Automatic control method and system for intelligent equipment
By integrating equipment status and environmental data, and utilizing improved optimization algorithms and convolutional neural networks, an intelligent automatic control system for equipment is constructed. This addresses the shortcomings of existing systems in multidimensional data processing and personalized needs, achieving precise control and efficient energy utilization, and enhancing user experience and system adaptability.
Patent Information
- Application Number
- CN202411520573.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-29
- Publication Date
- 2024-11-26
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing intelligent device control systems lack the ability to comprehensively analyze multi-dimensional data, making it difficult to adapt to dynamic environmental changes and user needs. Control decisions are not precise enough, and they lack personalized considerations, making them prone to abnormal data processing deviations and local optima problems.
By collecting equipment status and environmental data, fusing and processing them, and then inputting them into the control decision fitting model, the optimal control parameters are solved using an improved optimization algorithm. By combining convolutional neural networks and adaptive mechanisms, an intelligent automatic control system for equipment is constructed.
It enables comprehensive perception and precise control of smart home devices, improves data quality and model generalization ability, dynamically balances comfort and energy efficiency, and provides a smart, comfortable, and energy-saving user experience.
Smart Images

Figure CN119024713A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent device control technology, and more specifically, to an intelligent device automatic control method and system. Background Art
[0002] The patent with application publication number CN116414042A discloses an automatic control method and device for smart devices, including: receiving a device control request sent by a user terminal of a user; determining the user group to which the user belongs based on the user identity; determining the device that the user has the right to use from all the devices corresponding to the user group, and performing matching control operations on the device; being able to use the user group as a control unit of the smart home to achieve intelligent control of the device based on the user group, without the need for each user to register through a client one by one to obtain device permissions or for a single bound user of the smart home to share device permissions with other users one by one, which can simplify the device control process, effectively shorten the time required to control the device, thereby improving the control efficiency of the device, and also enrich the intelligent control methods of the device, improve the user experience, and thus improve the user stickiness of the smart home.
[0003] However, most existing smart device control systems only consider single devices or environmental factors and lack the ability to conduct comprehensive analysis of multi-dimensional data, resulting in incomplete and inaccurate control decisions. For example, the air-conditioning system may only adjust according to the room temperature, while ignoring factors such as humidity and air quality, which affects the actual comfort of users. Secondly, control models usually adopt fixed structures and parameters, which are difficult to adapt to the dynamic changes of the home environment and the diversity of user needs. For example, the same temperature setting may produce completely different experiences in different seasons or for different users. In addition, existing methods are insufficient in handling abnormal data, which can easily lead to control deviations. For example, a sudden change in temperature caused by temporarily opening a window may cause the system to overreact. At the same time, traditional methods are prone to fall into local optimality when looking for the best control parameters, and it is difficult to find the global optimal solution in complex multi-device collaborative control scenarios. Finally, existing methods often lack in-depth consideration of users' personalized needs and behavior patterns, and cannot achieve truly intelligent control.
[0004] In view of this, the present invention proposes an intelligent device automatic control method and system to solve the above problems. Summary of the invention
[0005] In order to overcome the above defects of the prior art and to achieve the above objectives, the present invention provides the following technical solutions: A method for automatic control of intelligent devices, comprising: S1, collecting device status data and environment data of intelligent home devices, fusing the device status data and environment data to obtain fused data; S2, inputting the fused data into the pre-built control decision fitting model to obtain a preliminary control decision; S3. Based on the preliminary control decision, the optimal control parameters are solved using the improved optimization algorithm; the optimal control parameters are applied to the actuators of the smart home devices.
[0006] Furthermore, the device status data includes the switch status of the device, device setting parameters, real-time power consumption of the device and device operating time; the environmental data includes indoor temperature, indoor humidity, indoor air quality, indoor light intensity, room personnel activities, room noise level, time information and geographic location information.
[0007] Furthermore, the fusion processing method includes: segmenting the device status data and the environmental data according to fixed time windows, and recording the data in each time window as a corresponding local time series; for each local time series, constructing a local box plot, calculating statistics for the local time series, and the statistics include the median, the upper quartile, the lower quartile and the interquartile range; for each local time series, calculating its kurtosis and skewness ; Kurtosis ;in, is the data point of the local time series, is the median of the local time series, is the number of data points in the local time series; is the interquartile range of the local time series, is the kurtosis correction coefficient; skewness ;in, is the skewness correction coefficient, is the deviation correction factor; deviation correction factor ; Define adaptive standard multiples based on kurtosis and skewness ;in, is the benchmark multiple; and is the adjustment factor, , and is a constant parameter; is the reference kurtosis; For each local box plot, use an adaptive standard multiple As a criterion, determine the outlier; about to exceed the upper quartile + ×Interquartile range or below the lower quartile - ×The data points with interquartile range are judged as outliers; the data points judged as outliers are replaced with the mean / median of the corresponding local time series; The local time series with replaced outliers are merged in the order of time windows and normalized to obtain the corresponding standard original data, namely, standard equipment status data and standard environment data; Mapping the standard device state data and the standard environment data to the same feature space to obtain a device state feature vector and an environment feature vector; constructing a device state core matrix based on the device state feature vector, and constructing an environment feature core matrix based on the environment feature vector in the same manner as the device state core matrix; The device state kernel matrix and the environment feature kernel matrix are fused at the matrix level to obtain a fused kernel matrix; a kernel method model is trained based on the fused kernel matrix; and the fused kernel matrix is constructed into a high-dimensional feature vector or tensor data structure, which is the fused data.
