Illumination big data monitoring method and illumination internet of things system thereof

Abstract

The present application relates to the technical field of illumination parameter detection method and automation device, disclose a kind of illumination big data monitoring method and its illumination Internet of Things system, set up illumination sensor group, multiple groups of environmental parameter sensor group under the illumination demand environment, real-time data value is collected and fused to obtain each environmental parameter real-time value, the ideal value of each real-time value is dynamically predicted to obtain illumination;Based on the ideal value of illumination and the real-time value of illumination, the real-time control quantity of illumination is obtained, based on the ideal value of illumination and the predicted value of illumination, the predicted control quantity of illumination is obtained, according to the real-time value of illumination and the actual control quantity of illumination, the disturbance quantity of illumination is obtained;According to the difference between the accumulation of the predicted control quantity of illumination and the real-time control quantity of illumination and the disturbance quantity of illumination as the actual control quantity of illumination.Compared with the prior art, the present application improves the accuracy of illumination adjustment, illumination adjustment benefit and illumination parameter monitoring effect.

Description

Technical Field

[0001] This invention relates to the technical field of illuminance parameter detection methods and automated devices, specifically to an illuminance big data monitoring method and its illuminance Internet of Things system. Background Technology

[0002] Illuminance monitoring methods are used in industries such as agriculture, manufacturing, transportation, healthcare, and food processing. However, most illuminance control methods rely on traditional PID control, resulting in low accuracy and instability. Furthermore, they lack dynamic optimization based on changes in environmental parameters and illuminance requirements, leading to low economic efficiency. This invention addresses these issues by developing a big data-driven illuminance monitoring method and its associated Internet of Things (IoT) system. This method collects large amounts of illuminance data, optimizes ideal illuminance values ​​under different conditions based on environmental parameter changes and the requirements of the target lighting object, designs an intelligent illuminance control system for adjusting illuminance, and implements an IoT system for intelligent monitoring and collection of illuminance parameters to meet the requirements of various industries for accuracy, stability, and economic efficiency in illuminance control. Summary of the Invention

[0003] Purpose of the invention: This invention provides a big data monitoring method for illuminance and its illuminance Internet of Things system, which realizes dynamic adjustment of illuminance based on changes in illuminance environmental parameters and different needs for illuminance, thereby improving the accuracy, efficiency, and monitoring effect of illuminance parameters.

[0004] Technical solution: This invention discloses a method for monitoring big data on illuminance, comprising the following steps:

[0005] Step 1: Set up light intensity sensor group, sensor group 1, sensor group 2 and sensor group 3, collect real-time data values ​​of each sensor group, and fuse them through the corresponding Adaline neural network model to obtain the real-time light intensity value, the real-time value of sensor group 1, the real-time value of sensor group 2 and the real-time value of sensor group 3.

[0006] Step 2: Use GWO's BiLSTM neural network model and GWO's JORDAN neural network model to dynamically predict each real-time value, and input the predicted value into GWO's fuzzy recurrent neural network model to obtain the ideal value of illuminance.

[0007] Step 3: Based on the ideal illuminance value and the real-time illuminance value, obtain the real-time illuminance control quantity through linear control; based on the ideal illuminance value and the predicted illuminance value, obtain the predicted illuminance control quantity through nonlinear control; and obtain the illuminance disturbance quantity after processing the real-time illuminance value and the actual illuminance control quantity through the JORDAN neural network model of GWO.

[0008] Step 4: Finally, the difference between the sum of the predicted illuminance control quantity and the real-time illuminance control quantity and the illuminance disturbance quantity is used as the actual illuminance control quantity. The actual illuminance control quantity is used as the input of the light source driver and the input of the JORDAN neural network model of GWO in step 3 for cyclic control. The output of the light source driver is used as the light source input to realize the dynamic adjustment of the illuminance of the light source adjustment environment.

[0009] Furthermore, the sensor group 1, sensor group 2 and sensor group 3 are groups of environmental parameters that affect illuminance, including but not limited to a visibility sensor group, a temperature and humidity sensor group, and a traffic flow sensor group.

[0010] Furthermore, in step 3, when obtaining the real-time illuminance control quantity through linear control based on the ideal illuminance value and the real-time illuminance value, the GWO fuzzy recurrent neural network model-PID controller is selected. The difference between the ideal illuminance value and the real-time illuminance value and the rate of change of the difference are used as the inputs of the GWO fuzzy recurrent neural network model-PID controller, and the GWO fuzzy recurrent neural network model-PID controller outputs the real-time illuminance control quantity.

[0011] Furthermore, when obtaining the illuminance prediction control quantity through nonlinear control based on the ideal illuminance value and the predicted illuminance value, the difference between the ideal illuminance value and the predicted illuminance value output by the GWO BiLSTM neural network model - GWO JORDAN neural network model and the rate of change of the difference are used as the input of the GWO fuzzy recurrent neural network model - GWO JORDAN neural network model. The illuminance is predicted and controlled using the GWO fuzzy recurrent neural network model - GWO JORDAN neural network model, and the output of the GWO fuzzy recurrent neural network model - GWO JORDAN neural network model is used as the illuminance prediction control quantity.

[0012] Furthermore, when obtaining the illuminance perturbation based on the real-time illuminance value and the actual illuminance control quantity processed by the JORDAN neural network model of GWO, the actual illuminance control quantity and the actual illuminance value output by the Adaline neural network model are used as the inputs to the JORDAN neural network model of GWO, and the output of the JORDAN neural network model of GWO is used as the illuminance perturbation.

