A street lamp energy-saving control method and system based on multi-modal data fusion

Abstract

This invention relates to the field of lighting control technology, and discloses a street light energy-saving control method and system based on multimodal data fusion. The method includes: predictively updating the expected traffic density index within a short future time window based on the rate of change of the dynamic traffic density index in a historical traffic condition feature queue; embedding the real-time traffic condition features of the street light area and the prediction into a multi-dimensional contextual feature vector; generating a dynamic brightness control benchmark value for the street light based on the multi-dimensional contextual feature vector, and distributing the dynamic brightness control benchmark value to the target street light; this invention can improve the efficiency of street light energy-saving control.

Description

Technical Field

[0001] This invention relates to the field of lighting control technology, and in particular to a street light energy-saving control method and system based on multimodal data fusion. Background Technology

[0002] Traditional street lighting control adopts a single control mode, which only adjusts brightness by pre-setting fixed time periods or collecting single environmental parameters such as light intensity. It does not integrate multi-dimensional data such as traffic conditions detected by millimeter-wave radar and slow traffic status recognized by temperature distribution images. It cannot combine real-time road traffic density, pedestrian and non-motorized vehicle traffic status and other real road traffic information to build lighting adjustment logic adapted to the actual scenario, and cannot achieve precise lighting adaptation control that is highly matched with road traffic needs.

[0003] Existing street light control technologies lack a short-time window traffic density index prediction and update system based on historical road condition feature queues, thus lacking the ability to predict future traffic conditions. The data processing process lacks dynamic calibration mechanisms such as queue length control and time coverage width maintenance. The brightness control benchmark value adopts a fixed and rigid generation method without dynamic adjustment and compensation links, failing to balance the dual needs of lighting safety and energy conservation. The overall efficiency of lighting control and its adaptation to actual scenarios have significant defects. Summary of the Invention

[0004] This invention provides a street light energy-saving control method and system based on multimodal data fusion to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a street light energy-saving control method based on multimodal data fusion, comprising: Based on the rate of change of the dynamic traffic density index in the historical road condition feature queue, the expected traffic density index in the future short time window is updated predictively. The real-time traffic conditions in the area where the streetlights are located are jointly embedded into a multi-dimensional contextual feature vector along with the predictions. Based on the multidimensional context feature vector, a dynamic brightness control reference value for the street light is generated, and the dynamic brightness control reference value is sent to the target street light.

[0006] In a preferred embodiment, the step of predictively updating the expected traffic density index within a future short time window based on the rate of change of the dynamic traffic density index in the historical traffic condition feature queue includes: By sequentially retrieving the dynamic traffic density index of multiple consecutive historical moments from a circularly stored historical data buffer, the traffic density sequence of the streetlights is obtained. Calculate the difference between the exponents of two adjacent time points in the traffic density sequence to obtain a rate of change sample; A weighted average is calculated from the rate of change samples, where the rate of change samples closer to the current time are given higher weights, to obtain a comprehensive trend value. Obtain the real-time dynamic traffic density index at the current moment, add the comprehensive trend value to the current real-time dynamic traffic density index, and generate the expected traffic density index for the next short time window.

[0007] In a preferred embodiment, before predictively updating the expected traffic density index for the next short time window based on the rate of change of the dynamic traffic density index in the historical traffic condition feature queue, the method further includes: At the beginning of each control cycle, the real-time dynamic traffic density index is obtained and added as a new element to the tail of the historical traffic condition feature queue. Determine whether the length of the historical traffic feature queue exceeds the preset queue capacity threshold after adding a new element. If it does, remove the oldest dynamic traffic density index at the head of the queue. After removing the oldest index at the head of the queue, the distribution span of all remaining indices in the historical road condition feature queue on the time axis is recalculated. If the distribution span is less than one control period, the last index in the queue is automatically copied and appended to the tail of the queue to maintain sufficient time coverage width of the queue.

