Intelligent light control system

Abstract

Disclosed in the present invention is an intelligent light control system. The system comprises a data acquisition module, a data preprocessing module, a feature selection module, an intelligent control model establishment module, a hyperparameter search module and a real-time intelligent control module. The present invention belongs to the technical field of intelligent light control, and specifically relates to the intelligent light control system. The present solution creatively combines the filtering method and the wrapper method, and thus can comprehensively consider the interaction between features and select feature subsets more suitable for intelligent control models, avoiding the problem of over-fitting in feature selection, thus improving the training effect and performance of intelligent control models. The present solution uses a hybrid circular segmented graph, a spiral flight search method and a mirror reflection method to improve an algorithm for acquiring built-in parameters for the intelligent control effect, so as to obtain optimal parameter combinations, thereby improving the stability of output results of intelligent control models, and accordingly solving the problem of inaccurate control performance of intelligent control models.

Description

Intelligent light control system TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent light control, specifically refers to an intelligent light control system. BACKGROUND

[0002] The intelligent light control system is a system for controlling light by using intelligent means to achieve precise control, energy-saving management and intelligent adjustment of light. However, in general intelligent light control systems, the feature selection method of the intelligent control model has the problems of easily ignoring the interaction between features, inaccurate selection of features, and overfitting in feature selection. In general intelligent light control systems, the improper setting of the built-in parameters of the intelligent control model leads to unstable output results of the intelligent control model, thereby resulting in inaccurate final intelligent control effect.

[0003] SUMMARY

[0004] To overcome the defects of the prior art, the present application provides an intelligent light control system. To overcome the problems of general feature selection methods, such as easily ignoring the interaction between features, inaccurate selection of features, and overfitting in feature selection, the present application creatively adopts a combination of filtering and packaging methods, which can comprehensively consider the interaction between features, select a feature subset more suitable for the intelligent control model, and avoid overfitting in feature selection, thereby improving the training effect and performance of the intelligent control model. To overcome the problem of improper setting of the built-in parameters of the general intelligent control model, which leads to unstable output results of the intelligent control model and inaccurate final intelligent control effect, the present application creatively adopts a hybrid circle segmentation diagram, a spiral flight search method, and a mirror reflection method to improve the algorithm for obtaining built-in parameters of the intelligent control effect, thereby obtaining an optimal parameter combination, improving the stability of the output results of the intelligent control model, and solving the problem of inaccurate control effect of the intelligent control model.

[0005] The technical solutions adopted by the present application are as follows: The present application provides an intelligent light control system, which comprises a data acquisition module, a data preprocessing module, a feature selection module, an intelligent control model establishment module, a hyperparameter search module, and a real-time intelligent control module.

[0006] The data acquisition module acquires digital signals in the light environment through a sensor, converts them into digital form through an analog-to-digital converter, obtains a light control original data set, and sends the light control original data set to the data preprocessing module.

[0007] The data preprocessing module is used for data preprocessing. Through data preprocessing, a light control preliminary data set is obtained, and the light control preliminary data set is sent to the feature selection module.

[0008] The feature selection module is configured to perform feature selection, obtain a feature optimization dataset through the feature selection, and send the feature optimization dataset to the intelligent control model establishment module.

[0009] The intelligent control model establishment module is configured to establish an intelligent control model, obtain a real-time intelligent control dataset through the establishment of the intelligent control model, and send the real-time intelligent control dataset to the real-time intelligent control module.

[0010] The hyperparameter search module is configured to perform hyperparameter search, obtain an intelligent control model parameter optimal solution combination through the hyperparameter search, and send the intelligent control model parameter optimal solution combination to the intelligent control model establishment module.

[0011] The real-time intelligent control module is configured to perform real-time light control, and obtain intelligent control of current light through monitoring of real-time light control data.

[0012] Further, in the data collection module, the data collection is configured to collect original data required for light control, specifically, through collection, a light control original dataset is obtained in the current intelligent light control system.

[0013] The light control original dataset specifically includes collected historical light control data and real-time light control data; the historical light control data and the real-time light control data both include power voltage data of a conduction angle, user interaction data, environmental parameter data, and light energy consumption data; the user interaction data includes light setting history records and light use habits; the environmental parameter data includes illumination intensity, indoor and outdoor temperature data, and indoor and outdoor humidity data; the light energy consumption data includes total power consumption, energy consumption per unit time, and energy consumption comparison data after intelligent control; the real-time light control data further includes light equipment state data and current user activity data.