[0013] Furthermore, the GWO fuzzy recurrent neural network model—the GWO JORDAN neural network model—consists of the GWO fuzzy recurrent neural network model, the GWO JORDAN neural network model, delay units, and a feedback layer. The input values ​​serve as the input to the GWO fuzzy recurrent neural network model and the input to delay unit 1. The output of delay unit 1 serves as the input to the GWO fuzzy recurrent neural network model. The output of the GWO fuzzy recurrent neural network model serves as the input to the GWO JORDAN neural network model and the input to delay unit 2. The output of delay unit 2 serves as the input to the GWO JORDAN neural network model. The input to the JORDAN neural network model is the output value Y(t) of the JORDAN neural network model of GWO, which is used as the input to delay unit 3. The output of delay unit 3 is used as the input to the feedback layer. The output of the feedback layer is used as the input to delay unit 5, delay unit 4, and the fuzzy recurrent neural network model of GWO. The outputs of delay unit 4 and delay unit 5 are used as the inputs to the JORDAN neural network model and the feedback layer of GWO, respectively. The inputs to the feedback layer are Y(t-1) and X(t-1), the connection weights of the feedback layer are w1 and w2, and the output of the feedback layer is X(t). The relationship between the output and input of the feedback layer is as follows:

[0014] .

[0015] Furthermore, the GWO BiLSTM neural network model—the GWO JORDAN neural network model—comprising a GWO BiLSTM neural network model, a GWO JORDAN neural network model, delay units, and a feedback layer. The input values ​​serve as the input to the GWO BiLSTM neural network model and the input to delay unit 6. The output of delay unit 6 serves as the input to the GWO BiLSTM neural network model. The output of the GWO BiLSTM neural network model serves as the input to the GWO JORDAN neural network model and the input to delay unit 7. The output of delay unit 7 serves as the input to the GWO JORDAN neural network model. The input to the network model is the JORDAN neural network model of GWO. The output value O(t) of the JORDAN neural network model is used as the input of delay unit 8. The output value O(t-1) of delay unit 8 is used as the input of the feedback layer. The output value Z(t) of the feedback layer is used as the input of delay unit 10, delay unit 9, and the BiLSTM neural network model of GWO. The output values ​​of delay unit 9 and delay unit 10 are used as the input of the JORDAN neural network model of GWO and the input of the feedback layer, respectively. The input of the feedback layer is O(t-1) and Z(t-1). The connection weights of the feedback layer are w3 and w4. The output of the feedback layer is Z(t). The relationship between the output and input of the feedback layer is as follows:

[0016] .

[0017] This invention also discloses a big data IoT system for illuminance, including an illuminance measurement and control node, an illuminance big data cloud platform, and an illuminance management terminal. The illuminance measurement and control node is responsible for detecting light source environmental parameters and adjusting illuminance. Data communication between the illuminance measurement and control node, the illuminance big data cloud platform, and the illuminance management terminal is realized through the wireless communication module of the illuminance measurement and control node and 5G. The illuminance management terminal monitors illuminance parameters and communicates with the illuminance big data cloud platform through 5G. The illuminance management terminal is equipped with an illuminance control system to implement the above-mentioned illuminance control method.

[0018] Compared with the prior art, the present invention has the following obvious advantages:

[0019] 1. The BiLSTM neural network model disclosed in this invention is a special type of recurrent neural network. The basic LSTM model can only process unidirectional illuminance time series data, while the BiLSTM model has bidirectional temporal information fusion capability compared to LSTM. It can integrate the bidirectional temporal features of illuminance parameter changes in forward and backward propagation, significantly enhancing the ability of the BiLSTM neural network model to capture long-term dependencies in the illuminance parameter time series. In the prediction of illuminance parameter changes, this bidirectional modeling mechanism can: (1) analyze the nonlinear coupling relationship between the historical cumulative effect of illuminance parameters and the potential future degradation mode, and more comprehensively capture the dynamic change law of illuminance parameters; (2) improve the ability to predict small changes in illuminance parameters through temporal context awareness of illuminance parameters, so as to capture the global dependencies in the time series data of illuminance parameters and improve the accuracy and robustness of the BiLSTM neural network model in predicting illuminance parameters.

[0020] 2. The advantages of the GWO fuzzy recurrent neural network model-GWO JORDAN neural network model designed in this invention are: the input of the GWO fuzzy recurrent neural network model has a recursive layer, which improves the accuracy, dynamics, and robustness of the input; the output of the GWO fuzzy recurrent neural network model is used as the input of the GWO JORDAN neural network model. Since the input of the GWO JORDAN neural network model is fed back to the input of the GWO JORDAN neural network model through a delay unit, the accuracy, dynamics, and robustness of the output value of the GWO JORDAN neural network model are improved; the feedback layer outputs of the GWO fuzzy recurrent neural network model-GWO JORDAN neural network model are respectively used as the input of the GWO fuzzy recurrent neural network. The input of the model and the input of the GWO JORDAN neural network model improve the accuracy, dynamics, and robustness of the GWO fuzzy recurrent neural network model-GWO JORDAN neural network model. Furthermore, the feedback layer and the input of the GWO fuzzy recurrent neural network model and the feedback of the GWO JORDAN neural network model constitute an inner and outer two-level feedback layer, which further improves the dynamics, accuracy, and robustness of the output value of the GWO fuzzy recurrent neural network model-GWO JORDAN neural network model. Moreover, the output value of the feedback layer is fed back to the input of the feedback layer through the delay unit 5, which further improves the dynamics, accuracy, and robustness of the GWO fuzzy recurrent neural network model-GWO JORDAN neural network model, thereby improving the accuracy of illuminance control.

