrates the terminal clocks using the NTP protocol to ensure that the time error for all terminals is less than 50ms. Simultaneously, the module monitors the terminal's voltage and current feedback data. If it detects that the actual brightness value of a terminal is 65% and has not reached the target for two seconds, it immediately generates an incremental adjustment signal (such as a +5% brightness instruction) and quickly reissues it through the edge node to synchronize the terminal's status with other terminals. To handle high-concurrency command distribution, the synchronization control module adopts a multi-threaded architecture. The main thread is responsible for command parsing and prioritizing (e.g., emergency fault commands take precedence), and sub-threads issue commands in parallel by terminal group.

[0056] For example, when a control involves 1,000 terminals, the module divides them into 10 thread groups, each with 100 terminals. Threads synchronize progress via semaphores to ensure that all terminals complete command reception within 2 seconds. During the distribution process, the module calculates the completion rate of each thread group in real time. If a group's completion rate falls below 90%, other thread resources are dynamically deployed to assist, preventing delays in individual terminals from impacting overall synchronization. The generation of real-time synchronous control signals also relies on environmental state awareness.

[0057] For example, the module connects to a meteorological data interface. If it detects a sudden drop in light intensity to 50 lux due to a sudden downpour, it immediately overwrites the original instruction set and generates a synchronization signal to immediately turn on the system to 100% brightness. The module also pre-calculates the startup delay for each terminal (for example, a high-pressure sodium lamp requires a 30-second warm-up period) through the edge node. Time-sharing startup parameters are embedded in the signal, allowing the terminals to start at staggered intervals of 0.5 seconds to prevent the grid's instantaneous load from exceeding the threshold. The synchronous control module's fault tolerance mechanism is implemented through heartbeat packets and status snapshots. Every 60 seconds, the module sends a heartbeat packet to the terminal, which responds with a snapshot containing data such as current brightness and energy consumption. If a terminal fails to respond three times in a row, the module marks it offline and automatically switches to redundant control mode with a neighboring terminal.

[0058] For example, if Terminal A fails, the module retrieves coverage data from Terminal B and increases its brightness from 80% to 95% to compensate for Terminal A's blind spot. It also records the abnormal status for subsequent maintenance. The real-time nature of command distribution and signal generation relies on hardware acceleration.

[0059] For example, the module's built-in FPGA chip performs hardware-level optimization for operations such as instruction set CRC checking and data encapsulation, reducing the processing time of a single instruction from 10ms in software to 1ms. Furthermore, the FPGA directly drives the Gigabit Ethernet port, keeping instruction distribution latency to less than 0.5ms, ensuring synchronization requirements can be met even in large-scale terminal scenarios. Dynamic load balancing is key to ensuring synchronization performance. The module monitors network traffic in real time and automatically switches some terminals to a backup frequency band when the load on a communication link exceeds 80%.

[0060] For example, when a group of terminals originally using the 470MHz band experiences congestion, the module temporarily switches 20% of them to the 868MHz band. Edge nodes then recalculate the signal synchronization offset between the two groups of terminals (for example, if the 470MHz group delay increases by 10ms, the 868MHz group's instructions are sent 10ms earlier). Ultimately, this keeps the action time difference for all terminals within ±20ms. To facilitate large-scale distribution of instruction sets, the module employs a hierarchical aggregation strategy.

[0061] For example, when an update involves 10,000 terminals, the module first divides the terminals into 100 clusters based on geographic region, with edge nodes acting as agents in each cluster. During the first round of synchronization, the module sends instructions only to the 100 agent nodes, which then distribute them to the terminals within the cluster via the local network. The synchronization control module aggregates terminal status through the agent nodes. If the synchronization failure rate for a cluster exceeds 5%, the module directly intervenes to redistribute instructions to that cluster, forming a three-level coordination mechanism: "center-edge-terminal." During signal generation, the module incorporates multi-dimensional constraints.

[0062] For example, when generating a synchronization signal to reduce brightness to 30% at midnight, the module considers parameters such as municipal regulations (brightness must not fall below 20%), terminal aging (old fixtures must support a minimum of 35%), and grid load (load must decrease by 40% during this period). A constraint-solving algorithm generates a compromise command to ensure the signal is both legal and executable. The module also pre-calculates the energy savings of the command (e.g., 50 kWh per hour) and embeds this data as metadata into the signal for statistical analysis by the operations and maintenance platform.

