An intelligent home control system based on artificial intelligence

Abstract

The present invention relates to the field of smart home technology and provides a smart home control system based on artificial intelligence. The energy consumption monitoring module is used to continuously monitor the energy consumption of the smart home, monitor the smart home usage data in real time, and obtain the user's historical smart home usage data. The standby analysis module determines the real standby state of the smart home according to the power consumption information of the smart home and calculates the standby time. The prediction module predicts the usage time and energy consumption of the smart home according to the historical monitoring data of the energy consumption monitoring module. The home control module performs relevant operations on the smart home according to the calculation results of the standby analysis module and the prediction results of the prediction module. The present invention accurately identifies the devices that are actually in the standby state through a clustering algorithm, which is beneficial to reducing energy waste and equipment loss. The time smoothing processing unit processes time series data, which solves the influence of complex environment on the chaotic and unpredictable collection time data.

Description

Technical Field

[0001] The present invention relates to the field of smart home technology, and in particular to a smart home control system based on artificial intelligence. Background Art

[0002] With the continuous development of society and the economy, people's pace of life is accelerating, and their demands for a higher quality of life are also increasing. Traditional home control methods require manual operation of various electrical devices, which is a cumbersome process. After a busy day, people need to turn on lights, air conditioners, curtains, and other devices separately. This operation method is gradually failing to meet people's demand for convenience. In this context, smart home control systems have emerged, allowing users to quickly and conveniently control home devices through simple commands or automated settings.

[0003] The rapid advancement of artificial intelligence (AI) technology provides solid technical support for smart home control systems. AI algorithms such as machine learning and deep learning enable systems to understand complex information such as user behavior patterns, voice commands, and environmental changes. By training deep learning algorithms on large amounts of voice data, the system can accurately recognize user voice commands and implement voice control.

[0004] The existing background technology still has the following deficiencies:

[0005] 1. Ignoring the energy consumption of devices in standby mode. Smart home systems include numerous smart devices, which still consume a certain amount of electricity when in standby mode. However, existing smart home control systems cannot accurately adapt to the standby energy consumption characteristics of all appliances when determining whether a device is in standby mode for a long period of time, and are prone to misjudgment. For example, some appliances may experience brief fluctuations in energy consumption when in standby mode. The control system may misinterpret this normal fluctuation as indicating that the device has entered the working state, and thus fail to automatically cut off the power supply, resulting in poor energy control. Or, conversely, some devices in normal working mode but with low energy consumption may be misinterpreted as being in standby mode and the power supply may be cut off, affecting the normal use of the devices. Over time, the accumulation of this standby energy consumption will also lead to energy waste and equipment damage.

[0006] 2. Existing smart home control systems often shut down devices in standby mode. This, in turn, requires users to restart devices when needed, resulting in delayed responses to user commands and impacting the user's long-term experience. Setting a timer or predicting user behavior can address some of these issues, but home environments are often complex and diverse, making it difficult for the system to accurately identify all scenario information, thus impacting the accuracy of predictive analysis. For example, on a winter morning, a user may wish to open the curtains later, but the system may only judge the user's waking habits based on previous times, ignoring seasonal factors and still recommending opening the curtains at the usual time. Summary of the Invention

[0007] In order to solve the above technical problems, the present invention provides an artificial intelligence-based smart home control system, which uses the clustering algorithm of the standby energy consumption monitoring unit to accurately judge the standby state of smart home devices, removes the influence of complex environments through data smoothing processing by the prediction module, and cooperates with the home usage prediction unit to accurately predict the time points when users use devices, so as to solve the problems in the existing technology.

[0008] An artificial intelligence-based smart home control system, comprising:

[0009] An energy consumption monitoring module, which is used to continuously monitor the energy consumption of the smart home, monitor the smart home usage data in real time, and obtain the user's historical smart home usage data;

[0010] A standby analysis module, which is data-connected to the energy consumption monitoring module and is used to determine the actual standby state of the smart home based on the power consumption information of the smart home and calculate the standby time;

[0011] A prediction module, which is data-connected to the energy consumption monitoring module and is used to predict the usage time and energy consumption of the smart home based on historical monitoring data of the energy consumption monitoring module;

[0012] The home control module is data-connected to the standby analysis module and the prediction module, and is used to perform relevant operations on the smart home according to the calculation results of the standby analysis module and the prediction results of the prediction module.