[0008] In a preferred embodiment, the real-time traffic features include: By using millimeter-wave radar sensors deployed on streetlight poles, electromagnetic waves are continuously emitted and echo signals are received. The relative speed and azimuth of all moving targets within the detection range are extracted from the echo signals. Based on the relative speed and azimuth angle, targets with continuous spatial positions and consistent movement directions within the same time window are grouped into one traffic participation unit; The total number of traffic participation units in the current control cycle is counted, and the total number is divided by the preset road segment detection area to generate a real-time dynamic traffic density index, which is used as a core component of real-time traffic characteristics. Synchronized temperature distribution images; Extract the connected regions in the temperature distribution image whose temperature is higher than the ambient background temperature and which show the outline of a human body or non-motorized vehicle. The connected region is spatially matched with the traffic participation unit. If the match is successful, the type of the traffic participation unit is marked as a slow-moving traffic participant. The number of slow-moving traffic participants marked within the current control period is counted, and a slow-moving traffic participant presence flag is generated as another independent component in the real-time traffic condition features.

[0009] In a preferred embodiment, the step of embedding the real-time traffic features of the streetlight area and the predicted co-embedded multi-dimensional contextual feature vector includes: The real-time dynamic traffic density index, the presence markers of slow-moving traffic participants, and the environmental interference coefficient collected during the current control cycle are assembled into a real-time feature tuple. The expected traffic density index within a short future time window, generated from the prediction of historical traffic feature queues, is used as a prediction feature element. Each element in the real-time feature tuple is concatenated with the predicted feature element according to a preset semantic order to form a dual-source information vector. The numerical ranges of each element in the dual-source information vector are normalized and aligned to obtain a multidimensional contextual feature vector.

[0010] In a preferred embodiment, after normalizing and aligning the numerical ranges of each element in the dual-source information vector to obtain a multidimensional contextual feature vector, the method further includes: The real-time dynamic traffic density index and its corresponding timestamp in the multi-dimensional context feature vector are analyzed, and the expected traffic density index and the endpoint of the future time window it points to are also analyzed. Determine whether the end point of the future time window is later than the timestamp of the real-time dynamic traffic density index. If it is later, then the time sequence logic is confirmed to be correct. If the end point of the future time window is not later than the real-time timestamp, the current expected traffic density index is discarded, and the real-time dynamic traffic density index is copied and replaced before being re-embedded to generate a corrected multi-dimensional contextual feature vector.

[0011] In a preferred embodiment, generating the dynamic brightness control reference value of the street light based on the multi-dimensional context feature vector includes: Obtain the real-time dynamic traffic density index, expected traffic density index, slow-moving traffic participant presence marker, and environmental interference coefficient from the multi-dimensional context feature vector; First, a basic brightness level value is determined based on the presence sign of slow-moving traffic participants and the environmental interference coefficient. The basic brightness level value is a first preset value when slow-moving traffic participants are present, and a second preset value when they are not present. Then, the first preset value or the second preset value is linearly adjusted according to the level of the environmental interference coefficient.

[0012] In a preferred embodiment, generating the dynamic brightness control reference value of the street light based on the multi-dimensional context feature vector further includes: Assess the presence of signs and environmental disturbance factors of slow-moving traffic participants to determine a basic safety lighting standard; Using the real-time dynamic traffic density index and the expected traffic density index as core inputs, an additional dynamic compensation coefficient above the basic safety lighting level is calculated through a non-linear mapping relationship. The basic safety lighting level is superimposed with an additional dynamic compensation coefficient to generate a dynamic brightness control baseline value.

[0013] In a preferred embodiment, the additional dynamic compensation coefficient includes: ; In the formula, For the additional dynamic compensation coefficient. The real-time dynamic traffic density index, To adjust the intensity of the impact of the expected traffic density index on the compensation coefficient, For the expected traffic density index, It is a saturation constant that approaches one as density increases.

[0014] To address the aforementioned problems, the present invention also provides a street light energy-saving control system based on multimodal data fusion, the system comprising: The density prediction module is used to predictively update the expected traffic density index within a short future time window based on the rate of change of the dynamic traffic density index in the historical road condition feature queue. The context fusion module is used to embed the real-time traffic features of the area where the streetlights are located and the predicted features into a multi-dimensional context feature vector. The dynamic dimming module is used to generate a dynamic brightness control reference value for the street light based on the multi-dimensional context feature vector, and to send the dynamic brightness control reference value to the target street light.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention completes the predictive update of traffic density index for short time windows through historical road condition feature queues, continuously maintains the time coverage and data validity of the data queue, accurately completes the prediction and processing of traffic conditions, and builds a complete real-time and predictive fusion feature system based on multimodal data collection to ensure the comprehensiveness and accuracy of lighting control basis.