[0014] Further, in the data preprocessing module, the data preprocessing is configured to perform data preprocessing on the original data, specifically including the following contents:

[0015] Feature encoding is configured to convert the original data into a digital form, specifically, the light control original dataset is encoded through a one-hot encoding method to obtain a light control encoded dataset;

[0016] Data normalization is configured to perform normalization processing on the data, specifically, the light control encoded dataset is normalized through a minimum-maximum method to obtain a light control preliminary dataset;

[0017] Data preprocessing is performed, specifically, data preprocessing is performed on the light control original data set through the feature encoding and the data normalization to obtain a light control preliminary data set.

[0018] Further, in the feature selection module, the feature selection is used for feature selection on the large-scale light control features, specifically, the feature selection is performed on the light control preliminary data set by using the combined filter and wrapper method to obtain a feature optimization data set, and specifically includes the following contents:

[0019] The filter objective function is calculated, and the formula is as follows:

[0020] In the formula, I(x, y) represents the correlation between the two variables x and y, p(x) and p(y) represent the marginal probability distribution of the features x and y respectively, p(x, y) represents the joint probability distribution of the features x and y, f1 represents the filter objective function, R(S) represents the redundancy of the feature subset S, D(S) represents the dependency of the feature subset S, S represents the size of the feature subset, that is, the number of features contained, x i and x j represent different features in the feature subset, I(x i , x j ) represents the correlation between the feature x i and the feature x j , and I(x i , y) represents the correlation between the feature x i and the target variable y.

[0021] The wrapper objective function is calculated, and the formula is as follows:

[0022] In the formula, f2 represents the wrapper objective function, which is used to evaluate the accuracy of the selected features, n represents the sample number of the feature subset, v represents the sample number of the validation feature subset, y i,t represents the true value of the sample in the i-th validation subset at the t-th time step, represents the predicted value of the sample in the i-th validation subset at the t-th time step given the feature subset S.

[0023] The minimum objective function value is obtained, and the formula is as follows:

[0024] In the formula, Minimize represents the minimum value of the filter and wrapper objective functions.

[0025] Feature selection is performed, and the best feature subset is selected as the final feature selection result according to Minimize to obtain a feature optimization data set.

[0026] Further, in the establishment of the intelligent control model, the establishment of the intelligent control model specifically includes the following contents:

[0027] Initialize the model parameters, set the number of model hidden units, input vector dimension, output vector dimension and training iteration times;

[0028] The forgetting gate calculation process uses the following formula:

[0029] In the formula, f t represents the output of the forgetting gate, X t represents the input data, H t-1 represents the hidden state at the last time, W xf and W hf represent the weight matrix corresponding to the forgetting gate, b f represents the bias term parameter of the forgetting gate, and sigma represents the Sigmoid function.

[0030] The input gate calculation process uses the following formula:

[0031] In the formula, i t represents the output of the input gate, W xi and W hi represent the weight matrix corresponding to the input gate, b i represents the bias term parameter of the input gate.

[0032] Calculate the new cell unit state candidate value The formula used is as follows:

[0033] In the formula, W xc and W hc represent the weight matrix corresponding to the cell unit state, b c represents the bias term parameter of the cell unit state; The cell unit state update process uses the following formula:

[0034] In the formula, c t represents the current time cell unit state, c t-1 represents the cell unit state at the last time.

[0035] The output gate calculation process uses the following formula:

[0036] In the formula, o t represents the output of the output gate, W xo and W ho represent the weight matrix corresponding to the output gate, b o represents the bias term parameter of the output gate.

[0037] The hidden state calculation process uses the following formula: h t = o t ⊙ tanh(c t );

[0038] In the formula, tanh represents the hyperbolic tangent function, h t represents the hidden state at the current moment;

[0039] The bidirectional LSTM is calculated using the following formula:

[0040] In the formula, represents the forward direction hidden state at the current moment, represents the forward direction hidden state at the previous moment, represents the reverse direction hidden state at the current moment, represents the reverse direction hidden state at the previous moment;

[0041] The final output y t is obtained using the following formula:

[0042] In the formula, represents the weight matrix connecting the forward state to the output, represents the weight matrix connecting the forward state to the output, g represents the activation function, and B y represents the bias term parameter of the output layer;

[0043] Iterative determination: if the anomaly detection model converges, stop the iterative training, obtain the intelligent control model, otherwise continue the iterative training.