[0013] Preferably, the historical smart home usage data monitored or acquired by the energy consumption monitoring module includes smart home energy consumption information, smart home usage time point information and smart home type information.

[0014] Preferably, the standby analysis module includes a judgment unit, a standby energy consumption monitoring unit and a standby time monitoring unit;

[0015] The judgment unit classifies the smart home devices into continuous operation type and non-continuous operation type according to the historical operation time data of the smart home monitored by the energy consumption monitoring module, by setting a threshold value for the operation time. Emergency type smart home devices are uniformly classified as continuous operation type.

[0016] The standby energy consumption monitoring unit is used to analyze the real-time power consumption information of the non-continuous operation type smart home in the judgment unit to determine whether the current smart home is in the standby state;

[0017] The standby time monitoring unit determines that the home appliance is in a standby state based on the standby energy consumption monitoring unit, counts the standby time of the home appliance, and transmits the standby time to the home control module.

[0018] Preferably, the standby energy consumption monitoring unit uses a clustering algorithm to eliminate judgment errors caused by normal energy consumption fluctuations of household appliances in the standby state;

[0019] Collect the energy consumption data samples of the equipment {E1, E2, E d ,...,};

[0020] The number of clusters is determined by the elbow rule so that each data point E d To the cluster center F j The Euclidean distance is the smallest. Since energy consumption data is one-dimensional data, the Euclidean distance becomes:

[0021] (E d ,F j )=|E d -F j |

[0022] Where d is the index of the energy consumption data sample, j is the sample index of the classified cluster center;

[0023] When determining whether a device is in standby mode, the cluster to which the energy consumption data belongs is observed. If the energy consumption data of the device always belongs to a specific cluster within a specified time and the energy consumption sample data at the center of the cluster is the smallest, the device is determined to be in standby mode.

[0024] Preferably, the prediction module includes a time smoothing processing unit, a prediction model building unit and a home use prediction unit;

[0025] The time smoothing processing unit is used to process the trend and periodic characteristics in the time series data, discover the periodic laws and overall change trends, and improve the prediction accuracy of the home use prediction unit;

[0026] The prediction model establishment unit establishes a decision tree prediction model based on the historical smart home usage data processed by the time smoothing processing unit;

[0027] The home use prediction unit is used to predict the time point of the user's future use of the smart home based on the user's smart home usage data.

[0028] Preferably, the time smoothing processing unit first obtains historical data of a certain smart home device and a time point at which the user uses the device according to the standby analysis module;

[0029] Assume that for this smart home device, the order of user's actions in turning it on is t, the first time it is turned on is t=1, and the second time it is turned on is t=2;

[0030] Assume that a series of time points at which users start using the device are recorded in the order of behavior t, and these time points are converted into a unified numerical measurement. The time points of each day are converted into minutes to form a minute value {y1, y2, y t ,...};

[0031] Initialization parameters: Select smoothing parameters α, β, and γ, which range from 0 to 1;

[0032] At the same time, the level L is initialized at t=1 t 、Trend H t The value of the cycle length c is as follows: for the value of the initialization level L1, the data y1 actually observed at the corresponding initial time is taken; for T1, it is initialized to 0; for the cycle length c, it is set to include but not limited to the number of times the smart home device is turned on within three days, a week, or half a month;

[0033] Level equation:

[0034] L t =α(y t -K t-c )+(1-α)(L t-1 +H t-1 )

[0035] L t It represents the basic level estimate after removing the periodic factors after t times of smoothing. When predicting the time point of user device usage, it is a basic time point reference value that does not include periodic fluctuations. The smoothing parameter α is used to control the trade-off between current data and historical trends. t-c Indicates the previous complete cycle, the cycle length is c corresponding to the cycle factor, y t is the actual recorded time point when the user uses the device for the tth time;

[0036] Trend equation:

[0037] H t =β(L t -L t-1)+(1-β)H t-1

[0038] H t The trend value at t indicates the increasing or decreasing trend of the behavior sequence. For the time point of user device usage, it reflects the trend of whether the time point is gradually advanced or delayed. β is used to control the trade-off between current trend changes and historical trends.