[0008] Furthermore, the method of constructing the device state core matrix includes: defining each row of the device state core matrix as a device state feature vector, and for any The device state feature vector and device state feature vectors, and calculate the weighted distance between them ;in, is a diagonal weight matrix; No. The weight of the device state feature vector when calculating the weighted distance ; The weights of all device state feature vectors when calculating weighted distances form a diagonal weight matrix, For the The normalized variance of the device state feature vector, and are two hyperparameters; based on weighted distance , calculate the first The device state feature vector and The distribution value between the device state feature vectors ; ;in, For the The adaptive bandwidth parameter of the device state feature vector, For the The adaptive bandwidth parameter of the device state feature vector, and is the corresponding weight parameter, For the The timestamp of the device status feature vector, No. The timestamp of the device status feature vector, For the The device type corresponding to the device status feature vector, For the The device type corresponding to the device status feature vector, is the indicator function; For the The device state feature vector and The cosine similarity between the device state feature vectors; Assign value As the element of the device state core matrix, The device state feature vector and The element value of the device state feature vector in the device state core matrix; the diagonal elements of the device state core matrix are 1, which means that the construction of the device state core matrix is completed.
[0009] Furthermore, the matrix-level fusion method includes: The environmental characteristic kernel matrix is used as the reference kernel matrix ; Calculate the device state kernel matrix and the reference kernel matrix The tensor product space of ; the basis of the tensor product space is and The element-wise product of the tensor product space is obtained by solving the characteristic equation The eigenvector matrix of and the eigenvalue matrix ; The characteristic equation is: ;in, for The transpose of Will Projection to the reference kernel matrix The tensor product space of is used to obtain the projected device state kernel matrix, which is recorded as ; ;in, for Before principal components, for The transpose of is the projection control parameter, is the nonlinear control parameter; is the element-wise product; is the hyperbolic tangent function; Fusion in Tensor Product Space and , get the fusion kernel matrix ;in, for The weight parameter, for The weight parameter of .
[0010] Furthermore, the method of training the kernel method model includes: Define the basic architecture of the kernel method model as a support vector machine; replace the original inner product calculation in the objective function of the support vector machine as ; The parameters of the kernel method model are obtained by optimizing the objective function, and the parameters are fixed to obtain the kernel method model.
[0011] Furthermore, the control decision fitting model is constructed in the following manner: Define the architecture of the control decision fitting model as a convolutional neural network, define the input dimension of the control decision fitting model to correspond to the dimension of the fusion data; set the network hierarchy of the convolutional neural network, including convolution layer, pooling layer, fully connected layer and output layer, and determine the parameters of the network hierarchy; add a random inactivation layer between the fully connected layers, and set the number of neurons in the output layer to the number of control parameters; Collect several groups of historical fusion data and corresponding parameter labels in a fixed historical time period, where the parameter labels are the adjustment parameters of smart home devices; divide the historical fusion data into training set, validation set and test set; define the loss function and optimizer, use the training set to train the control decision fitting model, and count the function value of the loss function on the validation set during the training process; Define the adaptive random dropout rate of the random dropout layer; the formula for the adaptive random dropout rate is: ;in, For the The adaptive dropout rate of the dropout layer during round training, It is The function value of the loss function on the validation set during the round of training, is the number of training rounds, , and are the polynomial coefficients, and is the exponential decay coefficient, , and is the periodic adjustment coefficient; during the training process, the adaptive random inactivation rate changes with the training rounds; Initialize and set a random dropout mask pool, which contains several random dropout masks, each of which corresponds to a random dropout pattern; the random dropout pattern is a different random dropout ratio or a different distribution of dropout neurons; During the training process, a random dropout mask is randomly selected from the random dropout mask pool, and the random dropout pattern corresponding to the random dropout is applied to perform a random dropout operation on the training data of the current training set; the training data after the random dropout operation is used for forward propagation to calculate the function value of the loss function, and back propagation is performed to update the parameters of the control decision fitting model; Record the function value of the loss function on the validation set corresponding to each random deactivation mask within a preset number of evaluation cycles; select the random deactivation mask with the smallest function value within the evaluation cycle as the best mask, and use the best mask before the next evaluation cycle starts until the next evaluation cycle ends; repeatedly select the best mask within each evaluation cycle, and use the best mask in the next cycle; Until the function value of the loss function of the control decision fitting model on the test set is the lowest; fix the parameters of the control decision fitting model at this time to obtain the constructed control decision fitting model.