[0021] 3. The advantage of the GWO BiLSTM neural network model designed in this invention is that the output of the GWO BiLSTM neural network model captures the temporal information of the two directions of illuminance parameter change, thereby improving the accuracy of the GWO BiLSTM neural network model in predicting illuminance parameters. The output of the GWO BiLSTM neural network model is used as the input of the GWO JORDAN neural network model. Since the output of the GWO JORDAN neural network model is fed back to the input of the GWO JORDAN neural network model through the delay unit, the accuracy, dynamics, and robustness of the output value of the GWO JORDAN neural network model are improved. The output of the feedback layer of the GWO BiLSTM neural network model-GWO JORDAN neural network model is used as the input of both the GWO BiLSTM neural network model and the GWO JORDAN neural network model, respectively, improving the accuracy, dynamics, and robustness of the GWO BiLSTM neural network model-GWO JORDAN neural network model. Furthermore, the feedback layer and the feedback of the GWO JORDAN neural network model constitute an inner and outer two-level feedback layer, which further improves the dynamics, accuracy, and robustness of the output value of the GWO BiLSTM neural network model-GWO JORDAN neural network model. Moreover, the output value of the feedback layer is fed back to the input of the feedback layer through the delay unit 10, which further improves the dynamics, accuracy, and robustness of the GWO BiLSTM neural network model-GWO JORDAN neural network model in predicting illuminance parameters.

[0022] 4. The on-demand lighting strategy of the illuminance control system designed in this invention saves energy. It fuses the outputs of multiple illuminance sensors, multiple visibility sensors, multiple temperature and humidity sensors, and multiple traffic flow sensors using an Adaline neural network model. These outputs are then used as inputs to GWO's BiLSTM neural network model and GWO's JORDAN neural network models 1-4, enabling dynamic prediction of illuminance, visibility, temperature and humidity, and traffic flow. The output values ​​of GWO's BiLSTM neural network model and GWO's JORDAN neural network models 1-4 serve as predicted values ​​for, but are not limited to, illuminance, visibility, temperature and humidity, and traffic flow. These predicted values ​​are then used as inputs to GWO's fuzzy recurrent neural network model. The output of GWO's fuzzy recurrent neural network model serves as the ideal illuminance value for the light source environment. The ideal illuminance value is dynamically adjusted based on actual environmental parameters and traffic flow, thus saving energy.

[0023] 5. The speed, accuracy, and robustness of illuminance regulation: GWO's fuzzy recurrent neural network model-PID controller and GWO's fuzzy recurrent neural network model-GWO's JORDAN neural network model achieve parallel regulation of illuminance. The GWO fuzzy recurrent neural network model is used to accurately and quickly tune the Kp, Ki, and Kd of the PID controller, achieving precise and real-time regulation of illuminance. The dynamic, accurate, and robust nature of the GWO fuzzy recurrent neural network model-GWO's JORDAN neural network model enables dynamic predictive regulation of illuminance with robustness, dynamism, and accuracy. The combination of real-time regulation, predictive regulation, and the disturbance amount output by the GWO JORDAN neural network model during the illuminance regulation process further improves the speed, robustness, and accuracy of illuminance regulation.

[0024] 6. The illuminance control system exhibits a dynamic interplay of speed, accuracy, and robustness, with each element interacting, influencing, and complementing the others. GWO's fuzzy recurrent neural network model rapidly infers the ideal illuminance value based on light source environmental parameters and the illuminance demand. The GWO fuzzy recurrent neural network model-PID controller utilizes this model to adjust the Kp and K values ​​of the PID controller. i and K d Precise and rapid tuning enables accurate and rapid real-time adjustment of illuminance; GWO's BiLSTM neural network model and GWO's JORDAN neural network models 1-4 provide robust, accurate, and rapid predictions of illuminance, visibility, temperature, and traffic flow; GWO's fuzzy recurrent neural network model and GWO's JORDAN neural network model provide dynamic, accurate, and robust prediction and adjustment of illuminance with robustness, speed, and accuracy; GWO's JORDAN neural network model outputs the perturbation during the illuminance adjustment process in a timely manner, and their combination, interaction, and complementarity further improve the accuracy, speed, and robustness of illuminance adjustment. Attached Figure Description

[0025] Figure 1 This is the illuminance control system of this patent;

[0026] Figure 2 This patent describes the fuzzy recurrent neural network model of GWO - the JORDAN neural network model of GWO.

[0027] Figure 3 This patent describes the GWO BiLSTM neural network model and the GWO JORDAN neural network model.

[0028] Figure 4 This patent is for a big data IoT system for illuminance.

[0029] Figure 5 This is the illuminance measurement and control node for this patent. Detailed Implementation

[0030] Combined with appendix Figure 1-5 The technical solution of this application will be further described below:

[0031] The illuminance control system designed in this invention includes a GWO fuzzy recurrent neural network model, a GWO fuzzy recurrent neural network model-PID controller, a GWO fuzzy recurrent neural network model-GWO JORDAN neural network model, a GWO JORDAN neural network model, a GWO BiLSTM neural network model-GWO JORDAN neural network model 1-4, and an Adaline neural network model 1-4.