[0039] Periodic equations:

[0040] K t =γ(y t -L t )+(1-γ)P t-m

[0041] Among them, K t represents the estimated value of the periodic factor at time t, which is used to capture the periodic patterns in the data, including users having similar device usage time points on the same day of the week or the same time of the day. γ is used to control the trade-off between current periodic changes and historical periodic patterns;

[0042] According to the above calculation, as t increases, the final smoothed minute value {y1,y2,y t ,...}with {L1,L2,L t ,...} to represent the original time series. These smoothed level values remove the interference of periodic fluctuations and some noises and serve as the basic data for the subsequent prediction model establishment unit.

[0043] Preferably, the prediction model building unit first smooths {L1, L2, L t ,...}Build a decision tree;

[0044] For {L1,L2,L t ,...}, that is, the time point when the user starts to use the device, first calculate its information entropy S(L t ), information entropy formula:

[0045]

[0046] Where n represents L t The total number of values in category i. If a day is divided into intervals of 10 minutes, the maximum value of n is 143. It's L t The probability of belonging to the i-th category is obtained by counting the frequency of time points falling into each interval in the past data. This information entropy value reflects the degree of uncertainty of the target variable before feature segmentation;

[0047] Extract L tThe derived features are used as input features, including the value of t and the period K in which t is located. t and trend value H t ; Take these three data as L t The calculation formula for the feature X in is:

[0048]

[0049] Where m is the number of values of feature X, and r is the index of feature X;

[0050]

[0051] Then calculate the information gain IG(L t ,X):

[0052] IG(L t ,X)=S(L t )-S(L t │X)

[0053] Select the feature with the largest information gain as the best feature for the current partition, partition at the root node, divide the data set into different subsets, repeat this process, calculate all historical data to build a decision tree, and the final subset is the leaf node for storing L t The corresponding time point y t .

[0054] Preferably, the home use prediction unit uses the constructed decision tree to make predictions;

[0055] Calculate the input feature vector X for a specific future time point, input the feature vector into the already constructed decision tree, start from the root node, select the corresponding branch according to the different subsets stored in the root node, and so on, until reaching a leaf node;

[0056] The leaf nodes are stored at the time points, expressed in minutes, which are the predicted time points when the user uses the device at that specific time point.

[0057] Preferably, the home control module includes an energy consumption calculation and confirmation unit and an execution unit;

[0058] The energy consumption calculation and confirmation unit calculates the energy consumption of the non-continuously running smart home based on the standby time calculated by the standby analysis module, and provides an operation interface for the user to set the standby energy consumption threshold. When the threshold is exceeded, the smart home in the non-standby state is automatically shut down;

[0059] The execution unit responds in advance according to the prediction result of the time point of the smart home that the user will use in the future by the prediction module, and turns on the non-continuous operation smart home that is turned off by the energy consumption calculation and confirmation unit.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] 1. The present invention adopts a clustering algorithm through the energy consumption monitoring module and the standby energy consumption monitoring unit. By collecting device energy consumption data samples, it determines whether the device is in standby state by judging the cluster center and cluster type, effectively avoiding the judgment error caused by the normal energy consumption fluctuation of household devices in standby state, so as to accurately identify the devices that are actually in standby state, which is beneficial to reducing energy waste and equipment loss.

[0062] 2. The present invention solves the problem of chaotic and unpredictable effects on collected time data in complex environments by processing time series data through a time smoothing processing unit, extracts relevant features and establishes a decision tree prediction model, which is used to predict specific time points in the future. The execution unit responds in advance according to the prediction results, and promptly turns on non-continuously running smart homes that were previously turned off due to energy consumption control when the user needs to use the device, avoiding the inconvenience caused by restarting the device due to premature shutdown, better meeting the user's usage needs in different scenarios, and improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a flow chart of the system module structure of the present invention;

[0064] Figure 2 This is a flowchart of the standby analysis module of the present invention;

