An intelligent recognition method and system for street lamp lighting in vehicle-road cooperation based on the Internet of Things

Abstract

The present application provides an intelligent recognition method and system for vehicle-road collaborative street lamp lighting based on the Internet of Things. The lighting rate is recognized by using the street lamp status images captured by the camera on the top of the vehicle, the geographical location information recorded by the vehicle GPS locator, and the unique GPS coordinate information of the street lamp GPS receiver. Specifically, when the vehicle is driving on the detection road, images along the road are captured, and the precise position of each street lamp determined by combining GPS information is used. Image recognition technology is adopted to analyze these images, and the lit street lamps are recognized to statistically obtain the lighting rate, and the analysis result is accurate. Further, when the present application uses machine learning technology for future lighting rate detection, the embedded mapping neural network is debugged based on the integrated information time series set, enabling the embedded mapping neural network to learn real information. When the embedded mapping neural network is used to estimate the time series set, the accuracy of the time series set estimation can be increased, that is, the accuracy of lighting monitoring can be improved.

Description

Technical Field

[0001] This application relates to the field of data processing, and particularly to an intelligent identification method and system for street lamp lighting in vehicle-road coordination based on the Internet of Things. Background Art

[0002] With the rapid development of Internet of Things technology, the concept of smart cities has gradually gained popularity. Among them, the vehicle-road coordination system, as an important part of smart city construction, is of great significance for improving the management efficiency of urban public lighting. In the vehicle-road coordination system, the intelligent monitoring and management of street lamps is a key link, which directly affects the safety of roads and the travel experience of residents. Traditional street lamp lighting detection mainly relies on manual inspections. This traditional inspection method is inefficient and difficult to cope with complex and changeable urban environments. In recent years, with the wide application of big data, cloud computing and artificial intelligence technologies, the intelligent identification and control technology of street lamps based on the Internet of Things has emerged, providing the possibility for accurate monitoring and dynamic analysis of street lamp lighting. Since the existing methods rely on the lighting data obtained from manual inspections when monitoring street lamp lighting, human errors will have a greater impact on the actual data during this detection period. In addition, although some advanced machine learning models (such as neural networks) have been applied to street lamp lighting detection, these models usually lack clear prior knowledge guidance, resulting in a decrease in the accuracy of street lamp lighting detection in the case of data loss or anomalies. Summary of the Invention

[0003] In view of this, at least one intelligent identification method and system for street lamp lighting in vehicle-road coordination based on the Internet of Things are provided in the embodiments of this application. The technical solution of this application is realized as follows:

[0004] On the one hand, an embodiment of the present application provides an intelligent recognition method for vehicle-road collaborative street lamp lighting based on the Internet of Things. The method includes: obtaining a time series set of past lighting information of the target number of lights to be lit; the time series set of past lighting information includes the lighting information of the target number of lights to be lit at each of multiple past moments, where the past lighting information is obtained through the street lamp status images captured by vehicle cameras, vehicle GPS positioning information, and street lamp GPS coordinate information; and at the same time, each of the lighting information is arranged in the order of the past moments of each; obtaining a prior label corresponding to each lighting information in the time series set of past lighting information, the prior label being a descriptive label indicating the evolution of the target number of lights to be lit, and the moment representing the evolution in the descriptive label is after the past moment of the lighting information corresponding to the prior label; extracting the semantic implicit representation of each prior label to obtain the semantic implicit representation of each prior label; integrating each lighting information in the time series set of past lighting information with the semantic implicit representation of the corresponding prior label to obtain integrated information, and obtaining a time series set of past integrated information; based on the time series set of past integrated information, detecting the lighting information of the target number of lights to be lit at each analysis moment among multiple sequentially distributed analysis moments, and obtaining a time series set of estimated lighting information; the analysis moment is after each of the past moments.

[0005] On the other hand, the present application provides a computer system, including a memory and a processor. The memory stores a computer program that can run on the processor, and when the processor executes the program, the steps in the above method are implemented.

[0006] The beneficial effects of the present application at least include: In the method provided by the present application, since the prior label is a descriptive label indicating the evolution of the target number of lights to be lit, and the moment representing the evolution in the descriptive label is after the past moment of the lighting information corresponding to the prior label, then integrating the lighting information in the sample with the semantic implicit representation of the corresponding prior label to obtain integrated information can make the integrated information in the time series set of integrated information include the information in the prior label. Since the prior label contains real information, debugging the embedding mapping neural network based on the time series set of integrated information can enable the embedding mapping neural network to learn real information, so that when the embedding mapping neural network is used for estimating the time series set, the accuracy of the time series set estimation can be increased, that is, the accuracy of lighting monitoring can be improved. Description of the Drawings

[0007] The drawings here are incorporated into the description and form a part of this description. These drawings show embodiments consistent with the present application and are used together with the description to explain the technical solutions of the present application.

[0008] Figure 1Schematic diagram of the implementation process of an intelligent identification method for vehicle-road collaborative street lamp lighting based on the Internet of Things provided by an embodiment of the present application.

[0009] Figure 2 Schematic diagram of the hardware entity of a computer system provided by an embodiment of the present application. Detailed implementation manners

[0010] An embodiment of the present application provides an intelligent identification method for vehicle-road collaborative street lamp lighting based on the Internet of Things, and this method can be executed by a processor of a computer system. Among them, the computer system may refer to devices with data processing capabilities such as servers, laptop computers, tablet computers, desktop computers, etc.

[0011] Figure 1 Schematic diagram of the implementation process of an intelligent identification method for vehicle-road collaborative street lamp lighting based on the Internet of Things provided by an embodiment of the present application, as Figure 1 shown, this method includes:

[0012] Step S100: Obtain a time series set of past lighting information of the target number of lights to be lit; the time series set of past lighting information includes the lighting information of the target number of lights to be lit at each of multiple past moments, where the past lighting information is obtained through the street lamp status images captured by vehicle cameras, vehicle GPS positioning information, and street lamp GPS coordinate information; and at the same time, each of the lighting information is arranged in the order of the sequence of their respective past moments.

[0013] In step S100, in order to construct the time series set of past lighting information and detect the street lamp lighting status, the computer system can access a comprehensive Internet of Things platform, which integrates a variety of sensors and data acquisition devices. In this scenario, the main data sources are the driving vehicles equipped with cameras, GPS locators, and display screens, and the street lamp poles with GPS coordinate information.

[0014] During the data collection process, a high-definition camera is installed on top of each vehicle participating in data collection to capture images of the street lamp status along the road. At the same time, the vehicle is equipped with a GPS locator to record the geographical location information of the vehicle in real time. Although the display screen does not directly participate in data collection in this scenario, it can be used to show relevant information to the driver or the remote monitoring center. GPS receivers are pre-installed on all street lamp poles, enabling each street lamp to have a unique GPS coordinate information, and these coordinate information are pre-imported into the computer system. In this way, the system can know the exact location of each street lamp. When the vehicle is driving on the detection road, the on-vehicle camera will continuously take pictures of the road along the way. The computer system uses image recognition technology (such as the convolutional neural network CNN based on deep learning) to analyze these images and identify the lit street lamps. This process is completed automatically without manual intervention. To improve the accuracy of recognition, the computer system also uses the dynamic change relationship between the GPS coordinates of the driving vehicle and the street lamp poles for auxiliary verification. Specifically, when the camera identifies a certain street lamp as being in the lit state, the system will immediately search for the GPS coordinates of the lamp poles near the current vehicle position and check whether these coordinates match the street lamp identified as being lit. This step effectively reduces the situation of misrecognition and improves the reliability of the data. For each frame of image, the system will record the number of street lamps identified as being lit and their corresponding timestamps (i.e., past moments). This information is transmitted to the database of the computer system in real time. As time goes by, the system accumulates a large amount of lit lamp data. In step S100, the computer system arranges these data in chronological order according to the timestamps to form an ordered chronological set of past lit lamp information. For example, this set may contain the following data items: {(2024-07-01 20:00, 80), (2024-07-01 21:00, 85),...}, where each tuple represents a time point and the number of lit lamps corresponding to that time point.

[0015] Step S200: Obtain the prior label corresponding to each lit lamp information in the chronological set of past lit lamp information, where the prior label is a descriptive label indicating the evolution of the target number of lamps that should be lit, and the moment representing the evolution in the descriptive label is after the past moment of the lit lamp information corresponding to the prior label.

[0016] In step S200, the computer system obtains the prior label corresponding to the lit lamp information at each time point (i.e., past moment) in the chronological set of past lit lamp information. These prior labels are descriptions of the evolution of the target number of lamps that should be lit within a specific time period, such as an increase, decrease, or remaining unchanged in the number, and the time point pointed to by the description of this evolution is after the actual lit lamp information recording time point.

[0017] First, the computer system has collected and organized an ordered chronological set of past lighting information through step S100, which contains multiple time points and their corresponding actual lighting quantities. However, simply knowing these specific lighting quantities is not sufficient to comprehensively understand the changing trend of the street lamp lighting status. Therefore, prior labels need to be introduced to further describe this changing trend. Prior labels are a set of predefined descriptive words or phrases used to represent the evolution of lighting quantities over time. In practical applications, these labels can be set according to specific requirements, such as "quantity increasing", "quantity decreasing", "quantity stable", etc. These labels not only reflect the changing direction of the lighting quantity but may also contain information about the changing amplitude. The process of obtaining prior labels usually relies on external data sources or domain knowledge.

[0018] Step S300: Extract the semantic implicit representation of each of the prior labels to obtain the semantic implicit representation of each of the prior labels.

[0019] In step S300, the computer system converts each prior label into a form that can be directly processed and understood by the computer, that is, extracts the semantic implicit representation of each prior label. Prior labels (such as "quantity increasing", "quantity decreasing") exist in natural language form, while when the computer system conducts data analysis or model training, it needs to convert this text information into numerical form for effective calculation and processing. The semantic implicit representation is the key means to achieve this conversion. It can capture the core semantic information in the label and encode it into numerical data in the form of vectors, matrices, or others.

[0020] In the intelligent identification scenario of vehicle-road collaborative street lamp lighting based on the Internet of Things, extracting the semantic implicit representation of prior labels can be achieved through various methods, including but not limited to:

[0021] Word Embedding: It is to use pre-trained word embedding models (such as Word2Vec, GloVe, FastText, etc.) to obtain the vector representation of the labels. These models are trained on a large-scale text corpus and can map each word to a high-dimensional vector space, making words with similar semantics closer in the vector space distance. For the prior labels in this scenario, these pre-trained models can be directly used to obtain their vector representations.

[0022] In some cases, the pre-trained word embedding model may not be directly applicable to the labels in a specific domain, or to further improve the accuracy of the representation, a custom feature extraction method can be designed. For example, a set of specific attributes (such as change trend, change amplitude, etc.) can be defined for each label, and corresponding numerical values can be manually assigned to each label. However, this method is usually more time-consuming and relies on the knowledge of domain experts. In this scenario, after the computer system extracts the semantic implicit representation of each prior label through step S300, these representations will be used in subsequent data integration (step S400) and model training (such as training of the embedding mapping neural network). For example, when training the embedding mapping neural network, these representations will be used as part of the network input to help the network learn the complex relationship between the number of lit lights and the prior labels.

[0023] Step S400: Integrate each lighting information in the time series set of the past lighting information with the semantic implicit representation of the corresponding prior label to obtain integrated information, and obtain a time series set of past integrated information.

[0024] In step S400, the computer system integrates each lighting information in the time series set of the past lighting information with the semantic implicit representation of its corresponding prior label, thereby generating a time series set of past integrated information containing richer information.

