Expressway service area heating and ventilation control method based on Internet of Things and neural network

Abstract

The invention discloses an expressway service area heating and ventilation control method based on the Internet of Things and a neural network, and the method comprises the steps: dividing an area in a building of a target service area according to functions, and deploying an Internet of Things sensor in each functional area for obtaining environment data, heating and ventilation equipment parameters and people flow distribution data; collecting access vehicle information at the access of the target service area, and collecting the number of access persons at the access of each functional area; taking the people flow distribution data, the vehicle information and the number of people entering and exiting from the current service area and the previous service area as a first data set, and inputting the first data set into a people flow prediction neural network to generate corresponding people flow prediction data; the environment data, the heating and ventilation equipment parameters, the people flow distribution data and the people flow prediction data of the current service area serve as a second data set, and a control strategy neural network is input to output a control strategy of each marshalling heating and ventilation equipment in the current service area in a future preset time period; and controlling each group of heating and ventilation equipment in a future preset time period according to the control strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of HVAC control, and in particular to a HVAC control method for a highway service area based on the Internet of Things and a neural network. Background Art

[0002] With the continuous expansion of the highway network, highway service areas have become important nodes that connect cities and villages and promote convenient transportation. The functions of service areas are not limited to providing infrastructure such as rest and catering, but also put forward higher requirements for the comfort of car owners and passengers. Especially in the case of frequent flow of people, how to maintain the stability and comfort of the temperature in the service area has become an important technical challenge.

[0003] The Chinese invention patent with publication number CN118729501B discloses an HVAC (heating, ventilation and air conditioning) optimization control method based on scene recognition and demand prediction. This method collects multi-source data in real time, uses the LSTM (long short-term memory network) model for scene recognition and demand prediction, and then generates an optimal control strategy to dynamically adjust the operating parameters of the HVAC system to improve system efficiency and reduce energy consumption, thereby achieving the goal of energy conservation and emission reduction. Although this solution performs well in the field of traditional building HVAC control, there are still several aspects that can be improved when it is applied to the HVAC control of highway service area buildings, including the following points:

[0004] 1. Limitations of data collection: Existing solutions only rely on the data of the target building at the current moment to make decisions. However, for special environments such as highway service areas, the traffic and passenger flows have a strong temporal correlation in space. For example, the probability of vehicles and personnel entering the previous service area entering the current service area is usually low. Even if they enter the current service area, and their stay time in the previous service area is longer, their stay time in the current service area is usually shorter. Therefore, based on the data comparison of the previous and next service areas, some irrelevant or noisy data can be effectively eliminated. In addition, the real-time data and predicted data of the previous service area can not only serve as a reference for the current service area, but also improve the accuracy of prediction and the effectiveness of strategy adjustment, thereby improving the response efficiency of the entire HVAC control system.

[0005] 2. Ignoring regional differences: The HVAC equipment adjustment methods in existing solutions are too uniform and lack personalized attention to different functional areas. However, highway service areas usually have multiple functional areas (such as toilet areas, dining areas, rest areas, etc.), and the characteristics and needs of people staying in these areas vary greatly. If a single prediction and strategy adjustment method is adopted, the equipment adjustment may not be sufficiently adaptable and cannot be optimized according to the actual needs of each area. Therefore, it is particularly important to formulate personalized control strategies for the flow of people and temperature control needs in different functional areas to ensure the accuracy and efficiency of the system.

[0006] 3. Singleness of data types: Existing prediction models mainly rely on time series feature data for prediction, so the conventional LSTM network is sufficient to meet the needs of most scenarios. However, the complexity of highway service areas exceeds the scope of simple time series data analysis, and the LSTM model that relies solely on time series data cannot fully mine and utilize diversified data features (such as personnel behavior patterns, external environmental changes, equipment status, etc.). In order to improve the accuracy and reliability of predictions, it is recommended to add more types of data input, such as environmental monitoring data, historical operating status, user behavior patterns, etc., and combine multimodal learning methods to build a more complex neural network model.

