Energy-saving and environmentally friendly control method and system for lamps based on AI algorithm

Abstract

The present application relates to the field of deep learning technology, and discloses a method and system for energy-saving and environmentally friendly control of lamps based on AI algorithms. The method includes: performing multi-dimensional data collection and preprocessing on target lamps to obtain a first multi-dimensional data set; performing fuzzy clustering analysis to obtain multiple lamp usage mode categories, and calculating the usage mode membership of each lamp usage mode category; performing projection analysis and redundancy elimination on the first multi-dimensional data set according to multiple lamp usage mode categories and usage mode memberships to obtain a second multi-dimensional data set; performing dynamic adjustment strategy analysis on the second multi-dimensional data set to generate a target dynamic adjustment strategy; predicting control parameters of the target lamp based on the target dynamic adjustment strategy to obtain dynamic adjustment control parameters; performing control path adjustment and parameter reinforcement learning optimization on the dynamic adjustment control parameters to obtain optimal adjustment control parameters. The present application improves the accuracy of energy-saving and environmentally friendly control of lamps.

Description

Technical Field

[0001] The present application relates to the field of deep learning technology, and in particular to an energy-saving and environmentally friendly control method and system for lamps based on AI algorithms. Background Art

[0002] As an important part of home lighting, lamps account for a considerable proportion of energy consumption. Traditional lamp control methods mainly rely on preset schedules and fixed brightness adjustment strategies, lacking dynamic adjustment capabilities, resulting in high energy consumption and unable to meet the changes in different scenarios and user needs. In addition, existing lamp control systems usually cannot fully utilize data from environmental sensors, user feedback, and device status, and fail to achieve intelligent energy-saving control.

[0003] The problems of the existing technology are as follows: fixed control strategies are difficult to cope with the ever-changing environmental conditions and user needs, resulting in waste of resources and poor user experience. The redundancy and complexity of data limit the performance of traditional control methods when processing high-dimensional data, making it difficult to extract valuable information from massive data. In other words, the accuracy of the existing technology is low. Summary of the invention

[0004] The present application provides a method and system for energy-saving and environmentally friendly control of lamps based on an AI algorithm, which is used to improve the accuracy of energy-saving and environmentally friendly control of lamps.

[0005] In a first aspect, the present application provides a lamp energy-saving and environmentally friendly control method based on an AI algorithm, and the lamp energy-saving and environmentally friendly control method based on an AI algorithm includes:

[0006] Performing multi-dimensional data collection and preprocessing on the target lamp to obtain a first multi-dimensional data set;

[0007] Performing fuzzy clustering analysis on the first multidimensional data set to obtain a plurality of lamp usage pattern categories, and calculating the usage pattern membership of each lamp usage pattern category;

[0008] According to the multiple lamp usage mode categories and the usage mode memberships, performing projection analysis and redundancy elimination on the first multidimensional data set to obtain a second multidimensional data set;

[0009] Performing dynamic adjustment strategy analysis on the second multidimensional data set to generate a target dynamic adjustment strategy;

[0010] Predicting control parameters of the target lamp based on the target dynamic adjustment strategy to obtain dynamic adjustment control parameters;

[0011] The control path is adjusted and the parameter reinforcement learning is optimized for the dynamic adjustment control parameters to obtain the optimal adjustment control parameters.

[0012] In a second aspect, the present application provides a lighting energy-saving and environmentally friendly control system based on an AI algorithm, and the lighting energy-saving and environmentally friendly control system based on an AI algorithm includes:

[0013] A collection module, used for collecting and preprocessing multi-dimensional data of the target lamp to obtain a first multi-dimensional data set;

[0014] a calculation module, configured to perform fuzzy clustering analysis on the first multidimensional data set to obtain a plurality of lamp usage pattern categories, and calculate a usage pattern membership degree of each lamp usage pattern category;

[0015] a redundancy elimination module, configured to perform projection analysis and redundancy elimination on the first multidimensional data set according to the plurality of lamp usage mode categories and the usage mode memberships, so as to obtain a second multidimensional data set;

[0016] An analysis module, configured to perform dynamic adjustment strategy analysis on the second multidimensional data set to generate a target dynamic adjustment strategy;

[0017] A prediction module, used to predict control parameters of the target lamp based on the target dynamic adjustment strategy to obtain dynamic adjustment control parameters;

[0018] The optimization module is used to perform control path adjustment and parameter reinforcement learning optimization on the dynamic adjustment control parameters to obtain the optimal adjustment control parameters.

[0019] In the technical solution provided by the present application, a high-quality first multidimensional data set is formed by collecting and preprocessing multidimensional data such as ambient brightness, usage frequency, energy consumption, user preferences and device status of the target lamp. Through data cleaning, normalization and set conversion, the accuracy and consistency of the data are ensured. The fuzzy clustering algorithm is used to analyze the first multidimensional data set, which can identify and divide multiple lamp usage mode categories and calculate the usage mode membership of each category. It can handle the fuzziness and uncertainty in the data and provide more accurate classification results, so as to better understand and predict the usage mode of the lamp. The first multidimensional data set is processed by projection analysis and redundancy elimination technology to generate a second multidimensional data set. Through feature extraction, principal component analysis and feature weight calculation, redundant information in the data is eliminated, the data dimension is reduced, and the efficiency and accuracy of data processing are improved. Through dynamic adjustment strategy analysis, a target dynamic adjustment strategy is generated according to different lamp usage modes and environmental scenarios. Using multi-scenario testing and strategy parameter initialization, the generated adjustment strategy can adapt to changes in the environment and user needs in real time to achieve intelligent energy-saving control. Based on the target dynamic adjustment strategy, the control parameters of the target lamp are predicted. Through the bidirectional and unidirectional long short-term memory networks, time series analysis and state prediction are performed to generate more accurate dynamic adjustment control parameters, which can better adjust the operating state of the lamps and optimize energy consumption. The optimal adjustment control parameters are obtained by adjusting the control path and optimizing the dynamic adjustment control parameters through reinforcement learning. Through environmental simulation, path planning and reward function calculation, the reinforcement learning algorithm can continuously optimize the control strategy to achieve the best energy-saving effect and user satisfaction, thereby improving the accuracy of energy-saving and environmental protection control of lamps. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0021] Figure 1 This is a schematic diagram of an embodiment of a lamp energy-saving and environmentally friendly control method based on an AI algorithm in an embodiment of the present application;

[0022] Figure 2 This is a schematic diagram of an embodiment of an energy-saving and environmentally friendly control system for lamps based on an AI algorithm in an embodiment of the present application. DETAILED DESCRIPTION

[0023] The embodiments of the present application provide a method and system for energy-saving and environmentally friendly control of lamps based on AI algorithms. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0024] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the energy-saving and environmentally friendly control method of a lamp based on an AI algorithm includes:

