Intelligent temperature control energy-saving system and method based on personnel position detection and prediction

Abstract

The invention discloses an intelligent temperature control energy-saving system and method based on personnel position detection and prediction. The system monitors indoor personnel distribution and temperature and humidity in real time through the human body and environment sensing module; the behavior prediction module utilizes an LSTM deep learning model to analyze the historical moving track of the user and predict the next target room and the arrival time of the user in combination with the environment context; the environment regulation and control module starts equipment such as an air conditioner in the target room in advance to adjust to a comfortable state; and the energy efficiency optimization module is used for implementing energy-saving control on the unmanned room. According to the method, the crossing from passive reaction to active prediction is realized, and the energy utilization efficiency of the home environment is remarkably improved on the premise of ensuring the comfort.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart home, and particularly relates to an intelligent temperature control energy-saving system based on personnel position detection and prediction. BACKGROUND

[0002] Most existing intelligent temperature control systems rely on environmental sensors to collect temperature, humidity and other parameters for simple control, lacking accurate perception of real-time personnel location and prediction of future personnel movement trends.

[0003] In actual application, the traditional temperature control system has the following defects: It cannot dynamically adjust temperature and humidity according to the current or upcoming location of the user, resulting in continuous energy consumption in unoccupied rooms; It lacks learning and prediction of user habits, and cannot achieve active pre-control; Temperature control response is lagging, affecting user comfort experience and causing energy waste.

[0004] Therefore, there is an urgent need for a new temperature control system that can detect personnel location in real time and perform intelligent control based on behavior prediction. SUMMARY

[0005] The present application is proposed to solve the problems in the prior art, and provides an intelligent temperature control energy-saving system based on personnel position detection and prediction.

[0006] To achieve the above-mentioned purpose, the present application adopts the following technical solutions: An intelligent temperature control energy-saving system based on personnel position detection and prediction, comprising: a human and environment perception unit for collecting personnel presence status and environmental parameters in multiple indoor partitions; A behavior prediction unit in communication connection with the perception unit, for predicting target partitions and estimated arrival time based on historical time series data, the historical time series data including personnel movement trajectory and environmental context information; An environmental control unit in communication connection with the behavior prediction unit, for starting and setting parameters of environmental adjustment devices in the target partitions before the estimated arrival time according to the prediction results; An energy efficiency optimization unit for executing energy-saving control strategies on devices in unoccupied partitions according to the personnel presence status collected by the perception unit.

[0007] Further, the behavior prediction unit comprises: A data management module for storing and processing personnel position data and environmental data uploaded by the perception unit in time stamp sequence; a deep learning prediction module, which is internally built with a trained recurrent neural network model, is configured to receive real-time sequence data provided by the data management module.

[0008] Further, the recurrent neural network model is a long short-term memory (LSTM) model.

[0009] Further, the input feature vector of the long short-term memory model comprises: a one-hot encoded position partition identifier; a periodically encoded timestamp feature; a normalized environmental context feature.

[0010] Further, the behavior prediction unit further comprises: a confidence judgment module, which is configured to compare the highest probability value output by the deep learning prediction module with a preset threshold, and only trigger the environmental control unit when the highest probability value is greater than the preset threshold.

[0011] Further, the deep learning prediction module is trained by minimizing a loss function, wherein the loss function L is a weighted sum of a position prediction loss and a time prediction loss, and is expressed as: L = L_cross-entropy + λ * L_MSE, wherein L_cross-entropy is a cross-entropy loss function for optimizing position classification accuracy; L_MSE is a mean square error loss function for optimizing time regression accuracy; and λ is a weight coefficient for balancing the two types of loss functions.

[0012] Further, the system supports a multi-user scenario, and different behavior prediction models are established and maintained for different users, or a single model supporting multi-task learning is used to distinguish the behavior patterns of different users.