[0007] 4. Disconnection between real-time prediction and HVAC system response delay: Most of the prediction and adjustment strategies in existing solutions rely on real-time conditions, that is, control is performed based on the current time node. However, the response of the HVAC system has a certain delay, especially in temperature regulation, and the adjustment of the equipment takes a certain amount of time to take effect. This means that if control is based only on real-time data, there may be a disconnect between the current adjustment and the expected state in the future, which in turn affects the stability and comfort of the system. Therefore, when adjusting the strategy, it is necessary to fully consider the response delay of the HVAC equipment and adopt a control strategy based on future predictions, rather than just limited to the data at the current moment. In this way, the system can be optimized and adjusted in advance to avoid instability problems caused by excessive or insufficient real-time adjustments. Summary of the invention

[0008] In order to solve the above technical problems, the present invention aims to optimize the operation of HVAC equipment through intelligent control strategies, and solve the problems of slow response, low energy efficiency, unstable comfort and so on of HVAC control system in highway service areas in the prior art. By integrating multi-source data, including environmental parameters of each area in the service area, personnel flow, vehicle information and data of the previous service area, combined with a neural network model, the operating parameters of the HVAC equipment are dynamically predicted and optimized to achieve efficient, accurate and energy-saving HVAC control.

[0009] In order to achieve the above-mentioned invention object, the technical solution provided by the present invention includes:

[0010] The method for controlling HVAC in a highway service area based on the Internet of Things and a neural network is characterized by comprising the following steps:

[0011] S1. Divide the building areas of the target service area by function, deploy IoT sensors in each functional area to obtain environmental data, HVAC equipment parameters and crowd distribution data, and group the HVAC equipment in the same building according to the functional area;

[0012] S2. An image acquisition device is respectively set at the entrance and exit of the target service area to collect information about vehicles entering and leaving the target service area; the vehicle information includes license plates and stay time; an infrared passenger flow counter is respectively set at the entrance and exit of each functional area of ​​the target service area to collect the number of people entering and leaving the target service area;

[0013] S3, taking the passenger flow distribution data, vehicle information and the number of people entering and leaving the current service area and the previous service area in the direction of the highway traffic flow as the first data set, inputting the pre-trained passenger flow prediction neural network, predicting the passenger flow of each functional area of ​​the target service area in the future preset time period, and generating corresponding passenger flow prediction data;

[0014] S4, taking the environmental data, HVAC equipment parameters, crowd distribution data and crowd prediction data of the current service area as the second data set, inputting the pre-trained control strategy neural network, and outputting the control strategy of each group HVAC equipment in the current service area in the future preset time period;

[0015] A5. Control the HVAC equipment of each group within the preset time period in the future according to the control strategy.

[0016] Preferably, before inputting the first data set into the pre-trained people flow prediction neural network in step S3, the method further includes the following steps:

[0017] Eliminate the vehicle information of vehicles that continuously enter the current service area and the previous service area in the direction of highway traffic and stay in the previous service area for more than a first preset threshold, and delete the pedestrian flow data of no more than 5 people in the pedestrian flow distribution data of any functional area of ​​the current service area and the number of people entering and leaving the current service area.

[0018] Preferably, the crowd flow prediction neural network includes:

[0019] A multiple input layer, comprising a plurality of separate input layers respectively set according to each category of input data of the first data set, outputting a feature representation of the first data set;

[0020] A temporal convolutional network, for extracting short-term dependency features in the feature representation through convolution expansion;

[0021] A long short-term memory network, used for extracting long-term dependency features from the short-term dependency features through a gating mechanism;

[0022] The output layer is used to merge the output of the long short-term memory network into a human flow prediction result.

[0023] Preferably, the control strategy neural network includes:

[0024] The time series input layer extracts the time series features of the crowd flow distribution data and the crowd flow prediction data in the second data set based on the long-short time neural network, and outputs a multi-dimensional time series feature vector;

[0025] The static input layer normalizes the environmental data and HVAC equipment parameters in the second data and outputs a one-dimensional numerical feature vector;

[0026] The feature fusion layers connected to the temporal input layer and the static input layer respectively are used to concatenate the temporal feature vector and the numerical feature vector into an input feature vector.

