A human body intelligent sensing method for water breathing circulation control in water heaters

Abstract

This invention belongs to the field of intelligent sensing, specifically disclosing a human body intelligent sensing method for controlling the water breathing cycle of a water heater. The method includes: transmitting signals, receiving signals, calculating intermediate frequency signals, signal sampling, filtering static background, identifying human activity states, and controlling the water breathing cycle of the water heater. This invention uses an FMCW radar to transmit a linear frequency modulated continuous wave sequence, establishes a transmission and reception signal model, and calculates the intermediate frequency signal. It utilizes moving target indication combined with exponentially weighted moving average to filter static background reflections, triggering the water breathing cycle only when a human is in a passing state. Through a deep neural network, combined with data augmentation techniques such as time offset, amplitude scaling, and randomized guided adaptation, as well as end-to-end Euclidean distance metric learning and out-of-distribution sample removal technology, it accurately extracts human distance, speed, and angle features and identifies human activity states, achieving a dual improvement in the anti-interference capability and recognition accuracy of human body sensing.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensing, specifically to a human body intelligent sensing method for controlling the water breathing circulation of a water heater. Background Technology

[0002] Water heater circulation control refers to the ability of water in a hot water system to circulate and regulate according to a certain pattern, achieving goals such as uniform water temperature, energy saving, and improved user experience. This ensures the stability of the hot water outlet temperature and enhances user comfort during hot water usage scenarios such as bathing. However, existing solutions, including temperature-controlled circulation, real-time circulation, and timed circulation, fail to consider actual water demand based on human activity levels. This results in the water heater repeatedly heating cold water and the recirculation system running idle, leading to unnecessary energy and water consumption and significant energy waste. Furthermore, traditional sensing methods often use infrared or ordinary microwave radar. Infrared sensors are easily affected by ambient light and temperature fluctuations, while ordinary microwave radar struggles to filter out reflected signals from static backgrounds such as furniture and walls. These sensors can only determine whether someone is present or not, failing to differentiate between human presence and activity levels. This can cause the water heater to erroneously activate circulation in non-use scenarios or fail to activate in time during use scenarios, severely impacting user comfort and system energy efficiency. Summary of the Invention

[0003] To address the above issues and overcome the shortcomings of existing technologies, this invention provides a human-based intelligent sensing method for water breathing circulation control in water heaters. Existing solutions, including temperature-controlled circulation, real-time circulation, and timed circulation, fail to consider actual water demand based on human activity levels. This leads to repeated heating of cold water by the water heater and idling of the recirculation system, resulting in unnecessary energy and water consumption and significant energy waste. This solution uses an FMCW radar to transmit a linear frequency modulated continuous wave sequence, establishes a transmission and reception signal model, and calculates the intermediate frequency signal. It utilizes moving target indication combined with exponentially weighted moving average filtering of static background reflections, triggering water breathing circulation only when a human is present. This achieves precise matching between water heater circulation control and human water demand, reducing unnecessary energy consumption. This addresses the limitations of traditional sensing methods. Most systems use infrared or ordinary microwave radar. Infrared radar is easily affected by ambient light and temperature fluctuations, while ordinary microwave radar struggles to filter out reflected signals from static backgrounds such as furniture and walls. It can only determine whether someone is present or not, but cannot distinguish between human states. This leads to water heaters mistakenly starting in non-water-use scenarios or failing to start in time during water-use scenarios, seriously affecting user comfort and system energy efficiency. This solution uses deep neural networks, combined with data augmentation techniques such as time offset, amplitude scaling, and randomized guided adaptation, as well as end-to-end Euclidean distance metric learning and out-of-distribution sample removal technology, to accurately extract human distance, speed, and angle features and identify human activity states. This achieves a dual improvement in the anti-interference capability and recognition accuracy of human body sensing, while accurately distinguishing different activity states to ensure timely hot water supply during water use, significantly improving the water user experience.