[0044] Further, in the hyperparameter search module, the hyperparameter search specifically includes the following contents:

[0045] Initialize parameters, swarm size N, search machine proportion P N , observation machine proportion S N , maximum iteration number T, search lower limit lb d and upper limit ub d of dimension d, early warning value R2, and safety threshold ST;

[0046] Initialize the population, generate N sequence values, initialize the random value z0, and use the following formula:

[0047] In the formula, represents the d-th dimension i-th sequence value, β represents a control parameter in the range of [0, 1], and rand represents a random number in the range of [0, 1];

[0048] The sequence value conversion search individual is expressed by the following formula:

[0049] In the formula, represents the initialization position of the ith individual in the dth dimension;

[0050] The fitness value fi is calculated: the performance of the intelligent control model established based on the individual position is taken as the fitness value of the individual, the individuals are sorted from good to bad according to the fitness values, the search aircraft and the wingman are divided, and the individual f b with the best fitness value and its position X b , the individual f n with the second best fitness value and its position X n , and the individual f w with the worst fitness value and its position X w are recorded;

[0051] The dynamic adaptive weight is obtained, and the used formula is as follows: r(i)=θ(i)+R·rand; x(i)=r(i)·sin(θ(i)) / max(|r(i)·sin(θ(i))|); y(i)=r(i)·cos(θ(i)) / max(|r(i)·cos(θ(i))|);

[0052] In the formula, represents the optimal individual position in the current iteration, α and R represent control parameters for the flight path, θ(i) and r(i) represent the polar angle and the polar diameter of the flight search, x(i) and y(i) represent the position in the polar coordinates, γ represents a dynamic adaptive weight parameter, and t represents the iteration number;

[0053] The search aircraft position is updated, and the used formula is as follows:

[0054] In the formula, represents the dth dimension position of the ith individual in the tth generation swarm, represents the dth dimension position of the ith individual in the tth generation swarm, R2 represents a warning value, ST represents a safety threshold, R2<ST indicates that there is no enemy aircraft nearby and the position is safe, R2≥ST indicates that there is an enemy aircraft nearby and the region needs to be changed for landing, and L represents a vector in which all elements are 1;

[0055] The wingman position is updated, and the used formula is as follows:

[0056] where n represents the number of individuals, represents the position of the individual with the best fitness value in dimension d in the t+1th iteration of the swarm, + represents the 1 x D dimensional pseudo-inverse matrix, the elements of which are randomly assigned 1 or -1;

[0057] Randomly select 10% of the individuals in the swarm as observers to update the position, using the following formula:

[0058] where X represents the position of the individual with the best fitness value in dimension d in the tth iteration of the swarm, represents the position of the individual with the worst fitness value in dimension d in the tth iteration of the swarm, r1 and δ represent standard normal distribution random numbers with a mean of 0 and a difference of 1, r2 represents a standard normal distribution random number with a mean of 0 and a variance of 0.7, and ∈ represents the minimum constant to avoid a zero denominator;

[0059] Perform a reflection operation, select the search individual with a poor fitness value to perform a reflection operation, using the following formula: X r = X m + α · (X b - X w );

[0060] where X m represents the midpoint position, and X r represents the reflection position;

[0061] Reflection point fitness value judgment, according to different fitness value relationship, adopt different position calculation mode to adjust the position of search individual, using the following formula:

[0062] where X ex represents the position of the expansion point, X ec represents the position of the outer contraction point, X ic represents the position of the inner contraction point, and f r represents the fitness value of the reflection point;

[0063] According to the adjustment of the position of the search individual, judge the effect, using the following formula:

[0064] where f ex represents the fitness value of the expansion point, f ec represents the fitness value of the outer contraction point, f ic represents the fitness value of the inner contraction point, and if the judgment in the formula is not met, then X w = X r ;

[0065] The search determination is provided with a fitness threshold value, when there is an individual fitness value higher than the fitness threshold value, the hyperparameter search is completed, the individual position is the optimal parameter of the intelligent control model, at this time the intelligent control model is established, if the maximum iteration number is reached, the swarm position is reinitialized, otherwise the position is continuously updated.

[0066] Further, in the real-time intelligent control module, the real-time intelligent control is based on the intelligent control model with set parameters, real-time light control data is acquired, and the intelligent control is performed on the current light based on the output of the intelligent control model module as a control signal.

[0067] The above scheme has the following beneficial effects:

[0068] (1) In view of the problems that the general feature selection method is easy to ignore the interaction between features, the selected features are not accurate, and overfitting occurs in feature selection, the scheme creatively adopts a combination method of filtering and packaging, can comprehensively consider the interaction between features, select a feature subset more suitable for the intelligent control model, avoid the problem of overfitting in feature selection, thereby improving the training effect and performance of the intelligent control model.

