Remote energy-saving illumination management and control system and method for landscape garden

Abstract

The invention provides an illumination remote energy-saving management and control system and method for a landscape garden, and belongs to the technical field of illumination remote energy-saving prediction and management and control. S2, data transmission; s3, data preprocessing; s4, constructing and training an illumination energy consumption prediction model; s5, generating a real-time energy-saving strategy decision; s6, a control instruction is issued and executed; and S7, performing feedback optimization and model updating. According to the method, by fusing the curved surface fitting, the self-organizing mapping and the mixed density network, the nonlinear relation between the environment data (illumination intensity, people flow, time and the like) and the illumination energy consumption is fully mined, the illumination energy consumption in different scenes can be predicted more accurately, the prediction error is reduced, and a reliable basis is provided for formulating an energy-saving strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of remote energy-saving prediction and control of lighting, and in particular to a remote energy-saving control system and method for lighting used in landscape gardens. Background Art

[0002] In the field of landscape garden lighting management, traditional lighting control methods usually use timed control or simple manual control. The timed control method only turns lighting equipment on or off according to fixed time settings, and cannot be flexibly adjusted according to actual environmental changes (such as cloudy or sunny weather, high or low traffic volume). For example, when the light intensity is low on cloudy days, the lights are still turned on at fixed times, or the lights are continuously turned on at full brightness during periods of minimal traffic at night, resulting in a large amount of energy waste. Manual control relies on manual on-site operation, which is not only inefficient but also difficult to achieve remote and unified management. It cannot meet the needs of intelligent and refined management of modern landscape gardens.

[0003] With the development of technology, some lighting systems have introduced single sensors (such as light intensity sensors) for control. However, this approach relies too much on single-dimensional data and lacks comprehensive perception of the environment. For example, if lighting is controlled solely based on light intensity, even if there is sufficient light during the day when there is a lot of traffic, the lighting may be turned off due to the inability to sense the flow of people, affecting the visitor experience. At night, when the flow of people is sparse, the lighting brightness cannot be reduced according to actual needs, resulting in high energy consumption. In addition, most existing lighting energy consumption prediction models use simple linear regression or rule-based methods, which make it difficult to accurately capture the complex nonlinear relationship between landscape lighting energy consumption and multi-dimensional environmental factors, resulting in a lack of precision in the formulation of energy-saving strategies. Summary of the Invention

[0004] The present invention provides a remote energy-saving control system and method for landscape garden lighting to solve one or more of the above problems.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A remote energy-saving control method for landscape garden lighting, comprising:

[0007] S1. Real-time collection of multi-dimensional environmental data: Light intensity sensors, pedestrian flow sensors, and time sensors deployed in landscape gardens collect real-time ambient light intensity data, pedestrian flow data, and time data.

[0008] S2. Data transmission: The collected light intensity data, pedestrian flow data, and time data are transmitted to the remote control center server via a wireless network;

[0009] S3. Data preprocessing: Preprocess the data transmitted to the remote control center server to remove outliers and fill in missing values ​​to form a standardized data set;

[0010] S4. Construction and Training of a Lighting Energy Consumption Prediction Model: Surface fitting is used to perform nonlinear fitting between the features in the standard dataset and the corresponding energy consumption values ​​to generate preliminary energy consumption predictions. These preliminary energy consumption predictions and the features in the standard dataset are combined into a high-dimensional vector, which is then input into a self-organizing map network for dimensionality reduction clustering to obtain cluster labels. Using the cluster labels as conditions and the preliminary energy consumption predictions and features in the standard dataset as input, a mixture density network is used to model the probability distribution of the preliminary energy consumption predictions, outputting the final predicted energy consumption value.

[0011] S5. Real-time energy-saving strategy decision generation: Based on the comparison of the final predicted energy consumption value with the preset energy-saving target, the control parameters of the brightness adjustment coefficient and the regional switching strategy are generated;

[0012] S6. Control command issuance and execution: Send control parameters to the field controller to adjust the brightness and area switch status of the LED lighting equipment, and collect actual energy consumption and equipment status data after execution;

[0013] S7. Feedback optimization and model update: The actual energy consumption and device status data are fed back to the server, and the error is calculated by comparing with the final predicted energy consumption value. If the error exceeds the threshold, the training dataset is updated and the lighting energy consumption prediction model is retrained.

[0014] In this specification, the light intensity sensor collects light intensity data at a fixed sampling period, and the pedestrian flow sensor collects pedestrian flow data using infrared sensing or video image analysis.

[0015] In this specification, in S2, a cyclic redundancy check algorithm is used to check the data before data transmission to ensure transmission accuracy.

[0016] In this specification, in S3, the 3σ principle is used to remove outliers and linear interpolation is used to fill missing values.

[0017] In this specification, in S4, surface fitting uses a cubic polynomial to fit the nonlinear relationship between features and energy consumption, the self-organizing mapping network maps high-dimensional vectors to two-dimensional space and generates cluster labels, and the mixed density network outputs the mean, variance and mixing coefficient of multiple Gaussian distributions. The final predicted energy consumption value is the weighted sum of each Gaussian distribution.

[0018] In this specification, in S4, surface fitting, self-organizing map and mixture density network are jointly trained through a unified loss function, and the loss function includes the preliminary prediction error, the final prediction error and the negative log-likelihood of the mixture density network.