[0012] Furthermore, the method for solving the optimal control parameters includes: Initialize optimization parameters, including population size N, maximum number of iterations, and aurora radius , Auroral attenuation coefficient , Aurora Weight , Dancer Weight and the dancer correction factor ; Taking the initial control decision as the center, N candidate solutions are randomly generated in the solution space near it to form the initial population, and each candidate solution corresponds to an initial dancer; N is a positive integer greater than 1; Defining dance functions ;in, For comfort, is the energy consumption of the equipment, is the target energy consumption, is the preset angular frequency, For time, , , and is the adaptive weight adjustment parameter, is the exponential parameter, is the noise intensity, is the noise that follows the standard normal distribution; is a positive weight parameter; is the historical average energy consumption of the equipment; For each dancer, the function value of its dance function is calculated and recorded as the dance value; iterative optimization is performed based on the calculated dance value; the iterative optimization is repeated until the maximum number of iterations is reached, and the candidate solution corresponding to the final optimal dancer is the optimal control parameters obtained by solving.
[0013] Furthermore, the process of iterative optimization includes: Determine the dancer with the largest dance value in the population at this time as the optimal dancer , based on the optimal dancer determined Generate Aurora Locations ;in, is the number of current iterations, is the exponential adjustment parameter, and ; For one dimensional standard normally distributed random vector, is the dimension of the control parameter; is the standard deviation of the dance value of the population at this time, This is the mean of the dance value of the population at this time; For every dancer , based on the generated auroral positions Computing Dancers Euclidean distance to the Aurora ; Based on Euclidean distance Calculating the gravitational pull of the aurora on dancers ; gravitational ;in, is the modulation item weight; is the angular frequency of the sine wave; is the phase of the sine wave; Computing Dancers The sum of the gravitational forces to the other dancers ; ;in, For the dancers Index of other dancers besides for and The Euclidean distance between for The dance value, for The dance value, is the gravitational attenuation parameter; Based on the gravitational pull of the aurora on the dancers and dancers The sum of the gravitational forces to the other dancers Update Dancer Get the updated location ,in For dancers who have not been updated location; Update Dancer After the position is determined, the dance value of each dancer is recalculated, the optimal dancer is re-determined and the current population is updated.
[0014] An intelligent device automatic control system, which is used to implement the intelligent device automatic control method, comprises: a data acquisition and fusion module, which is used to collect device status data and environment data of the intelligent home device, and fuse the device status data and environment data to obtain fused data; The model building module is used to input the fusion data into the pre-built control decision fitting model to obtain a preliminary control decision; The optimization application module uses an improved optimization algorithm to solve the optimal control parameters based on the preliminary control decision; and applies the optimal control parameters to the actuators of smart home devices.
[0015] The technical effects and advantages of the intelligent device automatic control method and system of the present invention are as follows: The present invention realizes comprehensive perception and precise control of smart home devices by integrating multi-dimensional data, adaptive model training and improved optimization algorithms, improves data quality by effectively processing abnormal data, and realizes efficient fusion of device status data and environmental data by constructing matrices and training models. A convolutional neural network structure is adopted, and an adaptive mechanism is introduced to enhance the generalization ability and robustness of the model. The improved optimization algorithm combines aurora search and dancer behavior to effectively avoid the problem of falling into local optimality. In addition, personalized factors such as user activity patterns, circadian rhythms, and geographic location are also considered. A dynamic balance between comfort and energy efficiency is achieved through mechanisms such as adaptive weight adjustment parameters and dynamic target energy consumption, which ensures user comfort while achieving efficient use of energy. The control strategy can be flexibly adjusted according to different scenarios to provide users with a more intelligent, comfortable, and energy-saving experience, effectively improving the overall performance, adaptability, and user satisfaction of the smart home system. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 A schematic diagram of an automatic control method for an intelligent device of the present invention; Figure 2 This is a schematic diagram of an intelligent device automatic control system of the present invention. DETAILED DESCRIPTION
[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0018] Example 1 See also Figure 1 As shown, the automatic control method of a smart device described in this embodiment includes: S1, collecting device status data and environment data of smart home devices, fusing the device status data and environment data to obtain fused data; S2, inputting the fused data into the pre-built control decision fitting model to obtain a preliminary control decision; S3. Based on the preliminary control decision, the optimal control parameters are solved using the improved optimization algorithm; the optimal control parameters are applied to the actuators of the smart home devices to achieve automatic control of the smart devices.