[0025] Step S101, performing multi-dimensional data collection and preprocessing on the target lamp to obtain a first multi-dimensional data set;

[0026] It is understandable that the execution subject of the present application can be a lighting energy-saving and environmental protection control system based on an AI algorithm, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0027] Specifically, the ambient brightness of the target lamp is collected through the light sensor to obtain the ambient brightness data. At the same time, the frequency of lamp use is collected for the target lamp to obtain its frequency of use data, and the energy consumption of the target lamp is analyzed to obtain energy consumption data. The user preference analysis of the target lamp is performed through the user feedback system to collect user preference data. The device status of the target lamp is monitored to monitor its operating status in real time to obtain device status data. Data cleaning and set conversion are performed on the ambient brightness data, frequency of use data, energy consumption data, user preference data and device status data respectively. The data cleaning process aims to remove noise and outliers to ensure the accuracy and reliability of the data, while the set conversion is to unify various types of data into a standard format for subsequent processing. After data cleaning and set conversion, various types of data are integrated into an initial multidimensional data set. The initial multidimensional data set is normalized to convert data of different dimensions to a unified scale to eliminate the impact caused by different data dimensions and make the data more comparable. After normalization, all data are scaled to the same range to obtain the first multidimensional data set. The specific steps of normalization processing include linearly transforming each type of data according to its maximum and minimum values ​​so that all data points fall within a predetermined interval (such as between 0 and 1).

[0028] Step S102, performing fuzzy clustering analysis on the first multidimensional data set to obtain a plurality of lamp usage pattern categories, and calculating the usage pattern membership of each lamp usage pattern category;

[0029] Specifically, the distance calculation is performed on the first multidimensional data set by the Euclidean distance algorithm to generate a similarity matrix. The Euclidean distance between each data point in the multidimensional data set is used to measure the similarity between the data points. The similarity matrix is ​​fuzzified to generate a fuzzy membership matrix. Each element in the similarity matrix is ​​converted into a membership value, which represents the possibility that each data point belongs to a different cluster center, thereby obtaining a fuzzy membership matrix. The initial cluster center is selected based on the fuzzy membership matrix to obtain an initial cluster center set. The selection of the initial cluster center is one of the key steps in fuzzy cluster analysis, and the rationality of its selection directly affects the accuracy and reliability of the final clustering result. The fuzzy C-means algorithm is applied to perform clustering iteration on the initial cluster center set to obtain the target cluster center set. The fuzzy C-means algorithm is a classic fuzzy clustering algorithm. By continuously iteratively updating the cluster center and adjusting the membership of the data point, it gradually approaches the optimal cluster center position and membership distribution until the convergence condition is met. The membership of the target cluster center set is calculated according to the membership function to obtain a membership data set. The calculation of the membership data set is based on the location of the target cluster center and the distance of each data point to these cluster centers. The calculated membership value represents the probability that each data point belongs to different cluster centers. The membership data set is divided into categories to obtain the first clustering result. Through the cluster optimization process, the cluster center and the category boundary are adjusted to obtain the second clustering result, and the accuracy and stability of clustering are improved, so that each data point can be more accurately attributed to its most suitable cluster center. The second clustering result is used to identify the use pattern through the pattern recognition algorithm to obtain multiple lamp use pattern categories. The pattern recognition algorithm uses the information in the second clustering result to identify and classify different use patterns to ensure that the obtained use pattern categories have high recognition and practicality. At the same time, the use pattern membership is calculated for multiple lamp use pattern categories to obtain the use pattern membership of each lamp use pattern category. By calculating the membership value of each lamp use pattern category, the belonging of each data point in different use pattern categories is clarified, and the entire fuzzy clustering analysis process is completed, and finally the clustering result and use pattern membership are obtained.

[0030] Step S103, performing projection analysis and redundancy elimination on the first multidimensional data set according to the multiple lamp usage mode categories and usage mode memberships to obtain a second multidimensional data set;

[0031] Specifically, feature extraction is performed on multiple lamp usage pattern categories to construct an initial feature vector. Feature weight calculation is performed on the initial feature vector according to the usage pattern membership to obtain a feature weight vector. The importance and weight of each feature are determined by weighted calculation of the membership of each feature in each usage pattern category. Principal component analysis is performed on the first multidimensional data set to obtain an initial principal component data set. The original multidimensional data set is converted into a set of new, unrelated principal components through linear transformation, and these principal components can retain the variation information of the original data set to the greatest extent, thereby simplifying the data structure and reducing the dimension. The feature weight of the initial principal component data set is adjusted according to the feature weight vector to obtain a weighted principal component data set. Linear projection is performed on the weighted principal component data set to obtain a linear projection data set. By mapping the weighted principal component data set to a new linear space, the data structure is further simplified while retaining as much original information as possible, so as to better discover the potential patterns and structures in the data set. Redundancy detection is performed on the linear projection data set to obtain a redundant information marker. Redundancy detection is to identify redundant information in the linear projection data set and mark those repeated or redundant data points so as to effectively eliminate redundancy in subsequent processing. According to the redundant information mark, the linear projection data set is de-redundanted to obtain the de-redundanted data set. The information marked as redundant in the linear projection data set is removed to simplify the data set structure and improve the compactness and effectiveness of the data set. The de-redundanted data set is subjected to data compaction processing and feature reconstruction to obtain the second multidimensional data set. Data compaction processing is to further optimize the data structure and ensure the simplicity and efficiency of the data set; feature reconstruction is to reconstruct the features of the data set so that it can accurately reflect the key information in the original data set.

[0032] Step S104, performing dynamic adjustment strategy analysis on the second multidimensional data set to generate a target dynamic adjustment strategy;

[0033] Specifically, the second multidimensional data set is subjected to scene classification according to multiple lamp usage mode categories to obtain scene classification results. The data set is divided into different usage scenarios, each of which represents a specific lamp usage mode. Feature selection is performed on the second multidimensional data set according to the scene classification results to obtain an adjustment feature data set, simplify the data set, highlight key features, and reduce computational complexity. Strategy parameters are initialized for the adjustment feature data set to obtain initial strategy parameters. Initial adjustment parameters are assigned to each feature, which will serve as the starting point for subsequent multi-scenario testing. Multi-scenario testing is performed on the initial strategy parameters to obtain target strategy parameters. The effects of the initial strategy parameters are tested and evaluated by simulating the operation of lamps in different usage scenarios. Through multiple iterations and optimizations, the strategy parameters are gradually adjusted so that the optimal effect can be achieved in various scenarios. A target dynamic adjustment strategy is generated according to the target strategy parameters. A specific adjustment plan is formulated according to the target strategy parameters obtained in the multi-scenario test. The strategy covers all possible usage scenarios to ensure that the lamps can achieve the best energy-saving effect and environmental adaptability under different conditions.