[0027] Preferably, the control strategy neural network also includes:

[0028] A reinforcement learning module connected to the feature fusion layer, wherein the reward function R of the reinforcement learning module t for:

[0029] R t =α1·R c +α2·R e +α3·R s +α4·R t ; Among them, α1, α2, α3 and α4 are preset weight parameters, R c is the comfort reward parameter, R e is the energy efficiency reward parameter, R s For the temperature control stability bonus, R t Parameters for traffic reward.

[0030] Preferably, the pre-training method comprises: using the length of the preset time period as an extraction window for feature data and / or input feature vectors.

[0031] Beneficial Effects

[0032] 1. Intelligent control strategy: By adopting a neural network model to predict passenger flow and optimize the HVAC equipment control strategy, the present invention can automatically adjust the operating status of the HVAC equipment according to the real-time data and predicted data of the service area, ensuring that the HVAC system can be accurately adjusted according to demand and avoiding energy waste.

[0033] 2. Reduce data noise and improve prediction accuracy: The present invention can effectively eliminate irrelevant noise data by combining the traffic and personnel data of the previous service area, thereby improving the accuracy of the prediction model. This method can further reduce the errors that may be caused by relying only on current data and increase the reliability of prediction and control strategies.

[0034] 3. Regionalized precise control: In view of the differences in the flow of people in different functional areas (such as toilet areas, rest areas, dining areas, etc.) in highway service areas, the present invention proposes a regionalized HVAC control strategy. Each functional area is intelligently adjusted according to its specific needs and personnel flow characteristics, achieving refined and efficient control of HVAC equipment.

[0035] 4. Effective control of response delay: Considering the time delay of the response of HVAC equipment, the present invention adopts a control strategy based on future predictions instead of relying solely on real-time data. This design can effectively avoid temperature control errors caused by the lag in real-time control strategy adjustment and improve the stability and comfort of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A schematic flow chart of a method for controlling HVAC in a highway service area based on the Internet of Things and a neural network provided in a preferred embodiment of the present invention;

[0037] Figure 2 A schematic diagram of a neural network structure for predicting human flow provided in another preferred embodiment of the present invention;

[0038] Figure 3 This is a schematic diagram of the neural network structure of the control strategy provided in another preferred embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention is further described below in conjunction with the accompanying drawings. In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside", "outside" and the like indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings, which are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention.

[0040] Embodiment 1

[0041] like Figure 1 As shown, this embodiment discloses a highway service area HVAC control method based on the Internet of Things and a neural network, comprising the following steps:

[0042] S1. Divide the building area of ​​the target service area by function, deploy IoT sensors in each functional area to obtain environmental data, HVAC equipment parameters and crowd distribution data, and group the HVAC equipment of the same building according to the functional area. The target service area refers to all service areas to be predicted in the direction of a specific lane of a certain section of highway. The current service area is one of the target service areas.

[0043] Those skilled in the art should know that for buildings in highway service areas, the internal space can be divided into several relatively independent areas according to different functions, for example:

[0044] Bathroom area: mainly includes men's toilets, women's toilets, washrooms, etc. The characteristics of its HVAC demand are that people usually stay for a short time, but due to frequent entry and exit, the air quality and temperature changes are more sensitive.

[0045] Entrance and exit area: mainly for personnel entrance and exit. Its HVAC demand is characterized by dense flow of people and the external climate has a great impact on the indoor environment, and the temperature control demand is relatively complex.

[0046] Dining area: The characteristics of its HVAC demand are that people stay longer and require better air quality and comfortable temperature, so the demand for HVAC equipment is greater.

[0047] Rest area: Its HVAC demand characteristics are similar to those of the dining area, but the flow of people may be relatively small, and the temperature control demand is relatively stable.