[0025] Step S100 generates a time series set of past lighting information, which contains the lighting number data at multiple time points, such as {(t1,n1),(t2,n2),…,(tm,nm)}, where ti represents the time point and ni represents the lighting number at the corresponding time point. Step S300 generates a semantic implicit representation for each prior label. For example, the semantic implicit representation of the label "quantity increase" may be a vector v 增加 = [v1,v2,…,vd], where d is the dimension of the vector. Next, in step S400, the computer system performs the following operations:

[0026] For each lighting information (ti,ni) in the time series set of the past lighting information, find the semantic implicit representation of its corresponding prior label. This is usually achieved by matching through a timestamp or some identifier. Integrate the found semantic implicit representation with the corresponding lighting information. The integration method can be simple concatenation or a more complex fusion method. For example, if the concatenation method is adopted, then for the lighting information (ti,ni) at time point ti and the semantic implicit representation v 增加 of its corresponding label "quantity increase", the integrated information can be (ti,ni,v 增加)。Arrange all the integrated information in chronological order to form a new chronological set, that is, the chronological set of past integrated information. This set not only contains the original information on the number of lights on, but also incorporates the semantic information of the prior labels.

[0027] For example, assume that the chronological set of past information on the number of lights on is \(\{(2024 - 07 - 01, 80), (2024 - 07 - 02, 85)\}\), and it is known that the prior label for July 1, 2024 is "stable quantity", and its semantic implicit representation is \(v\) 稳定 = [0.1, 0.0, -0.1]; the prior label for July 2, 2024 is "quantity increasing", and its semantic implicit representation is \(v\) 增加 = [0.2, 0.1, 0.1]. After integration, the chronological set of past integrated information may be as follows: \(\{((2024 - 07 - 01, 80), v\) 稳定 ), ((2024 - 07 - 02, 85), v\) 增加 )\}. In this set, each element is a tuple containing a timestamp, the number of lights on, and the corresponding semantic implicit representation.

[0028] Through the integration operation in step S400, the computer system not only retains the original data on the number of lights on, but also additionally introduces the semantic information contained in the prior labels. This enables subsequent data analysis and model training to more comprehensively consider the changes in the number of lights on and the reasons or trends behind them.

[0029] Step S500: Based on the chronological set of the past integrated information, estimate the information on the number of lights that should be on for the target at each analysis moment among multiple sequentially distributed analysis moments, and obtain a chronological set of estimated information on the number of lights; the analysis moments are after each of the past moments.

[0030] In step S500, the computer system uses the chronological set of past integrated information to estimate the number of lights that should be on for the target in a future period (i.e., the analysis moments) and generates a chronological set of estimated information on the number of lights. First, the computer system has obtained through the previous steps (especially step S400) a chronological set of past integrated information that includes the number of lights on and the semantic implicit representations of their corresponding prior labels. This set provides rich historical data and semantic information for detecting the number of lights on.

[0031] Next, the computer system will use one or more machine learning models (such as time series detection models, deep learning models, etc.) for prediction. These models can usually capture the time dependence and potential patterns in the data, thus accurately detecting future values. Common time series detection models include ARIMA (Autoregressive Integrated Moving Average), exponential smoothing, seasonal decomposition, etc. These models build detection models by analyzing trends, seasonality, and random components in historical data. However, in this scenario, due to the inclusion of semantic implicit representations of prior labels, these traditional models can combine additional feature engineering to effectively utilize this information. Considering the powerful representation learning ability of deep learning, models such as Recurrent Neural Network (RNN), Long Short-Term Memory Network (LSTM), Gated Recurrent Unit (GRU), or Transformer can be used for detection. These models can automatically learn complex non-linear relationships from the data and can handle long-term dependencies in sequence data. In particular, they can directly process the integrated information containing the number of lit lights and semantic implicit representations without complex feature engineering.

[0032] Taking the deep learning model as an example, the computer system will perform the following steps for prediction:

[0033] Data preprocessing: Divide the historical integrated information time series set into a training set and a test set (if available). For each integrated information (including timestamp, number of lit lights, and semantic implicit representation), preprocessing such as normalization, standardization, or other forms can be performed so that the model can learn better.

[0034] Model training: Use the training set data to train the selected deep learning model. During the training process, the model will learn how to map the historical number of lit lights and the semantic implicit representation of prior labels to the future number of lit lights. This usually involves minimizing a loss function (such as Mean Squared Error MSE), which measures the difference between the model's detected value and the actual value. The loss function is for example:

[0035] MSE = ;

[0036] where N is the number of samples, is the actual number of lit lights of the i-th sample, is the detected number of lit lights of the i-th sample by the model.

[0037] Model evaluation: Use the test set data (or a part of the training set data set aside) to evaluate the performance of the model. Evaluation metrics may include MSE, Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), etc.

[0038] Once the model is trained and evaluated, the computer system can use it to detect and analyze the number of lit lights. This typically involves applying the model to data containing future timestamps (but not the number of lit lights) and generating a sequence of estimated number of lit lights. For example, assume the computer system has trained an LSTM model to detect the number of streetlights lit each night in the coming week. For each analysis moment (e.g., each night in the future), the model will output an estimated number of lit lights. Arranging these estimated numbers in chronological order forms a chronological set of estimated lit light information, such as { 2024−07−08, 2024−07−09,…, 2024−07−14}, where t represents the estimated number of lit lights at time t.

[0039] Through step S500, the computer system can provide valuable detection information to the urban lighting management department, helping them better arrange maintenance work according to the streetlight lighting status, thereby improving the streetlight lighting rate and enhancing the quality of residents' lives.

[0040] As an implementation solution, in step S400, each lit light information in the chronological set of past lit light information is respectively integrated with the semantic implicit representation of the corresponding prior label to obtain integrated information, and a chronological set of past integrated information is obtained. Specifically, it may include:

[0041] Step S410: For each lit light information in the chronological set of past lit light information, combine the lit light information with the semantic implicit representation of the corresponding prior label to obtain integrated information corresponding to each lit light information;

[0042] Step S420: Arrange the integrated information corresponding to each lit light information according to the chronology of the lit light information in the chronological set of past lit light information to generate a chronological set of past integrated information.

[0043] In step S410, the computer system combines each lit light information item in the chronological set of past lit light information with the semantic implicit representation of its corresponding prior label, thereby generating integrated information items containing richer information. This step is the basis for data integration, ensuring that both the actual observed values of the number of lit lights and the related semantic information can be considered simultaneously in subsequent analysis processes.

[0044] There can be multiple specific ways to combine the lighted lamp information with the semantic implicit representation of the prior label, but the most common and direct method is concatenation. Concatenation means directly connecting the number of lighted lamps (a scalar value) with the semantic implicit representation of the prior label (a vector or matrix) to form a new data structure. This method is simple and intuitive and is easy to process subsequently.

[0045] Suppose a certain lighted lamp information item in the time series set of past lighted lamp information is (t1, n1), where t1 represents the time point (such as 8 pm on July 1, 2024), and n1 represents the number of lighted lamps at this time point (such as 80 lamps). At the same time, suppose the prior label corresponding to this lighted lamp information item is "stable quantity", and its semantic implicit representation is a 5-dimensional vector v 稳定 = [0.1, 0.0, −0.1, 0.05, −0.05].

[0046] In step S410, the computer system will perform the following operations:

[0047] Identify the correspondence: First, the computer system determines the correspondence between (t1, n1) and v 稳定 This is usually achieved through timestamps, identifiers, or other metadata.

[0048] Data concatenation: Next, the computer system concatenates n1 (a scalar value) with v 稳定 (a vector). Since n1 is a scalar, to maintain the consistency of the data structure, it can be regarded as a one-dimensional vector (or array), for example, [n1]. Then, these two vectors are directly connected together to form a new integrated information vector. In this example, the integrated information vector will be [80, 0.1, 0.0, −0.1, 0.05, −0.05].

[0049] Generate an integrated information item: Finally, the computer system associates this integrated information vector with the original timestamp t1 to generate a complete integrated information item, such as (t1, [80, 0.1, 0.0, −0.1, 0.05, −0.05]).

[0050] The integrated information item generated through step S410 not only contains the original information on the number of lighted lamps but also incorporates the relevant prior knowledge (i.e., the semantic implicit representation of the prior label). This integration method enables subsequent data analysis and model detection to simultaneously consider both the actual changes in the number of lighted lamps and the semantic meanings behind these changes, thereby improving the accuracy of analysis and detection.

[0051] In step S420, the task of the computer system is to arrange each integrated information item generated in step S410 according to its original time sequence in the time sequence set of past lighting information, so as to form an ordered time sequence set of past integrated information. The computer system only needs to traverse the list of integrated information items generated in step S410 and sort them according to the timestamps corresponding to each integrated information item (if they are already ordered in step S410, this step can be omitted). After sorting, an ordered time sequence set of past integrated information is obtained.

[0052] Continuing with the above example, assume there are already the following two integrated information items:

[0053] The first integrated information item: (t1, [80, 0.1, 0.0, -0.1, 0.05, -0.05]), represents the lighting situation at 8 pm on July 1, 2024 and the semantic implicit representation of the corresponding "quantity stable" prior label.

[0054] The second integrated information item (assumed to be generated through a process similar to step S410): (t2, [85, 0.2, 0.1, 0.1, -0.05, 0.0]), represents the lighting situation at 8 pm on July 2, 2024 and the semantic implicit representation of the corresponding "quantity increasing" prior label.

[0055] In step S420, the computer system will perform the following operations: First, the computer system identifies the timestamps (t1 and t2) in each integrated information item. Then, it sorts these integrated information items according to the order of the timestamps. In this example, since t1 is earlier than t2, the sorted order remains unchanged. Finally, the sorted integrated information items are combined into an ordered time sequence set of past integrated information, that is, {(t1, [80, 0.1, 0.0, -0.1, 0.05, -0.05]), (t2, [85, 0.2, 0.1, 0.1, -0.05, 0.0])}.

[0056] The ordered time sequence set of past integrated information provides a key data basis for subsequent data analysis and model detection. Since it retains the time sequence characteristics of the original lighting data and incorporates the prior knowledge related to it, it can more accurately reflect the changing pattern of the street lamp lighting state over time. Integrated information has a wide range of applications in the intelligent identification scenario of vehicle-road collaborative street lamp lighting based on the Internet of Things. The following are some specific application scenario examples:

[0057] Statistical street lamp lighting rate: By comparing the current number of lit lamps with the historical data in the integrated information time series set of the past, abnormal lighting states can be identified. Combining with the semantic information of the prior labels, the street lamp lighting rate of the entire road can be statistically calculated, providing timely lighting abnormality information and accurate assessment basis for urban lighting management departments.

[0058] Improve road safety: The intelligent identification and optimized management of the street lamp lighting state can not only improve the night lighting quality of the city, but also significantly enhance the road safety. By real-time monitoring the street lamp lighting state, lighting abnormality problems can be timely discovered and solved, which is helpful for reducing the risk of traffic accidents.

[0059] As an implementation solution, the method further includes: arranging the semantic implicit representations of the prior labels corresponding to each of the lighting information according to the time series of the lighting information in the past lighting information time series set, to generate a semantic implicit representation time series set; based on this, in step S410, for each lighting information in the past lighting information time series set, combining the lighting information with the semantic implicit representation of the corresponding prior label to obtain the integrated information corresponding to each of the lighting information, which may specifically include:

[0060] Step S411: Combining the lighting information and the semantic implicit representation at the same position in the past lighting information time series set and the semantic implicit representation time series set to obtain the integrated information corresponding to each of the lighting information.