[0069] (2) In view of the problem that the built-in parameter setting of the general intelligent control model is improper, resulting in unstable output results of the intelligent control model, thereby resulting in inaccurate final intelligent control effect, the scheme creatively adopts a mixed circle segmentation diagram, a spiral flight search method and a mirror reflection method to improve the algorithm for obtaining the built-in parameter of the intelligent control effect, obtains the optimal parameter combination, improves the stability of the output results of the intelligent control model, thereby solving the problem of inaccurate intelligent control model control effect. BRIEF DESCRIPTION OF DRAWINGS

[0070] Fig. 1 is a module schematic diagram of an intelligent light control system provided by the present application;

[0071] Fig. 2 is a flowchart of step S3 feature selection;

[0072] Fig. 3 is a flowchart of step S4 establishing an intelligent control model module;

[0073] Fig. 4 is a flowchart of step S5 hyperparameter search module;

[0074] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation of the present application. DETAILED DESCRIPTION

[0075] Clearly, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments; based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the protection scope of the present application.

[0076] In the description of the present application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0077] Embodiment one, referring to figure 1, the present application provides an intelligent light control system, comprising a data acquisition module, a data preprocessing module, a feature selection module, an intelligent control model module, a hyperparameter search module and a real-time intelligent control module;

[0078] The data acquisition module acquires digital signals in the light environment through sensors, converts them into digital form through an analog-to-digital converter, obtains a light control original data set, and sends the light control original data set to the data preprocessing module;

[0079] The data preprocessing module is used for data preprocessing, and through data preprocessing, a light control preliminary data set is obtained and sent to the feature selection module;

[0080] The feature selection module is used for feature selection, and through feature selection, a feature optimization data set is obtained and sent to the intelligent control model module;

[0081] The intelligent control model module is used for establishing an intelligent control model, and through the establishment of the intelligent control model, a real-time intelligent control data set is obtained and sent to the real-time intelligent control module;

[0082] The hyperparameter search module is used for hyperparameter search, and through hyperparameter search, an intelligent control model parameter optimal solution combination is obtained and sent to the intelligent control model module;

[0083] The real-time intelligent control module is used for real-time light control, and through monitoring of real-time light control data, intelligent control of the current light is obtained.

[0084] Embodiment two, referring to FIG. 1, based on the above embodiment, the light control original data set specifically includes historical light control data and real-time light control data; the historical light control data and the real-time light control data both include power voltage data of a conduction angle, user interaction data, environmental parameter data and light energy consumption data; the user interaction data includes light setting history records and light use habits; the environmental parameter data includes illumination intensity, indoor and outdoor temperature data and indoor and outdoor humidity data; the light energy consumption data includes total power consumption, energy consumption per unit time and energy consumption comparison data after intelligent control; the real-time light control data further includes light equipment state data and user current activity data.

[0085] Embodiment three, referring to FIG. 1, based on the above embodiment, in the data preprocessing module, the data preprocessing is used for data preprocessing of the original data, specifically including the following contents:

[0086] Feature encoding is used for converting the original data into a digital form, specifically through one-hot encoding method to encode the light control original data set to obtain a light control encoded data set;

[0087] Data normalization is used for normalizing the data, specifically through the minimum-maximum method to normalize the light control encoded data set to obtain a light control preliminary data set;

[0088] Data preprocessing is performed, specifically through the feature encoding and the data normalization to preprocess the light control original data set to obtain the light control preliminary data set.

[0089] Embodiment four, referring to FIG. 1 and FIG. 2, based on the above embodiment, in the feature selection module, the feature selection is used for feature selection of large-scale light control features, specifically through the combination of filtering method and packaging method to select features of the light control preliminary data set to obtain a feature optimization data set, specifically including the following contents:

[0090] The filter objective function is calculated, and the formula is as follows:

[0091] In the formula, I(x, y) represents the correlation between x and y, p(x) and p(y) represent the marginal probability distribution of features x and y respectively, p(x, y) represents the joint probability distribution of features x and y, f1 represents the filter objective function, R(S) represents the redundancy of the feature subset S, D(S) represents the dependency of the feature subset S, S represents the size of the feature subset, i.e. the number of features contained, x i and x j represent different features in the feature subset, I(xi x j ) represents the correlation between the characteristics x i and the characteristics x j , I(x i , y) represents the correlation between the characteristics x i and the target variable y;

[0092] The wrapper target function is calculated, and the formula is as follows:

[0093] In the formula, f2 represents the wrapper target function for evaluating the accuracy of the selected features, n represents the number of samples of the feature subset, v represents the number of samples of the validation feature subset, y i,t represents the true value of the sample in the ith validation subset at the tth time step, represents the predicted value of the sample in the ith validation subset at the tth time step given the feature subset S;

[0094] The minimum target function value is obtained, and the formula is as follows:

[0095] In the formula, Minimize represents the minimum value of the filter and wrapper target functions;

[0096] Feature selection is performed, and the best feature subset is selected as the final feature selection result according to Minimize, and the feature optimization dataset is obtained.