[0013] An intelligent temperature control and energy saving method based on personnel position detection and prediction, the method comprising: S1: collecting personnel position information and environmental information of each partition in the room in real time; S2: determining whether the personnel position has changed; S3: if the personnel position has changed, calculating the probability distribution and the predicted arrival time of the next target partition based on the historical moving trajectory sequence and the trained deep learning model; S4: if the highest probability value calculated exceeds the confidence threshold, performing environmental preprocessing on the target partition with the highest probability before the predicted arrival time; S5: continuously monitoring the occupancy state of each partition and performing energy saving control on the unoccupied partition.

[0014] Further, after the step S3 and before the step S4, the method further comprises: comparing the highest probability value output by the deep learning model with a preset threshold value; if the highest probability value is lower than the preset threshold value, a default regulation strategy is enabled, the default regulation strategy comprising no early regulation or only regulating the most recently active zone.

[0015] Further, the method further comprises: continuously collecting new moving trajectory data and actual results as new training samples, and performing online incremental learning on the deep learning model to adapt to changes in user behavior habits.

[0016] Compared with the prior art, the beneficial effects of the present application are: through the LSTM model, the moving intention of the personnel is accurately predicted, the "reactive control" is changed into "predictive control", the environmental regulation is completed before the personnel arrives, the waiting period is completely eliminated, and the comfort experience is significantly improved.

[0017] The method proposed by the present application realizes "on-demand regulation" by combining behavior prediction and occupancy monitoring, accurately provides services for the space to be used, and timely closes the services of the unoccupied space, avoids energy waste, and realizes global energy efficiency optimization.

[0018] The method proposed by the present application, the system can continuously learn the historical behavior patterns of the user, and fuse environmental context information (weather, season), so that the prediction model can adapt to changes in user habits and different seasonal demands, and the system becomes more intelligent with use.

[0019] The decision mechanism based on the confidence threshold is introduced, which effectively avoids the false triggering caused by low confidence prediction, and ensures the stability of the system operation. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 is a module composition diagram of the intelligent temperature control energy-saving system based on personnel position detection and prediction of the present application; Figure 2 is a control flow chart of the intelligent temperature control energy-saving method based on personnel position detection and prediction of the present application. DETAILED DESCRIPTION

[0021] In order to further understand the purpose, structure, features and functions of the present application, the following detailed description is provided with examples. EMBODIMENT

[0022] As Figure 1As shown, the present application provides a specific embodiment of an intelligent temperature control energy-saving system based on personnel position detection and prediction. The system mainly includes a human and environment perception unit, a behavior prediction unit, an environment regulation unit, and an energy efficiency optimization unit. Each unit is connected through wired or wireless communication protocols (such as Wi-Fi, ZigBee, or LoRa) to form a closed-loop control system that works cooperatively.

[0023] The human and environment perception unit is deployed in each subzone (such as a room or office area) in the indoor space. It usually includes personnel detection devices such as infrared sensors, cameras, or millimeter wave radars, as well as environmental sensors such as temperature and humidity, and illumination, for real-time collection of personnel presence status and environmental parameters in each subzone. The behavior prediction unit is communicatively connected with the perception unit and receives the time-series data uploaded by the perception unit. The built-in data management module stores and serializes the data, and the deep learning prediction module predicts the next target subzone and estimated arrival time of the personnel based on historical data. The environment regulation unit is connected with the behavior prediction unit and controls the environmental regulation devices such as air conditioners, heaters, and fresh air systems in the target subzone according to the prediction results, starting them in advance and setting appropriate parameters before the personnel arrives. The energy efficiency optimization unit executes the strategy of closing or adjusting to energy-saving mode for the devices in the unoccupied subzone based on the personnel presence status data from the perception unit.

[0024] Further, the behavior prediction unit includes: a data management module for storing and processing personnel position data and environmental data uploaded by the perception unit in time-stamped sequence; a deep learning prediction module with a trained recurrent neural network model for receiving real-time sequence data provided by the data management module.