[0061] Specifically, the computer system first traverses the past lighting information time series set, records the lighting information corresponding to each time point and the associated prior labels. This correspondence is usually established through timestamps or some unique identifiers. For each prior label, the computer system uses a pre-trained word embedding model (such as Word2Vec, GloVe, etc.) or a custom feature extraction method to extract its semantic implicit representation. These representations can be in vector form, and the values on each vector dimension reflect a certain semantic feature of the label.

[0062] According to the time series of the lighting information in the past lighting information time series set, the computer system arranges the corresponding semantic implicit representations in the same time order to generate a semantic implicit representation time series set. Each element in this set is a semantic implicit representation vector, which strictly corresponds to the lighting information in the past lighting information time series set in time.

[0063] Suppose the time - series set of past lighting information is \(\{(t_1,n_1),(t_2,n_2),\cdots,(t_m,n_m)\}\), where \(t_i\) represents the time point and \(n_i\) represents the number of lit lights at that time point. The corresponding prior label sequence is \(\{l_1,l_2,\cdots,l_m\}\), where each \(l_i\) is a prior label (such as "quantity increasing", "quantity decreasing" or "quantity stable") describing the change in the number of lit lights. For each prior label \(l_i\), the computer system uses the Word2Vec model to extract its semantic implicit representation vector \(v_{l_i}=[v_{i1},v_{i2},\cdots,v_{id}]\), where \(d\) is the dimension of the vector. Then, these vectors are arranged in chronological order to generate the time - series set of semantic implicit representations \(\{v_{l_1},v_{l_2},\cdots,v_{l_m}\}\).

[0064] After generating the time - series set of semantic implicit representations, the computer system can execute step S411 to combine each lighting information in the time - series set of past lighting information with the corresponding semantic implicit representation to generate integrated information.

[0065] The computer system first ensures that the elements in the time - series set of past lighting information and the time - series set of semantic implicit representations strictly correspond in time. This is usually achieved by comparing timestamps or unique identifiers.

[0066] For each element \((t_i,n_i)\) in the time - series set of past lighting information and the corresponding element \(v_{l_i}\) in the time - series set of semantic implicit representations, the computer system combines them into a new data structure. This combination can be a simple concatenation operation, connecting the number of lit lights \(n_i\) (which can be regarded as a one - dimensional vector or scalar) with the semantic implicit representation vector \(v_{l_i}\). After the combination, the computer system generates an integrated information item for each time point, and these items are arranged in chronological order to form the time - series set of past integrated information. Each integrated information item contains the number of lit lights and the related semantic information.

[0067] Continuing with the previous example, for the lighting information \((t_1,n_1)\) at time point \(t_1\) and the corresponding semantic implicit representation vector \(v_{l_1}\), the computer system combines them into the integrated information item \((t_1,[n_1,v_{l_1}])\). Here, \([n_1,v_{l_1}]\) is an extended vector that concatenates the number of lit lights \(n_1\) and the semantic implicit representation vector \(v_{l_1}\). The same operation is performed for each element in the time - series set of past lighting information, and finally a complete time - series set of past integrated information is generated.

[0068] The generated integrated information time series set provides a richer and more comprehensive data basis for subsequent data analysis and model detection. It not only contains the actual detection data of the number of lit lights, but also incorporates the related semantic information. This integration method enables the model to capture both the change patterns of the number of lit lights and the reasons or trends behind these changes during the training process, thereby improving the monitoring accuracy and generalization ability of the model.

[0069] As an implementation solution, in step S500, based on the past integrated information time series set, estimate the lit light information of the target expected number of lit lights at each of multiple sequentially distributed analysis times, and obtain an estimated lit light information time series set, which may specifically include:

[0070] Step S510: Decompose the past integrated information time series set according to the semantic implicit representation included in each integrated information in the past integrated information time series set, and obtain multiple segmented time series subsets; where, the semantic implicit representations included in the integrated information in the same segmented time series subset are the same, and the semantic implicit representations included in two adjacent segmented time series subsets are different;

[0071] Step S520: Based on the multiple segmented time series subsets, estimate the lit light information of the target expected number of lit lights at each of multiple sequentially distributed analysis times, and obtain an estimated lit light information time series set.

[0072] The objective of step S500 is to estimate the target expected number of lit lights at multiple future analysis times based on the past integrated information time series set, so as to form an estimated lit light information time series set.

[0073] In step S510, the computer system decomposes the past integrated information time series set to generate multiple segmented time series subsets. The purpose of this step is to aggregate data segments with similar semantic features based on the semantic implicit representation in the integrated information for more targeted analysis and detection in the subsequent process.

[0074] The computer system first traverses the past integrated information time series set and extracts the semantic implicit representations in each integrated information item. These semantic implicit representations usually exist in the form of vectors, and the values on each dimension reflect a certain semantic feature of the prior label. Next, the computer system calculates the similarity between these semantic implicit representations. This can be achieved through various methods, such as cosine similarity, Euclidean distance, etc. According to the calculated similarity, the computer system divides the past integrated information time series set into multiple subsets. The basis for division is whether the semantic implicit representations in the integrated information are similar. Specifically, the integrated information items with a similarity higher than a certain threshold can be grouped into the same subset. The purpose of doing this is to aggregate data segments with the same or similar semantic features for subsequent analysis.

[0075] Through the above process, the computer system generates multiple segmented time-series subsets. Each subset contains a series of integrated information items that are consecutive in chronological order and have similar implicit semantic representations.

[0076] Suppose the historical integrated information time-series set contains the following integrated information items:

[0077] Item 1: [80, v 稳定 , Item 2: [82, v 稳定 , Item 3: [78, v 减少 , Item 4: [75, v 减少 , Item 5: [85, v 增加 . Among them, v 稳定 , v 减少 and v 增加 are different implicit semantic representation vectors. By calculating the similarity and setting a suitable threshold, the computer system may segment the above integrated information items into the following three subsets: Subset 1: {[80, v 稳定 , [82, v 稳定}, Subset 2: {[78, v 减少 , [75, v 减少}, Subset 3: {[85, v 增加}.

[0078] [[ID=ID=37]]In step S520, the computer system uses the multiple segmented time-series subsets generated in step S510 to detect the target number of lights to be lit at multiple analysis times. This step usually involves the training and detection processes of a machine learning model. The computer system selects a suitable detection model to estimate the future number of lights to be lit. Since this is a time-series detection problem and the data has been segmented into subsets with similar semantic features, models suitable for processing time-series data, such as recurrent neural network models like LSTM (Long Short-Term Memory) and GRU (Gated Recurrent Unit), can be selected. If there is already a ready-made model trained for a similar problem, it can be directly used; otherwise, the model needs to be retrained based on the existing data.

[0079] If a model that needs to be trained (such as LSTM) is selected, the computer system will use each segmented time-series subset as training data to train the model. During the training process, the model will learn how to detect the future number of lights to be lit from the historical number of lights to be lit and their implicit semantic representations. This usually involves minimizing a loss function (such as Mean Squared Error MSE), which measures the difference between the model's detection value and the actual value.

[0080] After training is completed, the computer system uses the trained model to detect the number of lit lights at multiple future analysis times. For each segmented time series subset, a corresponding detection model can be trained, and then detections are performed separately and the results are combined.

[0081] Arrange the detection results of all analysis times in chronological order to form a time series set of detected lit light information.

[0082] Continuing with the previous example, assume that the computer system has trained an LSTM model for each segmented time series subset. For subset 1 (stable state), the model may learn that the number of lit lights remains relatively stable in a short period; for subset 2 (decreasing state), the model may learn the trend of the number of lit lights gradually decreasing; for subset 3 (increasing state), the model may learn the trend of the number of lit lights increasing.

[0083] Through the decomposition of the past integrated information time series set and the detection based on the segmented time series subsets, the computer system can more accurately detect the target number of lit lights at multiple analysis times. This process not only makes full use of the lit light number information in historical data but also deeply explores the related semantic features, thereby improving the accuracy and reliability of monitoring.

[0084] As an implementation solution, in step S520, based on the multiple segmented time series subsets, estimate the lit light information at each analysis time among multiple sequentially distributed analysis times for the target number of lit lights, and obtain an estimated lit light information time series set, which may specifically include:

[0085] Step S521: Perform embedding mapping on the multiple segmented time series subsets respectively to obtain the corresponding embedding mapping implicit representations of each segmented time series subset;

[0086] Step S522: Based on the corresponding embedding mapping implicit representations of each segmented time series subset, generate an embedding mapping implicit representation sequence corresponding to the past lit light information time series set;

[0087] Step S523: Based on the embedding mapping implicit representation sequence corresponding to the past lit light information time series set, estimate the lit light information at each analysis time among multiple sequentially distributed analysis times for the target number of lit lights, and obtain an estimated lit light information time series set.

[0088] In step S521, the computer system performs embedding mapping (encoding) on each segmented time series subset to generate their respective embedding mapping implicit representations. This process aims to convert the original time series data into a representation in a higher-dimensional or lower-dimensional space, which can capture the internal structure and features of the data and facilitate subsequent detection tasks.

[0089] First, the computer system selects a suitable embedding mapping method. Since the segmented time series subset contains a series of integrated information items (each item consists of the number of lit lights and semantic implicit representation), an embedding mapping method that can handle sequence data can be selected, such as Recurrent Neural Network (RNN), Long Short-Term Memory Network (LSTM), Gated Recurrent Unit (GRU), or Transformer, etc. These methods can capture the temporal dependencies and long-term dependencies in the sequence. Taking LSTM as an example, it is a special RNN that addresses the deficiencies of traditional RNNs in dealing with long-term dependencies by introducing a forget gate, an input gate, and an output gate. LSTM can transform each element in the sequence (here, the integrated information items in the segmented time series subset) into a vector representation of a fixed dimension, and this vector is the implicit representation of the embedding mapping.

[0090] Next, the computer system trains the embedding mapping model. This usually involves a supervised learning process, where the goal of the model is to minimize a certain loss function that measures the difference between the model output (the implicit representation of the embedding mapping) and the expected output (which may be a form of true label or reconstruction error). However, in this scenario, there is no direct expected output to supervise the embedding mapping process; instead, it is hoped to learn a useful representation through the embedding mapping that can play a role in subsequent detection tasks. Therefore, a commonly used strategy is unsupervised learning or self-supervised learning. For example, the structure of an autoencoder can be used to train the LSTM model, where the encoder part encodes the segmented time series subset into the implicit representation of the embedding mapping, and the decoder part attempts to reconstruct the original time series data from the implicit representation of the embedding mapping.

[0091] However, in this specific scenario, it is more likely to indirectly evaluate the effect of the embedding mapping model based on its performance in subsequent detection tasks. That is to say, these implicit representations of the embedding mapping will be used for detection in step S523, and the parameters of the embedding mapping model will be adjusted according to the accuracy of the detection results.

[0092] Once the embedding mapping model is trained (or reaches a certain predetermined number of training epochs), the computer system can apply it to all segmented time series subsets to generate their respective implicit representations of the embedding mapping. These representations will be used as inputs for subsequent tasks.

[0093] Suppose there is a segmented time series subset that contains the following integrated information items (simplified representation, omitting the timestamp and some details of the semantic implicit representation):

[0094] Subset 1: {[80, v1], [82, v1]} (indicating that the number of lit lights is stable); Subset 2: {[78, v2], [75, v2]} (indicating that the number of lit lights is decreasing). Here, v1 and v2 are different semantic implicit representation vectors.

[0095] The computer system uses the LSTM model to perform embedding mapping on this subset. For each integrated information item in Subset 1, the LSTM model will sequentially read their number of lit lights and semantic implicit representation (which can be combined into a vector), and generate an embedding mapping implicit representation vector of a fixed dimension, such as e1. Similarly, for Subset 2, a corresponding embedding mapping implicit representation vector e2 will be generated.