[0065] Figure 3 This is a flow chart for establishing a prediction module model of the present invention. DETAILED DESCRIPTION

[0066] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0067] The present invention provides an artificial intelligence-based smart home control system, comprising:

[0068] An energy consumption monitoring module, which is used to continuously monitor the energy consumption of the smart home, monitor the smart home usage data in real time, and obtain the user's historical smart home usage data;

[0069] A standby analysis module, which is data-connected to the energy consumption monitoring module and is used to determine the actual standby state of the smart home based on the power consumption information of the smart home and calculate the standby time;

[0070] A prediction module, which is data-connected to the energy consumption monitoring module and is used to predict the usage time and energy consumption of the smart home based on historical monitoring data of the energy consumption monitoring module;

[0071] The home control module is data-connected to the standby analysis module and the prediction module, and is used to perform relevant operations on the smart home according to the calculation results of the standby analysis module and the prediction results of the prediction module.

[0072] Example 1:

[0073] like Figure 1-3 As shown, in this embodiment, in an ordinary family, a smart home system is installed and operated, which includes a variety of smart devices such as a smart TV, a smart speaker, a smart microwave oven, a smart desk lamp and a smart smoke alarm.

[0074] The energy consumption monitoring module continuously collects energy consumption information, usage time information, and device type information for each smart device. For example, if a smart TV is turned on at 7 pm for viewing, the energy consumption monitoring module will record the TV's energy consumption during this period.

[0075] The standby analysis module includes a judgment unit, a standby energy consumption monitoring unit and a standby time monitoring unit;

[0076] The judgment unit classifies smart home devices into continuous operation and non-continuous operation categories based on the historical operation time data of the smart home monitored by the energy consumption monitoring module. The categories include continuous operation and non-continuous operation. Emergency-type smart home devices are uniformly classified into the continuous operation category; for example, smoke alarms are classified into the continuous operation category.

[0077] The standby energy consumption monitoring unit performs real-time power consumption analysis on non-continuously running devices such as smart TVs. It collects energy consumption data samples from smart TVs when they are in standby mode and uses a clustering algorithm and the elbow rule to determine the number of clusters.

[0078] Collect the energy consumption data samples of the equipment {E1, E2, E d ,...,};

[0079] The number of clusters is determined by the elbow rule so that each data point E d To the cluster center F j The Euclidean distance is the smallest. Since energy consumption data is one-dimensional data, the Euclidean distance becomes:

[0080] (E d ,F j )=|E d -F j |

[0081] Where d is the index of the energy consumption data sample, j is the sample index of the classified cluster center;

[0082] When determining whether a device is in standby mode, the cluster to which the energy consumption data belongs is observed. If the energy consumption data of the device always belongs to a specific cluster within a specified time and the energy consumption sample data at the center of the cluster is the smallest, the device is determined to be in standby mode.

[0083] For example, during ten minutes of continuous monitoring, the smart TV's energy consumption fluctuated slightly while in standby mode, remaining within the lowest cluster. The algorithm determined these fluctuations were within the normal standby energy consumption range. When a smart TV remains in standby mode for an extended period of time, with energy consumption remaining within the standby energy consumption cluster, the standby time monitoring unit measures the time and transmits it to the home control module.

[0084] The home control module's energy consumption calculation and confirmation unit calculates the smart TV's energy consumption based on its standby time and sets a reasonable standby energy consumption threshold through the user interface. If the smart TV's standby time exceeds the threshold, the energy consumption calculation and confirmation unit automatically shuts it down, avoiding unnecessary energy consumption while the smart TV is in standby mode and achieving energy conservation.

[0085] Example 2:

[0086] like Figure 1-3 As shown, in this embodiment, the smart TV in the home in Example 1 is taken as an example.

[0087] The prediction module includes a time smoothing processing unit, a prediction model building unit and a home use prediction unit;

[0088] The time smoothing unit obtains historical usage data for smart TVs, including the times when users turn on the TV to watch different programs. For example, a user might watch the news broadcast at 7:30 PM during the week and variety shows at 8 PM on weekends. These daily time points are converted into minutes, which are 450 and 480 minutes, respectively.