[0016] 2. This invention generates dynamic brightness control benchmark values ​​based on multi-dimensional contextual feature vectors, determines the basic lighting level by combining slow traffic conditions and environmental interference coefficients, and superimposes dynamic compensation coefficients to achieve precise adaptive adjustment of street light brightness, effectively improving the response efficiency and adaptation effect of lighting control, simultaneously optimizing lighting safety assurance capabilities and energy-saving control levels, and improving the overall lighting control operation efficiency. Attached Figure Description

[0017] Figure 1This is a flowchart illustrating a street light energy-saving control method based on multimodal data fusion, provided in an embodiment of the present invention. Figure 2 A functional block diagram of a street light energy-saving control system based on multimodal data fusion is provided in one embodiment of the present invention; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] This application provides a street light energy-saving control method based on multimodal data fusion. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the street light energy-saving control method based on multimodal data fusion can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0020] Reference Figure 1 The diagram shown is a flowchart illustrating a street light energy-saving control method based on multimodal data fusion according to an embodiment of the present invention. In this embodiment, the street light energy-saving control method based on multimodal data fusion includes: Based on the rate of change of the dynamic traffic density index in the historical road condition feature queue, the expected traffic density index in the future short time window is updated predictively. In this embodiment of the invention, the step of predictively updating the expected traffic density index within a short future time window based on the rate of change of the dynamic traffic density index in the historical road condition feature queue includes: By sequentially retrieving the dynamic traffic density index of multiple consecutive historical moments from a circularly stored historical data buffer, the traffic density sequence of the streetlights is obtained. Calculate the difference between the exponents of two adjacent time points in the traffic density sequence to obtain a rate of change sample; A weighted average is calculated from the rate of change samples, where the rate of change samples closer to the current time are given higher weights, to obtain a comprehensive trend value. Obtain the real-time dynamic traffic density index at the current moment, add the comprehensive trend value to the current real-time dynamic traffic density index, and generate the expected traffic density index for the next short time window.

[0021] Before predictively updating the expected traffic density index for the next short time window based on the rate of change of the dynamic traffic density index in the historical traffic condition feature queue, the method further includes: At the beginning of each control cycle, the real-time dynamic traffic density index is obtained and added as a new element to the tail of the historical traffic condition feature queue. Determine whether the length of the historical traffic feature queue exceeds the preset queue capacity threshold after adding a new element. If it does, remove the oldest dynamic traffic density index at the head of the queue. After removing the oldest index at the head of the queue, the distribution span of all remaining indices in the historical road condition feature queue on the time axis is recalculated. If the distribution span is less than one control period, the last index in the queue is automatically copied and appended to the tail of the queue to maintain sufficient time coverage width of the queue.

[0022] The historical data buffer, which is stored in a circular manner, has a fixed storage capacity. The buffer continuously stores the dynamic traffic density index in chronological order. According to the preset number of historical data extractions, the continuous dynamic traffic density indexes are extracted from the buffer in chronological order. The extracted indices are then arranged in chronological order to form the traffic density sequence of the streetlights.

[0023] Iterate through all adjacent combinations of dynamic traffic density indices in the traffic density sequence, subtract the dynamic traffic density index of the previous moment from the dynamic traffic density index of the next moment in each combination, and organize all the calculated differences into a rate of change sample.

[0024] Assign corresponding weights to the rate of change samples. The shorter the time interval between the historical moment and the current moment corresponding to the rate of change sample, the larger the weight value is assigned. Multiply each rate of change sample by its corresponding weight and sum the results. Then divide the sum by the total of all weights to obtain the comprehensive trend value.

[0025] The real-time dynamic traffic density index of the road segment corresponding to the streetlight at the current moment is collected. The comprehensive trend value is added to the real-time dynamic traffic density index, and the resulting value is the expected traffic density index in the short future time window.

[0026] From a pre-defined, fixed-length circular storage buffer of historical data, dynamic traffic density indices for consecutive historical moments that meet preset extraction rules are extracted in chronological order. The extracted indices are then arranged in chronological order to form a traffic density sequence corresponding to the streetlights.