[0032] 1. Design of GWO's BiLSTM Neural Network Model

[0033] (1) BiLSTM Neural Network Model Design

[0034] BiLSTM neural network models are a special type of recurrent neural network (RNN) that simultaneously run two LSTM neural network models along the time dimension of illuminance parameter changes: one predicting positive changes in illuminance parameters and the other predicting negative changes. These two models form the temporal information for predicting illuminance parameter changes in both directions. By integrating the temporal information from both directions, the model can more accurately predict and represent the time series data of illuminance parameter changes. To fully capture the time series information of illuminance parameters, BiLSTM neural network models utilize the powerful ability of hidden layers to read illuminance parameter information, simultaneously acquiring the temporal information features of historical illuminance parameter data in both positive and negative directions. This allows for in-depth exploration of the hidden intrinsic relationships between current illuminance data and past and future illuminance data, providing a comprehensive fitting analysis of the current illuminance data from a two-way perspective. This allows the BiLSTM neural network model to not only efficiently process longer time-series data of illuminance parameters, but also effectively avoid the vanishing or exploding gradient problems, greatly improving the accuracy and robustness of the BiLSTM neural network model in processing complex time-series data of illuminance parameters. The output expression of the BiLSTM neural network model is as follows:

[0035] (1)

[0036] In the formula: Let be the state value of the forward hidden layer of the BiLSTM neural network model at time t. Let be the inverse hidden layer state value of the BiLSTM neural network model at time t. and These are the weight matrix and bias term of the BiLSTM neural network model, respectively. For sigmoid activation function, This represents the output of the BiLSTM neural network model at time t.

[0037] BiLSTM neural network model is a special type of recurrent neural network. The basic LSTM model can only process unidirectional time series data, while the BiLSTM model has the ability to fuse bidirectional time series information compared to LSTM. It can integrate the bidirectional time series features of light intensity parameter changes in forward and backward propagation, which significantly enhances the ability of BiLSTM neural network model to capture long-term dependencies in the time series of light intensity parameters. In the prediction of changes in light intensity parameters, this bidirectional modeling mechanism can: (1) analyze the nonlinear coupling relationship between the historical cumulative effect of light intensity parameters and the potential future degradation mode, and capture the dynamic change law of light intensity parameters more comprehensively; (2) improve the ability to predict small changes in light intensity parameters through the temporal context awareness of light intensity parameters, so as to capture the global dependencies in the time series data of light intensity parameters and improve the accuracy and robustness of BiLSTM neural network model in predicting light intensity parameters.

[0038] In BiLSTM neural network models, hyperparameter settings such as the number of neurons in the hidden layers, dropout rate, and initial learning rate directly affect the network model's performance. However, determining these hyperparameters often relies on human experience and requires numerous trials, consuming significant manpower and time. To address the hyperparameter optimization problem in BiLSTM neural network models, the GWO algorithm is introduced. The number of neurons in the hidden layers, dropout rate, and initial learning rate are used as inputs to GWO. The root mean squared error (RMSE) between the predicted and actual values ​​is used as the fitness function for the optimal individual in the GWO optimization process, thus obtaining the optimal combination of hyperparameters for the prediction model.

[0039] (2) Design of Grey Wolf Optimizer (GWO) Algorithm

[0040] The GWO algorithm simulates the social hierarchy and hunting behavior of gray wolves in nature. It is characterized by its simplicity, parallelism, and ease of implementation. The GWO algorithm has few parameters, does not require gradient information from the problem, and possesses strong global search capabilities. To describe the social hierarchy of gray wolves, [the algorithm will be used to]... The wolf's position is considered the optimal solution. and The wolf's position is considered as the optimal and suboptimal solution, respectively. The wolf's location is considered as one of the remaining candidate solutions. The search is optimized by simulating the processes of wolf pack tracking, surrounding, pursuing, and attacking prey. Gray wolves need to surround their prey when hunting; the mathematical description of this surrounding behavior is as follows:

[0041] (2)

[0042] (3)

[0043] (4)

[0044] (5)

[0045] In the formula: t is the current iteration number; D is the distance vector between the gray wolf and the prey; A and C are the coordination coefficient vectors; The position vector of the prey; Let be the position vector of the gray wolf. and It is a random vector. The convergence factor decreases linearly from 2 to 0 as the number of iterations increases during the iteration process. Assume... For the optimal candidate solution, and With more knowledge about the prey's potential location, the formula for updating the location is as follows:

[0046] (6)

[0047] (7)

[0048] (8)

[0049] in: , and Representing candidates wolves and Wolf, wolves and Distance vector between wolves , and for Wolf, wolves and The wolf's position vector , and correspond Wolf, wolves and Wolf coordination coefficient vector , and right Wolf, wolves and Random perturbation coefficient of wolf distance perception. , and It's the updated version. Wolf, wolves and The wolf's position express The final location of the individual wolf.

[0050] (3) Design of GWO's BiLSTM Neural Network Model

[0051] The following steps were taken to optimize the number of hidden layer neurons, dropout rate, and initial learning rate of the BiLSTM neural network model using GWO to obtain the optimal hyperparameter combination and improve the accuracy of the prediction model:

[0052] Step 1: Set the initialization parameters of the GWO algorithm, determine the number of gray wolves in the population, and the position of each gray wolf represents a set of hyperparameter combinations, including the number of hidden layer neurons in the BiLSTM model, the dropout rate, and the initial learning rate. Also, determine the upper and lower boundaries of the hyperparameters to be optimized and set the maximum number of iterations for the gray wolf algorithm.

[0053] Step 2: Generate the cooperative coefficient vector A, the random perturbation factor C, and the convergence factor. The root mean square error of the BiLSTM model was used as the population fitness value of the GWO grey method.