[0025] Further, the recurrent neural network model is a long short-term memory (LSTM) model. It is specifically designed for processing time-series data, and its gating mechanism can capture long-term behavior patterns such as "from the living room to the bedroom" and can remember and combine historical context for prediction, rather than simply guessing. This model can accurately learn and predict the moving intention of personnel.

[0026] Further, the input feature vector of the long short-term memory network model includes: a position subzone identifier processed by one-hot encoding; a time stamp feature processed by periodic encoding, such as hour, weekday / weekend.

[0027] environmental context features processed by normalization. By integrating environmental context information, the prediction model can adapt to changes in user habits and different seasonal needs, achieving the beneficial effect of the system "getting smarter the more it is used".

[0028] Further, the behavior prediction unit further comprises: a confidence judgment module, configured to compare the highest probability value output by the deep learning prediction module with a preset threshold, and only trigger the environment control unit when the highest probability value is greater than the preset threshold. A reliability threshold is introduced. Only when the prediction is large enough, the early control strategy is executed. Otherwise, it falls back to the default energy saving mode. This effectively avoids the misadjustment of high energy consumption and disturbance when the user behavior is uncertain (such as indoor wandering), ensuring the rationality and credibility of the system behavior, ensuring the stability of the system operation and avoiding the mistriggering caused by low confidence prediction.

[0029] Further, the deep learning prediction module is trained by minimizing a loss function, and the loss function L is the weighted sum of the position prediction loss and the time prediction loss, expressed as: L = L_cross-entropy + λ * L_MSE, Wherein, L_cross-entropy is the cross-entropy loss function, which is used to optimize the position classification accuracy; L_MSE is the mean square error loss function, which is used to optimize the time regression accuracy; λ is the weight coefficient for balancing the two types of loss functions.

[0030] Further, the system supports multi-user scenarios, and distinguishes the behavior patterns of different users by establishing and maintaining independent behavior prediction models for different users, or using a single model supporting multi-task learning. In addition, the system supports online incremental learning, and continuously updates the model with new generated trajectory data and real results as new samples, so as to continuously learn the historical behavior patterns of users and continuously optimize the prediction accuracy. Embodiments

[0031] The intelligent temperature control energy saving method based on personnel position detection and prediction comprises: S1: Real-time acquisition of personnel position information and environment information of each partition in the room; S2: Determine whether the personnel position has changed; S3: If it has changed, calculate the probability distribution and predicted arrival time of the next target partition based on the historical moving trajectory sequence through the trained deep learning model; S4: If the highest probability value calculated exceeds the confidence threshold, pre-process the environment of the target partition with the highest probability before the predicted arrival time; S5: Continuously monitor the occupancy state of each partition, and perform energy saving control on the unoccupied partition.

[0032] Further, after step S3 and before step S4, it further comprises: The highest probability value output by the deep learning model is compared with a preset threshold value; If the preset threshold value is lower, a default regulation strategy is enabled, and the default regulation strategy includes no early regulation or only regulation on the most recently active partition.

[0033] Further, the method further includes: The continuously collected new mobile trajectory data and the actually occurred results are taken as new training samples, and online incremental learning is performed on the deep learning model to adapt to the change of user behavior habits.

[0034] The method realizes the core idea of "on-demand regulation" by combining behavior prediction and monitoring: on the one hand, accurate prediction is provided for the space to be used, and comfort is ensured; on the other hand, the service of the unoccupied space is closed in time, and energy waste is avoided. This two-pronged strategy finally realizes global energy efficiency optimization, improves user experience, and significantly reduces building operation energy consumption.

[0035] The present application has been described by the above-mentioned related embodiments, however, the above-mentioned embodiments are only examples for implementing the present application. It must be pointed out that the disclosed embodiments do not limit the scope of the present application. On the contrary, changes and modifications made without departing from the spirit and scope of the present application are within the scope of the patent protection of the present application.