[0019] Equipment status data includes the equipment's switch status (on / off), equipment setting parameters, equipment real-time power consumption, and equipment operation time; fully understand the actual operation of the equipment and provide the necessary status information for automatic control of the equipment; Environmental data includes indoor temperature, indoor humidity, indoor air quality (CO2 concentration), indoor light intensity, room activity (occupied / unoccupied), room noise level, date / time information, and geographic location information; it can sense changes in the environment and prepare to automatically adjust equipment based on environmental changes; it provides support for personalized intelligent control based on the user's home activities, day and night rhythms, and geographic location; the device status data and environmental data are usually encoded to adapt to the subsequent operation format of the system.
[0020] The device status data and environmental data are segmented according to fixed time windows (such as hours, days, weeks, etc.), and the data in each time window is recorded as the corresponding local time series; for each local time series, a local box plot is constructed respectively. Specifically, statistics are calculated for the local time series, and the statistics include the median, the upper quartile, the lower quartile, and the interquartile range (the upper quartile minus the lower quartile); the local box plot consists of a rectangular box and two extension lines. The upper and lower edges of the rectangular box represent the upper quartile and the lower quartile, respectively, the horizontal line in the rectangular box represents the median, and the upper and lower endpoints of the extension line represent the upper inner value and the lower inner value, respectively.
[0021] For each local time series, calculate its kurtosis and skewness ; Kurtosis ;in, is the data point of the local time series, is the median of the local time series, is the number of data points in the local time series; is the interquartile range of the local time series, is the kurtosis correction coefficient, which is used to adjust the scale of skewness; Skewness ;in, is the skewness correction coefficient, which is used to adjust the scale of kurtosis. is the deviation correction coefficient, which is used to correct the deviation of finite samples; deviation correction coefficient ; Define adaptive standard multiples based on kurtosis and skewness ;in, is the base multiple, usually 1.5; and is the adjustment factor, , and is a constant parameter, determined by data fitting; is the reference kurtosis, such as the kurtosis of the normal distribution3.
[0022] For each local box plot, use an adaptive standard multiple As a criterion, determine the outlier; about to exceed the upper quartile + ×Interquartile range or below the lower quartile - The data points with the interquartile range of × are judged as outliers; the data points judged as outliers are replaced with the mean / median of the corresponding local time series; the local time series that replaced the outliers are merged according to the order of time windows, and normalized to be in the same order of magnitude, and the corresponding standard original data are obtained; that is, standard equipment status data and standard environment data; The standard equipment status data and the standard environment data are mapped to the same feature space, that is, each type of data of the standard equipment status data is constructed into a vector and concatenated at the vector level to obtain the equipment status feature vector; similarly, each type of data of the standard environment data is constructed into a vector and concatenated at the vector level to obtain the environment feature vector.
[0023] The device state core matrix is constructed based on the device state feature vector. Specifically, each row of the device state core matrix defines a device state feature vector. The device state feature vector and device state feature vectors, and calculate the weighted distance between them ;in, is a diagonal weight matrix, which is used to assign different weights to different device status features; No. The weight of the device state feature vector when calculating the weighted distance ; The weights of all device state feature vectors when calculating weighted distances form a diagonal weight matrix, For the The normalized variance of the device state feature vector, and There are two hyperparameters that control the mapping relationship between variance and weight, thereby automatically assigning reasonable feature weights.
[0024] Based on weighted distance , calculate the first The device state feature vector and The distribution value between the device state feature vectors ; ;in, For the The adaptive bandwidth parameter of the device state feature vector, For the The adaptive bandwidth parameter of the device state feature vector, and is the corresponding weight parameter, controlling the importance of different prior knowledge, For the The timestamp of the device status feature vector (the time point when the data was collected), No. The timestamp of the device status feature vector, For the The device type (code) corresponding to the device status feature vector, For the The device type (code) corresponding to the device status feature vector, It is an indicator function. If the parameters in the indicator function are the same, the output function value is 1, otherwise it is 0; For the The device state feature vector and The cosine similarity between the device state feature vectors; Assign value As the element of the device state core matrix, The device state feature vector and The element value of the device state feature vector in the device state core matrix; the diagonal elements of the device state core matrix are 1, which means that the construction of the device state core matrix is completed.
[0025] The environment feature kernel matrix is constructed based on the environment feature vector in the same way as the device state kernel matrix is constructed; The device state kernel matrix and the environment feature kernel matrix are fused at matrix level to obtain a fused kernel matrix; a kernel method model is trained based on the fused kernel matrix.