[0034] Step S105: predicting control parameters of the target lamp based on the target dynamic adjustment strategy to obtain dynamic adjustment control parameters;

[0035] Specifically, an adjustment feature vector is generated based on the target dynamic adjustment strategy, and the current state of the target lamp is monitored to obtain a current state data set. Feature matching is performed on the current state data set according to the adjustment feature vector to obtain matching feature data. The current state data is compared with the adjustment feature vector to find the feature that best matches the adjustment strategy, ensuring that the feature data used can accurately reflect the current operating state of the lamp. Time series analysis is performed on the matching feature data through a bidirectional long short-term memory network to obtain a time series hidden feature vector. The bidirectional long short-term memory network can simultaneously consider the context of the data, capture complex time dependencies, and extract richer time series information. The future state of the time series hidden feature vector is predicted through a unidirectional long short-term memory network to obtain state prediction sequence data. The unidirectional long short-term memory network focuses on the data at the current and past time points, predicts future state changes, predicts the future operating state of the lamp, and generates a state prediction sequence data containing multiple time points. Adjustment parameter mapping is performed on the state prediction sequence data to obtain dynamic adjustment control parameters. The predicted state data is converted into specific control parameters, which are used to adjust the operating state of the lamp in real time to achieve the best energy-saving effect and environmental adaptability.

[0036] Step S106: perform control path adjustment and parameter reinforcement learning optimization on the dynamic adjustment control parameters to obtain optimal adjustment control parameters.

[0037] Specifically, control path planning is performed for the dynamic adjustment control parameters to generate multiple preliminary control paths. Control path planning is to formulate a series of feasible adjustment paths according to the current state and expected goals of the lamp, and each path represents a possible adjustment scheme. Environmental simulation is performed on multiple preliminary control paths to obtain environmental simulation data for each preliminary control path. The operation of the lamp under different environmental conditions is simulated by simulation technology to evaluate the effect of each control path in actual application. The simulation data includes parameters of multiple aspects such as energy consumption, brightness change, and user satisfaction of the lamp. Based on the reinforcement learning algorithm, the environmental simulation data of each preliminary control path is interacted with the environment to obtain an interaction data set for each preliminary control path. The control path is repeatedly adjusted and optimized in the simulation environment by the reinforcement learning algorithm, so that each path can gradually approach the optimal state. After obtaining the interaction data set, the reward function is calculated for the interaction data set of each preliminary control path to obtain the reward function data for each preliminary control path. The reward function calculation is to score each interaction result according to a predetermined evaluation standard (such as energy saving effect, user comfort, etc.). These score values ​​are the reward function data, which reflects the pros and cons of each control path. According to the reward function data, multiple preliminary control paths are selected to obtain the target control path. The control path selection is to screen out the path with the best performance according to the reward function data. The target control path is the adjustment scheme that can best achieve energy saving, environmental protection and user needs. According to the target control path, the dynamic adjustment strategy parameters are state modeled to obtain the state model, and the dynamic adjustment strategy parameters are action defined to obtain the action model. State modeling is a mathematical description of the operating state of the lamp, and action definition is a concretization of possible adjustment operations. The two together constitute the basic model of reinforcement learning. According to the state model and action model, the dynamic adjustment control parameters are optimized to obtain the optimal adjustment control parameters of the target lamp. By repeatedly adjusting the control parameters using the reinforcement learning algorithm under the framework of the state model and action model, a set of control parameters that can achieve the best energy saving effect and user satisfaction under various environmental conditions is finally found.

[0038] In the embodiment of the present application, a high-quality first multidimensional data set is formed by collecting and preprocessing multidimensional data such as ambient brightness, usage frequency, energy consumption, user preferences and device status of the target lamp. The accuracy and consistency of the data are ensured by data cleaning, normalization and set conversion. The first multidimensional data set is analyzed by a fuzzy clustering algorithm, which can identify and divide multiple lamp usage mode categories and calculate the usage mode membership of each category. It can handle the fuzziness and uncertainty in the data and provide more accurate classification results, so as to better understand and predict the usage mode of the lamp. The first multidimensional data set is processed by projection analysis and redundancy elimination technology to generate a second multidimensional data set. By feature extraction, principal component analysis and feature weight calculation, redundant information in the data is eliminated, the data dimension is reduced, and the efficiency and accuracy of data processing are improved. Through dynamic adjustment strategy analysis, a target dynamic adjustment strategy is generated according to different lamp usage modes and environmental scenarios. By using multi-scenario testing and strategy parameter initialization, the generated adjustment strategy can adapt to changes in the environment and user needs in real time to achieve intelligent energy-saving control. Based on the target dynamic adjustment strategy, the control parameters of the target lamp are predicted. Through the bidirectional and unidirectional long short-term memory networks, time series analysis and state prediction are performed to generate more accurate dynamic adjustment control parameters, which can better adjust the operating state of the lamps and optimize energy consumption. The optimal adjustment control parameters are obtained by adjusting the control path and optimizing the dynamic adjustment control parameters through reinforcement learning. Through environmental simulation, path planning and reward function calculation, the reinforcement learning algorithm can continuously optimize the control strategy to achieve the best energy-saving effect and user satisfaction, thereby improving the accuracy of energy-saving and environmental protection control of lamps.

[0039] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0040] (1) Collect the ambient brightness of the target lamp through the light sensor to obtain ambient brightness data; collect the usage frequency of the target lamp to obtain usage frequency data; perform energy consumption analysis on the target lamp to obtain energy consumption data; perform user preference analysis on the target lamp through the user feedback system to obtain user preference data; perform device status monitoring on the target lamp to obtain device status data;

[0041] (2) Perform data cleaning and set conversion on the ambient brightness data, usage frequency data, energy consumption data, user preference data, and device status data to generate an initial multidimensional data set;

[0042] (3) Normalize the initial multidimensional data set to obtain the first multidimensional data set.

