Intelligent home equipment intelligent control method based on deep learning

Abstract

The invention relates to an intelligent home equipment intelligent control method based on deep learning, and the method comprises the following steps: S1, collecting environment data and the operation state information of each piece of home equipment in real time through various sensors disposed in an intelligent home environment, the environment data including temperature, humidity, illumination intensity and personnel activity conditions; the operation state information of the home equipment comprises an on-off state, a working mode and an energy consumption parameter of the equipment; all the collected data are transmitted to the central control system; s2, according to the environment data and the home equipment operation state information obtained in the step S1, the current operation mode of the smart home equipment is analyzed, and the operation mode of the smart home equipment at least comprises a basic function mode, an energy efficiency management mode and a multi-equipment interaction linkage mode; and S3, based on the current operation mode analyzed and determined in the step S2, generating an optimization control instruction for one or more smart home devices, and selecting different energy consumption driving modes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart home, and particularly relates to a smart home device intelligent control method based on deep learning. BACKGROUND

[0002] With the rapid development of Internet of Things technology and the wide popularity of smart home market, more and more types of sensors and smart devices are deployed in the home environment, such as temperature and humidity sensors, light intensity detectors, human activity monitoring devices, and various network-controllable home appliances (air conditioners, lighting systems, water heaters, etc.). A complex home automation network is formed through the interconnection of the above-mentioned smart devices and sensors, aiming to improve living comfort, realize energy efficient utilization and support collaborative interaction among multiple devices.

[0003] However, the existing smart home control method usually adopts a fixed threshold trigger mechanism or a mode switching strategy based on simple logic rules (such as timing switch, single parameter linkage), which is difficult to dynamically perceive the complex changing relationship between the environment and the devices. For example, in the energy efficiency management mode, only relying on historical average data for coarse-grained adjustment cannot respond to sudden personnel activities or sudden changes in light in real time; in the multi-device linkage scene, the preset scene rules often lack self-learning ability, resulting in low efficiency of cross-device collaboration, and even control instruction conflicts.

[0004] Therefore, we propose a smart home device intelligent control method based on deep learning. SUMMARY

[0005] The present application proposes a smart home device intelligent control method based on deep learning, aiming to solve the problems of rigid mode switching, insufficient data processing capability and unbalanced multi-target optimization in the prior art.

[0006] To achieve the above-mentioned purpose, the present application proposes a smart home device intelligent control method based on deep learning, comprising the following steps: S1, using various types of sensors deployed in the smart home environment to collect real-time environmental data and running state information of each home device, the environmental data including temperature, humidity, light intensity and personnel activity; the running state information of the home device including the on-off state, working mode and energy consumption parameter of the device; and transmitting all collected data to the central control system; S2, according to the environmental data and home device running state information obtained in step S1, analyzing the running mode of the current smart home device, the running mode of the smart home device at least including: basic function mode, energy efficiency management mode and multi-device interactive linkage mode; The basic function mode is used for independent running state of a single smart home device, and is controlled according to a preset rule or a direct instruction of a user; the energy efficiency management mode is used for dynamically optimizing a device running parameter or a start-stop strategy to realize overall energy consumption minimization according to an associated analysis of environmental data and a device energy consumption parameter on the premise of meeting basic home function requirements; and the multi-device interactive linkage mode is used for coordinating multiple associated smart home devices to work cooperatively according to a change of environmental data, a state triggering of a specific device or a preset scene rule. S3, generating an optimized control instruction for one or more smart home devices based on the current running mode determined in step S2, and selecting different energy consumption driving modes.

[0007] As preferred, the step S2 comprises the following steps: Based on the environmental data and the device running state information collected in real time in step S1, a change amount of an environment and a device at adjacent time points is calculated; a current change rate of the environment and the device is calculated in combination with a sampling frequency or a time interval of a sensor; and a running mode of the smart home device is determined through the current change rate of the environment and the device.

[0008] As preferred, the running mode determination process of the smart home device is specifically: When the change rate of the environment and the device is higher than a preset first change threshold value, and a device state change intensity is higher than a preset second change threshold value, it is determined as the multi-device interactive linkage mode; when the change rate of the environment and the device is lower than a preset third change threshold value, or the monitored device state change intensity is lower than a preset fourth change threshold value, and the state lasts for more than a preset duration threshold value, it is determined as the energy efficiency management mode; when the determination conditions of the multi-device interactive linkage mode or the energy efficiency management mode are not triggered, it is determined as the basic function mode. The first change threshold value is a maximum value of the current change rate of the environment and the device, and is used for determining whether to enter the multi-device interactive linkage mode; the second change threshold value is a minimum value of the device change intensity, and is used for determining whether to enter the multi-device interactive linkage mode in combination with the first change threshold value; the third change threshold value is a minimum value of the current change rate of the environment and the device, and is used for determining whether to trigger the energy efficiency management mode; and the fourth change threshold value is a minimum value of the device change intensity, and is used for determining whether to enter the energy efficiency management mode in combination with the third change threshold value.

[0009] As preferred, the S2 further comprises the following steps: The environmental data and the device running state information collected in real time in step S1 are time-stamped and aligned, and are unified in dimension through normalization processing.

[0010] As preferred, the S3 comprises the following steps: According to the current operation mode determined in step S2, a corresponding energy consumption driving control strategy is selected; If it is a multi-device interactive linkage mode, a scene cooperative deep learning model is called to perform feature fusion on real-time environment data, a device state matrix, and a preset scene rule, construct a space-time feature tensor, output a control instruction probability distribution of each device within a preset time window through the scene cooperative deep learning model, and generate a device control instruction sequence according to the instructions with a selection confidence higher than a set threshold and according to a time window. If it is an energy efficiency management mode, an energy consumption optimization deep learning model is called to input historical energy consumption data and environment parameters into a time series encoder, extract cross-period energy consumption features, calculate the correlation weight of the current environment parameters and the historical features through an attention mechanism, predict the device energy consumption trend in a future short period according to the weighted feature vector, and dynamically adjust the power distribution coefficient of each device according to the real-time running load of the device to generate a device operation parameter adjustment instruction or a start-stop strategy. If it is a basic function mode, a control instruction corresponding to the current environment data and the device state is matched from a preset rule library.