[0026] Ways to perform matrix-level fusion include: The environmental characteristic kernel matrix is used as the reference kernel matrix ; Calculate the device state kernel matrix and the reference kernel matrix The tensor product space of ; the basis of the tensor product space is and The element-wise product of the tensor product space is obtained by solving the characteristic equation The eigenvector matrix of and the eigenvalue matrix ; The characteristic equation is: ;in, for The transpose of Will Projection to the reference kernel matrix The tensor product space of is used to obtain the projected device state kernel matrix, which is recorded as ; ;in, for Before principal components (eigenvectors), for The transpose of is the projection control parameter, which determines the weight of the fusion term in the final projection kernel matrix. is a nonlinear control parameter, which adjusts the degree of nonlinearity; is the element-wise product; is the hyperbolic tangent function; Fusion in Tensor Product Space and , get the fusion kernel matrix ;in, for The weight parameter, for The weight parameters are determined by cross-validation method; Based on the fusion kernel matrix Train the kernel method model and define the basic architecture of the kernel method model as a support vector machine; replace the original inner product calculation in the objective function of the support vector machine with ; The parameters of the kernel method model are obtained by optimizing the objective function, and the parameters are fixed to obtain the kernel method model. The kernel method model is used as the basic model for subsequent fusion processing to simplify the complex process of fusion features; the fusion kernel matrix is constructed into a high-dimensional feature vector or tensor data structure; that is, the fusion data.
[0027] Furthermore, the construction method of the control decision fitting model includes: Define the architecture of the control decision fitting model as a convolutional neural network, define the input dimension of the control decision fitting model to correspond to the dimension of the fusion data; set the network hierarchy of the convolutional neural network, including convolution layer, pooling layer, fully connected layer (map the extracted features to the space of the control parameters) and output layer, and determine the parameters of the network hierarchy (such as convolution kernel size, step size, activation function); add a random inactivation layer between the fully connected layers, and set the number of neurons in the output layer to the number of control parameters; Collect several groups of historical fusion data (the same data type as the fusion data) and corresponding parameter labels in a fixed historical time period. The parameter labels are the adjustment parameters of smart home devices (recorded based on human comfort); divide the historical fusion data into training set, validation set and test set; the division ratio can be 70% training set, 20% validation set and 10% test set; define the loss function (such as cross entropy loss, mean square error, etc.) and optimizer (such as Adam, SGD, etc.), use the training set to train the control decision fitting model, and count the function value of the loss function on the validation set during the training process; Define the adaptive random dropout rate of the random dropout layer; the formula for the adaptive random dropout rate is: ;in, For the The adaptive dropout rate of the dropout layer during round training, It is The function value of the loss function on the validation set during the round of training, is the number of training rounds, , and are the polynomial coefficients, and is the exponential decay coefficient, , and is the periodic adjustment coefficient; the exponential decay coefficient controls the exponential decay of the adaptive random inactivation rate as the number of training rounds increases; the periodic adjustment coefficient controls the periodic fluctuations during training; during training, the adaptive random inactivation rate changes with the number of training rounds.
[0028] Initialize a random inactivation mask pool, which contains several random inactivation masks, each of which corresponds to a random inactivation mode; the random inactivation mode is different random inactivation ratios or different distributions of inactivated neurons; during the training process, randomly select a random inactivation mask from the random inactivation mask pool, and apply the random inactivation mode corresponding to the random inactivation to perform a random inactivation operation on the training data of the current training set; use the training data after the random inactivation operation for forward propagation, calculate the function value of the loss function, and perform back propagation to update the parameters of the control decision fitting model; within a preset number of evaluation cycles (such as a certain number of steps or batches per training), record the function value of the loss function on the validation set corresponding to each random inactivation mask; Select the random deactivation mask with the smallest function value in the evaluation cycle and record it as the best mask. Before the next evaluation cycle starts, use the best mask until the end of the next evaluation cycle. Repeat the selection of the best mask in each evaluation cycle and use the best mask in the next cycle until the function value of the loss function of the control decision fitting model on the test set is the lowest. Fix the parameters of the control decision fitting model at this time to obtain the constructed control decision fitting model. Input the fused data into the constructed control decision fitting model to obtain a preliminary control decision. The preliminary control decision is the value of a discrete control parameter (such as set temperature, etc.).
[0029] The optimal control parameters can be solved by: Initialize optimization parameters, including population size N (number of candidate solutions), maximum number of iterations, and aurora radius , Auroral attenuation coefficient , Aurora Weight , Dancer Weight and the dancer correction factor ; Taking the initial control decision as the center, N candidate solutions (a combination of control parameter values) are randomly generated in the solution space near it to form the initial population. Each candidate solution corresponds to an initial dancer. N is a positive integer greater than 1. Defining dance functions ;in, For comfort, is the energy consumption of the equipment, is the target energy consumption, is the preset angular frequency, For time, , , and is the adaptive weight adjustment parameter, is the exponential parameter, is the noise intensity, is the noise that follows the standard normal distribution; is a positive weight parameter; It is the historical average energy consumption of the equipment. It enables the system to operate within a relatively stable energy consumption range to avoid excessive fluctuations, and can be dynamically adjusted over time to better adapt to diurnal rhythms and seasonal changes, such as expecting higher comfort during the day and lower energy consumption at night.
[0030] It needs to be explained that the measurement of comfort includes many aspects, such as temperature comfort, air quality comfort, noise level comfort, lighting comfort, personal preference comfort, etc. By setting the corresponding threshold range and manually measuring the corresponding comfort value, the overall comfort value can be obtained through weighted summation. The weight of the weighted summation is dynamically adjusted according to time, season and personal preference.