[0043] Specifically, the ambient brightness of the target lamp is collected through the light sensor to obtain the ambient brightness data. The light sensor is installed around the lamp to monitor the intensity change of the ambient light in real time. For example, in an office, the light sensor can record the ambient brightness value every minute, and these values ​​represent the changes in light at different time periods throughout the day. At the same time, the frequency of lamp use is collected for the target lamp to obtain the frequency of use data. For example, an intelligent lamp system can record the duration of the lamp on and the number of times it is turned on and off once an hour. Through long-term data accumulation, the usage habits and high-frequency usage periods of the lamp are analyzed. The energy consumption of the target lamp is analyzed to obtain energy consumption data. The collection of energy consumption data can be achieved through a smart meter or an energy consumption monitoring module to record the power consumption of the lamp under different usage states. For example, the power consumption of the lamp in different states such as on, off, and dimming is recorded every day. By analyzing these data, the energy consumption characteristics of the lamp are understood. The user preference analysis of the target lamp is performed through the user feedback system to obtain user preference data. The user feedback system can collect user preferences for lamp brightness, color temperature, switching time, etc. through questionnaires, mobile phone application feedback, etc. For example, users can adjust the brightness of lamps through mobile phone applications and give satisfaction ratings. These feedback data will be recorded and used to analyze user preferences and habits. At the same time, the target lamps are monitored for equipment status to obtain equipment status data. Equipment status monitoring uses sensors installed on lamps to monitor the working status of lamps in real time, such as current, voltage, temperature and other parameters. For example, when the temperature of the lamp rises abnormally, this state is recorded and an alarm is issued to prompt the user or maintenance personnel to check and handle it. Data cleaning and set conversion are performed on ambient brightness data, usage frequency data, energy consumption data, user preference data and equipment status data to generate an initial multidimensional data set. The data cleaning process includes removing noise data, filling missing values ​​and processing outliers. For example, if abnormally high or abnormally low values ​​appear when collecting ambient brightness data, these abnormal data are identified and eliminated through statistical methods. Set conversion is to convert data of different formats and sources into a unified format for subsequent processing. For example, different types of data such as ambient brightness data, usage frequency data, and energy consumption data are uniformly converted into a time series format to facilitate subsequent multidimensional data analysis. The initial multidimensional data set is normalized to obtain the first multidimensional data set. Data of different dimensions are converted to the same scale to eliminate the dimensional differences between the data. For example, the range of ambient brightness data from 0 to 1000 lux is normalized to the range of 0 to 1, and the range of usage frequency data from 0 to 24 hours is also normalized to the range of 0 to 1. Through normalization, all data are standardized, avoiding analysis deviations caused by different data dimensions, and improving data comparability and analysis accuracy.

[0044] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0045] (1) Using the Euclidean distance algorithm, calculate the distance of the first multidimensional data set to obtain a similarity matrix;

[0046] (2) Fuzzify the similarity matrix to obtain the fuzzy membership matrix;

[0047] (3) Select the initial cluster center of the fuzzy membership matrix to obtain the initial cluster center set;

[0048] (4) Using the fuzzy C-means algorithm, the initial cluster center set is clustered iteratively to obtain the target cluster center set;

[0049] (5) According to the membership function, the membership of the target cluster center set is calculated to obtain the membership data set;

[0050] (6) Classify the membership data set to obtain the first clustering result;

[0051] (7) Optimize the first clustering result, adjust the cluster center and category boundary, and obtain the second clustering result;

[0052] (8) Using a pattern recognition algorithm, the second clustering result is subjected to usage pattern recognition to obtain a plurality of lamp usage pattern categories;

[0053] (9) Calculate the usage mode membership of multiple lamp usage mode categories to obtain the usage mode membership of each lamp usage mode category.

[0054] Specifically, the Euclidean distance algorithm is used to calculate the distance of the first multidimensional data set to obtain a similarity matrix. Assume that there is a data set containing n data points, each of which has d features. The data points can be represented as . The Euclidean distance is defined as:

[0055] ;

[0056] By calculating the Euclidean distance between each pair of data points in the data set, an nxn similarity matrix is ​​constructed, where each element Represents data points and For example, suppose there is a dataset with 3 data points, each with 2 features:

[0057] ;

[0058] Compute the Euclidean distance between them:

[0059] ;

[0060] ;

[0061] ;

[0062] The obtained similarity matrix is:

[0063] ;

[0064] The similarity matrix is ​​fuzzified to obtain a fuzzy membership matrix. Fuzzification is to convert each distance value in the similarity matrix into a membership value, which is usually achieved using a membership function. A common membership function is the Gaussian membership function:

[0065] ;

[0066] in, is a tuning parameter that controls the decay rate of the membership value. Assume , fuzzify the similarity matrix above and get the fuzzy membership matrix:

[0067] ;

[0068] The initial cluster center is selected for the fuzzy membership matrix to obtain the initial cluster center set. The selection of the initial cluster center can be optimized by random selection or by some heuristic algorithm. Assume that two initial cluster centers are randomly selected:

[0069] ;

[0070] The initial cluster center set is clustered iteratively through the fuzzy C-means algorithm to obtain the target cluster center set. The goal of the fuzzy C-means algorithm is to update the cluster center and membership matrix by minimizing the objective function. The objective function is:

[0071] ;

[0072] in, It is a data point Cluster Center The membership degree is m, and m is the fuzzification parameter (usually m=2). The membership degree and cluster center are updated iteratively until convergence. Assume that after several iterations, the target cluster center set is:

[0073] ;

[0074] According to the membership function, the membership of the target cluster center set is calculated to obtain a membership data set. The membership function can also use the Gaussian membership function or other suitable function forms. The membership data set represents the possibility that each data point belongs to each cluster center. For example, for the data point

[0075] ;

[0076] The membership data set is divided into categories to obtain the first clustering result. Category division is to assign each data point to the corresponding cluster center according to the size of the membership value. For example, if ,but Belongs to the cluster center . Perform cluster optimization on the first clustering result, adjust the cluster center and category boundaries to obtain the second clustering result. Cluster optimization is to maximize the membership of each data point in its category by further iteration and adjustment of the cluster center position. For example, recalculate and Location:

[0077] ;

[0078] Through the pattern recognition algorithm, the second clustering result is subjected to usage pattern recognition to obtain multiple lamp usage pattern categories. The pattern recognition algorithm can use machine learning algorithms such as support vector machines and decision trees to classify the already divided clustering results and identify the usage pattern of each category. For example, based on the clustering results, "high-frequency usage pattern" and "low-frequency usage pattern" can be identified. The usage pattern membership is calculated for multiple lamp usage pattern categories to obtain the usage pattern membership of each lamp usage pattern category. The usage pattern membership indicates the probability of each lamp belonging to different usage pattern categories. For example, for lamp A, its membership in "high-frequency usage pattern" and "low-frequency usage pattern" is 0.7 and 0.3 respectively. Through these calculations and analyses, the identification and classification of lamp usage patterns are finally realized, providing data support and decision-making basis for energy-saving and environmental protection control.