[0048] Aisle area: The HVAC demand characteristics are that people usually pass through frequently, and the temperature control demand may be lower, but it also needs to maintain a certain degree of comfort.

[0049] IoT sensors are deployed for different functional areas to collect real-time environmental data, HVAC equipment parameters, and crowd distribution data. The types and functions of these sensors include but are not limited to:

[0050] 1. Environmental data sensors: including temperature sensors, humidity sensors, CO2 sensors, air quality sensors, etc., which can monitor the indoor environmental parameters of each area in real time. These data are crucial for intelligent HVAC control and help the system understand the environmental changes in each area in real time.

[0051] 2. HVAC equipment parameter sensors: These sensors are used to collect real-time operating status data of HVAC equipment, including the working status, operating power, wind speed, flow rate and other parameters of fans, air conditioners, heaters and other equipment. These data are used for subsequent control strategy optimization.

[0052] 3. Crowd distribution sensor: By deploying infrared crowd counters, video monitoring and analysis equipment, etc., real-time data such as the flow of people and the length of stay of people in each functional area can be obtained. Changes in crowd flow directly affect the HVAC demand in the area, so crowd distribution data is a key factor in determining the control strategy.

[0053] In each functional area, HVAC equipment is grouped according to the functional requirements and traffic characteristics of the area. Different areas require different types of HVAC equipment support, so when controlling, it is necessary to dispatch related equipment according to specific functional areas, and group HVAC equipment in the same functional area according to the area to which it belongs. For example, the HVAC equipment in the bathroom area may focus on air exchange and temperature regulation, while the dining area requires a more efficient air conditioning system to maintain appropriate temperature and humidity. According to the needs of the area, the system can adjust the operating status of each device in real time.

[0054] S2. Image acquisition devices are respectively set at the entrances and exits of the target service area to collect information about vehicles entering and leaving the area; the vehicle information includes license plates and stay time; infrared people flow counters are respectively set at the entrances and exits of each functional area of ​​the target service area to collect the number of people entering and leaving the area.

[0055] By installing image acquisition devices (such as high-definition cameras and license plate recognition systems) at the entrances and exits of the target service area, vehicle information entering and leaving the service area is collected, including license plate numbers, entry time, stay time, etc. The vehicle information entering the service area will be compared and correlated with the data of the previous service area to analyze the vehicle flow patterns and stay patterns. Under normal circumstances, the probability of vehicles and personnel entering the previous service area entering the current service area is low. Even if they enter the current service area, when their stay time in the previous service area is longer, their stay time in the current service area is usually shorter. This rule shows that vehicles or personnel who have stayed in the previous service area for a long time are likely to choose to leave the current service area quickly, so their impact on the current service area is small and they are "low-correlation" data. Therefore, in some preferred embodiments, preprocessing is performed in the following ways:

[0056] Eliminate the vehicle information of the vehicles that continuously enter the current service area and the previous service area in the direction of the highway traffic flow, and the vehicle information of the vehicle that stays in the previous service area for more than the first preset threshold, and delete the pedestrian flow data of no more than 5 people in the pedestrian flow distribution data of any functional area of ​​the current service area and the number of people entering and leaving the current service area. This operation can effectively eliminate those vehicles that have little impact on the flow of people in the current service area, and then eliminate the people accompanying the vehicle, reducing the interference of irrelevant data on subsequent prediction and control strategies. It should be noted that since most of the vehicles in the highway service area are ordinary vehicles, and usually private cars or small passenger cars, the number of people generally does not exceed 5. The actual number of people carried may be lower than the number of people carried, but it is usually not more than 5. Since it is difficult to monitor which functional area of ​​the service area the accompanying personnel go to for activities (it is impossible to accurately determine whether these people have entered the dining area, the rest area, or just stayed in the aisle), therefore, this embodiment performs data processing by random deletion. On the one hand, it avoids the deviation that may be caused by the fixed-pattern deletion method and ensures the naturalness and rationality of the deletion process. On the other hand, it retains other important data and reduces the interference of noise on data modeling. While ensuring that it does not affect the actual passenger flow prediction in the service area, it avoids data redundancy caused by a small number of short-staying people.