[0063] Step S107, constructing a historical data set containing timestamps, energy consumption and light intensity, including: collecting street lamp power supply monitoring data and ambient light data in real time through Internet of Things sensors, integrating city time zone adjustment records and seasonal time difference change data to construct an initial data set; if the timestamp of the initial data set is missing or deviated, using a time series interpolation algorithm to calibrate the timestamp to obtain a calibrated data set; using a sliding window algorithm to detect outliers for the energy consumption data of the calibration data set, marking energy consumption exceeding a preset threshold as an outlier, and obtaining an abnormal marked data set; using a clustering algorithm to divide the light intensity intervals according to the light intensity data of the abnormal marked data set, determining the light intensity level corresponding to each timestamp, and obtaining a hierarchical data set; extracting timestamps, energy consumption and light intensity levels from the hierarchical data set, constructing a time series feature matrix, and obtaining a feature data set.

[0064] IoT sensors collect real-time streetlight power supply monitoring data and ambient light data. The sensors record voltage, current, and light intensity once a minute. For example, a voltage fluctuation of 220V ± 5V, a current of 1.2A, and a light intensity of 300lux were recorded. City time zone adjustment records include daylight saving time switchover times (e.g., 2:00 AM to 3:00 AM on the last Sunday in March), while seasonal time difference data provides information on changes in sunrise and sunset times (e.g., dark one hour earlier in winter than in summer).

[0065] When fusing this data, the original timestamps of the sensors are aligned with the UTC time, and then the time zone offset (such as UTC+8) is added. This ultimately generates an initial dataset containing accurate local time, energy consumption (voltage × current), and light intensity. If the timestamps of the initial dataset are missing for 5 minutes due to network latency, a linear interpolation algorithm is used for correction: assuming that the timestamps before and after the missing segment are 08:00 (light intensity 200 lux) and 08:10 (light intensity 500 lux), the timestamp of 08:05 is complemented to 08:05. Light intensity is linearly interpolated to 350 lux. If a time zone adjustment results in duplicate timestamps (e.g., two 02:30 times due to a daylight saving time switch), the duplicate data is deleted and marked as a system event. For the calibrated energy consumption data, a sliding window algorithm uses a 30-minute window to detect anomalies. If the mean energy consumption within the window is 0.25 kWh and the standard deviation is 0.05 kWh, a threshold of 0.4 kWh (mean + 3σ) is set. Any value exceeding 0.6 kWh is marked as an anomaly, possibly due to a short circuit or sensor failure.

[0066] After anomaly marking, the original data is retained but an anomaly flag is added to facilitate subsequent analysis. K-means clustering is used to classify the light intensity data. Assuming the cluster centers are 50 lux (nighttime), 300 lux (dusk), and 1000 lux (daytime), the 350 lux at 08:05 belongs to the dusk level. Seasonal differences need to be considered when clustering. Dusk light can decrease by 20% in winter, so the cluster center threshold is dynamically adjusted. After extracting timestamps, energy consumption, and light levels from the hierarchical dataset, a 24-hour × 3 feature matrix (mean energy consumption, maximum energy consumption, and light level) is constructed. For example, the 08:00–09:00 period in the matrix corresponds to a mean energy consumption of 0.3 kWh, a peak of 0.5 kWh, and a light level of 2 (dusk). This matrix can reflect the correlation between diurnal energy consumption fluctuations and light intensity.

[0067] Step S108, using a neural network model to train the time difference change trend, including: obtaining historical time difference data from a storage system, using data cleaning technology to remove outliers, and obtaining a standardized time difference data set; training the standardized time difference data set through a long short-term memory neural network, adjusting network parameters, and generating an initial time difference change trend model; using cross-validation technology to evaluate the initial time difference change trend model, calculating the prediction offset and confidence interval, and determining the prediction accuracy; if the prediction accuracy is lower than a preset threshold, adjusting the hyperparameters of the long short-term memory neural network, retraining the standardized time difference data set, and obtaining an optimized time difference change trend model; based on the optimized time difference change trend model, generating a prediction result including the time difference offset and confidence interval.

[0068] Obtaining historical time difference data from the storage system is the first step in building a time difference trend model. Historical time difference data typically contains timestamps and corresponding time difference values. This data may come from multiple sensors or systems and may contain noise or outliers.