[0079] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0080] (1) Extract features from multiple lighting usage mode categories and construct initial feature vectors;

[0081] (2) Calculate the feature weight of the initial feature vector according to the usage pattern membership to obtain a feature weight vector;

[0082] (3) Performing principal component analysis on the first multidimensional data set to obtain an initial principal component data set;

[0083] (4) Adjust the feature weights of the initial principal component data set according to the feature weight vector to obtain a weighted principal component data set;

[0084] (5) Perform linear projection on the weighted principal component data set to obtain a linear projection data set;

[0085] (6) Perform redundancy detection on the linear projection data set to obtain redundant information labels;

[0086] (7) Eliminate the redundancy of the linear projection data set according to the redundant information mark to obtain a redundant eliminated data set;

[0087] (8) Perform data compaction and feature reconstruction on the redundancy-eliminated data set to obtain a second multidimensional data set.

[0088] Specifically, feature extraction is performed on multiple lamp usage pattern categories to construct an initial feature vector. Assume that there are three usage pattern categories: high-frequency usage pattern, medium-frequency usage pattern, and low-frequency usage pattern. Each pattern category has several features, such as usage frequency, energy consumption, user preference, etc. These features are extracted from each category and an initial feature vector is constructed. For example, for the high-frequency usage pattern, its feature vector can be expressed as ,in, Indicates the frequency of use, Indicates energy consumption, Indicates user preferences, Indicates the device status. Similarly, the feature vectors of the medium-frequency usage mode and the low-frequency usage mode are expressed as and . The feature weight vector is obtained by calculating the feature weight of the initial feature vector according to the usage pattern membership. The usage pattern membership represents the degree of belonging of each data point to different pattern categories. For example, suppose that the membership of a data point to high-frequency, medium-frequency and low-frequency patterns is 0.6, 0.3 and 0.1 respectively. The feature weight calculation formula is:

[0089] ;

[0090] in, represents the degree of membership of the data point to the i-th mode category, Represents the feature vector of the first pattern category. For the above example, the feature weight vector w is calculated as follows:

[0091] ;

[0092] The first multidimensional data set is subjected to principal component analysis to obtain an initial principal component data set. The goal of principal component analysis is to transform the original data set into a new coordinate system so that each axis (principal component) in the new coordinate system is a linear combination of the original data set and the principal components are mutually orthogonal. Principal component analysis achieves this goal by solving the eigenvalues ​​and eigenvectors of the covariance matrix. The initial principal component data set is subjected to feature weight adjustment according to the feature weight vector to obtain a weighted principal component data set. Feature weight adjustment is to apply the feature weight vector to the initial principal component data set to reflect the importance of each feature in the principal component. The weighted principal component data set is linearly projected to obtain a linear projection data set. The weighted principal component data set is mapped into a new low-dimensional space to further simplify the data structure. Redundancy detection is performed on the linear projection data set to obtain a redundant information marker. Redundant information is marked by analyzing the correlation between the features in the linear projection data set. Common methods include calculating the correlation coefficient matrix between features and identifying highly correlated feature pairs. For example, if the correlation coefficient of two features is close to 1, one of the features can be considered redundant. According to the redundant information mark, the linear projection data set is de-redundanted to obtain a de-redundanted data set. The information marked as redundant is removed to reduce the dimension of the data set and improve the compactness and effectiveness of the data. The de-redundanted data set is subjected to data compaction processing and feature reconstruction to obtain a second multidimensional data set. Data compaction processing is to further optimize the data structure to ensure the simplicity and efficiency of the data. Feature reconstruction is to reconstruct the relationship between features so that the data set can accurately reflect the key information in the original data. For example, by recalculating the weighted and combined features, a new feature vector is constructed to obtain a second multidimensional data set.

[0093] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0094] (1) classifying the second multidimensional data set according to the multiple lamp usage mode categories to obtain a scene classification result;

[0095] (2) performing feature selection on the second multidimensional data set according to the scene classification results to obtain an adjusted feature data set;

[0096] (3) Initialize the strategy parameters of the adjustment feature data set to obtain the initial strategy parameters;

[0097] (4) Conduct multi-scenario tests on the initial strategy parameters to obtain the target strategy parameters;

[0098] (5) Generate a target dynamic adjustment strategy based on the target strategy parameters.

[0099] Specifically, the second multidimensional data set is subjected to scene classification according to multiple lamp usage mode categories to obtain scene classification results. For example, these categories may include high-frequency usage mode, low-frequency usage mode, and energy-saving mode. Each usage mode category reflects different usage scenarios. For example, high-frequency usage mode may correspond to office daytime usage, while low-frequency usage mode may correspond to nighttime usage. By mapping each data point in the second multidimensional data set to these usage mode categories, the entire data set is subjected to scene classification. Assuming that there are 1,000 data points in the second multidimensional data set, they are classified into high-frequency usage mode, low-frequency usage mode, and energy-saving mode through analysis, and the scene classification results obtained are: 400 data points for high-frequency usage mode, 300 data points for low-frequency usage mode, and 300 data points for energy-saving mode. Feature selection is performed on the second multidimensional data set according to the scene classification results to obtain an adjustment feature data set. The most critical features for lamp control are extracted from the data set, and these features will directly affect the formulation and execution of the adjustment strategy. Assume that the data set contains multiple features, such as ambient brightness, usage frequency, energy consumption, user preference, and device status. Through feature selection, the most important features in each scene are determined. For example, in the high-frequency usage mode, the ambient brightness and the frequency of use may be the most important features, while in the energy-saving mode, the energy consumption and the device status may be the most critical features. Analyze the classification results of each scene, select the features that have the greatest impact on the adjustment strategy in each scene, and obtain the adjustment feature data set. Initialize the policy parameters of the adjustment feature data set to obtain the initial policy parameters. Assign initial adjustment parameters to each feature, which will serve as the starting point for subsequent multi-scenario testing. The rationality of the initialization policy parameters directly affects the final effect of the adjustment strategy, and they are set according to specific application scenarios and experience values. For example, in the high-frequency usage mode, the initial setting of the adjustment parameters of the ambient brightness is 0.8, and the adjustment parameters of the frequency of use is 0.6; in the energy-saving mode, the adjustment parameters of the energy consumption are initially set to 0.9, and the adjustment parameters of the device status are 0.7. Through these initial parameter settings, a reasonable starting point is provided for subsequent policy optimization. Perform multi-scenario tests on the initial policy parameters to obtain the target policy parameters. Test and evaluate the effect of the initial policy parameters by simulating the operation of lamps in different usage scenarios. Through multiple iterations and optimizations, the policy parameters are gradually adjusted to achieve the best results in various scenarios. Suppose there is a simulation environment that can simulate different office usage scenarios, such as daytime, nighttime, holidays, etc. By running the initial policy parameters in these simulation environments, recording indicators such as the energy consumption, brightness changes, and user satisfaction of the lamps, each parameter is gradually adjusted until the optimal parameter combination is found. For example, in a simulation test of a high-frequency usage mode, it may be found that adjusting the adjustment parameter of the ambient brightness to 0.7 and the adjustment parameter of the usage frequency to 0.5 can achieve better energy saving and user satisfaction.In this way, the target policy parameters are finally obtained. The target dynamic adjustment strategy is generated according to the target policy parameters. A specific adjustment plan is formulated according to the target policy parameters obtained in the multi-scenario test. The strategy will include all possible usage scenarios to ensure that the lamps can achieve the best energy-saving effect and environmental adaptability under different conditions. Assume that the target policy parameters are the ambient brightness adjustment parameter 0.7 and the usage frequency adjustment parameter 0.5 for the high-frequency usage mode; the energy consumption adjustment parameter 0.8 and the device status adjustment parameter 0.6 for the low-frequency usage mode. Based on these parameters, a specific adjustment strategy is generated. For example, when the ambient brightness is lower than a certain threshold, the brightness of the lamp is automatically adjusted to 70%; when the usage frequency is higher than a certain threshold, the brightness is reduced to 50% to save energy. In the energy-saving mode, when it is detected that the energy consumption exceeds a certain level, the brightness is automatically reduced or some lamps are turned off; when the device status is abnormal, an alarm is issued and maintenance is performed.