[0097] By performing the above operations, in view of the problems of ignoring the interaction between features, inaccurate selected features, and overfitting in feature selection in general feature selection methods, the scheme creatively adopts a combination method of filter method and wrapper method, which can comprehensively consider the interaction between features, select a feature subset more suitable for intelligent control model, avoid the problem of overfitting in feature selection, and thus improve the training effect and performance of the intelligent control model.

[0098] In the fifth embodiment, referring to FIGS. 1 and 3, the embodiment is based on the above-mentioned embodiments, and the establishment of the intelligent control model specifically includes the following contents:

[0099] Initialize the model parameters, set the number of hidden units, the dimension of the input vector, the dimension of the output vector, and the number of training iterations;

[0100] The forgetting gate calculation process is as follows:

[0101] In the formula, f t represents the output of the forgetting gate, X t represents the input data, H t-1denotes the hidden state of the previous time, W xf and W hf denote the weight matrix corresponding to the forget gate, b f denotes the bias term parameter of the forget gate, and σ denotes the Sigmoid function.

[0102] The input gate calculation process is as follows:

[0103] In the formula, i t denotes the output of the input gate, W xi and W hi denote the weight matrix corresponding to the input gate, b i denotes the bias term parameter of the input gate.

[0104] The new cell state candidate value is calculated The formula used is as follows:

[0105] In the formula, W xc and W hc denote the weight matrix corresponding to the cell state, b c denotes the bias term parameter of the cell state.

[0106] The cell state update process is as follows:

[0107] In the formula, c t denotes the cell state of the current time, and c t-1 denotes the cell state of the previous time.

[0108] The output gate calculation process is as follows:

[0109] In the formula, o t denotes the output of the output gate, W xo and W ho denote the weight matrix corresponding to the output gate, b o denotes the bias term parameter of the output gate.

[0110] The hidden state calculation process is as follows: h t = o t ⊙tanh(c t ).

[0111] In the formula, tanh denotes the hyperbolic tangent function, and h t denotes the hidden state of the current time.

[0112] The bidirectional LSTM is calculated as follows:

[0113] wherein, denotes the forward direction hidden state at the current time, denotes the forward direction hidden state at the previous time, denotes the backward direction hidden state at the current time, denotes the backward direction hidden state at the previous time;

[0114] get the final output y t , the formula is as follows:

[0115] wherein, denotes the weight matrix connecting the forward state to the output, denotes the weight matrix connecting the forward state to the output, g denotes the activation function, and B y denotes the bias term parameter of the output layer;

[0116] iterative determination, if the anomaly detection model converges, stop the iterative training, get the intelligent control model, otherwise continue the iterative training.

[0117] Embodiment six, referring to FIG. 1 and FIG. 4, this embodiment is based on the above-mentioned embodiments, in the hyperparameter search module, the hyperparameter search specifically includes the following contents:

[0118] initialize parameters, fleet size N, search machine proportion P N , observation machine proportion S N , maximum iteration number T, search lower limit lb of dimension d d and upper limit ub d , early warning value R2, safety threshold ST;

[0119] initialize population, generate N sequence values, initialize random value z0, the formula is as follows:

[0120] wherein, denotes the d-th dimension i-th sequence value, denotes the d-th dimension i+1-th sequence value, β denotes the control parameter in the range of [0, 1], and rand denotes the random number in the range of [0, 1];

[0121] sequence value conversion search individual, the formula is expressed as follows:

[0122] wherein, denotes the initialization position of the d-th dimension of the i-th individual;

[0123] Calculating fitness value fi: the performance of the intelligent control model established based on the individual position is taken as the fitness value of the individual, the individuals are ranked from good to bad according to the fitness value, the search aircraft and the wingman are divided, and the individual f with the best fitness and its position X are recorded b and the position X b of the individual f with the second best fitness n and the position X n of the individual f with the worst fitness w and the position X w ;

[0124] The dynamic adaptive weight is obtained, and the formula is as follows: r(i) = θ(i) + R·rand; x(i) = r(i)·sin(θ(i)) / max(|r(i)·sin(θ(i))|); y(i) = r(i)·cos(θ(i)) / max(|r(i)·cos(θ(i))|);