[0027] Read the dynamic traffic density index of each group of adjacent traffic density indices in the traffic density sequence in sequence, subtract the dynamic traffic density index of the previous time from the dynamic traffic density index of the next time, and summarize all the calculated differences to form a rate of change sample.

[0028] Weight values ​​are assigned according to the time interval between the historical moment and the current moment corresponding to the rate of change sample. The shorter the time interval, the higher the weight value is assigned to the rate of change sample. Each rate of change sample is multiplied by its corresponding weight value and then summed. The summation result is then divided by the sum of all weight values ​​to obtain the comprehensive trend value.

[0029] The real-time dynamic traffic density index is obtained by the on-site data collection equipment of the streetlights. The comprehensive trend value is added to the real-time dynamic traffic density index, and the final value obtained is the expected traffic density index in the short future time window.

[0030] The real-time traffic conditions in the area where the streetlights are located are jointly embedded into a multi-dimensional contextual feature vector along with the predictions. In this embodiment of the invention, the real-time traffic features include: By using millimeter-wave radar sensors deployed on streetlight poles, electromagnetic waves are continuously emitted and echo signals are received. The relative speed and azimuth of all moving targets within the detection range are extracted from the echo signals. Based on the relative speed and azimuth angle, targets with continuous spatial positions and consistent movement directions within the same time window are grouped into one traffic participation unit; The total number of traffic participation units in the current control cycle is counted, and the total number is divided by the preset road segment detection area to generate a real-time dynamic traffic density index, which is used as a core component of real-time traffic characteristics. Synchronized temperature distribution images; Extract the connected regions in the temperature distribution image whose temperature is higher than the ambient background temperature and which show the outline of a human body or non-motorized vehicle. The connected region is spatially matched with the traffic participation unit. If the match is successful, the type of the traffic participation unit is marked as a slow-moving traffic participant. The number of slow-moving traffic participants marked within the current control period is counted, and a slow-moving traffic participant presence flag is generated as another independent component in the real-time traffic condition features.

[0031] The step of embedding the real-time traffic features of the area where the streetlights are located and the predicted features into a multi-dimensional contextual feature vector includes: The real-time dynamic traffic density index, the presence markers of slow-moving traffic participants, and the environmental interference coefficient collected during the current control cycle are assembled into a real-time feature tuple. The expected traffic density index within a short future time window, generated from the prediction of historical traffic feature queues, is used as a prediction feature element. Each element in the real-time feature tuple is concatenated with the predicted feature element according to a preset semantic order to form a dual-source information vector. The numerical ranges of each element in the dual-source information vector are normalized and aligned to obtain a multidimensional contextual feature vector.

[0032] After normalizing and aligning the numerical ranges of each element in the dual-source information vector to obtain the multidimensional context feature vector, the process further includes: The real-time dynamic traffic density index and its corresponding timestamp in the multi-dimensional context feature vector are analyzed, and the expected traffic density index and the endpoint of the future time window it points to are also analyzed. Determine whether the end point of the future time window is later than the timestamp of the real-time dynamic traffic density index. If it is later, then the time sequence logic is confirmed to be correct. If the end point of the future time window is not later than the real-time timestamp, the current expected traffic density index is discarded, and the real-time dynamic traffic density index is copied and replaced before being re-embedded to generate a corrected multi-dimensional contextual feature vector.

[0033] The millimeter-wave radar sensor mounted on the street lamp pole faces the preset fixed road section monitoring area covered by the street lamp and continuously emits electromagnetic waves at a constant frequency. When the electromagnetic waves come into contact with moving targets such as vehicles and pedestrians within the monitoring range, they are reflected to form echo signals. The sensor uses hardware demodulation and signal analysis processing to accurately extract the relative velocity data of each moving target relative to the sensor, as well as the horizontal azimuth angle data of the target relative to the sensor's installation position, from the echo signals.

[0034] Based on all the extracted relative velocity and azimuth data, the spatial coordinates of all moving targets are calculated within a preset fixed collection time window. Multiple moving targets whose spatial coordinates are less than a preset spatial distance threshold and whose deviation angles in the direction of movement are less than a preset direction angle threshold are integrated into an independent traffic participation unit.