[0019] In this specification, in S5, if the final predicted energy consumption value exceeds the preset energy saving target, the brightness adjustment coefficient is calculated based on the difference between the final predicted energy consumption value and the preset energy saving target, and the regional switching strategy prioritizes closing areas with pedestrian flow below the threshold through a greedy algorithm.

[0020] In this specification, in S6, the LED brightness is controlled by a pulse width modulation signal, the duty cycle is determined by the brightness adjustment coefficient, and the regional switch state is controlled by a relay.

[0021] In this specification, in S7, if the error between the actual energy consumption and the predicted energy consumption exceeds a preset threshold, the latest data is added to the training data set and the earliest data is deleted to keep the size of the data set unchanged.

[0022] A remote energy-saving control system for landscape garden lighting, applying any one of the above-mentioned remote energy-saving control methods for landscape garden lighting, the remote energy-saving control system for landscape garden lighting comprising:

[0023] A multi-dimensional environmental data real-time acquisition module is used to collect real-time environmental light intensity data, human flow data, and time data through light intensity sensors, human flow sensors, and time sensors deployed in landscape gardens;

[0024] A data transmission module is used to transmit the collected light intensity data, pedestrian flow data and time data to the remote control center server via a wireless network;

[0025] The data preprocessing module is used to preprocess the data transmitted to the remote control center server, remove outliers and fill in missing values ​​to form a standardized data set;

[0026] The lighting energy consumption prediction model construction and training module is used to use surface fitting to perform nonlinear fitting between the features in the standard data set and the corresponding energy consumption values ​​to generate preliminary energy consumption prediction values. The preliminary energy consumption prediction values ​​and the features in the standard data set are combined into a high-dimensional vector, which is then input into a self-organizing map network for dimensionality reduction clustering to obtain cluster labels. Then, using the cluster labels as conditions and the preliminary energy consumption prediction values ​​and the features in the standard data set as input, the probability distribution of the preliminary energy consumption prediction values ​​is modeled through a mixture density network to output the final predicted energy consumption value.

[0027] A real-time energy-saving strategy decision generation module is used to generate the control parameters of the brightness adjustment coefficient and the regional switching strategy based on the comparison results of the final predicted energy consumption value and the preset energy-saving target;

[0028] The control instruction sending and execution module is used to send control parameters to the field controller, adjust the brightness and regional switch status of the LED lighting equipment, and collect the actual energy consumption and equipment status data after execution;

[0029] The feedback optimization and model update module is used to feed back actual energy consumption and device status data to the server, compare it with the final predicted energy consumption value, and calculate the error. If the error exceeds the threshold, the training data set is updated and the lighting energy consumption prediction model is retrained.

[0030] In summary, the present invention has at least the following beneficial effects:

[0031] Accurately predict energy consumption: By integrating surface fitting, self-organizing mapping, and mixed density networks, the nonlinear relationship between environmental data (light intensity, pedestrian flow, time, etc.) and lighting energy consumption is fully explored. This enables more accurate prediction of lighting energy consumption in different scenarios, reduces prediction errors, and provides a reliable basis for the formulation of energy-saving strategies.

[0032] Dynamic Intelligent Energy Saving: Based on real-time energy consumption forecasts and preset energy-saving targets, lighting control parameters (brightness adjustment coefficient, zone switching strategy) are dynamically generated. This reduces lighting energy consumption while ensuring a pleasant visitor experience. For example, during low-traffic periods at night, lighting brightness is automatically lowered or lighting in non-critical areas is turned off, effectively reducing energy waste.

[0033] Comprehensive environmental adaptability: Multi-dimensional sensor data collection and fusion analysis enable the system to fully perceive changes in the landscape environment. Whether it's sunny or cloudy, on holidays or weekdays, the system intelligently adjusts lighting modes based on the actual environment and foot traffic, meeting lighting needs in different scenarios while avoiding inefficient energy consumption.

[0034] Remote and Efficient Management: Remote data transmission and cloud-based processing enable unified remote management of landscape lighting. Managers can monitor lighting status and adjust energy-saving strategies in real time without having to be on-site, significantly improving management efficiency and reducing maintenance costs. Furthermore, the system's feedback optimization mechanism continuously updates the prediction model based on actual performance, continuously improving energy-saving management and ensuring long-term stable operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0036] Figure 1 Schematic diagram of the remote energy-saving control method for landscape garden lighting involved in the present invention.

[0037] Figure 2 It is a schematic diagram of the lighting energy consumption prediction model structure and process involved in the present invention.

[0038] Figure 3 It is a schematic diagram of the energy-saving control process involved in the present invention.

[0039] Figure 4 Schematic diagram of the remote energy-saving control system for landscape garden lighting involved in the present invention. DETAILED DESCRIPTION

[0040] Hereinafter, only certain exemplary embodiments are briefly described. As will be appreciated by those skilled in the art, the described embodiments may be modified in various ways without departing from the spirit or scope of the embodiments of the present invention. Therefore, the drawings and description are to be regarded as illustrative in nature and not restrictive.