[0054] Step 3: Calculate fitness value: from the current wolf's position vector The hyperparameters of the BiLSTM neural network model are decoded and initialized. The BiLSTM neural network model is trained using a strategy of fixed training rounds. The trained BiLSTM neural network model is used to predict the input illuminance data on the validation set to obtain the predicted value and calculate the root mean square error value. The calculated root mean square error value is used as the fitness value of the current wolf.

[0055] Step 4: According to Wolf, wolves and The position of the wolf is calculated according to formulas (6), (7) and (8) to obtain the final position of the updated gray wolf individual;

[0056] Step 5: Repeat steps 3 to 5 until the maximum number of iterations of the GWO algorithm is met.

[0057] The parameters of the BiLSTM neural network model were optimized using GWO. The search ranges for the number of hidden layers, initial learning rate, and dropout rate were set to [28, 511], [0.0001, 0.01], and [0, 0.5], respectively. The number of iterations of the BiLSTM neural network model was set to 290, and the maximum number of iterations of the GWO algorithm was set to 35. After each update of the gray wolf position, the wolf pack fitness value was calculated. The root mean square error of the illuminance parameter prediction value reached its minimum in the 24th iteration of the gray wolf optimization algorithm. In the subsequent 6 iterations, the root mean square error of the illuminance parameter prediction value and the parameter combination of the BiLSTM neural network model did not change. At this point, the parameter combination of the BiLSTM neural network model found was the optimal parameter combination. The optimal number of hidden layers in the BiLSTM neural network model is 31, indicating that the BiLSTM neural network model can capture the key temporal features of illumination parameters while avoiding complexity; the optimal initial learning rate is 0.0045, which helps the BiLSTM neural network model to converge stably in predicting complex illumination parameter time series data and avoid gradient oscillations; the optimal dropout rate is 0.4653.

[0058] The advantage of GWO's BiLSTM neural network model in illuminance parameter prediction lies in the fact that the BiLSTM neural network model learns past and future information from illuminance data simultaneously through forward and backward LSTM layers. The deep structure of the BiLSTM neural network model more accurately fits the complex nonlinear relationship between illuminance parameters. By simulating the hunting behavior of a pack of wolves, the GWO algorithm automatically optimizes the parameters of the BiLSTM neural network model, which overcomes the problem of traditional neural networks relying on experience to adjust parameters. This improves the accuracy, reliability, and robustness of the BiLSTM neural network model in predicting illuminance parameters. Furthermore, the swarm intelligence characteristics of the GWO algorithm make it more likely to find the global optimum of the BiLSTM neural network model, avoiding it from getting trapped in local optima, thereby improving the generalization ability of the BiLSTM neural network model.

[0059] 2. Design of the JORDAN neural network model for GWO

[0060] The Jordan neural network mainly consists of an input layer, hidden layers, an output layer, and a feedback layer. Let the neural network have n inputs, m outputs, m feedbacks, and q hidden units, where z represents the feedback delay of unit 1. Let the input of the Jordan neural network be:

[0061] (9)

[0062] The Jordan neural network output is:

[0063] (10)

[0064] Hidden layer outputs of the Jordan neural network:

[0065] (11)

[0066] The inputs and outputs of hidden layer neurons are:

[0067] (12)

[0068] in: It is the connection weight matrix from the input layer to the hidden layer. This is the hidden layer threshold matrix. is the activation function for neurons in the hidden layer.

[0069] The input and output of the Jordan neural network output layer are:

[0070] (13)

[0071] in: It is the connection weight matrix from the hidden layer to the output layer. For the output layer threshold matrix, is the activation function for the output layer neurons.

[0072] The design method of the JORDAN neural network model of GWO refers to the design method and steps of the BiLSTM neural network model of GWO in this patent.

[0073] 3. Design of GWO fuzzy recurrent neural network model

[0074] A fuzzy recurrent neural network model contains n input nodes. The signal transmission process within the fuzzy recurrent neural network model and the input-output relationships between each layer can be described as follows:

[0075] Layer 1: The nodes in this layer introduce input variables into the network. Feedback connections are added to this layer. The input-output relationship of Layer 1 can be described as follows:

[0076] (14)

[0077] in: For the i-th input of the first layer, This refers to the recursive layer weights for the i-th input in the first layer. Adding a recursive feedback loop to the first layer improves the robustness, accuracy, and reliability of the input values ​​in the recursive fuzzy neural network model. and This represents the output and input values ​​of the first-level node.

[0078] Layer 2: In this layer, the nodes fuzzify the input variables. Each node represents a membership function, using the Gaussian function as the membership function. The input-output relationship of Layer 2 can be described as follows:

[0079] (15)

[0080] in, and represents the center value and width of the Gaussian function, respectively; i is the input of the i-th node; and j is the language vocabulary set of the input of the i-th node. and These represent the input and output values ​​of the language vocabulary set of the i-th node in the second layer, respectively.

[0081] The third layer is the fuzzy inference layer, which uses multiplication to implement fuzzy "AND" operations. In this network, multiplication is used instead of the minimum operation.

[0082] The fourth layer is the output layer, which implements the deblurring function. The output of the fuzzy recurrent neural network model is:

[0083] (16)

[0084] in For Layer 4 network connectivity rights, Output values ​​for the 3rd layer nodes. This is the output value of the recursive fuzzy neural network model.

[0085] The design method of the GWO fuzzy recurrent neural network model refers to the design method and steps of the GWO BiLSTM neural network model in this patent.