[0096] In step S522, the task of the computer system is to combine the embedding mapping implicit representations of each segmented time series subset generated in step S521 to form a complete sequence of embedding mapping implicit representations. This sequence will be used as one of the inputs for subsequent tasks.

[0097] First, the computer system arranges the embedding mapping implicit representations of each segmented time series subset in the time order of the original past lit light information time series set. This ensures that the sequence of embedding mapping implicit representations can retain the time dependence of the original data. Then, the computer system performs a certain form of merging on the embedding mapping implicit representations of adjacent segmented time series subsets with similar semantic implicit representations.

[0098] If a merging operation is indeed required, one possible method is to use the attention mechanism (AttentionMechanism) to determine the relative importance between the embedding mapping implicit representations of different subsets and perform weighted merging accordingly. However, this method is usually used in more complex sequence-to-sequence (Seq2Seq) tasks and requires an additional model to learn the attention weights.

[0099] Finally, the computer system generates a complete sequence of embedding mapping implicit representations. Each element in this sequence is an embedding mapping implicit representation vector of a segmented time series subset, and these vectors are arranged in the time order of their appearance in the original past lit light information time series set.

[0100] Continuing with the previous example, assume there are two embedding mapping implicit representation vectors e1 and e2 of segmented time series subsets. The computer system arranges them in the time order of the original data (assuming Subset 1 is earlier in time than Subset 2) to generate the sequence of embedding mapping implicit representations [e1, e2].

[0101] In step S523, the final task of the computer system is to use the implicit representation sequence of the embedding mapping generated in step S522 to detect the lighting information at each of the multiple analysis times. This process typically involves the application of a detection model that can detect the number of lit lights based on the implicit representation of the embedding mapping of historical data.

[0102] First, the computer system selects a suitable detection model. Since this is a time series detection problem and there is already an implicit representation sequence of the embedding mapping of the data, some models specifically designed for time series detection can be selected, such as LSTM, GRU, Temporal Convolutional Network (TCN), or a simple fully connected neural network (if the implicit representation of the embedding mapping has already captured the internal structure of the data well). In this scenario, since LSTM has been used to generate the implicit representation sequence of the embedding mapping, it may be a natural choice to continue using LSTM as the detection model. However, this is not absolute; the most suitable model can be selected according to the specific characteristics of the problem and experimental results.

[0103] Next, the computer system uses the implicit representation sequence of the embedding mapping of historical data to train the detection model. This typically involves a supervised learning process where the goal of the model is to minimize the difference between the detected number of lit lights and the actual number of lit lights (measured by a certain loss function).

[0104] In this process, the number of lit lights at the last few time points in the chronological set of past lighting information can be used as the training target (i.e., the expected output), and the implicit representation sequence of the embedding mapping before them can be used as the input of the model. In this way, a model can be trained to detect the number of lit lights based on the implicit representation of the embedding mapping of historical data.

[0105] It should be noted that since it is working in an environment where new data is continuously collected (such as an Internet of Things-based vehicle-road collaborative street lamp system), the detection model can be retrained regularly to adapt to new data distributions and patterns. Once the detection model is trained and its performance is verified to meet the requirements, the computer system can use it to detect the lighting information at each of the multiple analysis times. This typically involves using the implicit representation sequence of the embedding mapping as the input of the model and obtaining the sequence of detected numbers of lit lights output by the model.

[0106] Continuing with the previous example, assume that an LSTM detection model has been trained and that this model can detect the number of future lit lights based on the implicit representation of the embedding mapping of historical data. Now, this model can be applied to the implicit representation sequence [e1, e2] generated in step S522 for detection.

[0107] The detection process is generally as follows: The embedded mapping implicit representation sequence [e1, e2] is used as the input of the LSTM detection model. The LSTM model sequentially reads each embedded mapping implicit representation vector in the sequence (i.e., e1 and e2), and uses its internal state to generate the detection of the number of lights to be turned on in the future. For each analysis moment (such as every night within the next week), the LSTM model outputs a detected number of lights to be turned on. Arranging the detection results of all analysis moments in chronological order forms a time series set of estimated light-on information.

[0108] As an implementation solution, the embedded mapping implicit representation is generated based on the embedded mapping neural network after debugging. The debugging process of the embedded mapping neural network includes:

[0109] Step S10: Obtain positive training example pairs. The positive training example pairs include training examples corresponding to the target number of lights to be turned on in several past time intervals with intersections. The training examples corresponding to the past time intervals include the light-on information of the target number of lights to be turned on at each past moment in the past time intervals. The light-on information in the training examples is arranged according to their respective chronological order.

[0110] Positive training examples are positive samples. In step S10, the computer system performs the following operations to obtain positive training example pairs:

[0111] First, the computer system determines several past time intervals with intersections. The selection of these time intervals should be based on the availability of actual data and the requirements of the application scenario.

[0112] For each selected time interval, the computer system collects the lighting information at each past moment within that interval. This information typically comes from the real-time monitoring data of Internet of Things devices (such as on-vehicle cameras, GPS locators, etc.), and after processing, the number of lit lights at each time point is obtained. For example, for time interval A (from July 1st to July 7th, 2024), the system may collect the following sequence of lighting information: {(tA1,nA1),(tA2,nA2),…,(tA7,nA7)}, where tAi represents the i-th time point within interval A, and nAi represents the number of lit lights at that time point. Next, the computer system combines the lighting information sequences corresponding to these time intervals with intersections into positive training example pairs. Each pair contains two or more training examples that are close in time and partially overlapping. The intersection part between these examples reflects the continuous change in the number of lit lights, while the non-intersection part may show sudden changes or periodic changes in the number of lit lights. For example, if time interval A overlaps partially with time interval B (from July 4th to July 10th, 2024), then the lighting information sequence of interval A can be combined with the lighting information sequence of the corresponding overlapping part in interval B to form a positive training example pair.

[0113] Suppose there are two time intervals:

[0114] Interval A: from July 1st to July 7th, 2024 (mainly weekdays);

[0115] Interval B: from July 4th to July 10th, 2024 (including weekdays and weekends);

[0116] For these two intervals, the computer system may collect the following sequence of lighted lamp information: Sequence of lighted lamp information for Interval A: {(2024-07-01, 80), (2024-07-02, 82), …, (2024-07-07, 85)}; Sequence of lighted lamp information for Interval B (note that only the overlapping part with Interval A and the subsequent part are listed here): {(2024-07-04, 83), (2024-07-05, 84), …, (2024-07-10, 90)}. Since Interval A and Interval B overlap during the period from July 4, 2024 to July 7, 2024, the computer system can combine the sequences of lighted lamp information for these two time periods into a positive training example pair: Positive training example pair: ({(2024-07-01, 80), …, (2024-07-07, 85)}, {(2024-07-04, 83), …, (at least including the corresponding data up to 2024-07-07, …)}). In actual operation, to maintain the consistency of the pair, it is usually necessary to ensure that the time points and the number of lighted lamps in the overlapping part of the two sequences are exactly the same (or at least very close), or align them by some means (such as interpolation, resampling, etc.).

[0117] Step S20: Obtain a negative training example pair, which includes training examples corresponding to the target number of lighted lamps in several non-overlapping past time intervals.

[0118] Negative training examples are negative samples. During the training process of the embedding mapping neural network, Step S20 is responsible for obtaining negative training example pairs. These sample pairs play a crucial role in training the neural network. They help the network distinguish dissimilar or irrelevant data patterns, thereby improving its generalization ability and robustness to unseen data. In Step S20, the computer system performs the following operations to obtain negative training example pairs:

[0119] Different from the positive training examples in Step S10, the time intervals in the negative training example pair should not have intersections. This means that the lighted lamp information within these intervals is not continuous in time and may represent completely different lighted lamp patterns or scenarios. The computer system carefully selects these time intervals to ensure that they can cover various lighted lamp situations that may be encountered in actual applications.

[0120] For each selected non - overlapping time interval, the computer system collects the lighting information at each past moment within that interval. These information also come from the real - time monitoring data of IoT devices. However, due to the discontinuity of the time intervals, there may not be an obvious correlation or trend among these data. After collecting the lighting information for all selected time intervals, the computer system constructs these information into negative training example pairs. Each pair contains two or more training examples that are discontinuous in time and may have completely different lighting patterns. The differences between these examples help the network learn how to distinguish different lighting situations and avoid making false detections on unseen data.

[0121] Suppose the following two non - overlapping time intervals are selected as the sources of negative training examples:

[0122] Interval C: During the Spring Festival in 2024 (the number of lit lights may increase significantly);

[0123] Interval D: An ordinary working day in summer 2024 (the number of lit lights is relatively stable);

[0124] For these two intervals, the computer system may collect the following lighting information sequences: Lighting information sequence for Interval C (during the Spring Festival): {(2024 - 02 - 10, 150), (2024 - 02 - 11, 160),...}, indicating that during the Spring Festival, due to celebration activities and other reasons, the number of lit lights increases significantly. Lighting information sequence for Interval D (ordinary working day): {(2024 - 07 - 15, 80), (2024 - 07 - 16, 82),...}, indicating that on an ordinary working day in summer, the number of lit lights remains relatively stable. Since Interval C and Interval D are discontinuous in time and have significantly different lighting patterns, the computer system can combine the lighting information sequences of these two time intervals into a negative training example pair: ({(2024 - 02 - 10, 150), (2024 - 02 - 11, 160),...}, {(2024 - 07 - 15, 80), (2024 - 07 - 16, 82),...}). This negative training example pair provides contrasting data support for the training of the embedding mapping neural network. By allowing the network to learn these dissimilar or unrelated data patterns, it can help the network make more accurate and robust detections when faced with unseen data.

[0125] Step S30: For each training example in the positive training example pair, integrate each lighting information in the training example with the semantic implicit representation of the corresponding prior label to obtain integrated information, and generate an integrated information sequence corresponding to each training example in the positive training example pair.

[0126] During the training process of the embedded mapping neural network, step S30 integrates the lighting information in the positive training examples with the corresponding prior label semantic implicit representations to generate an integrated information sequence containing rich context information. In this step, the computer system performs the following operations:

[0127] The computer system traverses all positive training example pairs. Each pair contains two or more training examples corresponding to time intervals with intersections, and these examples reflect the continuity and similarity of the change in the number of lit lights over time. For each training example in the pair, the computer system extracts the lighting information therein (i.e., the number of lit lights at each time point) and the corresponding prior label. The prior label is a descriptive label for the trend or state of the change in the number of lit lights, such as "quantity increasing", "quantity stable", etc. These labels are associated with the lighting information in some way (such as manual annotation or automatic extraction). Next, the computer system uses a pre-trained word embedding model or a custom feature extraction method to convert each prior label into a semantic implicit representation. This representation is usually a vector of a fixed dimension that can capture the semantic features of the label. For example, word embedding models such as Word2Vec, GloVe, etc. can be used to generate the vector representation of the label. For each lighting information item (including the time point and the number of lit lights) in the training example, the computer system integrates it with the corresponding prior label semantic implicit representation. The integration method can be a simple concatenation operation, connecting the number of lit lights (possibly converted into a vector form) with the semantic implicit representation vector to form a new integrated information vector. Finally, the computer system arranges all the integrated information vectors in chronological order to generate the corresponding integrated information sequence. This sequence not only contains the original lighting quantity information but also incorporates the semantic information of the prior labels, providing rich data support for the subsequent neural network training.

[0128] Suppose there is a positive training example pair, which contains two training examples corresponding to time intervals with intersections. For one of the training examples (taking time interval A as an example), the lighting information sequence is: {(tA1,nA1),(tA2,nA2),…,(tAk,nAk)}, where tAi represents the time point and nAi represents the number of lit lights at the corresponding time point. The prior label sequence associated with this training example may be: {lA1,lA2,…,lAk}, where each lAi is a label describing the trend of the change in the number of lit lights (such as "quantity increasing").