[0011] Preferably, the scene cooperative deep learning model adopts a time series convolution network structure, the input is aligned multi-sensor time series data and a device state matrix, and the output is a control instruction probability distribution of each device; the energy consumption optimization deep learning model adopts a Long Short-Term Memory (LSTM)-Attention Mechanism, the input is normalized environment parameters and a device energy consumption history sequence, and the output is an optimal operation parameter set of the device in the next period.

[0012] Preferably, when the energy consumption optimization deep learning model executes the start-stop strategy, a time-sharing operation instruction set is generated for interruptible devices, a parameter optimization instruction is generated for continuous running devices, and when the device operation parameter optimization instruction is output, the following constraints are met: An environment comfort constraint, the adjusted environment parameters are maintained in a user preset parameter interval; A device safe operation constraint, the power distribution coefficient αᵢ meets the device technical specification and the power change rate limit, the power distribution coefficient αᵢ is a ratio of the actual power to the rated power, and αᵢ ∈ [0, 1]; A power grid interaction constraint, the total electricity bill calculated based on the real-time electricity price cannot exceed the user preset daily budget upper limit; A device cooperation constraint: the total power when multiple devices are running cannot exceed the rated capacity of the total incoming line circuit breaker of the home.

[0013] Preferably, the preset scene rule comprises at least a trigger condition rule for defining an environment threshold combination, a device coordination rule for defining a device action logic relationship, and a rule for defining an instruction execution timing constraint.

[0014] Preferably, the S3 further comprises the following steps: The feasibility of the instruction is verified by a fuzzy control algorithm, and if there is a resource conflict or a target conflict in multiple instructions, arbitration is performed according to a preset priority rule.

[0015] Preferably, the priority rule is specifically: (a) Safety constraint priority: instructions that violate device safety operation constraints or grid interaction constraints are automatically downgraded; (b) User explicit demand priority: the priority of a device control instruction initiated by a user is higher than that of an optimization instruction automatically generated by the system; (c) Core function guarantee priority: the priority of a basic function mode instruction is higher than that of an energy efficiency management mode instruction; (d) Economic optimization grading: in a multi-device interactive mode, the device start-stop sequence is allocated according to the unit energy consumption cost saving rate, so as to satisfy: ; and the constraint condition is: , wherein, is a set of interruptible devices; represents the actual energy saving value of the device i; represents the energy consumption change of the device i; is the delayed start time of the device i; is the real-time electricity price in the delay period; is the expected energy saving value of the device i; is the maximum delay time allowed for the device i.

[0016] The technical scheme of the present application has the following advantages: By collecting real-time environment data and device operation state information, and calculating the comprehensive state change amount at adjacent time points and the comprehensive change rate obtained in combination with the sampling frequency, the dynamic change trend of the environment and the device state can be accurately captured. Compared with the traditional static threshold judgment method, for some sudden environmental changes such as temperature rise in a short time, a large number of people entering, etc., the system can accurately judge whether the operation mode of the device needs to be adjusted according to the real-time calculation of the comprehensive change rate, thereby avoiding problems such as device misoperation or response lag caused by unreasonable fixed threshold setting, and greatly improving the accuracy and timeliness of system decision-making.

[0017] Multi-device interaction linkage mode: when the system determines that the current is in the multi-device interaction linkage mode, the system will enter a highly coordinated and rapid response state to cope with sudden and significant environmental changes or user intensive interaction requirements, and provide an instant, smooth and comfortable overall experience through intelligent coordination of the actions of multiple devices.

[0018] Energy efficiency management mode: when the comprehensive change rate of the environment and the device is lower than the third change threshold, or the device state change intensity is lower than the fourth change threshold and lasts for a certain time, the system automatically switches to the energy efficiency management mode, and in this mode, the system optimizes and adjusts the running parameters of each device to reduce energy consumption as much as possible on the premise of ensuring basic functions. Compared with the traditional static threshold switching, the energy efficiency management mode switching mode compares the current environment and device comprehensive change rate with the preset minimum value of the current environment and device comprehensive change rate change rate, and then determines that the current environment state and device running tend to be stable or change extremely slowly; or in combination with the preset minimum value of the current environment and device comprehensive change rate change rate and the minimum value of the device change intensity, and then compared with the current environment and device comprehensive change rate, it is determined that the current environment state and device running tend to be stable, and the running state of each device is very stable, and automatically enters the energy efficiency management mode. In this mode, the system will optimize and adjust the running parameters of each device.

[0019] By introducing the duration threshold, the system can effectively filter the false triggering caused by sensor transient noise or environmental micro-disturbance. For example, when the light intensity fluctuates slightly due to temporary cloud changes, the traditional static threshold may repeatedly trigger mode switching, while the present scheme requires that the low change state lasts for more than a set time before switching to the energy efficiency mode, avoiding unnecessary mode oscillation.

[0020] And the basic function mode: when the system monitored environment and device comprehensive change rate and device state change intensity neither meet the stringent conditions for triggering the multi-device interaction linkage mode nor meet the conditions for entering the energy efficiency management mode, the system will maintain or switch to the basic function mode. In this mode, the smart home device runs according to the user's pre-set regular strategy or default parameters, responds to regular operation instructions and environmental changes, and maintains basic home environment comfort and functionality. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The logic flowchart of an embodiment of the present application; Figure 2 The running mode determination logic diagram of an embodiment of the present application.