[0057] S3. The passenger flow distribution data, vehicle information and number of people entering and leaving the current service area and the previous service area in the direction of highway traffic flow are used as the first data set, and input into the pre-trained passenger flow prediction neural network to predict the passenger flow of each functional area of ​​the target service area within a preset time period in the future, and generate corresponding passenger flow prediction data.

[0058] It should be understood that since the vehicles and personnel entering the target service area usually have a certain regularity in time, the behavior of the personnel staying in the previous service area has a strong reference for the prediction of the behavior of the personnel staying in the current service area (that is, the activity patterns of personnel in consecutive service areas are roughly the same or similar). Combining the data from the previous and subsequent service areas can help reduce data noise and improve the accuracy of passenger flow prediction.

[0059] For the crowd flow prediction neural network, the input data has many dimensions and contains long-term time series features and short-term time series features, such as Figure 2 As shown, in some preferred embodiments, the following people flow prediction neural network is considered to be constructed, specifically including:

[0060] The multi-input layer includes a plurality of separate input layers respectively set according to each category of input data of the first data set, and outputs the feature representation of the first data set. Considering the diversity of data sources (environmental data, vehicle information, personnel flow data), a multi-input neural network structure is constructed. Each input can be processed through different channels to capture the characteristics of different types of data. For example: vehicle information input layer: can extract the characteristics of vehicle flow patterns; environmental data input layer: processes environmental factors such as temperature and humidity that affect indoor human flow; human flow data input layer: specifically processes the mobility and density information of personnel. The multi-input layer is used to receive multiple different types of input data (such as environmental data, vehicle information, human flow distribution data, etc.), and pass each input data to the corresponding sub-network for processing. Each input layer will independently process its own data category, avoid interference between different types of data, and can efficiently extract features.

[0061] The temporal convolutional network is used to extract short-term dependency features in the feature representation through convolution expansion. The receptive field is expanded through the convolution expansion operation, so that it can capture data features with a long time span at a lower computational cost. Compared with the traditional steps, it can better handle long-term dependencies in time series data and is particularly suitable for processing complex time series data. Through this step, the network can identify rapid changes in the short term, such as fluctuations in personnel flow in a functional area in the short term.

[0062] Long short-term memory network is used to extract long-term dependency features from the short-term dependency features through a gating mechanism. It is used to process long-term dependent information in time series, such as the impact of long-term patterns such as seasonal changes and holidays on crowd flow prediction. The gating mechanism can maintain long-term memory by dynamically controlling the "forgetting" and "memory" of information flow, and is particularly suitable for processing long-term personnel flow patterns. For example, it can process periodic change trends in special periods such as holidays or peak periods that span several hours or days.

[0063] The output layer is used to merge the output of the long short-term memory network into the crowd flow prediction results. This layer is usually a fully connected layer, and its output is a probability value (classification task). These prediction results will be passed as input to the downstream control strategy neural network to adjust the control strategy of the HVAC equipment in the service area.

[0064] Obviously, the crowd flow prediction neural network can integrate different types of data sources and effectively handle local and long-term dependencies in time series data.

[0065] S4. The environmental data, HVAC equipment parameters, crowd distribution data and crowd prediction data of the current service area are used as the second data set, input into the pre-trained control strategy neural network, and the control strategy of each group of HVAC equipment in the current service area within the future preset time period is output.

[0066] Environmental data include temperature, humidity, air quality, wind speed and other data of each functional area in the current service area. These data help to assess the current comfort level and provide environmental background information for control strategies.

[0067] HVAC equipment parameters include the operating status, power, temperature setting value, etc. of each area HVAC equipment, which reflect the current status and equipment capacity of the HVAC system. This data helps determine whether each device needs to adjust its working status to achieve optimal energy efficiency and comfort.

[0068] The crowd distribution data reflects the crowd distribution in each functional area. The distribution of people in different areas directly affects the temperature control requirements of the area, so it is necessary to use it as input information to help the model make accurate control decisions.