[0004] The technical solution adopted by this invention is as follows: This invention provides a human intelligent sensing method for controlling the water breathing circulation of a water heater, specifically including the following steps:

[0005] Step S1: Transmit a signal. Install a sensor, specifically an FMCW radar, on the necessary route to the water source. The FMCW radar transmits a linear frequency modulated continuous wave sequence. Using the linear frequency modulated continuous wave sequence as the transmitted signal, establish a transmitted signal model. The time-domain expression of a single linear frequency modulated continuous wave sequence is as follows: ;

[0006] In the formula, It is a single linear frequency modulated continuous wave sequence. It is a time variable. It is pi. It is the initial frequency of a single linear frequency modulated continuous wave sequence. It is the frequency slope;

[0007] Step S2: Receive the signal. The linear frequency modulated continuous wave sequence is modulated by the radial distance and velocity of the human body, resulting in a time delay and Doppler shift, thus obtaining the received signal. The formula used is as follows: ; ;

[0008] In the formula, It's a Doppler shift. It's the speed of light. It is the target radial velocity. It is receiving signals. It is the round-trip time of the FMCW radar transmitting a signal and then reflecting it back to the FMCW radar;

[0009] Step S3: Calculate the intermediate frequency (IF) signal. After mixing the received and transmitted signals, extract the difference frequency component using a low-pass filter to obtain the IF signal. The formula used is as follows: ;

[0010] In the formula, It is an intermediate frequency signal. It is the amplitude of the intermediate frequency signal. It is a fixed phase term. It is a time-varying phase term;

[0011] Step S4: Signal sampling. After the difference frequency component is sampled by the analog-to-digital converter, each linear frequency modulated continuous wave sequence obtains 2048 sampling points. The multi-frame and multi-receiver channel data in the linear frequency modulated continuous wave sequence are integrated to obtain four-dimensional raw data. The four-dimensional raw data includes the frame time dimension, the receiver channel dimension, the linear frequency modulated continuous wave sequence dimension, and the sampling point dimension.

[0012] Step S5: Filter static background. The reflected signals from static backgrounds in the indoor environment are removed using moving target indicators. An exponentially weighted moving average is used to balance static suppression and signal smoothing to obtain the human target signal. The formula for the exponentially weighted moving average is as follows: ;

[0013] In the formula, It is the first Static background estimate of frame data after exponentially weighted moving average. It is the smoothing coefficient of the exponentially weighted moving average. It is the first The digitized intermediate frequency signal of the frame;

[0014] Step S6: Identify human activity status using a deep neural network, whereby human activity status includes rapidly approaching, slowly approaching, passing by, and moving away;

[0015] Step S7: Control the water breathing circulation of the water heater. When the water heater detects that a human activity state has passed by, it will automatically turn on to release water, circulating cold water to the water heater for heating, and circulating hot water to the return water device to complete one water breathing cycle.

[0016] Further, step S6, identifying the human activity state, specifically includes the following steps:

[0017] Step S61: Establish and initialize a deep neural network as a human motion recognition model. The deep neural network includes an input layer, a fast-time convolutional layer, a slow-time convolutional layer, a spatial convolutional layer, a global average pooling layer, a fully connected layer, a feature normalization layer, and an output layer. The input layer is used to receive four-dimensional raw data. The fast-time convolutional layer is used to extract distance-related features of the human target. The slow-time convolutional layer is used to extract speed-related features of the human target. The spatial convolutional layer is used to extract angle-related features of the human target. The global average pooling layer is used to perform feature dimensionality reduction. The fully connected layer is used to perform feature fusion. The feature normalization layer is used to output normalized features. The output layer is used to determine the activity state of the human target.

[0018] Step S62: Perform data augmentation on the four-dimensional raw data, including time offset, amplitude scaling, and random guided adaptation. The time offset randomly offsets the four-dimensional raw data by ±5% sampling points along the fast time dimension and the slow time dimension, respectively, to simulate the small changes in the distance and speed of the human target. The amplitude scaling randomly scales the signal amplitude of the four-dimensional raw data by 0.8 to 1.2 times. The random guided adaptation distorts the signal of the four-dimensional raw data along the frame time dimension to simulate the speed differences of different human targets.