[0031] For each dancer, the function value of the dance function is calculated and recorded as the dance value; iterative optimization is performed based on the calculated dance value; The process of iterative optimization includes: Determine the dancer with the largest dance value in the population at this time as the optimal dancer , based on the optimal dancer determined Generate Aurora Locations ;in, is the number of current iterations, is the exponential adjustment parameter, and , used to flexibly adjust the balance strategy between global and local search; For one dimensional standard normally distributed random vector, To control the dimension of the parameters; introducing different perturbations for each dimension is conducive to more thorough exploration in different dimensions; is the standard deviation of the dance value of the population at this time, This is the mean dance value of the population at this time.
[0032] For every dancer , based on the generated auroral positions Computing Dancers Euclidean distance to the Aurora ; Based on Euclidean distance Calculating the gravitational pull of the aurora on dancers ; ;in, is the modulation item weight, which controls the contribution of the modulation item; is the angular frequency of the sine wave, which controls the period of the sine wave; is the sine wave phase, which controls the displacement of the sine wave; Computing Dancers The sum of the gravitational forces to the other dancers ; ;in, For the dancers Index of other dancers besides for and The Euclidean distance between for The dance value, for The dance value, is the gravity attenuation parameter, which is used to control the attenuation rate of gravity between dancers with distance; Based on the gravitational pull of the aurora on the dancers and dancers The sum of the gravitational forces to the other dancers Update Dancer Get the updated location ,in For dancers who have not been updated location; Update Dancer After the position is determined, the dance value of each dancer is recalculated, the optimal dancer is re-determined and the current population is updated. The iterative optimization is repeated until the maximum number of iterations is reached. The candidate solution corresponding to the final optimal dancer is the optimal control parameter obtained by solving.
[0033] This embodiment realizes comprehensive perception and precise control of smart home devices by integrating multi-dimensional data, adaptive model training and improved optimization algorithms, improves data quality by effectively processing abnormal data, and realizes efficient fusion of device status data and environmental data by constructing matrices and training models. A convolutional neural network structure is adopted, and an adaptive mechanism is introduced to enhance the generalization ability and robustness of the model. The improved optimization algorithm combines aurora search and dancer behavior to effectively avoid the problem of falling into local optimality. In addition, personalized factors such as user activity patterns, circadian rhythms, and geographic locations are also considered. A dynamic balance between comfort and energy efficiency is achieved through mechanisms such as adaptive weight adjustment parameters and dynamic target energy consumption, which ensures user comfort while achieving efficient use of energy. The control strategy can be flexibly adjusted according to different scenarios to provide users with a more intelligent, comfortable, and energy-saving experience, effectively improving the overall performance, adaptability, and user satisfaction of the smart home system.
[0034] Example 2 See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, which provides an intelligent device automatic control system, including: The data collection and fusion module is used to collect the device status data and the environment data of the smart home device, and fuse the device status data and the environment data to obtain fused data; The model building module is used to input the fusion data into the pre-built control decision fitting model to obtain a preliminary control decision; The optimization application module uses an improved optimization algorithm to solve the optimal control parameters based on the preliminary control decision; the optimal control parameters are applied to the actuators of the smart home devices; and the modules are connected through wired and / or wireless means to realize data transmission between modules.
[0035] Example 3 This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the above-mentioned intelligent device automatic control method is implemented.
[0036] Since the electronic device introduced in this embodiment is an electronic device used to implement an intelligent device automatic control method in the embodiment of this application, based on the intelligent device automatic control method introduced in the embodiment of this application, the technical personnel of this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not described in detail here. As long as the technical personnel of this field implement the electronic device used in the intelligent device automatic control method in the embodiment of this application, it belongs to the scope of protection of this application.
[0037] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.
[0038] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.
Claims
1. A method for automatic control of intelligent equipment, characterized in that: include: S1, collecting device status data and environment data of smart home devices, fusing the device status data and environment data to obtain fused data; S2, inputting the fused data into the pre-built control decision fitting model to obtain a preliminary control decision; S3. Based on the preliminary control decision, the optimal control parameters are solved using the improved optimization algorithm; the optimal control parameters are applied to the actuators of the smart home devices.
2. The method for automatic control of intelligent equipment according to claim 1, characterized in that: The device status data includes the device's on / off status, device setting parameters, device real-time power consumption, and device operating time; the environmental data includes indoor temperature, indoor humidity, indoor air quality, indoor light intensity, room personnel activities, room noise level, time information, and geographic location information.