[0035] Within the system's preset street lighting control cycle, the total number of traffic participation units generated within that cycle is accurately counted. The total number of traffic participation units is divided by the preset detection area of ​​the road segment covered by the millimeter-wave radar, and the calculated value is the real-time dynamic traffic density index. This index is the core component in real-time traffic characteristics used to reflect road traffic density.

[0036] The thermal imaging acquisition device mounted on the same street light pole is synchronously triggered and acquired with the millimeter-wave radar sensor. At the same time as the radar data acquisition, the temperature distribution image of the entire road section is obtained, ensuring that the temperature distribution image and the radar data acquisition time are completely consistent.

[0037] The temperature value of each pixel in the temperature distribution image is compared point by point with the real-time collected road environment background reference temperature. Pixel areas with temperature values ​​higher than the road environment background reference temperature are selected. Then, the human body outline template and non-motor vehicle outline template are used for matching and recognition. From the selected area, the area with a completely matching outline and continuous pixels is extracted. This continuous pixel area is the target connected region.

[0038] The spatial coordinate data of the target connected area is compared one by one with the spatial coordinate data of the traffic participating unit. When the overlap ratio of the two coordinates reaches the preset spatial matching threshold, the spatial location is determined to be successfully matched. After the matching is completed, the type of the traffic participating unit is marked as a slow traffic participant.

[0039] Within the preset street light control cycle, the number of traffic participation units marked as slow traffic participants is counted. When the number of counted units exceeds the preset existence threshold of the system, a slow traffic participant existence flag is generated. This flag serves as an independent component in the real-time traffic features used to distinguish the slow traffic state.

[0040] According to the fixed data combination order within the street lighting control cycle preset by the system, the real-time dynamic traffic density index, the presence markers of slow traffic participants, and the environmental interference coefficient are collected and generated in a fixed order, and then encapsulated and integrated to form a fixed real-time feature tuple with fixed structure and elements.

[0041] From the prediction dataset corresponding to the historical traffic feature queue stored in a circular manner, extract the expected traffic density index for the next short time window obtained after updating the prediction of historical data change trends, and define this single prediction data as a prediction feature element.

[0042] According to the fixed semantic arrangement order preset by the system, each component element inside the real-time feature tuple is retrieved in turn, and then the predicted feature element is connected to the end position of all elements in the real-time feature tuple. After the orderly splicing and combination of all data elements is completed, a dual-source information vector containing real-time data and predicted data is formed.

[0043] The original values ​​of each data element within the dual-source information vector are uniformly mapped to the system's preset standard value range, ensuring that the value range of all elements within the vector remains completely consistent. After completing the normalization and alignment of the value range, a multi-dimensional contextual feature vector containing multi-dimensional road condition information is generated.

[0044] From the multidimensional context feature vector that has undergone numerical range normalization and alignment processing, the real-time dynamic traffic density index stored internally, as well as the precise collection timestamp data bound to the index, are extracted. Simultaneously, the expected traffic density index stored within the vector, as well as the future time window end time data corresponding to the index, are extracted.

[0045] The time sequence of the future time window endpoint is compared with the collection timestamp of the real-time dynamic traffic density index. When the future time window endpoint is in a period following the time corresponding to the collection timestamp, it is confirmed that the time sequence logic of the multi-dimensional context feature vector conforms to the system's preset specifications.

[0046] When the end time of the future time window is the same as the collection timestamp of the real-time dynamic traffic density index, or is in the period preceding the collection timestamp, the original expected traffic density index in the multidimensional context feature vector is deleted, the value of the real-time dynamic traffic density index is copied, and the copied value is embedded in the storage location of the original expected traffic density index in the vector. After data replacement and vector reconstruction are completed, the corrected multidimensional context feature vector is generated.

[0047] Based on the multidimensional context feature vector, a dynamic brightness control reference value for the street light is generated, and the dynamic brightness control reference value is sent to the target street light.

[0048] In this embodiment of the invention, generating the dynamic brightness control reference value of the street light based on the multi-dimensional context feature vector includes: Obtain the real-time dynamic traffic density index, expected traffic density index, slow-moving traffic participant presence marker, and environmental interference coefficient from the multi-dimensional context feature vector; First, a basic brightness level value is determined based on the presence sign of slow-moving traffic participants and the environmental interference coefficient. The basic brightness level value is a first preset value when slow-moving traffic participants are present, and a second preset value when they are not present. Then, the first preset value or the second preset value is linearly adjusted according to the level of the environmental interference coefficient.