[0019] Step S63: Learn using an end-to-end Euclidean distance metric, and optimize the feature distance using triplet loss. The objective function for optimizing the triplet loss is as follows: ;

[0020] In the formula, It is the objective function for triplet loss optimization. It is an anchor point sample. These are positive samples of the same type as the anchor samples. It is a negative sample that is different from the anchor sample. It is the inter-class interval;

[0021] Step S64: Predict the feature centers for each type of human activity, calculate the Euclidean distance between the four-dimensional original data sample and the feature centers. If the Euclidean distance is greater than half of the minimum distance between class centers, the four-dimensional original data sample is determined to be an out-of-distribution sample. After removing the out-of-distribution samples, learning is performed. The method for calculating the feature centers is as follows: ;

[0022] In the formula, It is the characteristic center of every type of human activity. It is a category The number of samples, Yes Traversal;

[0023] Step S65: The human motion recognition model uses the sum of cross-entropy classification loss and L2 regularization as the loss function to output the human activity state.

[0024] The beneficial effects achieved by the present invention using the above solution are as follows:

[0025] (1) Existing solutions include temperature control circulation, real-time circulation and timed circulation, but they do not combine the actual water demand with the human activity state, resulting in the water heater repeatedly heating cold water and the water return device running idle, generating unnecessary power and water consumption and huge energy waste. This solution uses FMCW radar to transmit linear frequency modulated continuous wave sequence, establishes the transmission and reception signal model and calculates the intermediate frequency signal, and uses moving target indication combined with exponential weighted moving average to filter static background reflection. Water breathing circulation is triggered only when the human activity state is passing by, so as to achieve precise matching between water heater circulation control and human water demand, and reduce unnecessary energy consumption.

[0026] (2) In view of the technical problems that traditional sensing methods mostly use infrared or ordinary microwave radar, infrared is easily affected by ambient light and temperature fluctuations, and ordinary microwave radar is difficult to filter out the reflected signals of static backgrounds such as furniture and walls. It can only determine whether there are people or not, but cannot distinguish the human body state, which leads to the water heater being mistakenly started in a cycle in non-water use scenarios or not being started in time in water use scenarios, seriously affecting the user's comfort and system energy saving. This solution uses deep neural networks, combined with data augmentation methods such as time offset, amplitude scaling, and random guidance adaptation, as well as end-to-end Euclidean distance metric learning and out-of-distribution sample removal technology to accurately extract human body distance, speed, and angle features and identify human body activity state, thereby achieving a dual improvement in the anti-interference ability and recognition accuracy of human body sensing. At the same time, it accurately distinguishes different activity states to ensure timely supply of hot water when water is used, greatly improving the water use experience. Attached Figure Description

[0027] Figure 1 The present invention provides a flowchart of a human body intelligent sensing method for controlling the water breathing circulation of a water heater.

[0028] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0029] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0030] Example 1, see Figure 1 This invention provides a human intelligent sensing method for controlling the water breathing circulation of a water heater. The method specifically includes the following steps:

[0031] Step S1: Transmit a signal. Install a sensor, specifically an FMCW radar, on the necessary route to the water source. The FMCW radar transmits a linear frequency modulated continuous wave sequence. Using the linear frequency modulated continuous wave sequence as the transmitted signal, establish a transmitted signal model. The time-domain expression of a single linear frequency modulated continuous wave sequence is as follows: ;

[0032] In the formula, It is a single linear frequency modulated continuous wave sequence. It is a time variable. It is pi. It is the initial frequency of a single linear frequency modulated continuous wave sequence. It is the frequency slope;

[0033] Step S2: Receive the signal. The linear frequency modulated continuous wave sequence is modulated by the radial distance and velocity of the human body, resulting in a time delay and Doppler shift, thus obtaining the received signal. The formula used is as follows: ; ;