[0022] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0023] The scheme in the embodiments of the present application will be described in detail below with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of protection of the present application.

[0024] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present application are only used to explain the relative position relationship, movement condition, etc. between components in a certain posture (as shown in the drawings), and if the certain posture changes, the directional indications also change accordingly.

[0025] It should also be noted that when an element is referred to as being "fixed" or "disposed" on another element, it can be directly on the other element or can have a middle element. When an element is referred to as being "connected" to another element, it can be directly connected to the other element or can have a middle element.

[0026] In addition, if the present application involves "first", "second", etc. descriptions, they are only for description purposes (such as for distinguishing the same or similar elements), and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the features limited by "first", "second" can explicitly or implicitly include at least one of the features. In addition, the technical solutions of each embodiment can be combined with each other, but it must be based on the realization of a person of ordinary skill in the art, and when the combination of technical solutions contradicts each other or cannot be realized, it should be considered that the combination of technical solutions does not exist and is not within the scope of protection claimed by the present application.

[0027] Referring to Figure 1 The present application proposes an intelligent home device intelligent control method based on deep learning, comprising the following steps: S1, using various sensors deployed in the smart home environment to collect environmental data and running state information of each home device in real time, the environmental data including temperature, humidity, light intensity and personnel activity; the running state information of the home device includes the on-off state, working mode and energy consumption parameter of the device; and all the collected data are transmitted to the central control system; S2, according to the environmental data and the running state information of the home device obtained in step S1, analyzing the running mode of the current smart home device, the running mode of the smart home device at least including: basic function mode, energy efficiency management mode and multi-device interactive linkage mode; The basic function mode is used for the independent running state of a single smart home device, and is controlled according to a preset rule or a direct instruction of a user; the energy efficiency management mode is used for dynamically optimizing a device running parameter or a start-stop strategy to realize the minimization of overall energy consumption according to the correlation analysis of environmental data and device energy consumption parameters on the premise of meeting basic home function requirements; and the multi-device interactive linkage mode is used for coordinating multiple associated smart home devices to work cooperatively according to the change of environmental data, the state triggering of a specific device or a preset scene rule. S3, generating an optimized control instruction for one or more smart home devices based on the current running mode determined in step S2, and selecting different energy consumption driving modes.

[0028] In this embodiment, by collecting environmental data (such as temperature, humidity, and light intensity) and personnel activity in real time, the actual environment and needs of the user can be accurately perceived. Thus, in hot summer, when it is detected that the indoor temperature is high and there is someone present, the system can automatically adjust the temperature setting value and the air speed of the air conditioner, so that the indoor environment remains cool and comfortable; and in a dimly lit area, the corresponding lighting device is turned on in time according to the light intensity data, so as to provide a suitable visual environment for the user, improve the comfort of the home environment, and make the home life more in line with the personalized needs of the user.

[0029] The multi-device interactive linkage mode further enhances the user experience. For example, when the user enters the bedroom to prepare for rest, the system can automatically turn off the lights and the television in the living room, dim the main light in the bedroom and turn on the bedside lamp, and pull up the curtains according to the preset scene rule, so as to create a quiet and warm sleep atmosphere. The cooperative work between different devices enables the home environment to automatically respond to the behavior and habits of the user, and brings the user a more convenient and comfortable life experience.

[0030] The energy efficiency management mode can accurately identify which device running parameter or start-stop strategy has optimization space through in-depth correlation analysis of environmental data and device energy consumption parameters. For example, in the case of sufficient natural light during the day, the system will reduce the use time and brightness of artificial lighting, thereby reducing the consumption of electric energy; for high-power electrical appliances such as air conditioners, the running frequency and cooling / heating power are reasonably adjusted according to the indoor and outdoor temperature difference and the personnel distribution, so as to avoid unnecessary energy waste. And by dynamically optimizing the device running parameter or the start-stop strategy, the minimization of overall energy consumption is realized.

[0031] In addition, as time goes by and the number of uses increases, the system continuously accumulates a large amount of running data, which provides rich training samples for the deep learning model, so that it can gradually learn and understand the living habits, preferences of the user, and the best device configuration scheme under different environments.

[0032] In one embodiment, step S2 includes the following steps: Based on the environmental data and equipment operating status information collected in real time in step S1, the comprehensive change in the environment and equipment status at adjacent time points is calculated; combined with the sampling frequency or time interval of the sensors, the current comprehensive change rate of the environment and equipment is calculated; and the operating mode of the smart home device is determined by the obtained current comprehensive change rate of the environment and equipment.

[0033] Specifically, the comprehensive environmental state change refers to the overall change value obtained by comprehensively calculating the changes in all environmental parameters (such as temperature, humidity, and light intensity) between two adjacent data collection points. It is used to reflect the overall change in the surrounding environment over a short period of time.

[0034] Specifically, suppose there are n environmental parameters (such as temperature, humidity, light intensity, etc.), denoted as: And the value of each parameter at time point t is... ; and on The parameters in the data are normalized, and the parameters are set as follows: The historical minimum value is The historical maximum value is The normalized value at time point t can be obtained as follows: ; Furthermore, for adjacent time points t and t+1, the parameters The change is: Weights are assigned based on the impact of temperature, humidity, and light on changes in environmental conditions. ,satisfy The weighted parameter change is: ; Finally, the expression for the overall change in environmental state is derived as follows: ; The overall change in the equipment's status is specifically represented as: a total change value obtained by comprehensively calculating the changes in various operating parameters of the equipment (such as power, wind speed, and setpoints) between two adjacent data collection points. It is used to reflect the adjustment of the equipment's operating status over a short period of time.