[0100] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0101] (1) Generate an adjustment feature vector based on the target dynamic adjustment strategy, and monitor the current state of the target lamp to obtain the current state data set;

[0102] (2) Perform feature matching on the current state data set according to the adjusted feature vector to obtain matching feature data;

[0103] (3) Perform time series analysis on the matching feature data through a bidirectional long short-term memory network to obtain a time series hidden feature vector;

[0104] (4) Predict the future state of the time series hidden feature vector through a unidirectional long short-term memory network to obtain state prediction sequence data;

[0105] (5) Map the adjustment parameters of the state prediction sequence data to obtain dynamic adjustment control parameters.

[0106] Specifically, an adjustment feature vector is generated based on the target dynamic adjustment strategy, and the current state of the target lamp is monitored to obtain the current state data set. Assume that the target dynamic adjustment strategy includes adjusting four features: ambient brightness, usage frequency, energy consumption and device status. These features constitute the adjustment feature vector. Sensors and monitoring devices are set in the system to monitor the operating status of the target lamp in real time and collect the current ambient brightness, usage frequency, energy consumption and device status data. For example, an intelligent office lamp control system can collect data every one minute, including ambient brightness (such as 300 lux), usage frequency (such as 8 hours of operation), energy consumption (such as 50 watt-hours) and device status (such as normal). According to the adjustment feature vector, feature matching is performed on the current state data set to obtain matching feature data. The current state data is compared with the adjustment feature vector to extract feature data related to the adjustment strategy. In this way, it is ensured that the analyzed and processed data set contains the most critical features for the control strategy. The matching feature data is analyzed in time series through a bidirectional long short-term memory network to obtain a time series hidden feature vector. The bidirectional long short-term memory network can simultaneously consider the context of the data, capture complex time dependencies, and extract richer time series information. Assuming that there is matching feature data within a period of time, including ambient brightness, usage frequency, energy consumption and device status in the past 24 hours, the model inputs this data and obtains time series hidden feature vectors through forward and backward analysis. These hidden feature vectors contain the trend and pattern of the current feature in the entire time series. The future state of the time series hidden feature vector is predicted through a unidirectional long short-term memory network to obtain state prediction sequence data. The unidirectional long short-term memory network focuses on the data at the current and past time points to predict future state changes. Assuming that the state of the lamp in the next 4 hours is predicted, the model inputs the time series hidden feature vector, and generates state prediction sequence data in the next 4 hours by learning and analyzing the feature data in the past 24 hours. These prediction data include the predicted values ​​of ambient brightness, usage frequency, energy consumption and device status every hour in the future. The state prediction sequence data is mapped with adjustment parameters to obtain dynamic adjustment control parameters. The predicted state data is converted into specific control parameters, which are used to adjust the operating state of the lamp in real time to achieve the best energy saving effect and environmental adaptability. Assume that in the forecast data for the next 4 hours, the predicted value of ambient brightness gradually decreases from 300 lux to 200 lux, the predicted value of energy consumption remains at 50 watt-hours, the predicted value of usage frequency gradually increases from 8 hours to 12 hours, and the device status remains normal. Based on these forecast data, dynamic adjustment control parameters are generated. For example, when the ambient brightness is lower than 250 lux, the brightness of the lamp is automatically increased; when the usage frequency is higher than 10 hours, the brightness or frequency of the lamp is reduced; when the energy consumption remains stable and the device status is normal, the operation mode of the lamp is optimized to achieve energy saving.

[0107] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0108] (1) Perform control path planning for dynamically adjusted control parameters to obtain multiple preliminary control paths;

[0109] (2) performing environmental simulation on multiple preliminary control paths to obtain environmental simulation data for each preliminary control path;

[0110] (3) Based on the reinforcement learning algorithm, the environmental simulation data of each preliminary control path is subjected to environmental interaction to obtain an interaction data set for each preliminary control path;

[0111] (4) Calculate the reward function for the interaction data set of each preliminary control path to obtain the reward function data of each preliminary control path;

[0112] (5) Selecting a control path from multiple preliminary control paths based on the reward function data to obtain a target control path;

[0113] (6) Performing state modeling on the dynamic adjustment strategy parameters according to the target control path to obtain a state model, and performing action definition on the dynamic adjustment strategy parameters to obtain an action model;

[0114] (7) Based on the state model and the action model, the dynamic adjustment control parameters are optimized to obtain the optimal adjustment control parameters of the target lamp.