[0049] The step of generating the dynamic brightness control benchmark value of the street light based on the multi-dimensional context feature vector further includes: Assess the presence of signs and environmental disturbance factors of slow-moving traffic participants to determine a basic safety lighting standard; Using the real-time dynamic traffic density index and the expected traffic density index as core inputs, an additional dynamic compensation coefficient above the basic safety lighting level is calculated through a non-linear mapping relationship. The basic safety lighting level is superimposed with an additional dynamic compensation coefficient to generate a dynamic brightness control baseline value.

[0050] The additional dynamic compensation coefficients include: ; In the formula, For the additional dynamic compensation coefficient. The real-time dynamic traffic density index, To adjust the intensity of the impact of the expected traffic density index on the compensation coefficient, For the expected traffic density index, It is a saturation constant that approaches one as density increases.

[0051] From the multidimensional contextual feature vector that has been normalized and aligned and passed the time sequence logic verification, four core data items stored in the vector are completely extracted through data positioning and precise reading operations: real-time dynamic traffic density index, expected traffic density index, presence markers of slow traffic participants, and environmental interference coefficient. The extraction process strictly follows the data storage location and format specifications within the vector to ensure that the extracted data is completely consistent with the original stored data of the vector without any deviation.

[0052] The system reads the final determination status of the presence sign for slow-moving traffic participants. When the sign determines that slow-moving traffic participants are present, it directly retrieves the first fixed brightness value pre-configured by the system as the initial basic brightness level value. When the sign determines that slow-moving traffic participants are not present, it directly retrieves the second fixed brightness value pre-configured by the system as the initial basic brightness level value. The environmental interference coefficient is precisely compared point by point with the system's preset standard interference coefficient range to determine the specific position and proportion of the environmental interference coefficient within the standard range. The initial basic brightness level value is then adjusted linearly according to this proportion. After the adjustment is completed, the final basic brightness level value adapted to the current environment and traffic conditions is obtained.

[0053] The presence status of slow-moving traffic participants within the multi-dimensional context feature vector is determined to clarify whether slow-moving traffic participants exist in the current road segment. At the same time, the environmental interference coefficient is matched step by step with the multi-level interference threshold range preset by the system to determine the specific interference level to which the environmental interference coefficient belongs. Combining the presence status of slow-moving traffic participants with the interference level corresponding to the environmental interference coefficient, a unique matching basic safety lighting level is selected from the multi-level lighting level library preset by the system.

[0054] Using the combined value of the real-time dynamic traffic density index and the expected traffic density index as the core input, the system calls the preset piecewise nonlinear correspondence rules to accurately match the combined value with the preset compensation coefficient range. Based on the corresponding point of the value in the piecewise rules, the additional dynamic compensation coefficient required above the basic safety lighting level is determined.

[0055] The value of the basic safety lighting level is directly added together with the value of the additional dynamic compensation coefficient. The final value obtained after the addition calculation is the benchmark value for dynamic brightness control of streetlights adapted to the current road conditions and environment.

[0056] The real-time dynamic traffic density index is generated by collecting data from millimeter-wave radar sensors mounted on streetlight poles. The sensors first extract the relative speed and azimuth data of all moving targets in the road segment, then classify targets that meet the spatial and motion conditions as traffic participation units, count the total number of traffic participation units in the current control cycle, and divide the total number by the preset road segment detection area to generate the index.

[0057] The expected traffic density index is generated based on a circularly stored historical data buffer. First, continuous historical dynamic traffic density indices are extracted from the buffer and arranged into a traffic density sequence. Then, the difference between adjacent indices in the sequence is calculated to obtain a rate of change sample. The rate of change sample is weighted according to the time interval between historical time and current time and a weighted average is calculated to obtain a comprehensive trend value. Finally, the comprehensive trend value is added to the real-time dynamic traffic density index to generate the index.

[0058] The value used to adjust the influence of the expected traffic density index is a fixed configuration value preset before the system leaves the factory. This value is used to accurately determine the proportion of the effect of the predicted traffic data on the dynamic compensation result.