[0035] Specifically, suppose there are m operating parameters for the equipment (such as power, wind speed, setpoint, etc.), denoted as: And the value of each parameter at time point t is... ; and on The parameters in the data are normalized, and the parameters are set as follows: The historical minimum value is The historical maximum value is The normalized value at time point t can be obtained as follows: ; Further, for adjacent time points t and t+1, the change amount of the parameter is: ; and the weight is set according to the influence of temperature, humidity, and light on the change of the environmental state, satisfying ; then the weighted change amount of the parameter is ; Finally, the expression of the comprehensive change amount of the device is: ; The comprehensive change rate of the environment specifically represents the result obtained by dividing the above-mentioned "comprehensive change amount of the environmental state" by the time experienced to produce the change. It is used to reflect the speed and intensity of the overall change of the environment.

[0036] The comprehensive change rate of the device specifically represents the result obtained by dividing the above-mentioned "comprehensive change amount of the device" by the time experienced to produce the change. It is used to reflect the speed and frequency of the adjustment of the running state of the device itself.

[0037] Referring to Figure 2 , the running mode determination process of the smart home device is specifically: When the comprehensive change rate of the environment and the device is higher than a preset first change threshold, and the intensity of the change of the device state monitored is higher than a preset second change threshold, it is determined as a multi-device interactive linkage mode; when the comprehensive change rate of the environment and the device is lower than a preset third change threshold, or the intensity of the change of the device state monitored is lower than a preset fourth change threshold, and the state continues to exceed a preset duration threshold, it is determined as an energy efficiency management mode; when the determination conditions of the multi-device interactive linkage mode or the energy efficiency management mode are not triggered, it is determined as a basic function mode. Further, the intensity of the change of the device state specifically represents the mathematical expression of the intensity index of the change of the device state based on weighted average within a preset time period T (containing N sampling points): ; wherein T is the preset length of the time window; N is the number of sampling points in the time window T, ; is the key running parameter of the device in the time window T, such as power, speed, working mode, and set value; is the weight of the parameter , used to reflect the importance of the parameter to the change of the device state; is the data acquisition time interval.

[0038] Specifically, a high device state change intensity indicates that the device is frequently adjusting itself to respond to some change (such as a user instruction or an environmental fluctuation); and a low intensity indicates that the device is in a stable, constant operating state.

[0039] The first change threshold is the maximum value of the current environment and device comprehensive change rate, and is used to determine whether to enter the multi-device interactive linkage mode; the second change threshold is the minimum value of the device change intensity, and is used to determine whether to enter the multi-device interactive linkage mode in combination with the first change threshold; the third change threshold is the minimum value of the current environment and device comprehensive change rate, and is used to determine whether to trigger the energy efficiency management mode; and the fourth change threshold is the minimum value of the device change intensity, and is used to determine whether to enter the energy efficiency management mode in combination with the third change threshold.

[0040] In this embodiment, by collecting environmental data and device operating state information in real time, and calculating the comprehensive state change amount of adjacent time points and the comprehensive change rate obtained in combination with the sampling frequency, the dynamic change trend of the environment and the device state can be accurately captured. Compared with the traditional static threshold determination method, for some sudden environmental changes such as a sudden temperature rise in a short time, a large number of people pouring in, etc., the system can accurately determine whether the operating mode of the device needs to be adjusted according to the real-time calculated comprehensive change rate, thereby avoiding problems such as device misoperation or response lag caused by unreasonable fixed threshold setting, and greatly improving the accuracy and timeliness of system decision-making. That is, when the environment and device comprehensive change rate is higher than the first change threshold and the device state change intensity is higher than the second change threshold, the multi-device interactive linkage mode is determined in time, and multiple devices are quickly coordinated to work together to meet the user's immediate needs and provide a more comfortable and convenient living environment.

[0041] Specifically, the multi-device interactive linkage mode: when the system determines that the current is in the multi-device interactive linkage mode (i.e., the environment and device comprehensive change rate is higher than the preset first change threshold and the device state change intensity is higher than the preset second change threshold), the system will enter a highly coordinated and rapid response state to cope with sudden and significant environmental changes or user intensive interaction needs, and provide an immediate, smooth and comfortable overall experience by intelligently coordinating the actions of multiple devices.

[0042] Energy efficiency management mode: when the comprehensive change rate of the environment and the device is lower than the third change threshold, or the device state change intensity is lower than the fourth change threshold and lasts for a certain time, the system automatically switches to the energy efficiency management mode, and in this mode, the system optimizes and adjusts the running parameters of each device to reduce energy consumption as much as possible on the premise of ensuring basic functions. Compared with the traditional static threshold switching, the current environment and device comprehensive change rate is compared with the preset current environment and device comprehensive change rate change rate minimum value, and it is judged that the current environment state and device running tend to be stable or change extremely slowly; or the preset current environment and device comprehensive change rate change rate minimum value and the minimum value of the device change intensity are combined, and then compared with the current environment and device comprehensive change rate, it is judged that the current environment state and device running tend to be stable, and the running state of each device (such as power, switching frequency, set value change amplitude, etc.) is very stable, and automatically enters the energy efficiency management mode. In this mode, the system will optimize and adjust the running parameters of each device. Taking air conditioning equipment as an example, the temperature setting value of refrigeration or heating will be dynamically adjusted according to the current indoor and outdoor temperature difference, personnel activity, etc., the temperature setting range will be appropriately increased, and the start-stop frequency of the compressor will be reduced; for the lighting system, the brightness of the lamps will be automatically reduced or some unnecessary lighting areas will be turned off according to the change of natural light intensity; the ventilation equipment will also reasonably adjust the wind speed and running time according to the indoor air quality index In addition, by introducing a duration threshold, the system can effectively filter false triggers caused by sensor transient noise or environmental micro-disturbances. For example, when the light intensity fluctuates slightly due to temporary cloud changes, the traditional static threshold may repeatedly trigger mode switching, while the present scheme requires that the low change state last for more than a certain time (such as > 10 minutes) before switching to the energy efficiency mode, avoiding unnecessary mode oscillation.