[0041] The disclosure below provides many different embodiments or examples for implementing different structures of the embodiments of the present invention. In order to simplify the disclosure of the embodiments of the present invention, the components and configurations of specific examples are described below. Of course, these are merely examples and are not intended to limit the embodiments of the present invention. In addition, the embodiments of the present invention may repeat reference numerals and / or reference letters in different examples. Such repetition is for the purpose of simplicity and clarity and does not in itself indicate the relationship between the various embodiments and / or configurations discussed.

[0042] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0043] like Figure 1 As shown, this embodiment provides a remote energy-saving control method for landscape garden lighting, including:

[0044] S1. Real-time collection of multi-dimensional environmental data: Light intensity sensors, pedestrian flow sensors, and time sensors deployed in landscape gardens collect real-time ambient light intensity data, pedestrian flow data, and time data.

[0045] S2. Data transmission: The collected light intensity data, pedestrian flow data, and time data are transmitted to the remote control center server via a wireless network;

[0046] S3. Data preprocessing: Preprocess the data transmitted to the remote control center server to remove outliers and fill in missing values ​​to form a standardized data set;

[0047] S4. Construction and Training of a Lighting Energy Consumption Prediction Model: Surface fitting is used to perform nonlinear fitting between the features in the standard dataset and the corresponding energy consumption values ​​to generate preliminary energy consumption predictions. These preliminary energy consumption predictions and the features in the standard dataset are combined into a high-dimensional vector, which is then input into a self-organizing map network for dimensionality reduction clustering to obtain cluster labels. Using the cluster labels as conditions and the preliminary energy consumption predictions and features in the standard dataset as input, a mixture density network is used to model the probability distribution of the preliminary energy consumption predictions, outputting the final predicted energy consumption value.

[0048] S5. Real-time energy-saving strategy decision generation: Based on the comparison of the final predicted energy consumption value with the preset energy-saving target, the control parameters of the brightness adjustment coefficient and the regional switching strategy are generated;

[0049] S6. Control command issuance and execution: Send control parameters to the field controller to adjust the brightness and area switch status of the LED lighting equipment, and collect actual energy consumption and equipment status data after execution;

[0050] S7. Feedback optimization and model update: The actual energy consumption and device status data are fed back to the server, and the error is calculated by comparing with the final predicted energy consumption value. If the error exceeds the threshold, the training dataset is updated and the lighting energy consumption prediction model is retrained.

[0051] In some embodiments, the light intensity sensor collects light intensity data at a fixed sampling period, and the pedestrian flow sensor collects pedestrian flow data using infrared sensing or video image analysis.

[0052] In some embodiments, in S2, a cyclic redundancy check algorithm is used to check the data before data transmission to ensure transmission accuracy.

[0053] In some embodiments, in S3, the 3σ principle is used to remove outliers and linear interpolation is used to fill missing values.

[0054] In some embodiments, in S4, surface fitting uses a cubic polynomial to fit the nonlinear relationship between features and energy consumption, the self-organizing mapping network maps high-dimensional vectors to two-dimensional space and generates cluster labels, and the mixed density network outputs the mean, variance and mixing coefficient of multiple Gaussian distributions. The final predicted energy consumption value is the weighted sum of each Gaussian distribution.

[0055] In some embodiments, in S4, the surface fitting, self-organizing map, and mixture density network are jointly trained using a unified loss function, where the loss function includes a preliminary prediction error, a final prediction error, and a negative log-likelihood of the mixture density network.

[0056] In some embodiments, in S5, if the final predicted energy consumption value exceeds the preset energy saving target, the brightness adjustment coefficient is calculated based on the difference between the final predicted energy consumption value and the preset energy saving target, and the regional switching strategy prioritizes closing areas with pedestrian flow below the threshold through a greedy algorithm.

[0057] In some embodiments, in S6, the LED brightness is controlled by a pulse width modulation signal, the duty cycle is determined by a brightness adjustment coefficient, and the regional switching state is controlled by a relay.

[0058] In some embodiments, in S7, if the error between the actual energy consumption and the predicted energy consumption exceeds a preset threshold, the latest data is added to the training data set and the earliest data is deleted to keep the size of the data set unchanged.

[0059] The technical concept of the present invention is as follows (the energy-saving control process is as follows Figure 3 shown):

[0060] S1. Real-time collection of multi-dimensional environmental data

[0061] The light intensity sensor, pedestrian flow sensor and time sensor deployed in the landscape garden collect data in real time. The light intensity sensor collects ambient light intensity data at a fixed frequency, which is denoted as I t (Unit: Lux, t represents time series); the pedestrian flow sensor uses infrared sensing or video analysis to obtain pedestrian flow data in each area, denoted as P t (Unit: person); time sensor records current time data T t (Unit: hours, minutes, seconds) The sensor collects data at a fixed sampling period Δt (unit: seconds), and after collection, the analog signal is converted into a digital signal through A / D.

[0062] S2. Data transmission

[0063] The collected light intensity data I t , traffic data P t and time data T t , transmitted to the remote control center server via wireless network (such as 4G, 5G, Wi-Fi). Before transmission, the data is packaged and a checksum is generated using the CRC (Cyclic Redundancy Check) algorithm and transmitted together with the data. At the sending end, the data is sent according to [Time Data T t , traffic data P t , light intensity data I t ] format, calculate D bmod G (D is the data sequence, G is the generating polynomial) to obtain the check code C and add it to the end of the data packet; the receiving end performs CRC check.