[0086] 4. GWO's Fuzzy Recurrent Neural Network Model - Design of GWO's JORDAN Neural Network Model

[0087] The GWO fuzzy recurrent neural network model—the GWO JORDAN neural network model—consists of the GWO fuzzy recurrent neural network model, the GWO JORDAN neural network model, delay units, and feedback layers. Its structure diagram is shown below. Figure 2As shown. The input values ​​are used as the inputs to the GWO fuzzy recurrent neural network model and the input to delay unit 1. The output of delay unit 1 is used as the input to the GWO fuzzy recurrent neural network model. The output of the GWO fuzzy recurrent neural network model is used as the input to the GWO JORDAN neural network model and the input to delay unit 2. The output of delay unit 2 is used as the input to the GWO JORDAN neural network model. The output value Y(t) of the GWO JORDAN neural network model is used as the input to delay unit 3. The output value Y(t-1) of delay unit 3 is used as the input to the feedback layer. The output value X(t) of the feedback layer is used as the input to delay unit 5, delay unit 4, and the GWO fuzzy recurrent neural network model. The output values ​​X(t-1) of delay unit 4 and delay unit 5 are used as the input to the GWO JORDAN neural network model and the feedback layer. The connection weights between X(t-1), Y(t-1), X(t), delay unit 5, and Y(t-1), X(t-1), and X(t) are used as the inputs. and The feedback layer is formed, and the relationship between its output and input is as follows:

[0088] (17)

[0089] Where: X(t) is the output value of the feedback layer, X(t-1) is the output value of the feedback layer at the previous time step, and Y(t-1) is the output value of the GWO fuzzy recurrent neural network model - the GWO JORDAN neural network model at the previous time step. and These are the connection weights between the input and output of the feedback layer.

[0090] The advantages of the GWO fuzzy recurrent neural network model—compared to the GWO JORDAN neural network model—are that the recurrent layers at the input of the GWO fuzzy recurrent neural network model improve the accuracy, dynamism, and robustness of the input; the output of the GWO fuzzy recurrent neural network model serves as the input of the GWO JORDAN neural network model, and because the input of the GWO JORDAN neural network model is fed back to the input of the GWO JORDAN neural network model through delay units, the accuracy, dynamism, and robustness of the output value of the GWO JORDAN neural network model are improved. The feedback layer outputs of the GWO fuzzy recurrent neural network model—compared to the GWO JORDAN neural network model—are used as inputs for the GWO fuzzy recurrent neural network model. By inputting into the network model and the GWO JORDAN neural network model, the accuracy, dynamics, and robustness of the GWO fuzzy recurrent neural network model-GWO JORDAN neural network model are improved. Furthermore, the feedback layer, together with the input recurrent layer and the feedback of the GWO fuzzy recurrent neural network model and the GWO JORDAN neural network model, forms an inner and outer two-level feedback layer, which further improves the dynamics, accuracy, and robustness of the output value of the GWO fuzzy recurrent neural network model-GWO JORDAN neural network model. Moreover, the output value of the feedback layer is fed back to the input end of the feedback layer through the delay unit 5, which further improves the dynamics, accuracy, and robustness of the GWO fuzzy recurrent neural network model-GWO JORDAN neural network model.

[0091] 5. GWO's BiLSTM Neural Network Model - Design of GWO's JORDAN Neural Network Model

[0092] The GWO BiLSTM neural network model – the GWO JORDAN neural network model – consists of the GWO BiLSTM neural network model, the GWO JORDAN neural network model, delay units, and feedback layers. Its structure diagram is shown below. Figure 3As shown. The input value serves as the input to the GWO BiLSTM neural network model and the input to delay unit 6. The output of delay unit 6 serves as the input to the GWO BiLSTM neural network model. The output of the GWO BiLSTM neural network model serves as the input to the GWO JORDAN neural network model and the input to delay unit 7. The output of delay unit 7 serves as the input to the GWO JORDAN neural network model. The output value O(t) of the GWO JORDAN neural network model serves as the input to delay unit 8. The output value O(t-1) of delay unit 8 serves as the input to the feedback layer. The output value Z(t) of the feedback layer serves as the input to delay unit 10, delay unit 9, and the GWO BiLSTM neural network model. The output values ​​Z(t-1) of delay unit 9 and delay unit 10 serve as the input to the GWO JORDAN neural network model and the feedback layer. The connection weights between Z(t-1), O(t-1), Z(t), delay unit 10, and O(t-1), Z(t-1), and Z(t) are used as the inputs. and The feedback layer is formed, and the relationship between its output and input is as follows:

[0093] (18)

[0094] Where: Z(t) is the current output value of the feedback layer, Z(t-1) is the output value of the feedback layer at the previous time step, and O(t-1) is the output value of the GWO BiLSTM neural network model - GWO JORDAN neural network model at the previous time step. and These are the connection weights between the input and output of the feedback layer.