[0125] In the formula, indicates the optimal individual position in the current iteration, α and R indicate the control parameters for the flight path, θ(i) and r(i) indicate the polar angle and polar diameter of the flight search, x(i) and y(i) indicate the position in polar coordinates, γ indicates the dynamic adaptive weight parameter, and t indicates the iteration number;

[0126] The position of the search aircraft is updated, and the formula is as follows:

[0127] In the formula, indicates the d-th dimension position of the i-th individual in the t-th generation swarm, indicates the d-th dimension position of the i-th individual in the t-th generation swarm, R2 indicates the early warning value, ST indicates the safety threshold, R2<ST indicates that there is no enemy aircraft nearby and it is in a safe position, R2≥ST indicates that there is an enemy aircraft nearby and it needs to land in another area, and L indicates a vector with all elements being 1;

[0128] The position of the wingman is updated, and the formula is as follows:

[0129] In the formula, n indicates the number of individuals, indicates the position of the individual with the best fitness value in the t+1 iteration of the swarm in dimension d, A + indicates the 1×D dimension pseudo-inverse matrix, and the elements are randomly assigned 1 or -1;

[0130] Randomly select 10% of the individuals in the swarm as observers to update the position, and the formula is as follows:

[0131] wherein, denotes the position of the individual with the best fitness in the swarm in the t-th iteration in dimension d, denotes the position of the individual with the worst fitness in the swarm in the t-th iteration in dimension d, r1 and δ denote both standard normal distributed random numbers with mean 0 and variance 1, r2 denotes a standard normal distributed random number with mean 0 and variance 0.7, and ∈ denotes a minimum constant to avoid a zero denominator;

[0132] Reflection operation, the search individual with poor fitness is selected for the reflection operation, and the formula used is as follows: X r = X m + α · (X b - X w ) ;

[0133] wherein, X m denotes the midpoint position, and X r denotes the reflection position;

[0134] Reflection point fitness value judgment, different position calculation methods are used to adjust the position of the search individual according to different fitness value relationships, and the formula used is as follows:

[0135] wherein, X ex denotes the position of the expansion point, X ec denotes the position of the outer shrink point, X ic denotes the position of the inner shrink point, and f r denotes the fitness value of the reflection point;

[0136] According to the adjustment of the position of the search individual, the effect is judged, and the formula used is as follows:

[0137] wherein, f ex denotes the fitness value of the expansion point, f ec denotes the fitness value of the outer shrink point, f ic denotes the fitness value of the inner shrink point, and if the judgment in the formula is not met, X w = X r ;

[0138] Search judgment, a fitness threshold is set in advance, when there is an individual fitness value higher than the fitness threshold, the hyperparameter search is completed, the individual position is the optimal parameter of the intelligent control model, at this time the intelligent control model is established, if the maximum iteration number is reached, the swarm position is reinitialized, otherwise the position is continuously updated.

[0139] By performing the above operation, for the general intelligent control model built-in parameter setting is improper, resulting in the instability of the intelligent control model output result, thereby resulting in the problem of inaccurate final intelligent control effect, the scheme creatively uses the mixed circle segmentation diagram, the spiral flight search method and the mirror reflection method to improve the algorithm for obtaining the built-in parameter of the intelligent control effect, obtains the optimal parameter combination, improves the stability of the intelligent control model output result, thereby solving the problem of inaccurate intelligent control model control effect.

[0140] Embodiment seven, refer to figure 1, this embodiment is based on the above-mentioned embodiment, in the real-time intelligent control module, the real-time intelligent control is based on the intelligent control model with set parameters, obtains real-time light control data, and based on the output of the established intelligent control model module as a control signal to intelligently control the current light.

[0141] It should be noted that in this paper, such as first and second relationship terms are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the term "include", "contain" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment.

[0142] Although the embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the present application.

[0143] The above describes the present application and its embodiments, which is not restrictive, and the drawings shown are only one of the embodiments of the present application, and the actual structure is not limited thereto. In general, if a person skilled in the art is inspired, without departing from the purpose of the present application, without creative design, similar structure and embodiments of the technical solution can be designed, which shall belong to the protection scope of the present application.

Claims

1. An intelligent lighting control system, characterized by: The system comprises a data acquisition module, a data preprocessing module, a feature selection module, an intelligent control model module, a hyperparameter search module and a real-time intelligent control module. The data acquisition module acquires digital signals in a light environment through a sensor, converts the digital signals into a digital form through an analog-to-digital converter, and obtains a light control original data set. The data preprocessing module performs data cleaning, data transformation and standardization processing on the collected data. The feature selection module selects features from historical light control data by combining a filtering method and a wrapping method, and obtains a representative feature subset. The intelligent control model module trains a machine learning algorithm based on the feature subset obtained by the feature selection module and the historical light control data, and obtains an intelligent control model. The hyperparameter search module obtains an optimal solution combination of model parameters by using a mixed circle segmentation diagram, a spiral flight search method and a mirror reflection method. The real-time intelligent control module intelligently controls the current light by monitoring real-time light control data based on the trained intelligent control model module.