[0064] S3. Data Preprocessing

[0065] After receiving the data, the server performs preprocessing. t and traffic data P t , the 3σ principle is used to remove outliers, that is, if the data point x satisfies x<μ-3σ or x>μ+3σ (μ is the mean, σ is the standard deviation), it is considered an outlier and removed; for missing values, linear interpolation is used to fill in the missing values. If t i The time data is missing, and the data before and after the time is x i-1 and x i+1 ,but After processing, a standard data set D is formed.

[0066] S4. Construction and training of lighting energy consumption prediction model integrating surface fitting, self-organizing map and hybrid density network) (structure and process as shown in Figure 2 shown)

[0067] Model building process:

[0068] Surface fitting: Based on the features in the preprocessed dataset D (I t 、P t 、T t ) and the corresponding energy consumption value E t , using nonlinear polynomial surface fitting, let the fitting function be Among them, a i1j1k1 are the polynomial coefficients. By minimizing the mean square error (N is the number of data samples), use the gradient descent method to solve the coefficient a i1j1k1 , and obtain the preliminary energy consumption forecast value E fit (I t ,P t ,T t ).

[0069] Self-organizing map (SOM): The initial prediction value E obtained by surface fitting fit (I t ,P t ,T t ) and the features in the preprocessed dataset D (I t 、P t 、T t ) into a new high-dimensional vector X t =[E fit (I t ,P t ,T t ),I t ,P t ,T t ], input to the SOM network. The SOM network consists of a two-dimensional neuron array, each neuron has a corresponding weight vector W s(s is the neuron number). During training, calculate the input vector X t and the weight vector W of each neuron s Euclidean distance (m is the vector dimension), find the winning neuron s with the smallest distance * According to W s (new)=W s (old)+α(t)·h s*s (t)·(X t -W s (old)) updates the weights of the winning neuron and its neighboring neurons, where α(t) is the learning rate that decreases over time, and h s*s (t) is the neighborhood function, Where σ(t) is the time-varying neighborhood radius. After training, the high-dimensional vector X t Mapped to two-dimensional space, we get the cluster label C t .

[0070] Mixture Density Network (MDN): Cluster label C output by SOM network t As a condition, the output E of the surface fitting fit (I t ,P t ,T t ) and the features in the preprocessed dataset D (I t 、P t 、T t ) as input and construct the MDN network. The MDN network outputs multiple Gaussian distribution parameters: mean μ k (C t ),variance and the mixing coefficient π k (C t )(k=1,2,…,K, K is the number of mixing components, let K=3), MDN mixing coefficient constraint: Ensure the probability is normalized. Final predicted energy consumption value

[0071] Model training process:

[0072] Define the comprehensive loss function Loss = MSE1 + MSE2 - NLL, where is the mean square error between the final predicted value and the actual value, is the negative log-likelihood of the output distribution of the mixture density network. Using the Adam optimizer, the learning rate is set to 0.001, and the coefficient a of the surface fitting is updated simultaneously through the back-propagation algorithm. i1j1k1 , the weight vector W of the SOM networks And the parameters μ of the MDN network k (C t ), and π k (C t ), and train for 100 rounds.

[0073] Model application process: The feature data collected and pre-processed in real time (I t 、P t 、T t ), and then pass through surface fitting, SOM network and MDN network in turn, and output the predicted energy consumption E pred (t), which is used for subsequent energy-saving decisions.

[0074] S5. Real-time energy-saving strategy decision generation

[0075] Based on the predicted energy consumption E pred (t) and preset energy saving target E target (such as 15% lower than the same period in history), the lighting control parameters are generated through the decision algorithm. pred (t)≤E target , maintain the current lighting mode; if E pred (t)>E target , start energy saving adjustment. Brightness adjustment coefficient α (range 0.5-1.0, default 1.0) according to the formula Calculation; Regional switching strategy β j (j is the lighting area number, β j =1 means open, β j =0 means closed) A greedy algorithm is used to prioritize closing traffic below the threshold θ people =0.3 (after normalization).

[0076] S6. Control instruction issuance and execution

[0077] The control parameters α,β generated by the decision j The command is sent to the field controller via TCP / IP protocol. After the controller analyzes the command, it controls the brightness of the LED lighting equipment through PWM (pulse width modulation) signal, with a duty cycle of D = α × 100%. The area switch is realized by controlling the on and off of the relay, corresponding to β j At the same time, the actual energy consumption E after execution is collected real (t) and device status S device (t) = [α exec ,β j,exec ], where α exec is the actual brightness coefficient, β j,exec It is the actual execution area switch status.

[0078] S7. Feedback Optimization and Model Update

[0079] E real (t) and S device (t) is fed back to the remote control center and compared with the predicted value E pred (t) comparison, calculation error ΔE=E real (t)-E pred (t). If |ΔE|>0.1×E target , the latest data [I t ,P t ,T t ,E real (t),S device (t)] Add the training dataset and delete the earliest data to keep the dataset size unchanged. Retrain the fusion model in S4 and update the model parameters.