[0059] The saturation constant is a fixed value preset by the system. This value is used to constrain the upper limit of the growth of the compensation coefficient, so that the compensation coefficient eventually approaches a fixed value as the traffic density index increases.

[0060] This calculation method uses real-time traffic density and expected traffic density as the core criteria. It generates dynamic compensation coefficients through numerical matching calculations and growth limit constraints. These coefficients are used to superimpose brightness compensation that matches traffic conditions on the basis of basic safety lighting levels, ultimately achieving dynamic adaptive adjustment of street light brightness according to the real-time traffic conditions of the road segment and the expected traffic conditions in the future short time window.

[0061] like Figure 2 The diagram shown is a functional block diagram of a street light energy-saving control system based on multimodal data fusion provided in an embodiment of the present invention.

[0062] The street light energy-saving control system based on multimodal data fusion described in this invention can be installed in electronic devices. Depending on the functions implemented, the street light energy-saving control system based on multimodal data fusion may include a density prediction module, a context fusion module, and a dynamic dimming module. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and is stored in the memory of the electronic device.

[0063] In this embodiment, the functions of each module / unit are as follows: The density prediction module is used to predictively update the expected traffic density index within a short future time window based on the rate of change of the dynamic traffic density index in the historical road condition feature queue. The context fusion module is used to embed the real-time traffic features of the area where the streetlights are located and the predicted features into a multi-dimensional context feature vector. The dynamic dimming module is used to generate a dynamic brightness control reference value for the street light based on the multi-dimensional context feature vector, and to send the dynamic brightness control reference value to the target street light.

[0064] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0065] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0066] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0067] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0068] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A street light energy-saving control method based on multimodal data fusion, characterized in that, The method includes: Based on the rate of change of the dynamic traffic density index in the historical road condition feature queue, the expected traffic density index in the future short time window is updated predictively. The real-time traffic conditions in the area where the streetlights are located are jointly embedded into a multi-dimensional contextual feature vector along with the predictions. Based on the multidimensional context feature vector, a dynamic brightness control reference value for the street light is generated, and the dynamic brightness control reference value is sent to the target street light.

2. The street light energy-saving control method based on multimodal data fusion as described in claim 1, characterized in that, The method of predictively updating the expected traffic density index within a short future time window based on the rate of change of the dynamic traffic density index in the historical road condition feature queue includes: By sequentially retrieving the dynamic traffic density index of multiple consecutive historical moments from a circularly stored historical data buffer, the traffic density sequence of the streetlights is obtained. Calculate the difference between the exponents of two adjacent time points in the traffic density sequence to obtain a rate of change sample; A weighted average is calculated from the rate of change samples, where the rate of change samples closer to the current time are given higher weights, to obtain a comprehensive trend value. Obtain the real-time dynamic traffic density index at the current moment, add the comprehensive trend value to the current real-time dynamic traffic density index, and generate the expected traffic density index for the next short time window.

3. The street light energy-saving control method based on multimodal data fusion as described in claim 2, characterized in that, Before predictively updating the expected traffic density index for the next short time window based on the rate of change of the dynamic traffic density index in the historical traffic condition feature queue, the method further includes: At the beginning of each control cycle, the real-time dynamic traffic density index is obtained and added as a new element to the tail of the historical traffic condition feature queue. Determine whether the length of the historical traffic feature queue exceeds the preset queue capacity threshold after adding a new element. If it does, remove the oldest dynamic traffic density index at the head of the queue. After removing the oldest index at the head of the queue, the distribution span of all remaining indices in the historical road condition feature queue on the time axis is recalculated. If the distribution span is less than one control period, the last index in the queue is automatically copied and appended to the tail of the queue to maintain sufficient time coverage width of the queue.

4. The street light energy-saving control method based on multimodal data fusion as described in claim 1, characterized in that, The real-time traffic features include: By using millimeter-wave radar sensors deployed on streetlight poles, electromagnetic waves are continuously emitted and echo signals are received. The relative speed and azimuth of all moving targets within the detection range are extracted from the echo signals. Based on the relative speed and azimuth angle, targets with continuous spatial positions and consistent movement directions within the same time window are grouped into one traffic participation unit; The total number of traffic participation units in the current control cycle is counted, and the total number is divided by the preset road segment detection area to generate a real-time dynamic traffic density index, which is used as a core component of real-time traffic characteristics. Synchronized temperature distribution images; Extract the connected regions in the temperature distribution image whose temperature is higher than the ambient background temperature and which show the outline of a human body or non-motorized vehicle. The connected region is spatially matched with the traffic participation unit. If the match is successful, the type of the traffic participation unit is marked as a slow-moving traffic participant. The number of slow-moving traffic participants marked within the current control period is counted, and a slow-moving traffic participant presence flag is generated as another independent component in the real-time traffic condition features.