[0069] For example, a system records hourly time deviation data, but due to equipment failure or network delay, the data at certain time points may deviate significantly from the normal range. In order to ensure the reliability of the data, data cleaning technology is needed to remove outliers. Data cleaning can be achieved through statistical methods, such as calculating the mean and standard deviation of the data, and treating data that deviates from the mean by more than three times the standard deviation as outliers and eliminating them. The cleaned data needs to be standardized to convert the time difference values ​​into a unified dimension, such as mapping the data to the range of 0 to 1 through normalization to facilitate subsequent neural network training. Training the standardized time difference dataset through a long short-term memory neural network is the core step in building the model. The long short-term memory neural network (LSTM) is a special recurrent neural network that can capture long-term dependencies in time series data. During the training process, the LSTM network gradually adjusts its internal parameters, such as weights and biases, to minimize prediction errors by learning the changing patterns of historical time difference data.

[0070] For example, assuming that historical time difference data exhibits periodic fluctuations, the LSTM network can remember this periodic pattern through its memory units and use this pattern to generate more accurate results during prediction. During the training process, it is necessary to set an appropriate learning rate and number of iterations to ensure that the model can fully learn the data features without overfitting. Using cross-validation technology to evaluate the initial time difference change trend model is an important step to ensure the model's generalization ability. Cross-validation divides the dataset into multiple subsets, and each subset is used as a validation set in turn, and the remaining subsets are used as training sets. The model is trained and validated multiple times.

[0071] For example, split the dataset into five parts, using four parts for training each time and one part for validation. After repeating this five times, calculate the average prediction accuracy. Through cross-validation, you can calculate the model's prediction bias and confidence interval. The prediction bias represents the difference between the model's predicted value and the actual value, while the confidence interval indicates the reliability range of the predicted value.

[0072] For example, the results of a certain cross-validation show that the average prediction offset of the model is 0.1 seconds, and the confidence interval is ±0.05 seconds, indicating that the model prediction results are relatively accurate. If the prediction accuracy is lower than the preset threshold, it is necessary to adjust the hyperparameters of the long short-term memory neural network and retrain the data set. Hyperparameters include the number of network layers, the number of neurons, the learning rate, etc. These parameters directly affect the performance of the model.

[0073] For example, if the initial model's prediction accuracy is 85% and the preset threshold is 90%, the model can be optimized by increasing the number of network layers or adjusting the learning rate. The optimized model needs to be retrained and cross-validated to ensure that its prediction accuracy meets the requirements.

[0074] For example, after adjustments, the model's prediction accuracy increased to 92%, meeting the preset threshold. Based on the optimized time difference trend model, the model generates a prediction result containing a time difference offset and a confidence interval. The time difference offset represents the predicted time difference value at a certain point in the future, while the confidence interval indicates the possible fluctuation range of the predicted value.

[0075] For example, the model predicts that the time difference offset in the next hour will be 0.2 seconds, with a confidence interval of ±0.1 seconds, indicating that the time difference may fluctuate between 0.1 seconds and 0.3 seconds. This prediction result can provide a reference for the system, helping it to adjust the time synchronization strategy in advance to avoid affecting system performance due to excessive time difference.

[0076] Step S109, using a filtering algorithm to smooth the predicted time difference offset, including: if the confidence interval of the prediction model output exceeds a preset threshold, obtaining the time difference offset from the prediction output to determine an initial offset sequence; smoothing the initial offset sequence through an adaptive Kalman filter algorithm to obtain a smoothed offset sequence; using a time series analysis method to extract the trend of the smoothed offset sequence to obtain a time difference change trend; if the fluctuation amplitude of the time difference change trend exceeds a preset fluctuation threshold, performing secondary smoothing on the time difference change trend through a moving average algorithm to generate a stable change sequence; calculating the autocorrelation coefficient based on the stable change sequence to determine the periodic time difference change pattern.

[0077] Step S1010, fusing the long-term time difference trend and the short-term light fluctuation to generate a comprehensive control signal, including: extracting the trend characteristics of the time difference change sequence by smoothing the time difference change sequence and the timestamp sequence using a sliding window method to obtain the long-term time difference trend; fusing the real-time environmental data based on the long-term time difference trend, and calculating the adjustment coefficient of the environmental dynamics to the trend using a weighted average algorithm to obtain the adjusted time difference trend; if the fluctuation amplitude of the adjusted time difference trend exceeds a preset threshold, smoothing the trend data using a low-pass filtering method to obtain a smoothed time difference trend; extracting the frequency characteristics of the short-term light fluctuation using a fast Fourier transform algorithm through the light intensity data to obtain the short-term light fluctuation; and generating a smooth control signal based on the smoothed time difference trend and the short-term light fluctuation using a linear interpolation method.