[0129] For each prior label lAi, the computer system uses the word embedding model to generate its semantic implicit representation vector vAi. Then, each lighting information item (tAi,nAi) is integrated with the corresponding semantic implicit representation vector vAi to generate an integrated information vector. For example, the integrated information vector can be obtained through a concatenation operation: , where is the process of converting the number of lit lights into a vector form (if itself is of numerical type, it can be encoded or embedded in some form first). Finally, all the integrated information vectors are arranged in chronological order to generate an integrated information sequence: .

[0130] This integrated information sequence will be used as one of the inputs to the embedding mapping neural network for the training process. By integrating the semantic implicit representation of the lit light information and the prior labels, the computer system provides the neural network with richer and more comprehensive learning materials, which helps to improve its detection accuracy and robustness.

[0131] Step S40: For each training example in the negative training example pair, integrate each lit light information in the training example with the semantic implicit representation of the corresponding prior label to obtain integrated information, and generate an integrated information sequence corresponding to each training example in the negative training example pair.

[0132] During the training process of the embedding mapping neural network, Step S40 is similar to Step S30 in operation, but it is for the negative training example pair. This step aims to integrate the lit light information in the negative training example with the semantic implicit representation of the corresponding prior label to generate an integrated information sequence, so as to provide the neural network with more diverse learning materials and enhance its ability to distinguish different lit light patterns. The following is a detailed explanation of Step S40, illustrated with a specific application scenario.

[0133] In this step, the computer system performs the following operations to process the negative training example pair:

[0134] First, the computer system traverses all negative training example pairs. Each pair contains training examples corresponding to two or more non - overlapping time intervals, and these examples may have very different lighting patterns, reflecting the street - lamp lighting conditions in different scenarios or conditions. For each training example in the pair, the computer system extracts the lighting information (i.e., the number of lit lamps at each time point) and the corresponding prior label. Similar to positive training examples, these prior labels describe a certain state or trend of change in the number of lit lamps, such as "decreasing quantity", "abnormal state", etc. Using a pre - trained word - embedding model or a custom feature - extraction method, the computer system converts each prior label into a semantic implicit representation. These representations are usually vectors of a fixed dimension, which can capture the semantic features of the label and serve as feature inputs in the subsequent network training process. For each lighting information item (including time point and number of lit lamps) in the training example, the computer system integrates it with the semantic implicit representation of the corresponding prior label. The integration method is usually a simple concatenation operation, connecting the number of lit lamps (which can be encoded or embedded first) with the semantic implicit representation vector to form an integrated information vector containing rich context information. Finally, the computer system arranges all the integrated information vectors in chronological order to generate the corresponding integrated information sequence. This sequence not only contains the actual observed values of the number of lit lamps but also incorporates the semantic information of the prior labels, providing more comprehensive training data for the neural network.

[0135] Suppose there is a negative training example pair that contains training examples corresponding to two non - overlapping time intervals. Taking one of the time intervals (such as interval D) as an example, its lighting information sequence may be: {(tD1,nD1),(tD2,nD2),…,(tDn,nDn)}, where tDi represents the time point and nDi represents the number of lit lamps at the corresponding time point. The prior label sequence associated with this training example is: {lD1,lD2,…,lDn}. Each lDi is a label describing the lighting state, such as "abnormal decrease in quantity". For each prior label lDi, the computer system uses a word - embedding model (such as Word2Vec) to generate its semantic implicit representation vector vDi. Then, each lighting information item (tDi,nDi) is integrated with the corresponding semantic implicit representation vector vDi to generate an integrated information vector. For example, through the concatenation operation, we get: , where, is the process of converting the number of lit lamps into vector form. Finally, all the integrated information vectors are arranged in chronological order to generate the integrated information sequence: This integrated information sequence will be used as one of the training data for the embedding mapping neural network, helping the network learn the differences between different lighting patterns, thereby improving its discrimination ability and detection accuracy when facing unseen data.

[0136] Step S50: Debug the embedding mapping neural network based on the integrated information sequences respectively corresponding to each training example in the positive training example binary group and the integrated information sequences respectively corresponding to each training example in the negative training example binary group.

[0137] During the training process of the embedding mapping neural network, Step S50 uses the integrated information sequences of positive and negative training examples to debug (i.e., train and optimize) the neural network. This step aims to adjust the parameters of the neural network so that it can accurately capture the pattern of changes in the number of lit lights and effectively distinguish the data features under different lighting patterns.

[0138] In this step, the computer system performs the following operations to debug the embedding mapping neural network:

[0139] The computer system first ensures that the integrated information sequences of positive and negative training examples have been generated according to steps S30 and S40. These sequences contain the integrated information of the number of lit lights and the implicit representation of the prior label semantics, providing rich learning materials for the neural network. Next, the computer system defines a suitable embedded mapping neural network structure. This network may consist of multiple layers, such as an input layer, hidden layers (which may contain recurrent neural network units such as LSTM, GRU, etc. to process sequence data), and an output layer. Each layer has its specific parameters (such as weights and biases), which will be optimized during the training process. To evaluate the performance of the neural network and guide parameter optimization, the computer system selects a suitable loss function (such as mean squared error MSE, cross-entropy loss, etc.) and an optimization algorithm (such as stochastic gradient descent SGD, Adam, etc.). The loss function is used to quantify the difference between the network detection value and the actual value, while the optimization algorithm is used to adjust the network parameters according to the gradient of the loss function. During the training process, the computer system first inputs the integrated information sequences into the neural network for forward propagation. During forward propagation, the input data passes through each layer of the network, and finally, a detection output is generated. Then, the system calculates the loss value between the detection output and the actual output, and propagates the loss value back to each layer of the network through the backpropagation algorithm to calculate the gradient of each layer. Using the gradient information obtained from backpropagation, the computer system updates the parameters (weights and biases) of the neural network. This step usually involves multiplying the gradient by a preset learning rate and subtracting the product from the current parameter value to optimize the parameters. The above process (forward propagation, loss calculation, backpropagation, and parameter update) will be repeated multiple times on the entire training set until a certain stopping condition is met (such as the loss value is less than a certain threshold, the number of iterations reaches a preset upper limit, etc.). Each iteration aims to further reduce the loss value and improve the detection accuracy of the neural network.

[0140] During the training process, the positive integrated information sequences are input into the network for forward propagation to obtain a detection output. Then, the MSE loss value between the detection output and the actual output is calculated. Next, the gradients of each layer are calculated through the backpropagation algorithm, and the weights and biases of the network are updated using the Adam optimization algorithm. This process will be repeated multiple times on the entire positive training set. At the same time, the integrated information sequences of negative training examples will also be used in the same way for the training process of the network. By considering the data characteristics of both positive and negative training examples simultaneously, the computer system can debug an embedded mapping neural network model that can accurately detect changes in the number of lit lights and distinguish different lighting patterns. Finally, the trained neural network model will be used in the actual vehicle-road collaborative street lamp lighting intelligent recognition task to achieve real-time monitoring and detection of the street lamp lighting status.

[0141] As an implementation solution, step S50, based on the integrated information sequences respectively corresponding to each training example in the positive training example pair and the integrated information sequences respectively corresponding to each training example in the negative training example pair, debugs the embedding mapping neural network, specifically including:

[0142] Step S51: Load the integrated information sequences respectively corresponding to each training example in the positive training example pair into the embedding mapping neural network respectively for embedding to obtain multiple embedding mapping implicit representations, and obtain the embedding mapping implicit representation sequences respectively corresponding to each training example in the positive training example pair.

[0143] In step S51, the computer system inputs the integrated information sequence corresponding to each training example in the positive training example pair into the embedding mapping neural network, and through the calculation process of the network, these sequences are converted into embedding mapping implicit representations, thereby generating an embedding mapping implicit representation sequence unique to each training example. This process is a key step in neural network training because it provides rich feature representations for subsequent model debugging and optimization. The computer system first traverses the positive training example pair, and for each training example therein, extracts its corresponding integrated information sequence. These sequences contain the integrated information of the number of lit lights and the prior label semantic implicit representation, and are the input data for subsequent embedding mapping. Before performing embedding mapping, a suitable neural network structure can be defined. This network usually includes multiple layers, such as an input layer, hidden layers (which may include recurrent neural network units such as LSTM and GRU to process sequence data), and an output layer. The specific structure of the network depends on the complexity of the problem and the characteristics of the data. In this scenario, since time series data is being processed, the hidden layer may adopt units such as LSTM or GRU that can capture sequence dependencies.

[0144] Taking the integrated information sequence of each training example as input, load it into the embedding mapping neural network. The network calculates and transmits information layer by layer through the forward propagation process, and finally generates a series of embedding mapping implicit representations. These representations are the mappings of the original integrated information sequences in the neural network feature space, and they capture the key information and patterns in the sequences. For each input integrated information sequence, the embedding mapping neural network outputs a corresponding embedding mapping implicit representation sequence. Each element in this sequence is an embedding vector, which represents the feature representation of a certain time point or segment in the original sequence.

[0145] For example, assume there is a positive training example pair, which contains two training examples A and B, corresponding to the lighting information of two time intervals with an intersection. The integrated information sequence of training example A: {[encode(nA1), vA1], [encode(nA2), vA2],...}; The integrated information sequence of training example B: {[encode(nB1), vB1], [encode(nB2), vB2],...}; where encode(nAi) and encode(nBi) are the encoded representations of the lighting quantities nAi and nBi respectively (if they are scalar values, numerical embedding may be performed first), and vAi and vBi are the semantic implicit representation vectors of the prior labels. These integrated information sequences are respectively loaded into the embedding mapping neural network, and the network processes these sequences through its internal structure (such as LSTM units) and outputs the corresponding embedding mapping implicit representation sequences: The embedding mapping implicit representation sequence of training example A: {eA1, eA2,...}; The embedding mapping implicit representation sequence of training example B: {eB1, eB2,...}; where eAi and eBi are the embedding vectors generated by the network, which capture the feature information of the corresponding time points or segments in the integrated information sequence.

[0146] Through step S51, the computer system successfully converts the integrated information sequences in the positive training examples into embedding mapping implicit representation sequences. These representations not only contain the key information of the original data but also introduce higher-level feature abstractions through the non-linear transformation of the neural network.

[0147] Step S52: The integrated information sequences respectively corresponding to each training example in the negative training example pair are respectively loaded into the embedding mapping neural network for embedding to obtain multiple embedding mapping implicit representations, and the embedding mapping implicit representation sequences respectively corresponding to each training example in the negative training example pair are obtained.

[0148] In step S52, the task of the computer system is similar to that in step S51, but the processing object changes from the positive training example pair to the negative training example pair. The goal of this step is to convert the integrated information sequences respectively corresponding to each training example in the negative training examples through the embedding mapping neural network to generate their respective unique embedding mapping implicit representation sequences.

[0149] The computer system traverses the negative training example pairs, and for each of these training examples, it extracts the corresponding integrated information sequence. These sequences also contain the integrated information of the number of lit lights and the implicit representation of the prior label semantics, but compared with the positive training examples, they represent different or relatively independent lit light patterns. Similar to step S51, the computer system uses the predefined embedding mapping neural network to process these negative training examples. This means that the structure and parameters of the network remain unchanged between steps S51 and S52, ensuring the consistency and comparability of data processing. The integrated information sequence of each negative training example is used as input and loaded into the embedding mapping neural network. The network processes the input sequence layer by layer through its internal structure and parameters, and finally generates a series of embedding mapping implicit representations. These representations capture the inherent characteristics and patterns of the lit light data in the negative training examples. For each input integrated information sequence of the negative training example, the embedding mapping neural network outputs a corresponding embedding mapping implicit representation sequence. Each embedding vector in the sequence is a point in the neural network feature space of the original data, representing the characteristics of the lit light state at a specific time point or segment. For example, assume there is a negative training example pair containing two training examples C and D, which respectively correspond to the lit light information of two non-overlapping time intervals.