[0080] The core of the present invention lies in the fusion processing logic of surface fitting, self-organizing map (SOM), and mixture density network (MDN), which is as follows:

[0081] 1. Overview of the Fusion Framework

[0082] The three form a closed loop through data dependencies and optimization goal coupling:

[0083] 1. Surface fitting first models the nonlinear relationship between the original features and energy consumption and outputs a preliminary prediction value;

[0084] 2. SOM combines the preliminary prediction value with the original features into a high-dimensional vector, reduces the dimension and clusters it to generate scene labels;

[0085] 3. MDN models the probability distribution of the preliminary prediction value based on the scene label and outputs the final prediction value;

[0086] 4. During training, the three parameters are optimized synchronously through a unified loss function to ensure that the prediction error is transmitted back to all modules.

[0087] 2. Specific Process of Fusion Processing

[0088] 1. Data flow and feature association

[0089] Input layer: Feature vector F(t) = [I t ,P t ,T t ](light intensity, traffic flow, time).

[0090] Surface fitting layer: takes F(t) as input and generates a preliminary energy consumption prediction value E fit (t), the formula is: where a i1j1k1are polynomial coefficients, by minimizing E fit (t) and actual energy consumption E true The mean square error MSE1 of (t) is solved.

[0091] SOM layer: [F fit (t), f(t)] are combined into a 4-dimensional vector X(t), which is input into the SOM network for dimensionality reduction clustering. The SOM maps X(t) to a 2D neuron array through competitive learning and outputs a cluster label C(t) (such as scene categories such as "high traffic during the day" and "low traffic at night"). Key formula: Neuron weight update rule where s * is the winning neuron, α(t) is the decay learning rate, h s*s (t) is the neighborhood function.

[0092] MDN layer: Take C(t) as the condition, fit The probability distribution model of (t) is used to output K=3 Gaussian distribution parameters (mean μ k ,variance Mixing coefficient π k ), the final predicted value is:

[0093] 2. Joint Parameter Optimization during Training

[0094] Unified loss function: Simultaneously optimize the accuracy of surface fitting, the clustering rationality of SOM, and the distribution fitting of MDN. The formula is: Loss = MSE1 + MSE2 - NLL;

[0095] in:

[0096] (surface fitting error);

[0097] (final prediction error);

[0098] (MDN negative log-likelihood).

[0099] Backward propagation path:

[0100] Loss vs. E pred (t) Derivative, update μ of MDN k ,σ k ,π k ;

[0101] Loss derives C(t) and adjusts the neuron weight W through the neighborhood update rule of SOM s ;

[0102] Loss vs. Efit (t) Derivative, update the coefficient a of the surface fitting i1j1k1 .

[0103] 3. Linked prediction in the inference phase

[0104] Input F(t) in real time and get E through surface fitting fit (t);

[0105] Combination [E fit (t), F(t)] input into SOM and query the current scene label C(t);

[0106] According to C(t), the corresponding Gaussian distribution parameters are obtained from MDN, and the weighted average prediction value E is calculated. pred (t).

[0107] 3. Integration of core advantages

[0108] 1. Surface fitting solves the nonlinear mapping problem of the original data and provides preliminary feature representation for SOM;

[0109] 2. SOM uses unsupervised clustering to divide the high-dimensional space into scenarios with clear physical meanings (such as "holiday night" and "weekday day"), narrowing the modeling scope of the MDN.

[0110] 3. MDN handles the uncertainty of energy consumption forecasts (such as sudden fluctuations in passenger flow) through probability distribution modeling, which is more robust than single-value predictions.

[0111] 4. The three are collaboratively optimized through a unified loss function to avoid the error accumulation problem of the traditional cascade model.

[0112] A remote energy-saving control system for landscape garden lighting, applying any one of the above-mentioned remote energy-saving control methods for landscape garden lighting, such as Figure 4 As shown, the remote energy-saving control system for landscape garden lighting includes:

[0113] A multi-dimensional environmental data real-time acquisition module is used to collect real-time environmental light intensity data, human flow data, and time data through light intensity sensors, human flow sensors, and time sensors deployed in landscape gardens;

[0114] A data transmission module is used to transmit the collected light intensity data, pedestrian flow data and time data to the remote control center server via a wireless network;

[0115] The data preprocessing module is used to preprocess the data transmitted to the remote control center server, remove outliers and fill in missing values ​​to form a standardized data set;

[0116] The lighting energy consumption prediction model construction and training module is used to use surface fitting to perform nonlinear fitting between the features in the standard data set and the corresponding energy consumption values ​​to generate preliminary energy consumption prediction values. The preliminary energy consumption prediction values ​​and the features in the standard data set are combined into a high-dimensional vector, which is then input into a self-organizing map network for dimensionality reduction clustering to obtain cluster labels. Then, using the cluster labels as conditions and the preliminary energy consumption prediction values ​​and the features in the standard data set as input, the probability distribution of the preliminary energy consumption prediction values ​​is modeled through a mixture density network to output the final predicted energy consumption value.