[0089] At the same time, the smoothing parameters and trend parameters are initialized, and a one-week cycle is set based on the user's viewing habits. For this smart home device, the order in which the user turns it on is recorded as t. On the first day of the cycle, the first 450 minutes begins, which is t = 1;

[0090] The data were processed using level equations, trend equations, and periodic equations;

[0091] Assume that a series of time points at which users start using the device are recorded in the order of behavior t, and these time points are converted into a unified numerical measurement. The time points of each day are converted into minutes to form a minute value {y1, y2, y t,...};

[0092] Initialization parameters: Select smoothing parameters α, β, and γ, which range from 0 to 1;

[0093] At the same time, the level L is initialized at t=1 t 、Trend H t The value of the cycle length c is as follows: for the value of the initialization level L1, the data y1 actually observed at the corresponding initial time is taken; for T1, it is initialized to 0; for the cycle length c, it is set to include but not limited to the number of times the smart home device is turned on within three days, a week, or half a month;

[0094] Level equation:

[0095] L t =α(y t -K t-c )+(1-α)(L t-1 +H t-1 )

[0096] L t It represents the basic level estimate after removing the periodic factors after t times of smoothing. When predicting the time point of user device usage, it is a basic time point reference value that does not include periodic fluctuations. The smoothing parameter α is used to control the trade-off between current data and historical trends. t-c Indicates the previous complete cycle, the cycle length is c corresponding to the cycle factor, y t is the actual recorded time point when the user uses the device for the tth time;

[0097] Trend equation:

[0098] H t =β(L t -L t-1 )+(1-β)H t-1

[0099] H t The trend value at t indicates the increasing or decreasing trend of the behavior sequence. For the time point of user device usage, it reflects the trend of whether the time point is gradually advanced or delayed. β is used to control the trade-off between current trend changes and historical trends.

[0100] Periodic equations:

[0101] K t =γ(y t -L t )+(1-γ)P t-m

[0102] Among them, K trepresents the estimated value of the periodic factor at time t, which is used to capture the periodic patterns in the data, including users having similar device usage time points on the same day of the week or the same time of the day. γ is used to control the trade-off between current periodic changes and historical periodic patterns;

[0103] According to the above calculation, as t increases, the final smoothed minute value {y1,y2,y t ,...}with {L1,L2,L t ,...} to represent the original time series. These smoothed level values remove the interference of periodic fluctuations and some noises and serve as the basic data for the subsequent prediction model establishment unit.

[0104] In this way, the periodic fluctuations and some noise interference in the original time series can be removed.

[0105] This approach is particularly useful for complex time series data on user behavior, by simultaneously processing trends, periodicity, and irregular fluctuations. For example, a user might use their device later in the evening on weekdays and earlier in the afternoon on weekends. This approach effectively captures both this periodicity (on a weekly basis) and any potential trends in the timing of use shifting later or earlier.

[0106] The prediction model building unit builds a decision tree based on the processed historical data. It calculates the information entropy of the TV on time point data, extracts derivative features such as the on sequence, cycle and trend value as input features, calculates the conditional entropy and information gain, selects the best features for division, and gradually builds a complete decision tree. The prediction model building unit first calculates the information entropy of the TV on time point data based on the {L1, L2, L t ,...}Build a decision tree;

[0107] For {L1,L2,L t ,...}, that is, the time point when the user starts to use the device, first calculate its information entropy S(L t ), information entropy formula:

[0108]

[0109] Where n represents L t The total number of values in category i. If a day is divided into intervals of 10 minutes, the maximum value of n is 143. It's L t The probability of belonging to the i-th category is obtained by counting the frequency of time points falling into each interval in the past data. This information entropy value reflects the degree of uncertainty of the target variable before feature segmentation;

[0110] Extract L tThe derived features are used as input features, including the value of t and the period K in which t is located. t and trend value H t ; Take these three data as L t The calculation formula for the feature X in is:

[0111]

[0112] Where m is the number of values of feature X, and r is the index of feature X;

[0113]

[0114] Then calculate the information gain IG(L t ,X):

[0115] IG(L t ,X)=S(L t )-S(L t │X)

[0116] Select the feature with the largest information gain as the best feature for the current partition, partition at the root node, divide the data set into different subsets, repeat this process, calculate all historical data to build a decision tree, and the final subset is the leaf node for storing L t The corresponding time point y t .