[0115] Specifically, control path planning is performed for the dynamic adjustment control parameters to obtain multiple preliminary control paths. Control path planning is to plan a series of possible adjustment paths based on the current dynamic adjustment control parameters. These paths represent the adjustment schemes of lamps under different environmental conditions. Assume that there is an office lamp system that needs to adjust the brightness and switch state of the lamps according to parameters such as ambient brightness, frequency of use and energy consumption. Design multiple preliminary control paths, such as path A gradually reduces the brightness, path B keeps the brightness unchanged, and path C gradually increases the brightness. Perform environmental simulation on multiple preliminary control paths to obtain environmental simulation data for each preliminary control path. Use simulation technology to simulate the operation of different control paths in the actual environment and evaluate the effect of each path. Assume that simulation software is used to simulate different time periods in an office during the day, and record parameters such as energy consumption, brightness changes and user satisfaction of lamps. For path A, you may see a reduction in energy consumption due to gradually reducing the brightness, but user satisfaction may decrease; for path B, the brightness remains unchanged, and energy consumption and user satisfaction remain at a high level; for path C, the brightness increases, energy consumption increases, but user satisfaction increases. Based on the reinforcement learning algorithm, the environmental simulation data of each preliminary control path is interacted with the environment to obtain the interaction data set of each preliminary control path. The reinforcement learning algorithm gradually approaches the optimal state by repeatedly adjusting and optimizing the control path in the simulation environment. Each control path is evaluated and optimized using a reinforcement learning algorithm such as a Q learning algorithm or a deep Q network. Each interaction process records the state and feedback information of the lamp under different environmental conditions to generate an interaction data set. After obtaining the interaction data set, the reward function is calculated for the interaction data set of each preliminary control path to obtain the reward function data of each preliminary control path. The reward function calculation is to score each interaction result according to a predetermined evaluation standard (such as energy saving effect, user comfort, etc.). These score values ​​are the reward function data, which reflects the pros and cons of each control path. According to the reward function data, the control path selection is performed on multiple preliminary control paths to obtain the target control path. The control path selection is to screen out the path with the best performance according to the reward function data. Assume that the reward value of path A is 50, the reward value of path B is 70, and the reward value of path C is 60. According to the reward value, path B is selected as the target control path because it performs best in terms of energy saving and user comfort. According to the target control path, the dynamic adjustment strategy parameters are state modeled to obtain the state model, and the dynamic adjustment strategy parameters are action defined to obtain the action model. State modeling is a mathematical description of the operating state of the lamp, and action definition is a concretization of possible adjustment operations. By defining the state transfer function and reward function, the operating mechanism and adjustment strategy of the lamp in different states are described. According to the state model and action model, the dynamic adjustment control parameters are optimized to obtain the optimal adjustment control parameters of the target lamp.By repeatedly adjusting the control parameters using a reinforcement learning algorithm under the framework of the state model and action model, we can eventually find a set of control parameters that can achieve the best energy-saving effect and user satisfaction under various environmental conditions. For example, through the optimization algorithm, we may find that when the ambient brightness is lower than 200 lux, we increase the brightness to 250 lux; when the usage frequency is higher than 10 hours, we reduce the brightness to 200 lux to save energy.

[0116] The above describes the energy-saving and environmentally friendly control method of lamps based on AI algorithm in the embodiment of the present application. The following describes the energy-saving and environmentally friendly control system of lamps based on AI algorithm in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the energy-saving and environmentally friendly control system for lamps based on AI algorithm includes:

[0117] The acquisition module 201 is used to collect and preprocess multi-dimensional data of the target lamp to obtain a first multi-dimensional data set;

[0118] A calculation module 202, configured to perform fuzzy clustering analysis on the first multidimensional data set to obtain a plurality of lamp usage pattern categories, and calculate a usage pattern membership degree of each lamp usage pattern category;

[0119] A redundancy elimination module 203, configured to perform projection analysis and redundancy elimination on the first multidimensional data set according to a plurality of lamp usage mode categories and usage mode memberships, to obtain a second multidimensional data set;

[0120] An analysis module 204 is used to perform dynamic adjustment strategy analysis on the second multidimensional data set to generate a target dynamic adjustment strategy;

[0121] Prediction module 205, used to predict control parameters of target lamps based on target dynamic adjustment strategy to obtain dynamic adjustment control parameters;

[0122] The optimization module 206 is used to perform control path adjustment and parameter reinforcement learning optimization on the dynamic adjustment control parameters to obtain the optimal adjustment control parameters.

[0123] Through the cooperation of the above components, the multi-dimensional data of the target lamps, such as ambient brightness, usage frequency, energy consumption, user preferences and device status, are collected and preprocessed to form a high-quality first multi-dimensional data set. Through data cleaning, normalization and set conversion, the accuracy and consistency of the data are ensured. The fuzzy clustering algorithm is used to analyze the first multi-dimensional data set, which can identify and divide multiple lamp usage mode categories and calculate the usage mode membership of each category. It can handle the fuzziness and uncertainty in the data and provide more accurate classification results, so as to better understand and predict the usage mode of lamps. The first multi-dimensional data set is processed by projection analysis and redundancy elimination technology to generate a second multi-dimensional data set. Through feature extraction, principal component analysis and feature weight calculation, redundant information in the data is eliminated, the data dimension is reduced, and the efficiency and accuracy of data processing are improved. Through dynamic adjustment strategy analysis, the target dynamic adjustment strategy is generated according to different lamp usage modes and environmental scenarios. Using multi-scenario testing and strategy parameter initialization, the generated adjustment strategy can adapt to changes in the environment and user needs in real time to achieve intelligent energy-saving control. Based on the target dynamic adjustment strategy, the control parameters of the target lamp are predicted. Through the bidirectional and unidirectional long short-term memory networks, time series analysis and state prediction are performed to generate more accurate dynamic adjustment control parameters, which can better adjust the operating state of the lamps and optimize energy consumption. The optimal adjustment control parameters are obtained by adjusting the control path and optimizing the dynamic adjustment control parameters through reinforcement learning. Through environmental simulation, path planning and reward function calculation, the reinforcement learning algorithm can continuously optimize the control strategy to achieve the best energy-saving effect and user satisfaction, thereby improving the accuracy of energy-saving and environmental protection control of lamps.

[0124] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0125] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0126] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A lamp energy-saving and environmental protection control method based on AI algorithm, characterized in that: The method comprises: Performing multi-dimensional data collection and preprocessing on the target lamp to obtain a first multi-dimensional data set; Performing fuzzy clustering analysis on the first multidimensional data set to obtain a plurality of lamp usage pattern categories, and calculating the usage pattern membership of each lamp usage pattern category; According to the multiple lamp usage mode categories and the usage mode membership, the first multidimensional data set is projected and analyzed and redundancy is eliminated to obtain a second multidimensional data set; specifically including: extracting features from the multiple lamp usage mode categories to construct an initial feature vector; calculating feature weights of the initial feature vector according to the usage mode membership to obtain a feature weight vector; performing principal component analysis on the first multidimensional data set to obtain an initial principal component data set; adjusting feature weights of the initial principal component data set according to the feature weight vector to obtain a weighted principal component data set; performing linear projection on the weighted principal component data set to obtain a linear projection data set; performing redundancy detection on the linear projection data set to obtain a redundant information tag; performing redundancy elimination on the linear projection data set according to the redundant information tag to obtain a redundant elimination data set; performing data compaction processing and feature reconstruction on the redundant elimination data set to obtain a second multidimensional data set; Performing dynamic adjustment strategy analysis on the second multidimensional data set to generate a target dynamic adjustment strategy; Predicting control parameters of the target lamp based on the target dynamic adjustment strategy to obtain dynamic adjustment control parameters; The control path adjustment and parameter reinforcement learning optimization are performed on the dynamic adjustment control parameters to obtain the optimal adjustment control parameters; specifically including: control path planning for the dynamic adjustment control parameters to obtain multiple preliminary control paths; environmental simulation is performed on the multiple preliminary control paths to obtain environmental simulation data for each preliminary control path; environmental interaction is performed on the environmental simulation data of each preliminary control path based on the reinforcement learning algorithm to obtain an interaction data set for each preliminary control path; reward function calculation is performed on the interaction data set of each preliminary control path to obtain reward function data for each preliminary control path; control path selection is performed on the multiple preliminary control paths according to the reward function data to obtain a target control path; state modeling is performed on the dynamic adjustment strategy parameters according to the target control path to obtain a state model, and action definition is performed on the dynamic adjustment strategy parameters to obtain an action model; parameter optimization is performed on the dynamic adjustment control parameters according to the state model and the action model to obtain the optimal adjustment control parameters for the target lamp.