[0117] A real-time energy-saving strategy decision generation module is used to generate the control parameters of the brightness adjustment coefficient and the regional switching strategy based on the comparison results of the final predicted energy consumption value and the preset energy-saving target;

[0118] The control instruction sending and execution module is used to send control parameters to the field controller, adjust the brightness and regional switch status of the LED lighting equipment, and collect the actual energy consumption and equipment status data after execution;

[0119] The feedback optimization and model update module is used to feed back actual energy consumption and device status data to the server, compare it with the final predicted energy consumption value, and calculate the error. If the error exceeds the threshold, the training data set is updated and the lighting energy consumption prediction model is retrained.

[0120] In a specific embodiment, a small landscape garden covers an area of ​​approximately 5,000 square meters and is deployed with 10 light intensity sensors, 5 pedestrian flow sensors, and corresponding time sensors. The following uses specific numerical values ​​to illustrate the implementation process of the remote energy-saving control method for lighting.

[0121] S1. Real-time collection of multi-dimensional environmental data

[0122] Data collection starts at 6:00 am, with a sampling period of Δt=60 seconds (i.e., data is collected once every minute).

[0123] S2. Data transmission

[0124] The collected data is stored in [Time Data T t , traffic data P t , light intensity data I t ] format, uses the CRC algorithm to generate a checksum, and then transmits it to the remote control center server via the 4G wireless network. Assume that the data transmitted is [6:00, 0, 10], and the checksum calculated by CRC is 123 (example value), the transmitted data packet is [6:00, 0, 10, 123].

[0125] S3. Data Preprocessing

[0126] The 3σ principle is used to remove outliers. For light intensity data, the mean is calculated. Standard deviation σ I ≈457.77. All data are in μ I -3σ I to μ I +3σ I There are no abnormal values.

[0127] If there are missing values, for example, the flow of people at 7:01 is missing, the flow of people at 7:00 is P 7:00 =2, passenger flow P at 7:02 7:02 =3, then calculate by linear interpolation After processing, a standardized data set is formed for subsequent operations.

[0128] S4. Construction and training of lighting energy consumption prediction model integrating surface fitting, self-organizing map and hybrid density network

[0129] Model building process:

[0130] Surface fitting: Assume that the highest degree of polynomial is 3, and adopt nonlinear polynomial surface fitting. The fitting function is Solve for coefficient a by least squares method i1j1k1 , assuming that a is obtained after calculation 000 =0.1, a 100 =0.001, a 010 =0.01, a 001 =0.05, etc. (other coefficient example values). Taking the data at 7:00 as an example, I 7:00 =50, P 7:00 =2, T 7:00 Assuming it is 0.3 after normalization (normalizing 24 hours a day to 0-1), then E fit (7:00) = 0.1 + 0.001 × 50 + 0.01 × 2 + 0.05 × 0.3 = 0.32 (kWh).

[0131] Self-Organizing Map (SOM): fit (t),I t ,P t ,T t ] is combined into a new high-dimensional vector X(t) and input into the SOM network. Assume that the SOM network is a 10×10 two-dimensional neuron array, and each neuron has a corresponding weight vector W s For X(7:00) = [0.32, 50, 2, 0.3], calculate the Euclidean distance to each neuron weight vector Find the winning neuron s with the smallest distance * Assume that the winning neuron is the neuron at position (3,4), according to Ws (new)=W s (old)+α(t)·h s*s (t)·(X(7:00)-W s (old)) update weights, where α(t) = 0.1 (initial learning rate), h s*s (t) is calculated according to the Gaussian function to obtain 0.8 (example value). Finally, the cluster label C(7:00) = 2 is obtained (assuming it represents the scene of "people flow in low light in the morning").

[0132] Mixture Density Network (MDN): With C(7:00)=2 as the condition, E fit (7:00) and the original features as input. Assume that the MDN network outputs three Gaussian distribution parameters: μ1(C(7:00))=0.3, π1(C(7:00))=0.3; μ2(C(7:00))=0.35, π2(C(7:00))=0.5; μ3(C(7:00))=0.4, π3(C(7:00))=0.2. Then the final predicted energy consumption value E pred (7:00) = 0.3 × 0.3 + 0.35 × 0.5 + 0.4 × 0.2 = 0.345 (kWh).

[0133] Model training process:

[0134] Define the comprehensive loss function Loss = MSE1 + MSE2 - NLL. Assume the number of training samples is 100. After 100 rounds of training, use the Adam optimizer with a learning rate of 0.001, and use the backpropagation algorithm to simultaneously update the parameters of the surface fitting, SOM network, and MDN network.

[0135] Model application process:

[0136] The feature data collected and preprocessed in real time are sequentially passed through surface fitting, SOM network and MDN network to output predicted energy consumption for subsequent energy-saving decisions.

[0137] S5. Real-time energy-saving strategy decision generation

[0138] Preset energy saving target E target =85% of the average energy consumption during the same period in history. Assuming that the average energy consumption at 7:00 during the same period in history is 0.5 kWh, then E target =0.5×0.85=0.425 kWh. pred (7:00) = 0.345 <E target , maintain the current lighting mode.

[0139] If at a certain moment E pred (t)>E target , for example, E pred (20:00) = 0.6 kWh, E target =0.5 kWh, then the brightness adjustment coefficient Regional switching strategy β j Through the greedy algorithm, prioritize closing the traffic flow below the threshold θ people =0.3 (after normalization).