[0117] One time, a popular variety show, usually broadcast at 8 p.m., was moved to 9 p.m. due to special arrangements. Based on this new situation, the home usage prediction unit combined the current time, cycle, and other information to calculate the input feature vector for this new time point and input it into the constructed decision tree.

[0118] Calculate the input feature vector X for a specific future time point, input the feature vector into the already constructed decision tree, start from the root node, select the corresponding branch according to the different subsets stored in the root node, and so on, until reaching a leaf node;

[0119] The leaf nodes are stored at the time points, expressed in minutes, which are the predicted time points when the user uses the device at that specific time point.

[0120] Since the system has learned the user's habit of watching the variety show and related surrounding information, the prediction results show that the user is likely to turn on the TV around 9 o'clock in the evening.

[0121] Based on this prediction result, the execution unit of the home control module turns on the power of the smart TV that was previously turned off by the energy consumption calculation confirmation unit for energy saving at around 9 o'clock, avoiding the inconvenience caused by restarting the device due to premature shutdown, better meeting the user's usage needs in different scenarios, and improving the user experience. At the same time, it avoids the energy consumption of the smart TV due to long-term standby, improves the user experience, and the TV's startup time can better adapt to complex situations such as changes in program time.

[0122] The embodiments of the present invention are provided for the purpose of illustration and description. Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations of the present invention. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. An artificial intelligence-based smart home control system, characterized in that: include: An energy consumption monitoring module, which is used to continuously monitor the energy consumption of the smart home, monitor the smart home usage data in real time, and obtain the user's historical smart home usage data; A standby analysis module, which is data-connected to the energy consumption monitoring module and is used to determine the actual standby state of the smart home based on the power consumption information of the smart home and calculate the standby time; A prediction module, which is data-connected to the energy consumption monitoring module and is used to predict the usage time and energy consumption of the smart home based on historical monitoring data of the energy consumption monitoring module; A home control module, which is data-connected to the standby analysis module and the prediction module, and is used to perform relevant operations on the smart home based on the calculation results of the standby analysis module and the prediction results of the prediction module; The prediction module includes a time smoothing processing unit, a prediction model building unit and a home use prediction unit; The time smoothing processing unit is used to process the trend and periodic characteristics in the time series data, discover the periodic laws and overall change trends, and improve the prediction accuracy of the home use prediction unit; The prediction model establishment unit establishes a decision tree prediction model based on the historical smart home usage data processed by the time smoothing processing unit; The home use prediction unit is used to predict the time point of the user's future use of the smart home based on the user's smart home usage data; The time smoothing processing unit first obtains the historical data of a certain smart home device and the time point at which the user uses the device according to the standby analysis module; Assume that for this smart home device, the order of user's activation is t, the first activation is t=1, and the second activation is t=2; Assume a series of time points recorded in the order of behavior t, when the user starts using the device, convert these time points into a unified numerical measurement, convert the time points of each day into minutes, and form a minute value ; Initialization parameters: Select smoothing parameters 、 and , the values of these parameters range from 0 to 1; At the same time, the level is initialized at t=1 ,trend and cycle length The value of the initialization level The value of is the data actually observed at the corresponding initial time. ;for Initialized to 0; the cycle length c is set to include but not limited to the number of times the smart home device is turned on within three days, a week, or half a month; Level equation: It represents the basic level estimate after removing the periodic factors after smoothing t times. When predicting the time point of user equipment usage, it is a basic time point reference value, which does not include periodic fluctuations and smoothing parameters. Used to control the balance between current data and historical trends; Indicates the previous complete cycle, the cycle factor corresponding to the cycle length c, is the actual recorded time point when the user uses the device for the tth time; Trend equation: The trend value at t indicates the increasing or decreasing trend of the behavior sequence. For the time point of user device usage, it reflects the trend of whether the time point is gradually advanced or delayed. Used to control the balance between current trend changes and historical trends; Periodic equations: in, Represents the estimated value of the periodic factor at t, which is used to capture the periodic patterns in the data, including users having similar device usage time points on the same day of the week or the same time of the day. Used to control the trade-off between current cyclical changes and historical cyclical patterns; According to the above calculation, as t continues to increase, the final smoothed minute value use To represent the original time series, these smoothed level values remove the interference of periodic fluctuations and some noise, and serve as the basic data for the subsequent prediction model establishment unit; The prediction model establishment unit first smooths the historical data Build a decision tree; for , that is, the time when the user starts to use the device, first calculate its information entropy , information entropy formula: Where n represents The total number of values in category i. If a day is divided into intervals of 10 minutes, the maximum value of n is 143. yes The probability of belonging to the i-th category is obtained by counting the frequency of time points falling into each interval in the past data. This information entropy value reflects the degree of uncertainty of the target variable before feature segmentation; Extract The derived features are used as input features, including the value of t and the period in which t is located. and trend values ; These three data are used as The calculation formula for the feature X in is: Where m is the number of values of feature X, and r is the index of feature X; Then calculate the information gain : Select the feature with the largest information gain as the best feature for the current partition, partition at the root node, divide the data set into different subsets, repeat this process, calculate all historical data to build a decision tree, and the final subset is the leaf node for storage Corresponding time point ; The home use prediction unit uses the constructed decision tree to make predictions; Calculate the input feature vector X for a specific future time point, input the feature vector into the already constructed decision tree, start from the root node, select the corresponding branch according to the different subsets stored in the root node, and so on, until reaching a leaf node; The leaf nodes are stored at the time points, expressed in minutes, which are the predicted time points when the user uses the device at that specific time point.