[0043] And the basic function mode: when the system monitored environment and device comprehensive change rate and device state change intensity neither meet the stringent conditions of triggering the multi-device interactive mode (i.e. the comprehensive change rate is higher than the first threshold and the device intensity is higher than the second threshold), nor meet the conditions of entering the energy efficiency management mode (i.e. the comprehensive change rate does not last below the third threshold or the device intensity does not last below the fourth threshold), the system will maintain or switch to the basic function mode. In this mode, the smart home devices run according to the user's pre-set regular strategy or default parameters, respond to regular operation instructions and environmental changes, and maintain basic home environment comfort and functionality. At the same time, the basic function mode ensures that the system runs stably in a dynamically changing environment, and prepares for responding to more intense changes (triggering the interactive mode) or longer calm periods (triggering the energy efficiency mode).

[0044] For example, the sensor sampling frequency is set to 1 min, so the time interval Δt = 1 min, the environmental data: temperature (range 10°C-30°C), humidity (range 20%-80%), light (range 0-1000 lux), device running state: device power (range 0-1000 W).

[0045] Further, the first change threshold: the maximum value of the comprehensive change rate of the environment and the device, is set to 0.5.

[0046] The second change threshold: the minimum value of the device state change intensity, is set to 0.3.

[0047] The third change threshold: the minimum value of the comprehensive change rate of the environment and the device, is set to 0.05.

[0048] The fourth change threshold: the minimum value of the device state change intensity, is set to 0.1.

[0049] The duration threshold: the low change duration for energy efficiency mode determination, is set to 3 min.

[0050] According to the above-mentioned raw data, time stamp alignment and normalization processing are performed, and the raw data and normalization results of all time points are as shown in Table 1:

[0051] Table 1 According to the content in Table 1, the comprehensive state change amount of the environment and the device is the absolute difference of the comprehensive state of adjacent time points, and the comprehensive state change rate of the environment and the device ; the comprehensive state change amount and the change rate of the environment and the device are calculated to obtain Table 2

[0052] Table 2 And use the following logic determination mode (determine at each time point t): Multi-device interactive linkage mode: when Rt>Rmax(0.5) and It>Iminlink(0.3), trigger immediately.

[0053] Energy efficiency management mode: when (Rt<Rmin(0.05) or It<Iminenergy(0.1)) and the state lasts more than Tduration(3 seconds), trigger.

[0054] Basic function mode: trigger when the above conditions are not met.

[0055] Duration handling: Energy efficiency mode requires low change state duration count. Design a counter lowchangecount, increase when condition (Rt < Rmin or It < Iminenergy) is true, otherwise reset to 0. Trigger energy efficiency mode when lowchangecount >= Tduration.

[0056] According to the above table 2, table 3 can be obtained:

[0057] Table 3 According to the above table, the multi-device interaction linkage mode trigger (t=4): due to environmental mutation (such as sudden temperature drop, light reduction), Rt=0.6375>Rmax=0.5 and It=0.70>Iminlink=0.3, the system immediately enters the linkage mode, and coordinates the devices (such as air conditioner, lighting) to quickly respond.

[0058] Energy efficiency management mode trigger (t=8): low change state (Rt < Rmin and It < Iminenergy) lasts for 3 seconds (t=6 to t=8), the system switches to energy efficiency mode to avoid false triggering caused by transient noise (such as small change at t=6).

[0059] Basic function mode: at time points t=1-3, t=5-7, etc., the system maintains the default operation and responds to regular changes.

[0060] In one of the embodiments, the S3 comprises the following steps: According to the current operation mode determined in step S2, a corresponding energy consumption driving control strategy is selected; If it is a multi-device interaction linkage mode, a scene coordination deep learning model is called, real-time environment data, device state matrix and preset scene rules are fused, a space-time feature tensor is constructed, control instruction probability distribution of each device in a preset time window is output through the scene coordination deep learning model, instructions with probability distribution selection confidence higher than a set threshold are selected, and device control instruction sequence is generated according to time window. If it is an energy efficiency management mode, an energy consumption optimization deep learning model is called, historical energy consumption data and environment parameters are input into a time sequence encoder, cross-period energy consumption features are extracted, an attention mechanism is used to calculate the correlation weight of current environment parameters and historical features, device energy consumption trend in a future short period is predicted according to the weighted feature vector, and device power distribution coefficient is dynamically adjusted according to real-time running load of the device to generate device running parameter adjustment instruction or start-stop strategy. If it is a basic function mode, control instructions corresponding to the current environment data and device state are matched from a preset rule library.

[0061] In the multi-device interaction linkage mode, the scene coordination deep learning model is used to realize the feature fusion of real-time environment data, device state matrix and preset scene rules, the constructed spatiotemporal feature tensor comprehensively captures the dynamic characteristics of the system, the output control instruction probability distribution has high accuracy and pertinence, the device control instruction sequence generated according to the time window can accurately coordinate the operation rhythm of each device, ensures that the collaborative work between devices is more efficient and smooth, effectively avoids conflicts or resource waste caused by improper control, and significantly improves the flexibility and response ability of the entire system to complex scenes.