[0034] In the formula, It's a Doppler shift. It's the speed of light. It is the target radial velocity. It is receiving signals. It is the round-trip time of the FMCW radar transmitting a signal and then reflecting it back to the FMCW radar;

[0035] Step S3: Calculate the intermediate frequency (IF) signal. After mixing the received and transmitted signals, extract the difference frequency component using a low-pass filter to obtain the IF signal. The formula used is as follows: ;

[0036] In the formula, It is an intermediate frequency signal. It is the amplitude of the intermediate frequency signal. It is a fixed phase term. It is a time-varying phase term;

[0037] Step S4: Signal sampling. After the difference frequency component is sampled by the analog-to-digital converter, each linear frequency modulated continuous wave sequence obtains 2048 sampling points. The multi-frame and multi-receiver channel data in the linear frequency modulated continuous wave sequence are integrated to obtain four-dimensional raw data. The four-dimensional raw data includes the frame time dimension, the receiver channel dimension, the linear frequency modulated continuous wave sequence dimension, and the sampling point dimension.

[0038] Step S5: Filter static background. The reflected signals from static backgrounds in the indoor environment are removed using moving target indicators. An exponentially weighted moving average is used to balance static suppression and signal smoothing to obtain the human target signal. The formula for the exponentially weighted moving average is as follows: ;

[0039] In the formula, It is the first Static background estimate of frame data after exponentially weighted moving average. It is the smoothing coefficient of the exponentially weighted moving average. It is the first The digitized intermediate frequency signal of the frame;

[0040] Step S6: Identify human activity status using a deep neural network, whereby human activity status includes rapidly approaching, slowly approaching, passing by, and moving away;

[0041] Step S7: Control the water breathing circulation of the water heater. When the water heater detects that a human activity state has passed by, it will automatically turn on to release water, circulating cold water to the water heater for heating, and circulating hot water to the return water device to complete one water breathing cycle.

[0042] Example 2, see Figure 1 This embodiment is based on the above embodiment. Step S6, identifying the human activity state, specifically includes the following steps:

[0043] Step S61: Establish and initialize a deep neural network as a human motion recognition model. The deep neural network includes an input layer, a fast-time convolutional layer, a slow-time convolutional layer, a spatial convolutional layer, a global average pooling layer, a fully connected layer, a feature normalization layer, and an output layer. The input layer is used to receive four-dimensional raw data. The fast-time convolutional layer is used to extract distance-related features of the human target. The slow-time convolutional layer is used to extract speed-related features of the human target. The spatial convolutional layer is used to extract angle-related features of the human target. The global average pooling layer is used to perform feature dimensionality reduction. The fully connected layer is used to perform feature fusion. The feature normalization layer is used to output normalized features. The output layer is used to determine the activity state of the human target.

[0044] Step S62: Perform data augmentation on the four-dimensional raw data, including time offset, amplitude scaling, and random guided adaptation. The time offset randomly offsets the four-dimensional raw data by ±5% sampling points along the fast time dimension and the slow time dimension, respectively, to simulate the small changes in the distance and speed of the human target. The amplitude scaling randomly scales the signal amplitude of the four-dimensional raw data by 0.8 to 1.2 times. The random guided adaptation distorts the signal of the four-dimensional raw data along the frame time dimension to simulate the speed differences of different human targets.

[0045] Step S63: Learn using an end-to-end Euclidean distance metric, and optimize the feature distance using triplet loss. The objective function for optimizing the triplet loss is as follows: ;

[0046] In the formula, It is the objective function for triplet loss optimization. It is an anchor point sample. These are positive samples of the same type as the anchor samples. It is a negative sample that is different from the anchor sample. It is the inter-class interval;

[0047] Step S64: Predict the feature centers for each type of human activity, calculate the Euclidean distance between the four-dimensional original data sample and the feature centers. If the Euclidean distance is greater than half of the minimum distance between class centers, the four-dimensional original data sample is determined to be an out-of-distribution sample. After removing the out-of-distribution samples, learning is performed. The method for calculating the feature centers is as follows: ;

[0048] In the formula, It is the characteristic center of every type of human activity. It is a category The number of samples, Yes Traversal;

[0049] Step S65: The human motion recognition model uses the sum of cross-entropy classification loss and L2 regularization as the loss function to output the human activity state.