[0078] Step S1011, using a weight fusion method to generate a comprehensive control factor, including: obtaining a smoothed time difference change sequence and real-time light intensity data, using data sequence analysis and environmental data processing to obtain a time difference offset and a light processing result; using a weight fusion method to perform weighted calculation on the time difference offset and the light processing result to generate a comprehensive control factor; if the comprehensive control factor is greater than a preset threshold, determining the switching time; based on the comprehensive control factor, using a brightness adjustment factor calculation formula L=w1F+w2I, where L is the brightness adjustment factor, F is the comprehensive control factor, I is the light processing result, w1 and w2 are preset weights, to obtain the brightness adjustment factor; if the brightness adjustment factor exceeds the preset range, adjusting the light processing result and recalculating the comprehensive control factor.

[0079] Step S1012, using a dynamic programming algorithm to generate an optimization control instruction set including time points and brightness levels, including: obtaining ambient light intensity and traffic flow data through sensors, and using a preprocessing method to denoise the data to obtain a first data set; if the light intensity of the first data set is lower than a preset threshold, using a dynamic programming algorithm to generate a control strategy including time points and brightness levels for the light intensity and traffic flow, and determining a first control strategy; extracting the time points and brightness levels according to the first control strategy, generating a first optimization instruction, and outputting a first instruction set; if the time points of the first instruction set overlap with a preset power-saving time period, adjusting the brightness level through factor analysis, generating a second optimization instruction, and outputting a second instruction set.

[0080] Step S1013, distributing the optimized control instruction set to the street lamp terminal through the synchronous control module, including: obtaining the optimized control instruction set through the synchronous control module, collecting feedback data from the street lamp terminal, and determining the instruction distribution status; if the feedback delay is greater than the preset threshold, processing the terminal feedback data through the edge computing node to generate a local control signal; adjusting the optimized control instruction set according to the local control signal to obtain an updated control instruction; distributing the updated control instruction to the street lamp terminal through the synchronous control module to obtain new terminal feedback data; using the K-means algorithm to perform cluster analysis on the terminal feedback data, determine the abnormal status of the feedback data, and generate a synchronous control signal.

Claims

1. A street lamp power supply monitoring and control system based on the Internet of Things, characterized in that: include: Obtain streetlight power supply monitoring data and ambient light data collected by IoT sensors, integrate city time zone adjustment records and seasonal time difference change data, and build a historical dataset containing timestamps, energy consumption, and light intensity; Based on historical data sets, a neural network model is used to train the time difference change trend and generate a prediction model that includes the predicted time difference offset and confidence interval; According to the output of the prediction model, the predicted time difference offset is smoothed by using a filtering algorithm to obtain a smoothed time difference change sequence; Based on the smoothed time difference change sequence and real-time environmental data, the long-term time difference trend and short-term light fluctuation are integrated to generate a comprehensive control signal; According to the comprehensive control signal, the weight fusion method is used to generate the comprehensive control factor to determine the switching time and brightness adjustment factor; According to the comprehensive control factors, a dynamic programming algorithm is used to generate an optimized control instruction set including time points and brightness levels; The optimized control instruction set is distributed to the street lamp terminal through the synchronous control module to generate real-time synchronous control signals.

2. The street lamp power supply monitoring and control system based on the Internet of Things according to claim 1 is characterized in that: Construct a historical dataset containing timestamps, energy consumption, and light intensity. This includes: collecting real-time streetlight power supply monitoring data and ambient light data through IoT sensors, integrating city time zone adjustment records and seasonal time difference change data to construct an initial dataset; If the timestamps of the initial dataset are missing or deviated, the timestamps are calibrated using a time series interpolation algorithm to obtain a calibrated dataset; For the energy consumption data of the calibration dataset, a sliding window algorithm is used to detect outliers, and energy consumption exceeding the preset threshold is marked as an outlier to obtain an anomaly marked dataset; According to the light intensity data of the abnormal marked dataset, a clustering algorithm is used to divide the light intensity intervals, determine the light intensity level corresponding to each time stamp, and obtain a graded dataset; The timestamps, energy consumption and light intensity levels are extracted from the hierarchical dataset, and the time series feature matrix is ​​constructed to obtain the feature dataset.

3. The street lamp power supply monitoring and control system based on the Internet of Things according to claim 1 is characterized in that: Using a neural network model to train time difference change trends, including: obtaining historical time difference data from the storage system, using data cleaning technology to remove outliers, and obtaining a standardized time difference data set; The standardized time difference data set is trained through a long short-term memory neural network, the network parameters are adjusted, and an initial time difference change trend model is generated; Cross-validation techniques were used to evaluate the initial time difference trend model, calculate the prediction offset and confidence interval, and determine the prediction accuracy; If the prediction accuracy is lower than the preset threshold, the hyperparameters of the long short-term memory neural network are adjusted, and the standardized time difference dataset is retrained to obtain an optimized time difference change trend model; Based on the optimized time difference change trend model, a prediction result including the time difference offset and confidence interval is generated.