3. The method for automatic control of intelligent equipment according to claim 2, characterized in that: The method of performing the fusion processing includes: The device status data and environmental data are segmented according to fixed time windows, and the data in each time window is recorded as the corresponding local time series; for each local time series, a local box plot is constructed, and statistics are calculated for the local time series, including the median, upper quartile, lower quartile and interquartile range; for each local time series, its kurtosis is calculated and skewness ; Kurtosis ;in, is the data point of the local time series, is the median of the local time series, is the number of data points in the local time series; is the interquartile range of the local time series, is the kurtosis correction coefficient; Skewness ;in, is the skewness correction coefficient, is the deviation correction factor; deviation correction factor ; Define adaptive standard multiples based on kurtosis and skewness ;in, is the benchmark multiple; and is the adjustment factor, , and is a constant parameter; is the reference kurtosis; For each local box plot, use an adaptive standard multiple As a criterion, determine the outlier; about to exceed the upper quartile + ×Interquartile range or below the lower quartile - ×The data points with interquartile range are judged as outliers; the data points judged as outliers are replaced with the mean / median of the corresponding local time series; The local time series with replaced outliers are merged in the order of time windows and normalized to obtain the corresponding standard original data, namely, standard equipment status data and standard environment data; Mapping the standard device state data and the standard environment data to the same feature space to obtain a device state feature vector and an environment feature vector; constructing a device state core matrix based on the device state feature vector, and constructing an environment feature core matrix based on the environment feature vector in the same manner as the device state core matrix; The device state kernel matrix and the environment feature kernel matrix are fused at the matrix level to obtain a fused kernel matrix; a kernel method model is trained based on the fused kernel matrix; and the fused kernel matrix is constructed into a high-dimensional feature vector or tensor data structure, which is the fused data.
4. The method for automatic control of intelligent equipment according to claim 3, characterized in that: The method of constructing the device status core matrix includes: Define each row of the device state kernel matrix as a device state feature vector. The device state feature vector and device state feature vectors, and calculate the weighted distance between them ;in, is a diagonal weight matrix; No. The weight of the device state feature vector when calculating the weighted distance ; The weights of all device state feature vectors when calculating weighted distances form a diagonal weight matrix, For the The normalized variance of the device state feature vector, and are two hyperparameters; based on weighted distance , calculate the first The device state feature vector and The distribution value between the device state feature vectors ; ;in, For the The adaptive bandwidth parameter of the device state feature vector, For the The adaptive bandwidth parameter of the device state feature vector, and is the corresponding weight parameter, For the The timestamp of the device status feature vector, No. The timestamp of the device status feature vector, For the The device type corresponding to the device status feature vector, For the The device type corresponding to the device status feature vector, is the indicator function; For the The device state feature vector and The cosine similarity between the device state feature vectors; Assign value As the element of the device state core matrix, that is, The device state feature vector and The element value of the device state feature vector in the device state core matrix; the diagonal elements of the device state core matrix are 1, which means that the construction of the device state core matrix is completed.
5. The method for automatic control of intelligent equipment according to claim 4, characterized in that: Ways to perform matrix-level fusion include: The environmental characteristic kernel matrix is used as the reference kernel matrix ; Calculate the device state kernel matrix and the reference kernel matrix The tensor product space of ; the basis of the tensor product space is and The element-wise product of the tensor product space is obtained by solving the characteristic equation The eigenvector matrix of and the eigenvalue matrix ; The characteristic equation is: ;in, for The transpose of Will Projection to the reference kernel matrix The tensor product space of is used to obtain the projected device state kernel matrix, which is recorded as ; ;in, for Before principal components, for The transpose of is the projection control parameter, is the nonlinear control parameter; is the element-wise product; is the hyperbolic tangent function; Fusion in Tensor Product Space and , get the fusion kernel matrix ;in, for The weight parameter, for The weight parameter of .
6. The method for automatic control of intelligent equipment according to claim 5, characterized in that: The method of training the kernel method model includes: Define the basic architecture of the kernel method model as a support vector machine; replace the original inner product calculation in the objective function of the support vector machine as ; The parameters of the kernel method model are obtained by optimizing the objective function, and the parameters are fixed to obtain the kernel method model.