[0140] S6. Control instruction issuance and execution

[0141] The control parameters (α, β j ) is sent to the field controller via TCP / IP protocol. If α = 0.92, the brightness of the LED lighting device is controlled by PWM signal, and the duty cycle D = 0.92 × 100% = 92%; according to β j Control the relay on and off to realize regional switching. After execution, collect the actual energy consumption E real (t) and device status S device (t). Assume that E is executed at 20:00 real (20:00) = 0.55 kWh, S device (20:00) = [0.92, β1 = 1, β2 = 0, ...] (indicates the brightness coefficient and the switching status of each area).

[0142] S7. Feedback Optimization and Model Update

[0143] Calculation error ΔE=E real (20:00)-E pred (20:00) = 0.55 - 0.6 = -0.05 kWh, |ΔE| = 0.05 < 0.1 × E target =0.05, no model update is triggered. If |ΔE|>0.1×E target , then add the latest data to the training dataset, delete the earliest data, retrain the fusion model in S4, and update the model parameters.

[0144] The above embodiments are intended to illustrate the present invention, not to limit the present invention. Therefore, changes in illustrative values ​​or substitutions of equivalent components should still fall within the scope of the present invention.

[0145] From the above detailed description, it will be clear to those skilled in the art that the present invention can indeed achieve the aforementioned objectives and is in compliance with the provisions of the Patent Law.

[0146] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they become aware of the basic inventive concepts. Therefore, the appended claims are intended to be interpreted as covering the preferred embodiments and all changes and modifications that fall within the scope of the invention. The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

[0147] It should be noted that the above description of the relevant processes is for illustration and purpose only and does not limit the scope of application of this specification. For those skilled in the art, various modifications and changes can be made to the processes under the guidance of this specification. However, such modifications and changes are still within the scope of this specification.

[0148] The basic concepts have been described above. It will be apparent to those skilled in the art after reading this application that the above disclosures are merely illustrative and do not constitute limitations on this application. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and amendments to this application. Such modifications, improvements, and amendments are suggested in this application and remain within the spirit and scope of the exemplary embodiments of this application.

[0149] At the same time, this application uses specific terms to describe the embodiments of this application. For example, "one embodiment," "an embodiment," and / or "some embodiments" refer to a certain feature, structure, or characteristic related to at least one embodiment of this application. Therefore, it should be emphasized and noted that "one embodiment," "an embodiment," or "an alternative embodiment" mentioned twice or more in different places in this specification does not necessarily refer to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of this application may be appropriately combined.

[0150] In addition, it will be understood by those skilled in the art that various aspects of the present application can be illustrated and described by a number of patentable categories or situations, including any new and useful combination of processes, machines, products or substances, or any new and useful improvements thereto. Therefore, various aspects of the present application can be implemented entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software can all be referred to as "units", "modules" or "systems". In addition, various aspects of the present application can take the form of a computer program product embodied in one or more computer-readable media, wherein computer-readable program code is contained therein.

[0151] The computer program code required for the operation of each part of the application can be written in any one or more programming languages, including object-oriented programming languages ​​such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, conventional procedural programming languages ​​such as C programming language, VisualBasic, Fortran2103, Perl, COBOL2102, PHP, ABAP, dynamic programming languages ​​such as Python, Ruby and Groovy or other programming languages. The program code can be run completely on the user's computer, or run on the user's computer as an independent software package, or run partly on the user's computer and partly on a remote computer, or run completely on a remote computer or server. In the latter case, the remote computer can be connected to the user's computer through any network form, such as a local area network (LAN) or a wide area network (WAN), or be connected to an external computer (such as by the Internet), or in a cloud computing environment, or used as a service such as software as a service (SaaS).

[0152] In addition, unless expressly stated in the claims, the order of the processing elements and sequences described in this application, the use of alphanumeric characters, or the use of other names are not intended to limit the order of the processes and methods of this application. Although the above disclosure discusses some embodiments of the invention that are currently considered useful through various examples, it should be understood that such details are only for illustrative purposes, and the attached claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that are consistent with the essence and scope of the embodiments of this application. For example, although the implementation of the various components described above can be embodied in a hardware device, it can also be implemented as a pure software solution, for example, installation on an existing server or mobile device.

[0153] Similarly, it should be noted that in order to simplify the presentation of this disclosure and thereby facilitate understanding of one or more of the invention's embodiments, the foregoing descriptions of the embodiments of this disclosure sometimes combine multiple features into a single embodiment, figure, or description thereof. However, this approach should not be interpreted as reflecting an intention that the claimed subject matter requires more features than expressly recited in each claim. Rather, the subject matter of the invention may possess fewer features than the single embodiment described above.