[0062] In the energy efficiency management mode, by extracting the cross-period energy consumption characteristics and calculating the correlation weight of the current environment parameters and the historical characteristics, the device energy consumption trend in the future short period can be accurately predicted. On this basis, combined with the real-time running load of the device, the power distribution coefficient of each device is dynamically adjusted, and the generated device running parameter adjustment instruction or start-stop strategy can make the energy distribution more reasonable, which helps to reduce the overall energy consumption level of the system and reduce unnecessary energy consumption, so as to realize the goal of energy saving and emission reduction, and meet the concept of sustainable development.

[0063] When in the basic function mode, the control instruction corresponding to the current environment data and device state is matched from the preset rule library. Through the preset rule library, various common basic scenes and working conditions are covered, which can ensure the stable operation of the device at the basic function level. Whether it is daily use or emergency treatment in special circumstances, appropriate control instructions can be quickly and accurately obtained, effectively avoiding delays or errors caused by complex algorithms.

[0064] In one of the embodiments, the scene coordination deep learning model adopts a time sequence convolution network structure, the input is the aligned multi-sensor time sequence data and device state matrix, and the output is the control instruction probability distribution of each device; the energy consumption optimization deep learning model adopts a long short-term memory network-attention mechanism, the input is the normalized environment parameters and device energy consumption history sequence, and the output is the optimal running parameter set of the device in the next period.

[0065] Specifically, during the running of the scene coordination deep learning model, when the time sequence convolutional network receives the aligned multi-sensor time sequence data and the device state matrix, it will first extract features from the input data through multiple convolutional layers. The above-mentioned convolutional layers have different receptive fields and can capture data patterns and associated information at different time scales. For example, the convolutional kernel at the shallow layer can capture short-term local change features, such as instantaneous fluctuations in sensor data; while the convolutional kernel at the deep layer can mine long-term global trends and periodic rules. The feature maps obtained after convolution operation will be sent to the pooling layer for dimension reduction processing to reduce the amount of calculation and highlight the key features. Subsequently, the high-dimensional features extracted are mapped to a low-dimensional space through a fully connected layer, and then the control instruction probability distribution of each device is generated. This probability distribution is not a simple discrete value, but a comprehensive evaluation result considering various factors in the current scene, providing a rich information foundation for subsequent device decision-making.

[0066] For the energy consumption optimization deep learning model, the long short-term memory network-attention mechanism it adopts has unique advantages. The LSTM unit can effectively handle the long-term dependencies in the normalized environment parameters and device energy consumption history sequence. It selectively remembers or forgets information through the gating structure (input gate, forget gate and output gate), thereby solving the gradient vanishing and explosion problems faced by traditional recurrent neural networks (RNN) when processing long sequences. At the same time, the introduction of the attention mechanism enables the model to assign different weights at different time steps, focusing on key time periods that have a greater impact on the prediction results. For example, when the environment parameters change dramatically or the device energy consumption appears abnormal peaks, the corresponding time steps will be assigned higher attention weights. Based on such design, the model can more accurately learn the complex nonlinear relationship between environmental factors and device energy consumption, and output the optimal set of device running parameters for the next period. Each parameter in the set is finely adjusted to minimize the device's energy consumption while meeting system performance requirements.

[0067] Further, the preset scene rules at least include: trigger condition rules for defining environment threshold combinations, device coordination rules for defining device action logic relationships, and instruction execution timing constraint rules.

[0068] Among them, the trigger condition rules for defining environment threshold combinations can be, for example, illumination <100 lux and temperature >28 ; or humidity >70% and concentration >1500ppm; or PM2.5 index >150 and wind speed <0.5 m / s, etc. These composite threshold settings can accurately capture the abnormal state or special demand scenario in a specific environment, thereby starting the corresponding linkage mechanism. For example, when it is detected that the indoor light is insufficient and the temperature is too high, the system can automatically dim the lighting devices in unnecessary areas to reduce the heat load, while enhancing the cooling power of the air conditioning system; if the humidity and carbon dioxide concentration are both excessive, it indicates that the indoor air circulation is poor and relatively hot and humid, at which time the fresh air system should be opened for ventilation and dehumidification, and the operation mode of the heating and ventilation equipment should be adjusted appropriately.

[0069] In terms of device coordination rules of device action logic relationships, the rules specify the interaction mode and priority order between different devices in detail. For example, when the fire alarm is triggered, according to the preset safety protocol, all elevators immediately stop running and return to the first floor, the smoke exhaust fan runs at full power, the emergency lighting system is fully bright, and the fireproof rolling shutter door sequentially descends to form a fire compartment. For example, in an intelligent office scenario, half an hour before the meeting starts, the projector automatically preheats, the curtain slowly descends, the sound system tests whether the sound source is normal, at the same time the air conditioner adjusts to the appropriate temperature and humidity interval suitable for human comfort, the lights gradually dim to create a focused atmosphere - this series of actions are realized based on the precise coordination and scheduling between devices.

[0070] As for the instruction execution timing constraint rule, it is a time management framework established to ensure that each control instruction can be executed in order and efficiently. It specifies the sequence and minimum interval time requirements between different types of instructions, for example, when the night mode is started, first turn off the unnecessary public area lighting, after 5 minutes, gradually dim the lights in the corridor and auxiliary function area, and finally switch to the soft night light mode of the private space such as the bedroom.