[0095] The advantages of the GWO BiLSTM neural network model compared to the GWO JORDAN neural network model are: the output of the GWO BiLSTM model captures temporal information from both directions of illuminance parameter changes, improving the accuracy of the GWO BiLSTM model in predicting illuminance parameters; the output of the GWO BiLSTM model serves as the input to the GWO JORDAN neural network model, and because the output of the GWO JORDAN model is fed back to the input through delay units, the accuracy, dynamics, and robustness of the output values ​​of the GWO JORDAN neural network model are improved. The output of the feedback layer serves as the input to both the GWO BiLSTM neural network model and the GWO JORDAN neural network model, improving the accuracy, dynamics, and robustness of the GWO BiLSTM-JORDAN neural network model. Furthermore, the feedback layer and the feedback from the GWO JORDAN neural network model form a two-tiered feedback layer, further enhancing the dynamics, accuracy, and robustness of the output values ​​of the GWO BiLSTM-JORDAN neural network model. The output values ​​of the feedback layer are also fed back to the input of the feedback layer through the delay unit 10, further improving the dynamics, accuracy, and robustness of the GWO BiLSTM-JORDAN neural network model.

[0096] 6. Design of GWO's Fuzzy Recurrent Neural Network Model - PID Controller

[0097] In the process of illuminance regulation, since there are many factors affecting illuminance, a PID controller is combined with a GWO fuzzy recurrent neural network model. The three parameters output by the GWO fuzzy recurrent neural network model are used as parameters for the PID controller. , and The GWO fuzzy recurrent neural network model-PID controller uses the illuminance error e and the error change rate ec as inputs, which can meet the self-tuning requirements of the PID controller parameters for illuminance e and error change rate ec at different times. The output of the PID controller can be represented by the following function:

[0098] (19)

[0099] 7. Design of the Adaline Neural Network Model

[0100] The Adaline neural network model adaptively adjusts the weights by utilizing the error between the neuron's output value and the desired output value, ensuring that the neuron's output tracks the desired output, thus achieving the fusion of multiple illuminance parameters. The input-output relationship of the Adaline neural network model is as follows:

[0101] (20)

[0102] in: , and Let be the weights, input signal, and output of the Adaline neural network model, respectively. If we let... The expected output for the actual target of the illuminance parameter. Let be the convergence step size factor for the weights. Then, the weight adjustment algorithm for the Adaline neural network model based on least mean square (LMS) is as follows:

[0103] (twenty one)

[0104] 8. Performance test of the illuminance control system

[0105] In this illuminance control system, the fuzzy recurrent neural network model of GWO—the PID controller—is designated as A, the fuzzy recurrent neural network model of GWO—the JORDAN neural network model of GWO—is designated as B, and the illuminance disturbance controller GWO—the JORDAN neural network model—is designated as C. During this test, the illuminance control system using only the A+C combination is designated as 1, the illuminance control system using only the B+C combination is designated as 2, and the illuminance control system using the A+B+C combination is designated as 3. This test used a street lighting scenario in Huai'an City. Sensor group 1 was a visibility sensor group, sensor group 2 was a temperature and humidity sensor group, and sensor group 3 was a traffic flow sensor group. The full power consumption was 1.4 kWh. The specific energy consumption data and lighting effect data after using different types of illuminance control systems are shown in Table 1 below.

[0106] Table 1 Energy consumption optimization test results

[0107] 1 (A+C) 0.88 37.14% Larger 5.25 2 (B+C) 0.84 40.00% smaller 4.57 3 (A+B+C) 0.69 50.71% very small 3.62

[0108] As shown in Table 1, the illuminance control system of the present invention has the most outstanding energy consumption optimization performance with 3, and it is also superior to other algorithms in terms of stability and response time. The overall performance of the illuminance control system of this patent is the best, which will promote the widespread application of intelligent lighting systems in future cities.

[0109] 9. Illuminance Big Data Internet of Things System

[0110] The illuminance big data IoT system comprises illuminance measurement and control nodes, an illuminance big data cloud platform, and an illuminance management terminal. The illuminance measurement and control nodes are responsible for detecting illuminance environmental parameters and adjusting illuminance. Communication of illuminance parameters between the measurement and control nodes, the illuminance big data cloud platform, and the illuminance management terminal is achieved through their wireless communication modules. The illuminance management terminal contains an illuminance control system that monitors illuminance parameters and communicates with the illuminance big data cloud platform via 5G. See [link to illuminance big data IoT system] for details. Figure 4 As shown.

[0111] The illuminance measurement and control node uses the ESP8266 wireless communication module as the terminal for measuring and controlling illuminance parameters. The node communicates bidirectionally with the illuminance big data cloud platform via the ESP8266. The node includes sensors for acquiring ambient illuminance and temperature, sensor 1, sensor 2, sensor 3 and corresponding signal conditioning circuits, an STM32 microprocessor, a power acquisition module, a light source driver, and the ESP8266 wireless communication module. The software of the illuminance measurement and control node mainly implements bidirectional communication between the node and the illuminance big data cloud platform, as well as the acquisition and control of illuminance parameters. The software is designed in C language, offering high compatibility and significantly improving the efficiency of software design and development, while enhancing the reliability, readability, and portability of the program code. The structure of the illuminance measurement and control node is shown below. Figure 5 .