2. The artificial intelligence-based smart home control system according to claim 1, characterized in that: The historical smart home usage data monitored or acquired by the energy consumption monitoring module includes smart home energy consumption information, smart home usage time point information and smart home type information.

3. The artificial intelligence-based smart home control system according to claim 1, characterized in that: The standby analysis module includes a judgment unit, a standby energy consumption monitoring unit and a standby time monitoring unit; The judgment unit classifies the smart home devices into continuous operation type and non-continuous operation type according to the historical operation time data of the smart home monitored by the energy consumption monitoring module, by setting a threshold value for the operation time. Emergency type smart home devices are uniformly classified as continuous operation type. The standby energy consumption monitoring unit is used to analyze the real-time power consumption information of the non-continuous operation type smart home in the judgment unit to determine whether the current smart home is in the standby state; The standby time monitoring unit determines that the home appliance is in a standby state based on the standby energy consumption monitoring unit, and counts the standby time of the home appliance, and transmits the standby time to the home control module.

4. The artificial intelligence-based smart home control system according to claim 3, characterized in that: The standby energy consumption monitoring unit uses a clustering algorithm to eliminate judgment errors caused by normal energy consumption fluctuations of household appliances in the standby state; Collect energy consumption data samples of equipment ; The number of clusters is determined by the elbow rule so that each data point To the center of its cluster The Euclidean distance is the smallest. Since energy consumption data is one-dimensional data, the Euclidean distance becomes: Where d is the index of the energy consumption data sample, j is the sample index of the classified cluster center; When determining whether a device is in standby mode, the cluster to which the energy consumption data belongs is observed. If the energy consumption data of the device always belongs to a specific cluster within a specified time and the energy consumption sample data at the center of the cluster is the smallest, the device is determined to be in standby mode.

5. The artificial intelligence-based smart home control system according to claim 1, characterized in that: The home control module includes an energy consumption calculation and confirmation unit and an execution unit; The energy consumption calculation and confirmation unit calculates the energy consumption of the non-continuously running smart home based on the standby time calculated by the standby analysis module, and provides an operation interface for the user to set the standby energy consumption threshold. When the threshold is exceeded, the smart home in the non-standby state is automatically shut down; The execution unit responds in advance according to the prediction result of the time point of the smart home that the user will use in the future by the prediction module, and turns on the non-continuous operation smart home that is turned off by the energy consumption calculation and confirmation unit.

Images