Claims

1. An intelligent temperature control energy saving system based on personnel location detection and prediction, characterized in that, The system comprises: a human and environment perception unit configured to collect personnel presence states and environmental parameters in multiple indoor sub-zones; a behavior prediction unit in communication connection with the perception unit and configured to predict a target sub-zone and a predicted arrival time based on historical time-series data including personnel movement trajectories and environmental context information; an environment regulation unit in communication connection with the behavior prediction unit and configured to start and set parameters of environmental regulation devices in the target sub-zone before the predicted arrival time according to the prediction result; an energy efficiency optimization unit configured to perform energy-saving control strategies on devices in unoccupied sub-zones according to the personnel presence states collected by the perception unit.

2. The intelligent temperature control energy saving system based on personnel location detection and prediction as claimed in claim 1, wherein: The behavior prediction unit comprises: a data management module configured to store and process personnel position data and environmental data uploaded by the perception unit in a time-stamped sequence; a deep learning prediction module internally provided with a trained recurrent neural network model and configured to receive real-time sequence data provided by the data management module for forward calculation and output probability distribution prediction and time prediction of a next target sub-zone.

3. The intelligent temperature control energy saving system based on personnel location detection and prediction as claimed in claim 2, wherein: The recurrent neural network model is a long short-term memory network model.

4. The intelligent temperature control energy saving system based on personnel location detection and prediction of claim 3, wherein: The input feature vector of the long short-term memory network model comprises: a position sub-zone identifier processed by one-hot encoding; a time-stamp feature processed by periodic encoding; an environmental context feature processed by normalization.

5. The intelligent temperature control energy saving system based on personnel location detection and prediction as claimed in claim 2, wherein: The behavior prediction unit further comprises: a confidence judgment module configured to compare a highest probability value output by the deep learning prediction module with a preset threshold, and only trigger the environment regulation unit when the highest probability value is greater than the preset threshold.

6. The intelligent temperature control energy saving system based on personnel location detection and prediction as claimed in claim 2, wherein: The deep learning prediction module is trained by minimizing a loss function, and the loss function L is a weighted sum of a position prediction loss and a time prediction loss, expressed as: L = L_cross-entropy + λ * L_MSE, where L_cross-entropy is a cross-entropy loss function for optimizing position classification accuracy; L_MSE is a mean square error loss function for optimizing time regression accuracy; and λ is a weight coefficient for balancing the two types of loss functions.

7. The intelligent temperature control energy saving system based on personnel location detection and prediction as claimed in claim 1, wherein: The system supports a multi-user scenario, and different behavior patterns of different users are distinguished by establishing and maintaining independent behavior prediction models for different users or by using a single model supporting multi-task learning.

8. The intelligent temperature control energy saving method based on personnel position detection and prediction, characterized in that: The method is applied to the system of any one of claims 1-7, and the method comprises: S1: collecting personnel position information and environmental information in each sub-zone in real time; S2: determining whether the personnel position has changed; S3: if the personnel position has changed, calculating a probability distribution of a next target sub-zone and a predicted arrival time based on a historical movement trajectory sequence by using a trained deep learning model; S4: if a highest probability value calculated exceeds a confidence threshold, performing environmental preprocessing on a target sub-zone with the highest probability before the predicted arrival time; S5: continuously monitoring occupancy states of each sub-zone and performing energy-saving control on unoccupied sub-zones.

9. The intelligent temperature control energy saving method based on personnel position detection and prediction according to claim 8, characterized in that: After step S3 and before step S4, the method further comprises: comparing a highest probability value output by the deep learning model with a preset threshold. If the preset threshold is not reached, a default regulation strategy is enabled, the default regulation strategy including no early regulation or regulation only on the most recently active partition.

10. The intelligent temperature control energy saving method based on personnel position detection and prediction as claimed in claim 8, wherein, The method further includes: The new mobile trajectory data continuously collected and the actually occurred results are taken as new training samples, and online incremental learning is performed on the deep learning model to adapt to changes in user behavior habits.

Images