4. The street lamp power supply monitoring and control system based on the Internet of Things according to claim 1, characterized in that: A filtering algorithm is used to smooth the predicted time difference offset, including: if the confidence interval of the prediction model output exceeds a preset threshold, the time difference offset is obtained from the prediction output to determine the initial offset sequence; The initial offset sequence is smoothed by an adaptive Kalman filter algorithm to obtain a smoothed offset sequence; The time series analysis method is used to extract the trend of the smoothed offset sequence and obtain the time difference change trend; If the fluctuation amplitude of the time difference change trend exceeds the preset fluctuation threshold, the time difference change trend is secondary smoothed by the moving average algorithm to generate a stable change sequence; The autocorrelation coefficient is calculated based on the stable change sequence to determine the periodic time difference change pattern.

5. The street lamp power supply monitoring and control system based on the Internet of Things according to claim 1 is characterized in that: The long-term time difference trend and short-term light fluctuation are integrated to generate a comprehensive control signal, including: smoothing the time difference change sequence and the timestamp sequence, and using a sliding window method to extract the trend characteristics of the time difference change sequence to obtain the long-term time difference trend; According to the long-term time difference trend, the real-time environmental data is integrated and the weighted average algorithm is used to calculate the adjustment coefficient of the environmental dynamics to the trend to obtain the adjusted time difference trend; If the fluctuation amplitude of the adjusted time difference trend exceeds the preset threshold, the trend data is smoothed by a low-pass filtering method to obtain a smoothed time difference trend; Through the light intensity data, the fast Fourier transform algorithm is used to extract the frequency characteristics of short-term light fluctuations to obtain short-term light fluctuations; According to the smoothed time difference trend and short-term light fluctuation, a linear interpolation method is used to generate a smooth control signal.

6. The street lamp power supply monitoring and control system based on the Internet of Things according to claim 1, characterized in that: A weight fusion method is used to generate a comprehensive control factor, including: obtaining a smoothed time difference change sequence and real-time light intensity data, and using data sequence analysis and environmental data processing to obtain the time difference offset and light processing results; The weight fusion method is used to perform weighted calculation on the time difference offset and illumination processing results to generate a comprehensive control factor; If the comprehensive control factor is greater than the preset threshold, the switching time is determined; According to the comprehensive control factor, the brightness adjustment factor is calculated using the formula L=w1F+w2I, where L is the brightness adjustment factor, F is the comprehensive control factor, I is the lighting processing result, and w1 and w2 are preset weights to obtain the brightness adjustment factor; If the brightness adjustment factor exceeds the preset range, the lighting processing result is adjusted and the comprehensive control factor is recalculated.

7. The street lamp power supply monitoring and control system based on the Internet of Things according to claim 1, characterized in that: A dynamic programming algorithm is used to generate an optimized control instruction set containing time points and brightness levels, including: Acquire ambient light intensity and traffic flow data through sensors, and use a preprocessing method to denoise the data to obtain a first data set; If the light intensity of the first data set is lower than a preset threshold, a dynamic programming algorithm is used to generate a control strategy including time points and brightness levels based on the light intensity and traffic flow, and determine a first control strategy; Extracting the time point and the brightness level according to the first control strategy, generating a first optimization instruction, and outputting a first instruction set; If the time point of the first instruction set overlaps with the preset power-saving time period, the brightness level is adjusted through factor analysis, a second optimization instruction is generated, and the second instruction set is output.

8. The street lamp power supply monitoring and control system based on the Internet of Things according to claim 1, characterized in that: Distributing the optimized control instruction set to the street lamp terminal through the synchronous control module, including: obtaining the optimized control instruction set through the synchronous control module, collecting feedback data from the street lamp terminal, and determining the instruction distribution status; If the feedback delay is greater than the preset threshold, the terminal feedback data is processed by the edge computing node to generate a local control signal; Adjust and optimize the control instruction set according to the local control signal to obtain an updated control instruction; Distribute the updated control instructions to the street lamp terminals through the synchronous control module to obtain new terminal feedback data; The K-means algorithm is used to perform cluster analysis on the terminal feedback data, determine the abnormal state of the feedback data, and generate synchronous control signals.

Images