5. The street light energy-saving control method based on multimodal data fusion as described in claim 4, characterized in that, The step of embedding the real-time traffic features of the area where the streetlights are located and the predicted features into a multi-dimensional contextual feature vector includes: The real-time dynamic traffic density index, the presence markers of slow-moving traffic participants, and the environmental interference coefficient collected during the current control cycle are assembled into a real-time feature tuple. The expected traffic density index within a short future time window, generated from the prediction of historical traffic feature queues, is used as a prediction feature element. Each element in the real-time feature tuple is concatenated with the predicted feature element according to a preset semantic order to form a dual-source information vector. The numerical ranges of each element in the dual-source information vector are normalized and aligned to obtain a multidimensional contextual feature vector.

6. The street light energy-saving control method based on multimodal data fusion as described in claim 5, characterized in that, After normalizing and aligning the numerical ranges of each element in the dual-source information vector to obtain the multidimensional context feature vector, the process further includes: The real-time dynamic traffic density index and its corresponding timestamp in the multi-dimensional context feature vector are analyzed, and the expected traffic density index and the endpoint of the future time window it points to are also analyzed. Determine whether the end point of the future time window is later than the timestamp of the real-time dynamic traffic density index. If it is later, then the time sequence logic is confirmed to be correct. If the end point of the future time window is not later than the real-time timestamp, the current expected traffic density index is discarded, and the real-time dynamic traffic density index is copied and replaced before being re-embedded to generate a corrected multi-dimensional contextual feature vector.

7. The street light energy-saving control method based on multimodal data fusion as described in claim 1, characterized in that, The step of generating the dynamic brightness control benchmark value of the street light based on the multi-dimensional context feature vector includes: Obtain the real-time dynamic traffic density index, expected traffic density index, slow-moving traffic participant presence marker, and environmental interference coefficient from the multi-dimensional context feature vector; First, a basic brightness level value is determined based on the presence sign of slow-moving traffic participants and the environmental interference coefficient. The basic brightness level value is a first preset value when slow-moving traffic participants are present, and a second preset value when they are not present. Then, the first preset value or the second preset value is linearly adjusted according to the level of the environmental interference coefficient.

8. The street light energy-saving control method based on multimodal data fusion as described in claim 7, characterized in that, The step of generating the dynamic brightness control benchmark value of the street light based on the multi-dimensional context feature vector further includes: Assess the presence of signs and environmental disturbance factors of slow-moving traffic participants to determine a basic safety lighting standard; Using the real-time dynamic traffic density index and the expected traffic density index as core inputs, an additional dynamic compensation coefficient above the basic safety lighting level is calculated through a non-linear mapping relationship. The basic safety lighting level is superimposed with an additional dynamic compensation coefficient to generate a dynamic brightness control baseline value.

9. The street light energy-saving control method based on multimodal data fusion as described in claim 8, characterized in that, The additional dynamic compensation coefficients include: ； In the formula, For the additional dynamic compensation coefficient. The real-time dynamic traffic density index, To adjust the intensity of the impact of the expected traffic density index on the compensation coefficient, For the expected traffic density index, It is a saturation constant that approaches one as density increases.

10. A street light energy-saving control system based on multimodal data fusion, used to implement the street light energy-saving control method based on multimodal data fusion as described in claim 1, the system comprising: The density prediction module is used to predictively update the expected traffic density index within a short future time window based on the rate of change of the dynamic traffic density index in the historical road condition feature queue. The context fusion module is used to embed the real-time traffic features of the area where the streetlights are located and the predicted features into a multi-dimensional context feature vector. The dynamic dimming module is used to generate a dynamic brightness control reference value for the street light based on the multi-dimensional context feature vector, and to send the dynamic brightness control reference value to the target street light.

Images