[0150] Integrated information sequence of training example C: {[encode(nC1), vC1], [encode(nC2), vC2],...};

[0151] Integrated information sequence of example D: {[encode(nD1), vD1], [encode(nD2), vD2],...};

[0152] Similar to step S51, these integrated information sequences are respectively loaded into the embedding mapping neural network. The network processes these sequences through its recurrent neural network units such as LSTM or GRU, and outputs the corresponding embedding mapping implicit representation sequences:

[0153] Embedding mapping implicit representation sequence of training example C: {eC1, eC2,...};

[0154] Embedding mapping implicit representation sequence of training example D: {eD1, eD2,...};

[0155] These embedding vectors eCi and eDi capture the characteristics of the lit light data in the negative training examples and are distinguishable from the embedding vectors of the positive training examples in the feature space.

[0156] Through step S52, the computer system converts the integrated information sequences in the negative training examples into an embedded mapping implicit representation sequence. This process not only deepens the network's understanding of different lighting patterns but also provides key data support for subsequent model debugging and optimization.

[0157] Step S53: Determine the commonality measurement results between the embedded mapping implicit representation sequences respectively corresponding to each training example in the positive training example pair, and obtain the commonality measurement result corresponding to the positive training example pair.

[0158] Commonality measurement is a quantitative metric used to evaluate the similarity or consistency between different data sequences. In the context of positive training examples, commonality measurement can reflect the continuity or periodicity of the lighting data changes within different time intervals. Commonly used commonality measurement methods include cosine similarity, Pearson correlation coefficient, dynamic time warping (DTW), etc. The computer system first traverses each pair of training examples in the positive training example pair. For each pair of training examples, they each correspond to an embedded mapping implicit representation sequence. Then, the system uses the selected commonality measurement method to calculate the commonality measurement result between these two sequences. Since the positive training example pair may contain multiple pairs of training examples, the computer will obtain multiple commonality measurement results. To obtain an overall commonality measurement result to represent the entire pair, the computer system can use an average value, median, or other aggregation functions to summarize these results.

[0159] For example, assume that the positive training example pair contains two training examples A and B, which respectively correspond to the lighting data in two time intervals with an intersection. Through step S51, the embedded mapping implicit representation sequences corresponding to these two training examples have been obtained:

[0160] Embedded mapping implicit representation sequence of training example A: {eA1, eA2, …, eAn};

[0161] Embedded mapping implicit representation sequence of training example B: {eB1, eB2, …, eBm};

[0162] Next, the computer system uses cosine similarity as the commonality measurement method to calculate the similarity between these two sequences. Assume that the calculated cosine similarity is 0.8, which indicates a high similarity between the embedded mapping implicit representation sequences of training examples A and B, reflecting the continuity of the lighting data changes within these two time intervals. If the positive training example pair contains multiple pairs of training examples, the computer system will repeat the above calculation process and aggregate all the obtained commonality measurement results (such as taking the average value) to obtain the commonality measurement result of the entire pair.

[0163] Through step S53, the computer system can quantitatively evaluate the commonalities or similarities between the implicit representation sequences of the embedding mappings corresponding to each training example in the positive training example pairs. This process not only helps to understand the variation patterns of the lit data but also provides important feedback information for subsequent network debugging and optimization.

[0164] Step S54: Determine the commonality measurement results between the implicit representation sequences of the embedding mappings respectively corresponding to each training example in the negative training example pairs, and obtain the commonality measurement result corresponding to the negative training example pairs.

[0165] In step S54, the goal of the computer system is to evaluate the differences or lack of commonalities between the implicit representation sequences of the embedding mappings corresponding to each training example in the negative training example pairs. Different from step S53 which focuses on the commonalities of positive training examples, this step focuses on identifying the differences between negative training examples.

[0166] In the context of negative training examples, the commonality measurement no longer directly reflects similarity but is used to quantify the differences or distinguishability between different sequences. Therefore, the computer system may adopt a measurement method similar to that in step S53 (such as cosine similarity, Pearson correlation coefficient, etc.). However, here, a lower commonality measurement value (i.e., a larger difference) will be regarded as a positive result. For each pair of training examples in the negative training example pairs, the computer system first obtains their respective implicit representation sequences of the embedding mappings. Then, the selected commonality measurement method is used to calculate the difference value between these two sequences. Since negative training examples usually represent different lit patterns or conditions, it is expected that there will be a lower commonality measurement result between their implicit representation sequences of the embedding mappings. Similar to step S53, the computer system aggregates the commonality measurement results of all training example pairs in the negative training example pairs to obtain an overall difference measurement result. This can be achieved by calculating the average, median or other statistics of all difference values. However, different from positive training examples, the aggregated result here will be used to evaluate the overall distinguishability between negative training examples.

[0167] For example, assume that the negative training example pair contains two training examples C and D, which respectively correspond to the lit data of two non - overlapping time intervals. Through step S52, the implicit representation sequences of the embedding mappings corresponding to these two training examples have been obtained:

[0168] The implicit representation sequence of training example C: {eC1, eC2, …, eCn};

[0169] The implicit representation sequence of training example D: {eD1, eD2, …, eDm};

[0170] To quantify the difference between these two sequences, the computer system can use the reciprocal of the cosine similarity (or other appropriate difference metrics) to calculate the difference between them. However, a more direct approach is to use a metric that can directly reflect the difference, such as the square of the Euclidean distance:

[0171] ;

[0172] Suppose the calculated square value of the Euclidean distance is a relatively large number (indicating a significant difference between the sequences), which reflects that the lit lamp data patterns of training examples C and D have a large distinguishability in the feature space after the embedding mapping. If the negative training example pair contains multiple training example pairs, the computer system will repeat the above calculation process and aggregate all the obtained difference metric results (such as taking the average) to obtain the difference metric result of the entire pair. This result will be used in the subsequent network debugging and optimization process to ensure that the neural network can learn to effectively distinguish between different lit lamp patterns.

[0173] Through step S54, the computer system quantifies the difference between the implicit representation sequences of the embeddings corresponding to each training example in the negative training example pair. This process is of great significance for improving the distinguishability and generalization performance of the neural network in the intelligent identification scenario of vehicle-road collaborative street lamp lighting based on the Internet of Things. By ensuring that the negative training examples have sufficient distinguishability in the feature space, the neural network can better distinguish different lit lamp patterns and make more accurate detections.

[0174] Step S55: Debug the embedding mapping neural network based on the commonality metric result corresponding to the positive training example pair and the commonality metric result corresponding to the negative training example pair.

[0175] To quantify the difference between the network detection and the actual data and thereby guide the adjustment of network parameters, the computer system first defines a loss function. In this scenario, the design of the loss function takes into account the commonality metric results of positive and negative training examples. A possible method is to design a composite loss function that simultaneously considers two aspects: one is the commonality (i.e., similarity) between positive training examples, and the other is the difference between negative training examples.

[0176] For example, a loss function can be defined that includes two terms: one is the negative log-likelihood of the commonality metric result of positive training examples (encouraging commonality), and the other is the exponential function of the commonality metric result of negative training examples (penalizing commonality, i.e., encouraging difference). However, a more common practice may be to use a more direct metric, such as directly incorporating the commonality metric result into the loss function, or indirectly influencing the loss function through other mechanisms (such as regularization terms).

[0177] After defining the loss function, the computer system will use optimization algorithms (such as gradient descent, Adam, etc.) to minimize this loss function. During the optimization process, the computer will calculate the gradient of the loss function with respect to the network parameters and update the network parameters based on the gradient information. This step will be repeated multiple times until the value of the loss function is reduced to an acceptable level or the preset number of iterations is reached.

[0178] During the debugging process, the computer system ensures that the impacts of positive training examples and negative training examples on the adjustment of network parameters are balanced. This is usually achieved by adjusting the weights of each term in the loss function. For example, if the commonality metric result of positive training examples is already very high, while the difference metric result of negative training examples is not yet ideal, then the weight of the negative example term can be increased in the loss function so that more attention is paid to the differences of negative examples during the optimization process.

[0179] For example, assume that the commonality metric results corresponding to positive training example pairs and negative training example pairs have been obtained. To simplify the explanation, assume that the commonality metric result of positive training examples has been converted into a loss term (encouraging commonality) in a certain way, and the commonality metric result of negative training examples (which is actually the difference metric result) has also been converted into another loss term (penalizing commonality, i.e., encouraging differences). The loss function may be similar to the following form:

[0180]

[0181] where is the loss term calculated based on the commonality metric result of positive training examples, is the loss term calculated based on the difference metric result of negative training examples (note that here it is actually the reverse application of the commonality metric result), is a hyperparameter used to balance the impacts of the two losses. During the optimization process, the computer will calculate the gradient of this composite loss function with respect to the network parameters and use an optimization algorithm (such as Adam) to update the network parameters. Through multiple iterations, the network will gradually learn how to maintain the commonality of positive training examples while increasing the differences between negative training examples, thereby improving its performance in the intelligent identification task of vehicle-road collaborative street lamp lighting.

[0182] Through step S55, the computer system can debug the embedded mapping neural network according to the commonality metric results of positive training examples and negative training examples. This process involves defining the loss function, adjusting the network parameters, and balancing the impacts of different examples on the loss function. By continuously optimizing the network parameters to minimize the value of the loss function, the computer system can train a neural network model that performs excellently in the intelligent identification scenario of vehicle-road collaborative street lamp lighting.

[0183] As an implementation solution, in step S55, based on the commonality measurement results corresponding to the positive training example pairs and the commonality measurement results corresponding to the negative training example pairs, the embedding mapping neural network is debugged, which specifically includes:

[0184] Step S551: Perform a first pre-designed calculation on the commonality measurement results corresponding to multiple positive training example pairs respectively to obtain a first calculated commonality measurement result;

[0185] Step S552: Perform the first pre-designed calculation on the commonality measurement results corresponding to multiple negative training example pairs respectively to obtain a second calculated commonality measurement result;

[0186] Step S553: Debug the embedding mapping neural network according to the error between the first calculated commonality measurement result and the second calculated commonality measurement result.

[0187] In step S551, the computer system performs a summary processing on the commonality measurement results corresponding to multiple positive training example pairs respectively. The commonality measurement result reflects the similarity or continuity of the lighting data in different time intervals in the positive training examples, and is a key indicator for evaluating the network's ability to learn positive patterns.

[0188] First, the computer system traverses all positive training example pairs. For each pair of training examples within each pair, it uses a pre-defined commonality measurement method (such as cosine similarity, Pearson correlation coefficient, etc.) to calculate the commonality measurement results between their corresponding embedding mapping implicit representation sequences. These results form a set of commonality measurement values.

[0189] Next, the computer system performs a first pre-designed calculation on these commonality measurement results, that is, adopts a certain summary method (such as addition, averaging, etc.) to obtain a comprehensive commonality measurement result, which is called the first calculated commonality measurement result. The purpose of this result is to quantify the overall performance of all positive training example pairs in terms of commonality. Suppose there are three positive training example pairs, and each pair contains two training examples. For the first pair, the commonality measurement result is 0.8; for the second pair, it is 0.75; for the third pair, it is 0.85. If averaging is selected as the first pre-designed calculation method, then the first calculated commonality measurement result will be 0.8. This result reflects the average level of all positive training examples in terms of commonality and is an important indicator for measuring the network's ability to learn positive patterns.