7. The method for automatic control of intelligent equipment according to claim 6, characterized in that: The construction method of the control decision fitting model includes: Define the architecture of the control decision fitting model as a convolutional neural network, define the input dimension of the control decision fitting model to correspond to the dimension of the fusion data; set the network hierarchy of the convolutional neural network, including convolution layer, pooling layer, fully connected layer and output layer, and determine the parameters of the network hierarchy; add a random inactivation layer between the fully connected layers, and set the number of neurons in the output layer to the number of control parameters; Collect several groups of historical fusion data and corresponding parameter labels in a fixed historical time period, where the parameter labels are the adjustment parameters of smart home devices; divide the historical fusion data into training set, validation set and test set; define the loss function and optimizer, use the training set to train the control decision fitting model, and count the function value of the loss function on the validation set during the training process; Define the adaptive random dropout rate of the random dropout layer; the formula for the adaptive random dropout rate is: ;in, For the The adaptive dropout rate of the dropout layer during round training, It is The function value of the loss function on the validation set during the round of training, is the number of training rounds, , and are the polynomial coefficients, and is the exponential decay coefficient, , and is the periodic adjustment coefficient; during the training process, the adaptive random inactivation rate changes with the training rounds; Initialize and set a random dropout mask pool, which contains several random dropout masks, each of which corresponds to a random dropout pattern; the random dropout pattern is a different random dropout ratio or a different distribution of dropout neurons; During the training process, a random dropout mask is randomly selected from the random dropout mask pool, and the random dropout pattern corresponding to the random dropout is applied to perform a random dropout operation on the training data of the current training set; the training data after the random dropout operation is used for forward propagation to calculate the function value of the loss function, and back propagation is performed to update the parameters of the control decision fitting model; Record the function value of the loss function on the validation set corresponding to each random deactivation mask within a preset number of evaluation cycles; select the random deactivation mask with the smallest function value within the evaluation cycle as the best mask, and use the best mask before the next evaluation cycle starts until the next evaluation cycle ends; repeatedly select the best mask within each evaluation cycle, and use the best mask in the next cycle; Until the function value of the loss function of the control decision fitting model on the test set is the lowest; fix the parameters of the control decision fitting model at this time to obtain the constructed control decision fitting model.
8. The method for automatic control of intelligent equipment according to claim 7, characterized in that: The method for solving the optimal control parameters includes: Initialize optimization parameters, including population size N, maximum number of iterations, and aurora radius , Auroral attenuation coefficient , Aurora Weight , Dancer Weight and the dancer correction factor ; Taking the initial control decision as the center, N candidate solutions are randomly generated in the solution space near it to form the initial population, and each candidate solution corresponds to an initial dancer; N is a positive integer greater than 1; Defining dance functions ;in, For comfort, is the energy consumption of the equipment, is the target energy consumption, is the preset angular frequency, For time, , , and is the adaptive weight adjustment parameter, is the exponential parameter, is the noise intensity, is the noise that follows the standard normal distribution; is a positive weight parameter; is the historical average energy consumption of the equipment; For each dancer, the function value of its dance function is calculated and recorded as the dance value; iterative optimization is performed based on the calculated dance value; the iterative optimization is repeated until the maximum number of iterations is reached, and the candidate solution corresponding to the final optimal dancer is the optimal control parameters obtained by solving.
9. The method for automatic control of intelligent equipment according to claim 8, characterized in that: The process of iterative optimization includes: Determine the dancer with the largest dance value in the population at this time as the optimal dancer , based on the optimal dancer determined Generate Aurora Locations ;in, is the number of current iterations, is the exponential adjustment parameter, and ; For one dimensional standard normally distributed random vector, is the dimension of the control parameter; is the standard deviation of the dance value of the population at this time, This is the mean of the dance value of the population at this time; For every dancer , based on the generated auroral positions Computing Dancers Euclidean distance to the Aurora ; Based on Euclidean distance Calculating the aurora's gravitational pull on dancers ; gravitational ;in, is the modulation item weight; is the angular frequency of the sine wave; is the phase of the sine wave; Computing Dancers The sum of the gravitational forces to the other dancers ; ;in, For the dancers Index of other dancers besides for and The Euclidean distance between for The dance value, for The dance value, is the gravitational attenuation parameter; Based on the gravitational pull of the aurora on the dancers and dancers The sum of the gravitational forces to the other dancers Update Dancer Get the updated location ,in For dancers who have not been updated location; Update Dancer After the position is determined, the dance value of each dancer is recalculated, the optimal dancer is re-determined and the current population is updated.
10. An intelligent device automatic control system, used to implement an intelligent device automatic control method according to any one of claims 1 to 9, characterized in that: include: The data collection and fusion module is used to collect the device status data and the environment data of the smart home device, and fuse the device status data and the environment data to obtain fused data; The model building module is used to input the fusion data into the pre-built control decision fitting model to obtain a preliminary control decision; The optimization application module uses an improved optimization algorithm to solve the optimal control parameters based on the preliminary control decision; and applies the optimal control parameters to the actuators of smart home devices.
Citation Information
Patent Citations
Intelligent home environment cooperative control method and system based on emotion robot
CN104965552A
Smart home multi-sensor data fusion method based on neural network
CN110084358A
Intelligent home environment monitoring method and system based on artificial neural network
CN113867167A
Intelligent home emotion interaction feedback system
CN117348428A
Air conditioner energy-saving control method and system based on reinforcement learning and digital twinborn model
CN118031385A
Cited By
Building energy-saving intelligent integrated control system and method
CN119644751A
Intelligent delay closing electric energy meter control platform and method
CN119787653A
Internet of Things big data fusion processing platform
CN119967037A
An internet of things big data fusion processing platform
CN119967037B
Intelligent household equipment adaptive control method and system based on artificial intelligence
CN120196003A