[0050] Example 3 is based on the above examples. In Example 1, the FMCW radar transmits a linear frequency modulated continuous wave sequence. The carrier frequency of the linear frequency modulated continuous wave sequence is 77 GHz, the single bandwidth is 2 GHz, the effective linear frequency modulated continuous wave sequence time is 100 microseconds, and the frequency slope is 2× Hz / s, with a repetition period of 1 millisecond, and each frame contains 64 linear frequency modulated continuous wave sequences.

[0051] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0052] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0053] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A human body intelligent sensing method for controlling the water breathing circulation of a water heater, characterized in that, Specifically, the following steps are included: Step S1: Transmit a signal. Install a sensor on the only way to the water supply point. The sensor is an FMCW radar. The FMCW radar transmits a linear frequency modulated continuous wave sequence. Use the linear frequency modulated continuous wave sequence as the transmitted signal to establish a transmitted signal model. Step S2: Receive signal. The linear frequency modulated continuous wave sequence is modulated by the radial distance and velocity of the human body, resulting in time delay and Doppler frequency shift, thus obtaining the received signal. Step S3: Calculate the intermediate frequency signal. After mixing the received and transmitted signals, extract the difference frequency component through a low-pass filter to obtain the intermediate frequency signal. Step S4: Signal sampling. After the difference frequency component is sampled by the analog-to-digital converter, each linear frequency modulated continuous wave sequence obtains 2048 sampling points. The multi-frame and multi-receiver channel data in the linear frequency modulated continuous wave sequence are integrated to obtain four-dimensional raw data. Step S5: Filter static background. Remove the reflection signal of static background in the indoor environment by using moving target indicator. Use exponential weighted moving average to balance static suppression and signal smoothing to obtain human target signal. Step S6: Identify human activity status using a deep neural network, whereby human activity status includes rapidly approaching, slowly approaching, passing by, and moving away; Step S7: Control the water breathing circulation of the water heater. When the water heater detects that a human activity state has passed by, it will automatically turn on to release water, circulating cold water to the water heater for heating, and circulating hot water to the return water device to complete one water breathing cycle.

2. The human body intelligent sensing method for water breathing circulation control in a water heater according to claim 1, characterized in that, Step S6, identifying human activity status, specifically includes the following steps: Step S61: Establish and initialize a deep neural network as a human motion recognition model. The deep neural network includes an input layer, a fast-time convolutional layer, a slow-time convolutional layer, a spatial convolutional layer, a global average pooling layer, a fully connected layer, a feature normalization layer, and an output layer. Step S62: Perform data augmentation on the four-dimensional raw data, including time offset, amplitude scaling, and random guided adaptation. The time offset randomly offsets the four-dimensional raw data by ±5% sampling points along the fast time dimension and the slow time dimension, respectively, to simulate the small changes in the distance and speed of the human target. The amplitude scaling randomly scales the signal amplitude of the four-dimensional raw data by 0.8 to 1.2 times. The random guided adaptation distorts the signal of the four-dimensional raw data along the frame time dimension to simulate the speed differences of different human targets. Step S63: Learn using an end-to-end Euclidean distance metric and optimize feature distance through triplet loss; Step S64: Predict the feature center of each type of human activity, calculate the Euclidean distance between the four-dimensional original data sample and the feature center. If the Euclidean distance is greater than half of the minimum distance between class centers, the four-dimensional original data sample is determined to be an out-of-distribution sample. After removing the out-of-distribution samples, learning is performed. Step S65: The human motion recognition model uses the sum of cross-entropy classification loss and L2 regularization as the loss function to output the human activity state.

Images