[0071] In one embodiment, the energy consumption optimization deep learning model generates a set of time-sharing operation instructions for interruptible devices when executing the start-stop strategy; generates parameter optimization instructions for continuous operation devices; and when outputting the device operation parameter optimization instructions, the constraint conditions need to be met: environmental comfort constraint, the adjusted environmental parameters are maintained within the user's preset parameter interval; the expression is: ; wherein T, H, and L represent temperature, humidity, and light intensity, respectively, and the subscript min / max is the preset threshold; device safety operation constraint, the power distribution coefficient αᵢ satisfies the device technical specifications and the power change rate limit, and the power distribution coefficient αᵢ is the ratio of the actual power to the rated power, and αᵢ ∈ [0, 1]; grid interaction constraint, the total electricity bill calculated based on the real-time electricity price should not exceed the user's preset daily budget upper limit; the expression is: ; wherein, t is the electricity price at time t, Daily energy consumption budget; Device coordination constraint: the total power of multiple devices does not exceed the rated capacity of the total incoming line circuit breaker of the home; the expression is: ; wherein the D is the current set of linked devices, is the maximum carrying power of the home circuit.

[0072] In this embodiment, the model is built-in fuzzy PID controller, real-time monitoring of the living room, bedroom and other areas of T / H / L indicators. When detecting that the temperature of a certain area exceeds the user-set temperature, humidity, and light intensity threshold, the fresh air system is preferentially started for pre-adjustment, and the air conditioner is linked to gradually approach the target value in a stepwise temperature difference compensation strategy. For example, if the current room temperature reaches 30°C, the maximum air volume is first opened to reduce the temperature to 29°C, and then the air conditioning cooling mode is used to avoid sudden changes that cause discomfort.

[0073] At the same time, for the variable load devices of the home, αᵢ is allowed to break through 1.0 (overclocking operation) for a short time during the valley electricity period, but the cumulative duration does not exceed 5% of the total daily operation duration; for high-power devices that are turned on for a long time, such as refrigerators, αᵢ is limited to ≤0.8 to prevent the compressor from overheating, and when the power grid frequency fluctuates more than ±0.5 Hz, the model will automatically reduce the αᵢ value of all unnecessary devices, and the lighting basic load is prioritized.

[0074] In summary, through the hierarchical and progressive control strategy, the single-device efficiency maximization is realized to the global optimization of the home micro-grid level, which reduces the overall energy consumption cost on the premise of ensuring user comfort.

[0075] Further, the S3 further comprises the following steps: Verify the feasibility of the instructions through the fuzzy control algorithm, and if there is a resource conflict or a target conflict among multiple instructions, then arbitrate according to the preset priority rules.

[0076] The priority rules are as follows: (a) Safety constraint priority: instructions that violate device safety operation constraints or power grid interaction constraints are automatically downgraded; (b) User explicit demand priority: the priority of the device control instructions initiated by the user is higher than that of the optimization instructions generated by the system; (c) Core function priority: the priority of the basic function mode instructions is higher than that of the energy efficiency management mode instructions; (d) Economic optimization classification: in the multi-device interactive linkage mode, the device start-stop sequence is allocated according to the unit energy consumption cost saving rate, which satisfies: ; and the constraint condition is: , wherein is the set of interruptible devices; actual energy saving value of device i; energy variation of device i; delay start time of device i; real-time electricity price during the delay period; expected energy saving value of device i; maximum delay time allowed for device i; For example, safety constraint priority: the system calculates the real-time power and rated power of each device in real time (i.e. αᵢ), when αᵢ>1 or the power change rate exceeds the safety slope specified in the device manual, the three-level protection response is triggered immediately: 1. Yellow warning prompts the user to confirm the operation rationality; 2. Automatically insert a 5-second buffer delay to prevent accidental touch; 3. Force αᵢ to be limited in the interval [0,1] and smoothly reduce the load to a safe level.

[0077] User explicit demand priority: when the user directly issues a command contrary to the optimization suggestion (such as forcibly turning off the air conditioner that is executing the energy saving cycle), the system will freeze all current automation processes and enter manual mode, and record event logs for subsequent analysis. At this time, only the basic safety protection function is retained to continue running, ensuring the highest authority of manual intervention.

[0078] Core function priority: limit the mode switching path to only one-way flow (basic mode→energy efficiency mode→linkage mode). In any case, if a fault reset occurs, it will default to the basic function mode to ensure basic availability. And preset an important load whitelist that cannot be interrupted, ensuring that the control authority of the whitelist devices is always higher than other optimization targets.

[0079] The above only describes some or preferred embodiments of the present application, neither the text nor the drawings can limit the scope of protection of the present application, any equivalent structural transformation using the contents of the present application specification and drawings, or direct / indirect application in other related technical fields are included in the scope of protection of the present application.

Claims

1. A method for intelligent control of smart home devices based on deep learning, characterized in that, Includes the following steps: S1. Real-time collection of environmental data and operating status information of various home devices using various sensors deployed in the smart home environment, and transmission of all collected data to the central control system; the environmental data includes temperature, humidity, light intensity, and human activity; the operating status information of the home devices includes the device's on / off status, working mode, and energy consumption parameters; S2. Based on the environmental data and home device operating status information obtained in step S1, analyze the current operating mode of the smart home device. The operating mode of the smart home device includes at least: basic function mode, energy efficiency management mode, and multi-device interaction and linkage mode. The basic function mode is used to correspond to the independent operating state of a single smart home device and to control it according to preset rules or direct user commands; the energy efficiency management mode is used to dynamically optimize device operating parameters or start-stop strategies to minimize overall energy consumption, based on the correlation analysis of environmental data and device energy consumption parameters, while meeting basic home function requirements; the multi-device interaction and linkage mode is used to coordinate multiple related smart home devices to work together based on changes in environmental data, the state trigger of specific devices, or preset scene rules. S3. Based on the current operating mode determined in step S2, generate optimized control commands for one or more smart home devices and select different energy consumption drive modes.

2. The intelligent control method for smart home devices based on deep learning according to claim 1, characterized in that, Step S2 includes the following steps: Based on the environmental data and equipment operating status information collected in real time in step S1, the comprehensive change in the environment and equipment status at adjacent time points is calculated; combined with the sampling frequency or time interval of the sensors, the current comprehensive change rate of the environment and equipment is calculated; and the operating mode of the smart home device is determined by the obtained current comprehensive change rate of the environment and equipment.