2. The energy-saving and environmentally friendly control method for lamps based on AI algorithm according to claim 1 is characterized in that: The multi-dimensional data collection and preprocessing of the target lamps to obtain a first multi-dimensional data set includes: The ambient brightness of the target lamp is collected through a light sensor to obtain ambient brightness data; the usage frequency of the target lamp is collected to obtain usage frequency data; the energy consumption of the target lamp is analyzed to obtain energy consumption data; the user preference of the target lamp is analyzed through a user feedback system to obtain user preference data; the device status of the target lamp is monitored to obtain device status data; Respectively performing data cleaning and set conversion on the ambient brightness data, the usage frequency data, the energy consumption data, the user preference data, and the device status data to generate an initial multidimensional data set; The initial multidimensional data set is normalized to obtain a first multidimensional data set.

3. The energy-saving and environmentally friendly control method for lamps based on AI algorithm according to claim 1 is characterized in that: The performing fuzzy cluster analysis on the first multidimensional data set to obtain a plurality of lamp usage pattern categories, and calculating the usage pattern membership of each lamp usage pattern category, includes: Performing distance calculation on the first multidimensional data set by using a Euclidean distance algorithm to obtain a similarity matrix; Performing fuzzy processing on the similarity matrix to obtain a fuzzy membership matrix; Performing initial cluster center selection on the fuzzy membership matrix to obtain an initial cluster center set; Performing clustering iteration on the initial cluster center set by using a fuzzy C-means algorithm to obtain a target cluster center set; According to the membership function, the membership of the target cluster center set is calculated to obtain a membership data set; Classify the membership data set into categories to obtain a first clustering result; Performing cluster optimization on the first clustering result, adjusting cluster centers and category boundaries, and obtaining a second clustering result; Using a pattern recognition algorithm, performing usage pattern recognition on the second clustering result to obtain a plurality of lamp usage pattern categories; The usage mode membership is calculated for the multiple lamp usage mode categories to obtain the usage mode membership of each lamp usage mode category.

4. The energy-saving and environmentally friendly control method for lamps based on AI algorithm according to claim 1 is characterized in that: The step of performing dynamic adjustment strategy analysis on the second multidimensional data set to generate a target dynamic adjustment strategy includes: Performing scene classification on the second multidimensional data set according to the multiple lamp usage mode categories to obtain a scene classification result; Performing feature selection on the second multidimensional data set according to the scene classification result to obtain an adjusted feature data set; Initializing the strategy parameters of the adjustment feature data set to obtain initial strategy parameters; Performing multi-scenario tests on the initial strategy parameters to obtain target strategy parameters; A target dynamic adjustment strategy is generated according to the target strategy parameters.

5. The energy-saving and environmentally friendly control method for lamps based on AI algorithm according to claim 1, characterized in that: The predicting of control parameters of the target lamp based on the target dynamic adjustment strategy to obtain dynamic adjustment control parameters includes: Generate an adjustment feature vector based on the target dynamic adjustment strategy, and monitor the current state of the target lamp to obtain a current state data set; Performing feature matching on the current state data set according to the adjustment feature vector to obtain matching feature data; Performing time series analysis on the matching feature data through a bidirectional long short-term memory network to obtain a time series hidden feature vector; Performing future state prediction on the time series hidden feature vector through a unidirectional long short-term memory network to obtain state prediction sequence data; The state prediction sequence data is subjected to adjustment parameter mapping to obtain dynamic adjustment control parameters.

6. A lighting energy-saving and environmental protection control system based on AI algorithm, characterized in that: The system comprises: A collection module, used for collecting and preprocessing multi-dimensional data of the target lamp to obtain a first multi-dimensional data set; a calculation module, configured to perform fuzzy clustering analysis on the first multidimensional data set to obtain a plurality of lamp usage pattern categories, and calculate a usage pattern membership degree of each lamp usage pattern category; A redundancy elimination module is used to perform projection analysis and redundancy elimination on the first multidimensional data set according to the multiple lamp usage mode categories and the usage mode membership to obtain a second multidimensional data set; specifically comprising: extracting features from the multiple lamp usage mode categories to construct an initial feature vector; calculating feature weights on the initial feature vector according to the usage mode membership to obtain a feature weight vector; performing principal component analysis on the first multidimensional data set to obtain an initial principal component data set; adjusting feature weights on the initial principal component data set according to the feature weight vector to obtain a weighted principal component data set; performing linear projection on the weighted principal component data set to obtain a linear projection data set; performing redundancy detection on the linear projection data set to obtain a redundant information tag; performing redundancy elimination on the linear projection data set according to the redundant information tag to obtain a redundant elimination data set; performing data compaction processing and feature reconstruction on the redundant elimination data set to obtain a second multidimensional data set; An analysis module, configured to perform dynamic adjustment strategy analysis on the second multidimensional data set to generate a target dynamic adjustment strategy; A prediction module, used to predict control parameters of the target lamp based on the target dynamic adjustment strategy to obtain dynamic adjustment control parameters; An optimization module is used to perform control path adjustment and parameter reinforcement learning optimization on the dynamic adjustment control parameters to obtain the optimal adjustment control parameters; specifically comprising: performing control path planning on the dynamic adjustment control parameters to obtain multiple preliminary control paths; performing environmental simulation on the multiple preliminary control paths to obtain environmental simulation data for each preliminary control path; performing environmental interaction on the environmental simulation data of each preliminary control path based on a reinforcement learning algorithm to obtain an interaction data set for each preliminary control path; performing reward function calculation on the interaction data set for each preliminary control path to obtain reward function data for each preliminary control path; performing control path selection on the multiple preliminary control paths according to the reward function data to obtain a target control path; performing state modeling on the dynamic adjustment strategy parameters according to the target control path to obtain a state model, and performing action definition on the dynamic adjustment strategy parameters to obtain an action model; performing parameter optimization solution on the dynamic adjustment control parameters according to the state model and the action model to obtain the optimal adjustment control parameters for the target lamp.

Images