[0069] The crowd flow prediction data is based on the prediction results of the previous step, showing the flow of people in each functional area in the future preset time period. This data enables the control strategy to foresee future changes in crowd flow and adjust the working mode of HVAC equipment in advance.

[0070] The control strategy neural network combines all the above data to achieve intelligent control of HVAC equipment. Specifically, after receiving the information in the second data set and calculating it through several layers of neural networks, a control strategy is generated according to the preset goals (such as comfort, energy efficiency, energy saving, etc.). The specific control strategy includes but is not limited to: the operation mode of HVAC equipment in each functional area (such as switch status, power regulation, etc.); the specific adjustment value of the equipment operation time and power output.

[0071] For the second data set, there are time series data with significant time dependence (such as crowd distribution data, temperature and humidity changes, etc.), and there are also static data without obvious time dependence (such as equipment status, environmental data, etc.). If a single input layer is used for fusion and input, although the structure can be simplified and the uniformity of the data can be increased, it is easy to cause data loss, resulting in poor feature extraction effect, and it is difficult to distinguish data types and it is not easy to optimize for different types of data. Therefore, in some preferred embodiments, such as Figure 3 As shown, a method of partially using multiple input layers and partially directly fusion is considered to construct the input layer of the control strategy neural network, including:

[0072] The time series input layer extracts the time series features of the crowd distribution data and crowd prediction data in the second data set based on the long-short time neural network, and outputs a multi-dimensional time series feature vector; the main function of this layer is to process and extract the time series features in the input data. These time series data reflect the dynamic changes in the flow of people and are crucial to the control strategy of HVAC equipment, because the change in the number of people is closely related to the regional temperature control requirements.

[0073] The static input layer normalizes the environmental data and HVAC equipment parameters in the second data and outputs a one-dimensional numerical feature vector; these data are usually static, that is, there is no time series change, but they are crucial to the decision-making of the control strategy.

[0074] The feature fusion layer connected to the time series input layer and the static input layer respectively is used to splice the time series feature vector and the numerical feature vector into an input feature vector. The splicing can use a simple concatenation operation, or it can use weighted summation, average pooling and other methods for splicing. The specific method can be optimized by those skilled in the art in the art in the art, and the present invention does not make further limitations. The data of the second data set processed by this method, on the one hand, reduces the interference between the data by separately processing the time series data and the static data, so that each data type can use the most suitable processing method, thereby improving the performance of the model. On the other hand, the processing method and hyperparameters of each type of data can be adjusted separately, making the model more adjustable and interpretable.

[0075] For control strategy tasks, since the control tasks involve real-time optimization and adjustment of strategies, in some preferred embodiments, the use of reinforcement learning (RL) methods is considered to further enhance the learning ability of the network. Design a reinforcement learning module to control the HVAC equipment in each area so that it can make the best decision under changing environmental and human flow conditions. The goal of the reinforcement learning module is to dynamically adjust the control strategy based on the current environmental data and status to improve the overall efficiency and comfort of the system. Those skilled in the art will know that the reward function is the most important component in reinforcement learning, which determines the pros and cons of an action taken by an agent in a certain state. Reasonable reward design can guide the reinforcement learning agent to choose the correct control strategy. In other preferred embodiments, the reward function R of the reinforcement learning module t for:

[0076] R t =α1·R c +α2·R e +α3·R s +α4·R t ; Among them, α1, α2, α3 and α4 are preset weight parameters, and the importance of each part can be adjusted by technicians in this field according to actual needs;

[0077] R c It is a comfort reward parameter that measures whether the temperature in each functional area meets the comfort requirements in a certain period of time. In the future period of time, the system needs to keep the temperature within a predetermined comfort range. If the temperature fluctuates greatly, the comfort score will be low.