2. The intelligent light control system of claim 1, wherein: In the feature selection module, the feature selection is used for feature selection of large-scale light control features, specifically, a combination of a filtering method and a wrapping method is used to select features from the preliminary light control data set to obtain an optimized feature data set, and the following steps are included: The filter objective function is computed, which is expressed as follows: where I(x, y) denotes the correlation between the two variables x and y, p(x) and p(y) denote the marginal probability distribution of the features x and y, respectively, p(x, y) denotes the joint probability distribution of the features x and y, fi denotes the filter objective function, R(S) denotes the redundancy of the feature subset S, D(S) denotes the dependency of the feature subset S, S denotes the size of the feature subset, i.e., the number of features included, x i and x j denote different features in the feature subset, I(x i , x j ) denotes the correlation between the feature x i and the feature x j , and I(x i , y) denotes the correlation between the feature x i and the target variable y; The computation of the wrapper objective function is formulated as follows: where f2represents a wrapper objective function to evaluate the accuracy of the selected features, n represents the number of samples of the feature subset, v represents the number of samples of the validation feature subset, y i,t represents the true value of the sample in the i-th validation subset at the t-th time step, represents the prediction value of the tth time step of the ith validation subset under a given feature subset S. The minimum objective function value is obtained, which is expressed by the following formula: In the formula, Minimize represents the minimum value of the filter and wrapper objective function. The feature selection is performed, and the best feature subset is selected as the final feature selection result according to Minimize, to obtain an optimized feature data set.

3. The intelligent light control system of claim 1, wherein: In the hyperparameter search module, the hyperparameter search includes the following steps: Initialization parameters, swarm size N, searchers ratio P N , observers ratio S N , maximum number of iterations T, search lower bound lb for dimension d d and upper bound ub d , warning value R2, safety threshold ST; Initialize the population, generate N sequence values, initialize random value z0, using the following formula: In the formulae, represents the i-th sequence value in the d-th dimension, represents the (i+1)th sequence value in the dth dimension, β represents a control parameter in the range of [0, 1], and rand represents a random number in the range of [0, 1]; The sequence value conversion search individual is expressed by the following formula: In the formulae, represents the initialization position of the ith individual in the dth dimension; Calculating fitness value fi: the performance of intelligent control model based on individual position is taken as the fitness value of individual, the individuals are ranked from good to bad according to fitness value, the search machine and wingman are divided, and the individual f with the best fitness is recorded b and its position X b , the individual f with the second best fitness n and its position X n , and the individual f with the third best fitness Individual f with worst fitness w and its position X w ; The dynamic adaptive weight is obtained, and the formula used is as follows: r(i)=θ(i)+R·rand; x(i)=r(i)·sin(θ(i)) / max(|r(i)·sin(θ(i))|); y(i) = r(i) - cos(0(i)) / max(|r(i) - cos(0(i))|); In the formulae, represents the optimal individual position in the current iteration, α and R represent control parameters of a flight path, θ(i) and r(i) represent a polar angle and a polar diameter of flight search, x(i) and y(i) represent positions in polar coordinates, γ represents a dynamic adaptive weight parameter, and t represents the number of iterations; The search machine position is updated using the following formula: In the formulae, represents the dthdimensional position of the ithindividual in the tthgeneration of the population, represents the dth dimension position of the ith individual in the tth generation swarm, R2 represents a warning value, ST represents a safety threshold, R2<ST represents that there is no enemy aircraft nearby and the position is safe, R2≥ST represents that there is an enemy aircraft nearby and the position needs to be changed for landing, and L represents a vector with all elements being 1. The wingman position is updated using the following formula: In the formula, n represents the number of individuals, A represents the position of the individual with the best fitness value in the machine population in the t+1 iteration in dimension d + A represents the position of the individual with the best fitness value in the machine population in the t+1 iteration in dimension d In the formulae, represents the dth dimension position of the individual with the best fitness in the tth iteration of the swarm Place, r1 and δ represent both standard normal distribution random numbers with mean 0 and difference 1, r2 indicates a standard normal distribution random number with mean 0 and variance 0.7, and ε represents a minimum constant to avoid a zero denominator; The reflection operation is performed on the search individual with poor fitness, and the formula used is as follows: X r = X m + a · (X b - X w ); wherein X m represents the midpoint position, X r represents the reflection position; The reflection point fitness value judgment adopts different position calculation modes to adjust the position of the search individual according to different fitness value relations, and the used formula is as follows: wherein X ex represents the position of the expansion point, X ec represents the position of the outer pinch point, X ic represents the position of the inner pinch point, f r represents the fitness value of the reflection point; According to the adjustment of the search individual position, the judgment effect, the used formula is as follows: where f ex represents the fitness value of the extension point, f ec represents the fitness value of the outer contraction point, f ic represents the fitness value of the inner contraction point, and represents X if the determination made in the formula does not hold w = X r ; The search determination is provided with an adaptive threshold value, when there is an individual adaptive value higher than the adaptive threshold value, the hyperparameter search is completed, the individual position is the optimal parameter of the intelligent control model, at this time, the intelligent control model is established, if the maximum iteration number is reached, the swarm position is reinitialized, otherwise the position is continuously updated.