3. The intelligent control method for smart home devices based on deep learning according to claim 2, characterized in that, The process for determining the operating mode of the smart home device is as follows: When the combined rate of change of the environment and equipment is higher than a preset first change threshold, and the intensity of the detected change in equipment status is higher than a preset second change threshold, it is determined to be a multi-device interactive linkage mode; when the combined rate of change of the environment and equipment is lower than a preset third change threshold, or the intensity of the detected change in equipment status is lower than a preset fourth change threshold, and the status continues for more than a preset duration threshold, it is determined to be an energy efficiency management mode; when the conditions for determining the multi-device interactive linkage mode or the energy efficiency management mode are not triggered, it is determined to be a basic function mode. Wherein, the first change threshold is the maximum value of the current comprehensive change rate of the environment and equipment, used to determine whether to enter the multi-device interactive linkage mode; the second change threshold is the minimum value of the equipment change intensity, used in combination with the first change threshold to determine whether to enter the multi-device interactive linkage mode; the third change threshold is the minimum value of the current comprehensive change rate of the environment and equipment, used to determine whether to trigger the energy efficiency management mode; the fourth change threshold is the minimum value of the equipment change intensity, used in combination with the third change threshold to determine whether to enter the energy efficiency management mode.

4. The intelligent control method for smart home devices based on deep learning according to claim 3, characterized in that, S2 further includes the following steps: The environmental data and equipment operating status information collected in real time in step S1 are timestamped and then normalized to unify the units.

5. The intelligent control method for smart home devices based on deep learning according to claim 3, characterized in that, S3 includes the following steps: Based on the current operating mode determined in step S2, select the corresponding energy consumption drive control strategy; In the case of a multi-device interactive linkage mode, the scene collaboration deep learning model is invoked to fuse real-time environmental data, device status matrix and preset scene rules to construct a spatiotemporal feature tensor. The scene collaboration deep learning model outputs the probability distribution of control instructions of each device within a preset time window. Instructions with confidence levels higher than a set threshold are selected from the probability distribution and a sequence of device control instructions is generated according to the time pane. If energy efficiency management mode is in use, the energy consumption optimization deep learning model is invoked. Historical energy consumption data and environmental parameters are input into the time-series encoder to extract cross-cycle energy consumption features. The association weight between current environmental parameters and historical features is calculated through an attention mechanism. Based on the weighted feature vector, the energy consumption trend of the equipment in the short term is predicted. Combined with the real-time operating load of the equipment, the power allocation coefficient of each device is dynamically adjusted to generate equipment operating parameter adjustment instructions or start-stop strategies. If it is in basic function mode, the control command corresponding to the current environment data and device status will be matched from the preset rule base.

6. The intelligent control method for smart home devices based on deep learning according to claim 5, characterized in that, The scenario-coordinated deep learning model adopts a temporal convolutional network structure. The input is aligned multi-sensor temporal data and device state matrix, and the output is the probability distribution of control commands for each device. The energy consumption optimization deep learning model adopts a long short-term memory network-attention mechanism. The input is normalized environmental parameters and device energy consumption history sequence, and the output is the set of optimal operating parameters for the device in the next cycle.

7. The intelligent control method for smart home devices based on deep learning according to claim 1, characterized in that, When the energy consumption optimization deep learning model executes the start-stop strategy, it must satisfy the following constraints when generating time-sharing operation instruction sets for interruptible devices, generating parameter optimization instructions for continuously running devices, and outputting device operation parameter optimization instructions: Environmental comfort constraints ensure that the adjusted environmental parameters remain within the user-preset parameter range. The equipment safety operation constraints are that the power allocation coefficient αᵢ meets the equipment technical specifications and power change rate limits, and the power allocation coefficient αᵢ is the ratio of its actual power to its rated power, and αᵢ∈[0,1]. Grid interaction constraints stipulate that the total electricity expenditure calculated based on real-time electricity prices must not exceed the user's preset daily budget limit; Equipment coordination constraint: The total power of multiple devices operating at the same time shall not exceed the rated capacity of the main incoming circuit breaker of the household.

8. The intelligent control method for smart home devices based on deep learning according to claim 1, characterized in that, The preset scenario rules include at least: trigger condition rules for defining combinations of environmental thresholds, device coordination rules for defining logical relationships of device actions, and rules for defining instruction execution timing constraints.

9. The intelligent control method for smart home devices based on deep learning according to claim 1, characterized in that, S3 further includes the following steps: The feasibility of instructions is verified by fuzzy control algorithm. If multiple instructions have resource conflicts or target conflicts, arbitration is carried out according to preset priority rules.

10. The intelligent control method for smart home devices based on deep learning according to claim 6, characterized in that, The priority rule is specifically as follows: (a) Safety constraints take precedence: Instructions that violate equipment safety operation constraints or power grid interaction constraints are automatically downgraded; (b) Prioritize explicit user needs: User-initiated device control commands take precedence over system-generated optimization commands. (c) Core function protection takes priority: Basic function mode instructions have higher priority than energy efficiency management mode instructions; (d) Economic Optimization Classification: In the multi-device interactive linkage mode, the equipment start-up and shutdown sequence is allocated according to the unit energy consumption cost saving rate, satisfying: And the constraints are: ,in, A set of interruptible devices; This represents the actual energy-saving value of device i; This represents the change in energy consumption of device i; Let i be the delayed startup time of device i; The real-time electricity price during the delay period; The expected energy saving value of device i; This represents the maximum allowed delay time for device i.

Images