[0112] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for monitoring illuminance big data, characterized in that, Includes the following steps: Step 1: Set up light intensity sensor group, sensor group 1, sensor group 2 and sensor group 3, collect real-time data values ​​of each sensor group, and fuse them through the corresponding Adaline neural network model to obtain the real-time light intensity value, the real-time value of sensor group 1, the real-time value of sensor group 2 and the real-time value of sensor group 3. Step 2: Use GWO's BiLSTM neural network model and GWO's JORDAN neural network model to dynamically predict each real-time value, and input the predicted value into GWO's fuzzy recurrent neural network model to obtain the ideal value of illuminance. Step 3: Based on the ideal illuminance value and the real-time illuminance value, obtain the real-time illuminance control quantity through intelligent control; obtain the predicted illuminance control quantity based on the ideal illuminance value and the predicted illuminance value; and obtain the illuminance perturbation quantity after processing the real-time illuminance value and the actual illuminance control quantity through the JORDAN neural network model of GWO. When obtaining the illuminance prediction control quantity based on the ideal illuminance value and the predicted illuminance value, the difference between the ideal illuminance value and the predicted illuminance value output by the GWO BiLSTM neural network model - GWO JORDAN neural network model and the rate of change of the difference are used as the input of the GWO fuzzy recurrent neural network model - GWO JORDAN neural network model. The illuminance is predicted and controlled using the GWO fuzzy recurrent neural network model - GWO JORDAN neural network model, and the output of the GWO fuzzy recurrent neural network model - GWO JORDAN neural network model is used as the illuminance prediction control quantity. The GWO fuzzy recurrent neural network model – the GWO JORDAN neural network model – consists of the GWO fuzzy recurrent neural network model, the GWO JORDAN neural network model, delay units, and a feedback layer. The input value serves as the input to the GWO fuzzy recurrent neural network model and the input to delay unit 1. The output of delay unit 1 serves as the input to the GWO fuzzy recurrent neural network model. The output of the GWO fuzzy recurrent neural network model serves as the input to the GWO JORDAN neural network model and the input to delay unit 2. The output of delay unit 2 serves as the input to the GWO JORDAN neural network model. The output value Y(t) of the GWO JORDAN neural network model serves as the input to delay unit 3. The output of delay unit 3 serves as the input to the feedback layer. The output value of the feedback layer serves as the input to delay unit 5, delay unit 4, and the GWO fuzzy recurrent neural network model. The outputs of delay unit 4 and delay unit 5 serve as the inputs to the GWO JORDAN neural network model and the feedback layer, respectively. The inputs to the feedback layer are Y(t-1) and X(t-1), the connection weights of the feedback layer are w1 and w2, and the output of the feedback layer is X(t). The relationship between the output and input of the feedback layer is as follows: ； The GWO BiLSTM neural network model – GWO JORDAN neural network model – includes a GWO BiLSTM neural network model, a GWO JORDAN neural network model, delay units, and a feedback layer. The input values ​​serve as the input to both the GWO BiLSTM neural network model and delay unit 6. The output of delay unit 6 serves as the input to the GWO BiLSTM neural network model. The output of the GWO BiLSTM neural network model serves as the input to both the GWO JORDAN neural network model and delay unit 7. The output of delay unit 7 serves as the input to the GWO JORDAN neural network model. The input to the network model is the output value O(t) of the JORDAN neural network model of GWO, which is used as the input to delay unit 8. The output value O(t-1) of delay unit 8 is used as the input to the feedback layer. The output value Z(t) of the feedback layer is used as the input to delay unit 10, delay unit 9, and the BiLSTM neural network model of GWO. The outputs of delay unit 9 and delay unit 10 are used as the inputs to the JORDAN neural network model of GWO and the feedback layer, respectively. The inputs to the feedback layer are O(t-1) and Z(t-1), the connection weights of the feedback layer are w3 and w4, and the output of the feedback layer is Z(t). The relationship between the output and input of the feedback layer is as follows: ； Step 4: Finally, the difference between the sum of the predicted illuminance control quantity and the real-time illuminance control quantity and the illuminance disturbance quantity is used as the actual illuminance control quantity. The actual illuminance control quantity is used as the input of the light source driver and the input of the JORDAN neural network model of GWO in step 3 for cyclic control. The output of the light source driver is used as the light source input to realize the dynamic adjustment of the illuminance of the light source adjustment environment.

2. The method for monitoring illuminance big data according to claim 1, characterized in that, The sensor group 1, sensor group 2 and sensor group 3 are groups of environmental parameters that affect illuminance, including but not limited to the visibility sensor group, temperature and humidity sensor group and traffic flow sensor group.

3. The method for monitoring illuminance big data according to claim 1, characterized in that, In step 3, when obtaining the real-time control quantity of illuminance through intelligent control based on the ideal value of illuminance and the real-time value of illuminance, the GWO fuzzy recurrent neural network model-PID controller is selected. The difference between the ideal value of illuminance and the real-time value of illuminance and the rate of change of the difference are used as the input of the GWO fuzzy recurrent neural network model-PID controller, and the GWO fuzzy recurrent neural network model-PID controller outputs the real-time control quantity of illuminance.

4. The method for monitoring illuminance big data according to claim 1, characterized in that, When obtaining the illuminance perturbation after processing the real-time illuminance value and the actual illuminance control quantity through the JORDAN neural network model of GWO, the actual illuminance control quantity and the actual illuminance value output by the Adaline neural network model are used as the inputs of the JORDAN neural network model of GWO, and the output of the JORDAN neural network model of GWO is used as the illuminance perturbation.

5. A big data Internet of Things system for illumination, characterized in that, The system includes an illuminance measurement and control node, an illuminance big data cloud platform, and an illuminance management terminal. The illuminance measurement and control node is responsible for detecting light source environmental parameters and adjusting illuminance. Data communication between the illuminance measurement and control node, the illuminance big data cloud platform, and the illuminance management terminal is achieved through the wireless communication module of the illuminance measurement and control node and 5G. The illuminance management terminal monitors illuminance parameters and communicates with the illuminance big data cloud platform through 5G. The illuminance management terminal is equipped with an illuminance control system, thereby realizing the illuminance big data monitoring method as described in any one of claims 1 to 4.

Images