4. The intelligent light control system of claim 1, wherein In the intelligent control model, the intelligent control model is established, specifically including the following contents: Initialize the model parameters, set the number of model hidden units, input vector dimension, output vector dimension and training iteration number; The forget gate computation process uses the following formula: In the formula, f t represents the output of the forget gate, X t represents the input data, H t-1 represents the hidden State, W xf and W hf denotes the weight matrix corresponding to the forget gate, b f denotes the bias term parameter of the forget gate, and σ denotes the Sigmoid function. The input gate calculation process uses the following formula: where i t denotes the output of the input gate, W xi and W hi denote the corresponding input gate weight matrix, b i denotes the bias term parameter of the input gate; Computing new cell element state candidate values The formula used is as follows: where W xc and W hc denote the corresponding cell state weight matrix, b c denotes the bias term parameter for the cell state. The cell unit state update process is expressed by the following formula: In the formula, c t represents the current time cell unit state, c t-1 represents the previous time cell unit state; The output gate calculation process uses the following formula: where o t represents the output of the output gate, W xo and W ho represent the weight matrix corresponding to the output gate, b o represents the bias term parameter of the output gate; The formula used in the hidden state calculation process is as follows: h t = o t ⊙ tanh(c t ); wherein tanh denotes the hyperbolic tangent function, h t denotes the hidden state at the current time instant; A bidirectional LSTM is computed, using the following formula: In the formulae, a forward direction hidden state representing a current time instant, denotes the forward direction hidden state of the previous time instant, a reverse direction hidden state representing a current time, r1 and δ represent both standard normal distribution random numbers with mean 0 and difference 1, r2 indicates a standard normal distribution random number with mean 0 and variance 0.7, and ε represents a minimum constant to avoid a zero denominator; obtaining a final output y t The formula used is as follows: In the formulae, a weight matrix representing the connection of the forward state to the output, denotes the weight matrix connecting the forward state to the output, g denotes the activation function, B y denotes the bias term parameter of the output layer; The iteration determination is used to stop iteration training if the anomaly detection model converges, and obtain the intelligent control model, otherwise continue iteration training.

5. The intelligent light control system of claim 1, wherein: In the data acquisition module, the data acquisition is used to acquire the original data required for light control, specifically in the current intelligent light control system, through acquisition, the light control original data set is obtained; The light control original data set specifically includes historical light control data and real-time light control data; the historical light control data and real-time light control data both include power voltage data of conduction angle, user interaction data, environmental parameter data and light energy consumption data; the user interaction data includes light setting history record and light use habit; the environmental parameter data includes illumination intensity, indoor and outdoor temperature data and indoor and outdoor humidity data; the light energy consumption data includes total power consumption, energy consumption per unit time and energy consumption comparison data after intelligent control; the real-time light control data further includes light equipment state data and user current activity data.

6. The intelligent light control system of claim 1, wherein: In the data preprocessing module, the data preprocessing is used to preprocess the original data, specifically including the following contents: Feature encoding is used to convert the original data into digital form, specifically the light control original data set is encoded through one-hot encoding method to obtain light control encoded data set; Data normalization is used to normalize the data, specifically the light control encoded data set is normalized through the minimum-maximum method to obtain the light control preliminary data set; Data preprocessing is performed, specifically the light control original data set is preprocessed through the feature encoding and the data normalization to obtain the light control preliminary data set.

7. The intelligent light control system of claim 1, wherein: In the real-time intelligent control module, the real-time intelligent control is based on the intelligent control model with set parameters, acquires real-time light control data, and outputs the established intelligent control model module as a control signal for intelligent control of the current light.

Images