Claims

1. A remote energy-saving control method for landscape garden lighting, characterized in that: include: S1. Real-time collection of multi-dimensional environmental data: Light intensity sensors, pedestrian flow sensors, and time sensors deployed in landscape gardens collect real-time ambient light intensity data, pedestrian flow data, and time data. S2. Data transmission: The collected light intensity data, pedestrian flow data, and time data are transmitted to the remote control center server via a wireless network; S3. Data preprocessing: Preprocess the data transmitted to the remote control center server to remove outliers and fill in missing values ​​to form a standardized data set; S4. Construction and Training of a Lighting Energy Consumption Prediction Model: Surface fitting is used to perform nonlinear fitting between the features in the standard dataset and the corresponding energy consumption values ​​to generate preliminary energy consumption predictions. These preliminary energy consumption predictions and the features in the standard dataset are combined into a high-dimensional vector, which is then input into a self-organizing map network for dimensionality reduction clustering to obtain cluster labels. Using the cluster labels as conditions and the preliminary energy consumption predictions and features in the standard dataset as input, a mixture density network is used to model the probability distribution of the preliminary energy consumption predictions, outputting the final predicted energy consumption value. S5. Real-time energy-saving strategy decision generation: Based on the comparison between the final predicted energy consumption value and the preset energy-saving target, the control parameters of the brightness adjustment coefficient and the regional switching strategy are generated; S6. Control command issuance and execution: Send control parameters to the field controller to adjust the brightness and area switch status of the LED lighting equipment, and collect actual energy consumption and equipment status data after execution; S7. Feedback optimization and model update: The actual energy consumption and device status data are fed back to the server, and the error is calculated by comparing with the final predicted energy consumption value. If the error exceeds the threshold, the training dataset is updated and the lighting energy consumption prediction model is retrained.

2. The remote energy-saving control method for landscape garden lighting according to claim 1, characterized in that: The light intensity sensor collects light intensity data at a fixed sampling period, and the pedestrian flow sensor collects pedestrian flow data using infrared sensing or video image analysis.

3. The remote energy-saving control method for landscape garden lighting according to claim 1, characterized in that: In S2, a cyclic redundancy check algorithm is used to check the data before data transmission to ensure transmission accuracy.

4. The remote energy-saving control method for landscape garden lighting according to claim 1, characterized in that: In S3, the 3σ principle is used to remove outliers and linear interpolation is used to fill missing values.

5. The remote energy-saving control method for landscape garden lighting according to claim 1, characterized in that: In S4, surface fitting uses cubic polynomials to fit the nonlinear relationship between features and energy consumption. The self-organizing map network maps high-dimensional vectors to two-dimensional space and generates cluster labels. The mixture density network outputs the mean, variance and mixing coefficient of multiple Gaussian distributions. The final predicted energy consumption value is the weighted sum of each Gaussian distribution.

6. The remote energy-saving control method for landscape garden lighting according to claim 1, characterized in that: In S4, surface fitting, self-organizing maps, and mixture density networks are jointly trained using a unified loss function, which includes the preliminary prediction error, the final prediction error, and the negative log-likelihood of the mixture density network.

7. The remote energy-saving control method for landscape garden lighting according to claim 1, characterized in that: In S5, if the final predicted energy consumption value exceeds the preset energy saving target, the brightness adjustment coefficient is calculated based on the difference between the final predicted energy consumption value and the preset energy saving target, and the regional switching strategy prioritizes closing areas with pedestrian flow below the threshold through a greedy algorithm.

8. The remote energy-saving control method for landscape garden lighting according to claim 1, characterized in that: In S6, the LED brightness is controlled by a pulse width modulation signal, the duty cycle is determined by the brightness adjustment coefficient, and the regional switching state is controlled by a relay.

9. The remote energy-saving control method for landscape garden lighting according to claim 1, characterized in that: In S7, if the error between the actual energy consumption and the predicted energy consumption exceeds a preset threshold, the latest data is added to the training dataset and the earliest data is deleted to keep the dataset size unchanged.

10. A remote energy-saving control system for landscape garden lighting, characterized in that: The remote energy-saving control method for landscape garden lighting according to any one of claims 1 to 9 is applied, and the remote energy-saving control system for landscape garden lighting comprises: A multi-dimensional environmental data real-time acquisition module is used to collect real-time environmental light intensity data, human flow data, and time data through light intensity sensors, human flow sensors, and time sensors deployed in landscape gardens; A data transmission module is used to transmit the collected light intensity data, pedestrian flow data and time data to the remote control center server via a wireless network; The data preprocessing module is used to preprocess the data transmitted to the remote control center server, remove outliers and fill in missing values ​​to form a standardized data set; The lighting energy consumption prediction model construction and training module is used to use surface fitting to perform nonlinear fitting between the features in the standard data set and the corresponding energy consumption values ​​to generate preliminary energy consumption prediction values. The preliminary energy consumption prediction values ​​and the features in the standard data set are combined into a high-dimensional vector, which is then input into a self-organizing map network for dimensionality reduction clustering to obtain cluster labels. Then, using the cluster labels as conditions and the preliminary energy consumption prediction values ​​and the features in the standard data set as input, the probability distribution of the preliminary energy consumption prediction values ​​is modeled through a mixture density network to output the final predicted energy consumption value. A real-time energy-saving strategy decision generation module is used to generate the control parameters of the brightness adjustment coefficient and the regional switching strategy based on the comparison results of the final predicted energy consumption value and the preset energy-saving target; The control instruction sending and execution module is used to send control parameters to the field controller, adjust the brightness and regional switch status of the LED lighting equipment, and collect the actual energy consumption and equipment status data after execution; The feedback optimization and model update module is used to feed back actual energy consumption and device status data to the server, compare it with the final predicted energy consumption value, and calculate the error. If the error exceeds the threshold, the training data set is updated and the lighting energy consumption prediction model is retrained.

Images