[0190] Step S552 is similar to step S551. However, in this step, the computer system processes the commonality measurement results of negative training example pairs. The commonality measurement results of negative training examples reflect the differences or independence of the lighting data in different time intervals, and are key indicators for evaluating the network's ability to distinguish different lighting patterns.

[0191] The computer system traverses all negative training example pairs. For each pair of training examples within each pair, it uses the same commonality measurement method as for positive training examples to calculate the commonality measurement results between their corresponding embedded mapping implicit representation sequences. Since negative training examples represent different lighting patterns, it is expected that their commonality measurement results are low. Similar to step S551, the computer system performs a first pre-designed calculation on these commonality measurement results to obtain a comprehensive commonality measurement result, which is called the second calculated commonality measurement result. The purpose of this result is to quantify the overall performance of all negative training example pairs in terms of differences. Suppose there are also three negative training example pairs, and the commonality measurement results of each pair are 0.3, 0.25, and 0.35 respectively. If the average is still chosen as the first pre-designed calculation method, then the second calculated commonality measurement result will be 0.3. This result reflects the average level of all negative training examples in terms of differences and is an important indicator for measuring the network's ability to distinguish different lighting patterns.

[0192] In step S553, the computer system uses the first calculated commonality measurement result and the second calculated commonality measurement result obtained in the previous two steps to debug the embedded mapping neural network. The purpose of debugging is to reduce the error between the two calculated results by adjusting the network parameters, thereby improving the network's ability to distinguish between positive and negative patterns. First, the computer system calculates the error between the first calculated commonality measurement result and the second calculated commonality measurement result. This error reflects the deficiency of the network in simultaneously learning the positive pattern and distinguishing the negative pattern. The larger the error, the more unbalanced the network's performance in these two aspects. To guide the network debugging process, the computer system designs a loss function, which should be able to reflect the magnitude of the above error. A possible design is to directly use the error as part of the loss function, or combine the error with other types of losses (such as detection error) to form a composite loss function. After having the loss function, the computer system can apply optimization algorithms (such as gradient descent, Adam, etc.) to minimize the value of the loss function. During the optimization process, the computer will update the network parameters according to the gradient of the loss function with respect to the network parameters, thereby gradually reducing the error and improving the network performance. The above debugging process will be repeated multiple times until the value of the loss function is reduced to an acceptable level or the preset number of iterations is reached. In each iteration, the computer system will recalculate the commonality measurement results, the error, and the value of the loss function, and update the network parameters according to the new gradient of the loss function.

[0193] Suppose a simple loss function is defined, which consists of two parts: one part is the detection error (omitted here for simplicity), and the other part is the commonality measurement error, that is, the weighted sum of the difference between the first calculated commonality measurement result and the ideal positive commonality measurement (assumed to be 1, indicating complete commonality) and the second calculated commonality measurement result (the weights can be adjusted according to the actual situation).

[0194] During the debugging process, the computer system continuously adjusts the network parameters through an optimization algorithm to reduce the value of this loss function. As the iteration progresses, the first calculated commonality measurement result may gradually approach 1 (such as increasing to 0.9), while the second calculated commonality measurement result may further decrease (such as decreasing to 0.2). In this way, the value of the loss function will also decrease accordingly, indicating that the network's ability to learn positive patterns and distinguish negative patterns has been improved. Through the detailed analysis and example of step S'55', it can be seen how to use the commonality measurement results of positive training examples and negative training examples to debug the embedding mapping neural network. This process not only improves the network's ability to distinguish between positive and negative patterns, but also provides a more accurate and robust detection model for the subsequent intelligent identification task of vehicle-road collaborative street lamp lighting.

[0195] As another implementation solution, in step S55, based on the commonality measurement results corresponding to the positive training example pairs and the commonality measurement results corresponding to the negative training example pairs, the embedding mapping neural network is debugged, specifically including:[[]]

[0196] Step S55a: Analyze the concentration degree of the commonality measurement results corresponding to multiple positive training example pairs respectively to obtain the first analysis result;

[0197] Step S55b: Analyze the concentration degree of the commonality measurement results corresponding to multiple negative training example pairs respectively to obtain the second analysis result;

[0198] Step S55c: Debug the embedding mapping neural network according to the first analysis result and the second analysis result.

[0199] In step S55a, the computer system performs a concentration analysis on the commonality measurement results of multiple positive training example pairs. Concentration is an indicator that measures the tightness of data distribution and is usually obtained by calculating the variance, standard deviation, or coefficient of variation of the data. For the commonality measurement results, a low concentration (i.e., small variance, small standard deviation, or small coefficient of variation) means that the commonality measurement results among different pairs are relatively concentrated and the differences are small, which reflects the stability of the network in positive pattern recognition. First, the computer system collects the commonality measurement results of all positive training example pairs. These results may be obtained by calculating the cosine similarity, Pearson correlation coefficient, or other commonality measurement methods between the implicit representation sequences of the embedded mappings of each pair of training examples.

[0200] Next, the computer system calculates the variance, standard deviation, or coefficient of variation of these commonality measurement results. Finally, the computer system takes the obtained variance (or other concentration indicator) as the first analysis result. The smaller this value is, the more concentrated the commonality measurement results of the positive training example pairs are, and the more stable the network's performance in identifying positive patterns is.

[0201] Suppose there are three positive training example pairs with commonality measurement results of 0.80, 0.81, and 0.82 respectively. Calculating the average of these results gives μ = 0.81, and the variance is 0.000333. Since the variance is very small, it indicates that the commonality measurement results of these three positive training example pairs are very concentrated, and the network shows high stability in identifying these positive patterns.

[0202] Step S55b is similar to step S55a, but this step is for negative training example pairs. The task of the computer system is to perform a concentration analysis on the commonality measurement results of these pairs to evaluate the network's ability to distinguish different negative patterns. Different from positive patterns, there should be greater differences between negative patterns, so it is expected that their commonality measurement results are more dispersed. Similar to step S55a, but the data is from negative training example pairs. The purpose of the concentration analysis is to evaluate whether the network can effectively identify the differences between these negative patterns.

[0203] In step S55c, the computer system will debug the embedded mapping neural network based on the first analysis result (the concentration of the commonality measurement results of positive training example pairs) and the second analysis result (the concentration of the commonality measurement results of negative training example pairs). The purpose of debugging is to optimize these two results by adjusting the network parameters, so that the network can maintain the stability of positive pattern recognition while improving the ability to distinguish the differences of negative patterns.

[0204] To incorporate the concentration analysis results into the network debugging process, the computer system designs a loss function that can reflect these two analysis results. A possible design is to include concentration as part of the loss function, guiding the adjustment of network parameters by penalizing situations with too high concentration (for negative patterns) or too low concentration (for positive patterns). With the loss function, the computer system can apply an optimization algorithm to minimize the value of the loss function. During the optimization process, the computer updates the network parameters according to the gradient of the loss function with respect to the network parameters, thereby gradually optimizing the concentration analysis results and improving network performance. The debugging process will be repeated for multiple iterations until the value of the loss function is reduced to an acceptable level or the preset number of iterations is reached. In each iteration, the computer system recalculates the commonality metric results, concentration analysis results, and loss function value, and updates the network parameters according to the new loss function gradient.

[0205] Suppose a simplified loss function consists of two parts: one part reflects the concentration (variance) of the commonality metric results of positive patterns, and the other part reflects the concentration of the commonality metric results of negative patterns (also represented by variance, but expected to be large to show differences). To balance these two parts of the loss, different weights can be assigned to them, and the concentration values can be transformed (such as taking the reciprocal) to conform to the direction of the decreasing loss function. However, in practical applications, more complex loss functions and more refined debugging strategies may be preferred. In practical applications, the loss function can be designed and adjusted more precisely. During the debugging process, the computer optimizes the network parameters according to this loss function. As the iteration progresses, the variance of the commonality metric results of positive patterns may gradually decrease (concentration increases), while the variance of the commonality metric results of negative patterns may remain at a certain level or increase slightly (to show differences), thus making the value of the entire loss function gradually decrease.

[0206] By analyzing the concentration of the commonality metric results of positive and negative training example pairs and debugging the embedding mapping neural network accordingly, the performance of the network in identifying positive patterns and distinguishing negative patterns can be effectively improved. This method not only considers the numerical values of the commonality metric results but also pays attention to their distribution, providing a more comprehensive perspective for network optimization.

[0207] As an implementation scheme, the estimated lighting information time series set is obtained by estimating a neural network based on the time series set after debugging. The debugging process of the time series set estimation neural network includes:

[0208] Step S1: Obtain the training lighting information time series set; the training example lighting information time series set includes the lighting information of the target lighting quantity at each past moment in the first past time interval, and the lighting information in the training example lighting information time series set is arranged according to the order of their respective past moments;

[0209] Step S2: Integrate each lighting information in the training example lighting information time series set with the semantic implicit representation of the corresponding prior label respectively to obtain integrated information, and generate a training integrated information time series set;

[0210] Step S3: Perform embedding mapping on the training example integrated information sequence according to the embedded mapping neural network that has been debugged to generate a training embedded mapping implicit representation time series set;

[0211] Step S4: Load the training example embedded mapping implicit representation sequence into the time series prediction neural network to be debugged, and predict the lighting information of the target lighting quantity at each past moment in the second past time interval to generate a prediction time series set;

[0212] Step S5: Debug the time series prediction neural network based on the prediction time series set and the lighting information time series set label.

[0213] Before debugging the time series prediction neural network, the computer system first obtains a training lighting information time series set. This set contains the lighting information of the target lighting quantity at each past moment within the first past time interval, and this information is arranged in chronological order.

[0214] Suppose we want to detect the lighting situation of street lamps on a certain road in the next week (the second past time interval). First, collect the lighting quantity data of each night on this road in the past month (the first past time interval) as training data. This data may come from the real-time monitoring and recording of Internet of Things devices (such as in-vehicle cameras, GPS locators, etc.), and after processing, form an ordered lighting information time series set. For example, the training lighting information time series set may contain the following data items: {(2024 - 07 - 01, 75), (2024 - 07 - 02, 78), …, (2024 - 07 - 31, 82)}; where each tuple represents a time point (date) and the lighting quantity at that time point.

[0215] After obtaining the training sequence set of lighting information, the computer system integrates each lighting information with the corresponding prior label semantic implicit representation to generate a training integrated information sequence set. This process aims to combine the actual observed data with prior knowledge to provide richer and more comprehensive information for subsequent neural network training. For each lighting information item in the training sequence set of lighting information, the computer system looks up its corresponding prior label (such as "quantity stable", "quantity increasing", etc.) and uses a pre-trained word embedding model or a custom feature extraction method to convert these labels into semantic implicit representation vectors. Then, these vectors are concatenated or fused with the lighting quantity (which can be numerically embedded first) to generate integrated information.

[0216] Suppose the lighting quantity at a certain time point is 80 lights, and its corresponding prior label is "quantity stable", and the semantic implicit representation vector of this label is v 稳定 =[0.1, 0.0, -0.1, …]. Then, the integrated information may be [80, 0.1, 0.0, -0.1, …] (the dimension of the vector is simplified here to save space). After performing this operation on each element in the training sequence set of lighting information, the training integrated information sequence set can be generated.

[0217] After generating the training integrated information sequence set, the computer system performs an embedding mapping process on these integrated information using the already debugged embedding mapping neural network. The purpose of this step is to convert the integrated information into an implicit representation in a higher-dimensional or lower-dimensional space so that the subsequent sequence set prediction neural network can better capture the internal features and time dependencies of the data. Each integrated information item in the training integrated information sequence set is sequentially input into the embedding mapping neural network. The network calculates the embedding mapping implicit representation of each integrated information through its internal structure and parameters, and arranges these representations in chronological order to generate a training embedding mapping implicit representation sequence set.