[0078] Re is the energy efficiency reward parameter, which evaluates the energy efficiency performance of the current control strategy. The energy efficiency reward is inversely proportional to the energy consumption of the HVAC equipment, that is, the higher the energy efficiency (the lower the energy consumption), the higher the reward. By analyzing the relationship between the power consumption of the HVAC equipment and the temperature control requirements in the future time period, the reward system should encourage the reduction of unnecessary energy consumption.

[0079] R s The temperature control stability reward measures the stability of the temperature control system, that is, whether the temperature remains within a stable range in different time periods and regions. To avoid frequent temperature fluctuations, the system should encourage a stable temperature control strategy. Reinforcement learning should try to avoid over-adjustment of the system due to short-term temperature adjustments.

[0080] R t The reward parameter for the flow of people. Since the prediction of flow of people will directly affect the control strategy of the HVAC system, the reward function should take into account the impact of flow of people on temperature control. Areas with high flow of people require more energy support, while areas with low flow of people should reduce energy consumption.

[0081] The reinforcement learning module is directly connected to the aforementioned feature fusion layer. The multi-dimensional input feature vector provided by the feature fusion layer, including time series data (such as crowd distribution data and prediction data) and static data (such as environmental data and equipment parameters), is the basis for the reinforcement learning module to make decisions. On this basis, the reinforcement learning module will dynamically adjust the control strategy according to the reward function feedback received.

[0082] S5. Control the HVAC equipment of each marshaling group within the preset time period in the future according to the control strategy. Control of the HVAC equipment of each marshaling group includes but is not limited to the following aspects:

[0083] 1. Real-time adjustment of equipment parameters: According to the control strategy output, the working status of the HVAC equipment in each group is adjusted in real time. For example, if a functional area is expected to have a large influx of people in the next 30 minutes, the control strategy may require the air conditioning equipment in the area to increase the operating power to adapt to the temperature requirements in advance.

[0084] 2. Adjust equipment operation mode: For example, when the flow of people is small and the temperature is relatively stable, the HVAC equipment in some areas may be adjusted to energy-saving mode (such as low-power operation mode or temporarily shut down).

[0085] 3. Coordinated equipment operation: The control strategy may also require HVAC equipment in different areas to work in coordination to avoid wasting resources due to excessive operation of a single device. For example, if the equipment in one area is overloaded, the system can adjust the equipment in nearby areas to share the load.

[0086] The core of this step is to convert the control strategy into specific equipment operation instructions, and through the commonly used adjustment methods in the existing technology (automatic or manual), the HVAC equipment in each functional area can operate according to the optimal strategy in the future preset time period. Through this precise control, the system can not only improve comfort and stability, but also effectively reduce energy consumption and achieve the purpose of energy conservation and emission reduction.

[0087] Embodiment 2

[0088] This embodiment is developed on the basis of the above-mentioned embodiment 1, and aims to illustrate the pre-training process of the crowd prediction neural network and the control strategy neural network to avoid the disconnection problem between the real-time prediction and the response delay of the HVAC system.

[0089] Those skilled in the art can know that in time series data, the input feature vector is usually generated by historical data within a period of time, specifically, it is usually generated by historical data within a period of time, and a time window of a specific length is selected from the entire data sequence as input data to capture the regularity and change trend within a specific time period. In this embodiment, the preset time period, that is, the length of the future preset time period for the flow prediction of each functional area of ​​the target service area and the control strategy of each group of HVAC equipment in the current service area in the previous text, is considered as the extraction window of feature data and / or input feature vector, which can avoid the adjustment disconnection problem caused by relying solely on real-time data. Specifically, the preset time period (such as 30 minutes, 1 hour, etc.) is the basic time unit of the control strategy neural network, which not only covers the instant feedback of the equipment, but also can span the response delay of the HVAC system. By using this time period as a window for feature extraction, the model can take into account the delay of equipment adjustment when performing data analysis to ensure that the adjustment result and system response can be synchronized. For example, if the current indoor temperature is 24°C, and the temperature demand of a functional area is expected to increase in the next 30 minutes based on the flow of people forecast and environmental data analysis, the control strategy neural network will adjust the equipment operation parameters (such as starting the air conditioner, adjusting the temperature, etc.) in advance according to this forecast. In this way, even if there is a delay in the equipment adjustment, the final temperature adjustment effect can still match the predicted demand, thereby improving the stability and comfort of the system.