[0218] Suppose the embedding mapping neural network is an LSTM network. For each integrated information item in the training integrated information sequence set (such as [80, v 稳定 ), the LSTM network will sequentially read these information items and generate the corresponding embedding mapping implicit representation vectors through its internal state update mechanism (the internal calculation details of LSTM are not elaborated here). Finally, all the embedding mapping implicit representation vectors are arranged in chronological order to form a training embedding mapping implicit representation sequence set.

[0219] After obtaining the training embedding mapping implicit representation sequence set, the computer system loads it into the sequence set prediction neural network to be debugged to predict the lighting information at each past moment of the target lighting quantity in the second past time interval. This process is a key step for the neural network to perform forward propagation and output detection results.

[0220] The time series set prediction neural network to be debugged may be an RNN, LSTM, GRU, or other neural network models suitable for processing time series data. The computer system will train the embedded mapping implicit representation of the time series set as input and pass it to the network for forward propagation calculation. The network generates a predicted sequence of lit lamp numbers, i.e., the predicted time series set, based on the input data and its own parameters.

[0221] Suppose the time series set prediction neural network is an LSTM network. When the training embedded mapping implicit representation of the time series set is passed to the LSTM network as input, the network processes each embedded mapping implicit representation vector in sequence and generates the corresponding detected lit lamp numbers through its internal state transfer mechanism. Finally, all the predicted lit lamp numbers are arranged in chronological order to form the predicted time series set.

[0222] Finally, the computer system calculates the value of the loss function based on the predicted time series set and the actual lit lamp information time series set label, and adjusts the network parameters through the backpropagation algorithm to minimize the value of the loss function. This process will be repeated multiple times until the stop condition is met (such as the value of the loss function is less than a certain threshold or the preset number of iterations is reached).

[0223] During the debugging process, the computer system defines a suitable loss function to evaluate the difference between the predicted time series set and the actual lit lamp information time series set. Commonly used loss functions include mean squared error (MSE), mean absolute error (MAE), etc. These loss functions can quantify the deviation between the predicted result and the actual result and provide guidance for adjusting the network parameters.

[0224] In each iteration, the computer system calculates the gradients of the network parameters based on the value of the loss function and backpropagates these gradients to each layer of the network through the backpropagation algorithm. Then, the network parameters (such as weights and biases) are updated according to the gradient information to reduce the value of the loss function. This process will be continuously repeated until the stop condition is met.

[0225] Suppose there is a large deviation between the detection result of the time series set prediction neural network and the actual result in the initial state. After calculating the value of the MSE loss function and using the backpropagation algorithm to adjust the network parameters, the accuracy of the predicted result gradually improves. As the number of iterations increases, the MSE value gradually decreases until it converges to a relatively low level. At this time, it can be considered that the network has been fully trained and can be used for actual lit lamp detection tasks.

[0226] Figure 2 A schematic diagram of the hardware entity of a computer system provided by an embodiment of this application is shown in Figure 2As shown, the hardware entities of the computer system 1000 include: a processor 1001 and a memory 1002. Among them, the memory 1002 stores a computer program that can run on the processor 1001. When the processor 1001 executes the program, it implements the steps in the method of any of the above embodiments.

[0227] As described above, only the embodiments of the present application are concerned, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application.

Claims

1. A method for intelligently identifying streetlights in vehicle-road collaboration based on the Internet of Things, characterized in that: The method comprises: Obtaining a time series set of past lighting information for a target number of lights to be on; the time series set of past lighting information includes lighting information for the target number of lights to be on at each of a plurality of past moments, with each piece of lighting information arranged in a chronological order according to its respective past moments; wherein the past lighting information is obtained using a street light status image captured by a vehicle camera, vehicle GPS positioning information, and street light GPS coordinate information; Obtaining a priori labels corresponding to each light-on information in the past light-on information time series set, where the priori labels are descriptive labels representing the evolution of the number of target lights that should be on, and the time representing the evolution in the descriptive labels is after the past time of the light-on information corresponding to the priori labels; Extracting the semantic implicit representation of each of the prior labels to obtain the semantic implicit representation of each of the prior labels; Integrating each piece of lighting information in the past lighting information time series set with the semantic implicit representation of the corresponding prior label to obtain integrated information, thereby obtaining a past integrated information time series set; Based on the past integrated information time series set, the lighting information of the target number of lights to be lit at each analysis moment in a plurality of sequentially distributed analysis moments is estimated to obtain an estimated lighting information time series set; the analysis moment is located after each of the past moments.

2. The method according to claim 1, characterized in that The step of integrating each piece of lighting information in the past lighting information time series set with the semantic implicit representation of the corresponding priori label to obtain integrated information and obtain the past integrated information time series set includes: For each light-on information in the past light-on information time series set, combining the light-on information with the semantic implicit representation of the corresponding priori label to obtain integrated information corresponding to each light-on information; According to the time sequence of the lighting information in the past lighting information time sequence set, the integrated information corresponding to each lighting information is arranged to generate a past integrated information time sequence set.

3. The method according to claim 2, characterized in that The method further comprises: Arranging the semantic implicit representations of the priori labels corresponding to the respective lighting information according to the time sequence of the lighting information in the past lighting information time sequence set to generate a semantic implicit representation time sequence set; For each piece of light-on information in the past light-on information time series set, combining the light-on information with the semantic implicit representation of the corresponding priori label to obtain integrated information corresponding to each piece of light-on information, including: The lighting information and semantic implicit representation located at the same position in the past lighting information time series set and the semantic implicit representation time series set are combined to obtain integrated information corresponding to each lighting information.

4. The method according to claim 1, wherein The step of estimating the lighting information of the target number of lights to be lit at each of a plurality of sequentially distributed analysis moments based on the past integrated information time series set to obtain the estimated lighting information time series set includes: Decomposing the past integrated information time series set according to the semantic implicit representation included in each integrated information in the past integrated information time series set to obtain a plurality of segmented time series subsets; wherein the semantic implicit representation included in the integrated information in the same segmented time series subset is the same, and the semantic implicit representation included in the integrated information in two adjacent segmented time series subsets is different; According to the multiple segmented time series subsets, the lighting information of the target number of lights to be lit at each analysis moment in a plurality of sequentially distributed analysis moments is estimated to obtain an estimated lighting information time series set.

5. The method according to claim 4, characterized in that The step of estimating lighting information of the target number of lights to be turned on at each of a plurality of sequentially distributed analysis moments based on the plurality of segmented time series subsets to obtain an estimated lighting information time series set includes: Performing embedding mapping on the plurality of segmented time series subsets respectively to obtain an embedded mapping implicit representation corresponding to each of the segmented time series subsets; Based on the embedded mapping implicit representation corresponding to each of the segmented time series subsets, generating an embedded mapping implicit representation sequence corresponding to the past lighting information time series set; According to the embedded mapping implicit representation sequence corresponding to the past lighting information time series set, the lighting information of the target number of lights to be lit at each analysis moment in multiple sequentially distributed analysis moments is estimated to obtain an estimated lighting information time series set.

6. The method according to claim 5, characterized in that The embedding-mapping implicit representation is generated based on the debugged embedding-mapping neural network, and the debugging process of the embedding-mapping neural network includes: Obtaining a positive training example binary, the positive training example binary comprising training examples corresponding to the target number of lights that should be on in a plurality of intersecting past time intervals, the training examples corresponding to the past time intervals comprising lighting information of the target number of lights that should be on at each past moment in the past time intervals, the lighting information in the training examples being arranged in a sequential order; Obtaining a negative training example binary group, wherein the negative training example binary group includes training examples corresponding to the number of target lights that should be on in a plurality of non-intersecting past time intervals; For each training example in the active training example binary group, integrating each light-up information in the training example with the semantic implicit representation of the corresponding prior label to obtain integrated information, and generating an integrated information sequence corresponding to each training example in the active training example binary group; For each training example in the negative training example binary group, integrating each light-up information in the training example with the semantic implicit representation of the corresponding prior label to obtain integrated information, and generating an integrated information sequence corresponding to each training example in the negative training example binary group; The embedding mapping neural network is debugged based on the integrated information sequence corresponding to each training example in the positive training example binary group and the integrated information sequence corresponding to each training example in the negative training example binary group.

7. The method according to claim 6, characterized in that The debugging of the embedding mapping neural network based on the integrated information sequence corresponding to each training example in the positive training example binary group and the integrated information sequence corresponding to each training example in the negative training example binary group includes: Loading the integrated information sequence corresponding to each training example in the active training example binary group into the embedding mapping neural network to embed multiple embedding mapping implicit representations, and obtaining the embedding mapping implicit representation sequence corresponding to each training example in the active training example binary group; Loading the integrated information sequence corresponding to each training example in the negative training example binary group into the embedding mapping neural network to embed multiple embedding mapping implicit representations, and obtaining the embedding mapping implicit representation sequence corresponding to each training example in the negative training example binary group; Determine the commonality measurement results between the embedding mapping implicit representation sequences corresponding to each training example in the active training example binary group, and obtain the commonality measurement results corresponding to the active training example binary group; Determining a commonality measurement result between embedding mapping implicit representation sequences corresponding to each training example in the negative training example binary group, and obtaining a commonality measurement result corresponding to the negative training example binary group; The embedding mapping neural network is debugged based on the commonality measurement results corresponding to the positive training example binary groups and the commonality measurement results corresponding to the negative training example binary groups.

8. The method according to claim 7, characterized in that The debugging of the embedding mapping neural network based on the commonality measurement results corresponding to the positive training example binary and the commonality measurement results corresponding to the negative training example binary includes: Performing a first preset calculation on the commonality measurement results corresponding to the plurality of active training example binary groups to obtain a first calculated commonality measurement result; Performing the first preset calculation on the commonality measurement results corresponding to the plurality of negative training example binary groups, respectively, to obtain a second calculated commonality measurement result; Debugging the embedding mapping neural network according to an error between the first calculated commonality metric result and the second calculated commonality metric result; Alternatively, debugging the embedding mapping neural network based on the commonality measurement results corresponding to the positive training example binary and the commonality measurement results corresponding to the negative training example binary includes: Analyzing the concentration of commonality measurement results corresponding to a plurality of positive training example binary groups to obtain a first analysis result; Analyzing the concentration of the commonality measurement results corresponding to the plurality of negative training example binary groups to obtain a second analysis result; The embedding mapping neural network is debugged according to the first analysis result and the second analysis result.

9. The method according to claim 1, characterized in that The estimated lighting information timing set is obtained based on the debugged timing set estimation neural network. The debugging process of the timing set estimation neural network includes: Obtaining a training lighting information time series set; the training lighting information time series set includes lighting information of the target number of lights to be lit at each past moment in a first past time interval, and each lighting information in the training lighting information time series set is arranged according to the order of the respective past moments; Integrate each piece of light-up information in the training light-up information time series set with the semantic implicit representation of the corresponding priori label to obtain integrated information, thereby generating a training integrated information time series set; Embedding and mapping the training integrated information time series set according to the debugged embedding mapping neural network to generate a training embedding mapping implicit representation time series set; Loading the training embedding mapping implicit representation time series set into the time series set estimation neural network to be debugged, estimating the lighting information of the target number of lights to be lit at each past moment in the second past time interval, and generating an estimated time series set; Based on the estimated timing set and the lighting information timing set mark, the timing set estimation neural network is debugged.

10. A computer system comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, wherein: When the processor executes the program, the steps of the method according to any one of claims 1 to 9 are implemented.

Images