[0090] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A highway service area HVAC control method based on the Internet of Things and neural networks, characterized in that: The following steps are involved: S1. Divide the building areas of the target service area by function, deploy IoT sensors in each functional area to obtain environmental data, HVAC equipment parameters and crowd distribution data, and group the HVAC equipment in the same building according to the functional area; S2. An image acquisition device is respectively provided at the entrance and exit of the target service area to collect information of vehicles entering and leaving the target service area; the vehicle information includes the license plate and the length of stay; Infrared people flow counters are set up at the entrances and exits of each functional area of ​​the target service area to collect the number of people entering and leaving; S3, taking the passenger flow distribution data, vehicle information and the number of people entering and leaving the current service area and the previous service area in the direction of the highway traffic flow as the first data set, inputting the pre-trained passenger flow prediction neural network, predicting the passenger flow of each functional area of ​​the target service area in the future preset time period, and generating corresponding passenger flow prediction data; S4, taking the environmental data, HVAC equipment parameters, crowd distribution data and crowd prediction data of the current service area as the second data set, inputting the pre-trained control strategy neural network, and outputting the control strategy of each group HVAC equipment in the current service area in the future preset time period; S5. Control the HVAC equipment of each group within a preset time period in the future according to the control strategy.

2. The method for controlling HVAC in a highway service area based on the Internet of Things and a neural network as claimed in claim 1, characterized in that: Before inputting the first data set into the pre-trained people flow prediction neural network in step S3, the following steps are also included: Eliminate the vehicle information of vehicles that continuously enter the current service area and the previous service area in the direction of highway traffic and stay in the previous service area for more than a first preset threshold, and delete the pedestrian flow data of no more than 5 people in the pedestrian flow distribution data of any functional area of ​​the current service area and the number of people entering and leaving the current service area.

3. The method for controlling HVAC in a highway service area based on the Internet of Things and a neural network as claimed in claim 1, characterized in that: The crowd flow prediction neural network includes the following connected in sequence: A multiple input layer, comprising a plurality of separate input layers respectively set according to each category of input data of the first data set, outputting a feature representation of the first data set; A temporal convolutional network, for extracting short-term dependency features in the feature representation through convolution expansion; A long short-term memory network, used for extracting long-term dependency features from the short-term dependency features through a gating mechanism; The output layer is used to merge the output of the long short-term memory network into a human flow prediction result.

4. The method for controlling HVAC in a highway service area based on the Internet of Things and a neural network as claimed in claim 1, characterized in that: The control strategy neural network includes: The time series input layer extracts the time series features of the crowd flow distribution data and the crowd flow prediction data in the second data set based on the long-short time neural network, and outputs a multi-dimensional time series feature vector; The static input layer normalizes the environmental data and HVAC equipment parameters in the second data and outputs a one-dimensional numerical feature vector; The feature fusion layers connected to the temporal input layer and the static input layer respectively are used to concatenate the temporal feature vector and the numerical feature vector into an input feature vector.

5. The method for controlling HVAC in a highway service area based on the Internet of Things and a neural network as claimed in claim 4, characterized in that: The control strategy neural network also includes: A reinforcement learning module connected to the feature fusion layer, wherein the reward function R of the reinforcement learning module t for: R t =α1·R c +α2·R e +α3·R s +α4·R t ; Among them, α1, α2, α3 and α4 are preset weight parameters, R c is the comfort reward parameter, R e is the energy efficiency reward parameter, R s For temperature control stability bonus, R t Parameters for traffic reward.

6. The method for controlling HVAC in a highway service area based on the Internet of Things and a neural network as claimed in claim 3 or 4, characterized in that: The pre-training method includes: using the length of the preset time period as an extraction window for feature data and / or input feature vectors.

Images