A method for controlling LED underwater lights based on remote communication integrated management

Abstract

The present invention discloses a method for controlling an LED underwater light based on integrated remote communication management, comprising the following steps: real-time collection of a variety of underwater environment data and real-time working status data of an underwater light, and compression to obtain a variety of underwater environment compressed data and real-time working status compressed data; nesting and associating the compressed data to form a nested underwater data packet; generating a real-time transmission strategy based on the nested underwater data packet using a control frame interval period, and transmitting the strategy; generating an environment-state comprehensive dictionary for the nested underwater data packet, and inputting an underwater light normal operating state model to obtain the real-time operating state of the underwater light; generating the environmental adaptability of the underwater light based on the environment-state comprehensive dictionary; inputting the real-time operating state of the underwater light and the environmental adaptability into the underwater light control model, and generating an underwater control instruction to transmit to the underwater light. The present invention significantly improves the adaptability, control efficiency and control accuracy of the underwater light to a dynamic environment.

Description

Technical Field

[0001] The present invention relates to the technical field of underwater lighting control, and in particular to a method for controlling an underwater LED light based on remote communication integrated management. Background Art

[0002] With the rapid development of marine resource development, marine exploration, underwater operations and underwater tourism, the demand for underwater lighting technology is growing. LED underwater lights have gradually become the main choice for underwater lighting due to their high efficiency, energy saving, long life and strong environmental protection. However, the complexity and dynamic changes of the underwater environment have posed severe challenges to the performance, control methods and adaptability of LED underwater lights.

[0003] (1) Impact of underwater environment complexity on lighting system

[0004] The underwater environment has unique physical properties, including light attenuation, refraction, scattering, and variability of environmental parameters (such as temperature, pressure, light intensity, turbidity, salinity, etc.). In deep water environments, light attenuation and dispersion are particularly obvious, and the penetration depth of light of different wavelengths in water varies greatly, affecting the effective coverage of the light. In addition, different water qualities, different depths, and environmental changes have different requirements for light brightness, spectral distribution, and energy consumption. If the operating status of underwater lights cannot be adjusted in real time according to environmental changes, the lighting effect may be reduced or energy consumption may be too high.

[0005] (2) Insufficient monitoring and control of the real-time operating status of underwater lights

[0006] The long-term operation of LED underwater lights will be affected by many factors, such as the dynamic changes of the underwater environment, the performance fluctuations of the electrical system, etc. If the operating status of the underwater lights cannot be monitored and analyzed in real time, problems such as equipment overheating, light attenuation, abnormal power consumption, etc. may occur, which will in turn affect the service life and operational reliability of the equipment. However, most of the current control systems of underwater lights only rely on simple preset modes, lack the ability of real-time monitoring and status adjustment, and are difficult to meet the needs of complex underwater environments.

[0007] (3) Limitations of data transmission and remote control

[0008] Due to the limitations of underwater communication conditions, data transmission efficiency and reliability have always been important factors affecting the intelligent control of underwater equipment. Therefore, how to achieve efficient transmission of real-time status data and environmental data of underwater lights under limited communication conditions and perform remote control based on these data is an urgent problem to be solved by current technology.

[0009] (4) Insufficient environmental adaptability and intelligent control

[0010] Existing underwater light control technologies are mainly based on fixed parameters or simple environmental adaptation algorithms, which make it difficult to achieve intelligent adaptation to dynamic environments. For example, under different depths and water quality conditions, underwater lights may need to adjust brightness, spectral distribution, and operating mode to achieve the best effect, and traditional methods cannot achieve these dynamic adjustments.

[0011] (5) Insufficient data processing and comprehensive analysis capabilities

[0012] Underwater environmental data and underwater light operating status data are multidimensional and highly temporally and spatially correlated, which poses great challenges to data collection, compression, transmission and analysis. Existing control methods are often unable to effectively integrate and analyze large amounts of real-time data, resulting in lagging control strategies, poor environmental adaptability and serious waste of resources. Summary of the invention

[0013] In order to overcome the shortcomings of the prior art, the present invention provides a LED underwater light control method based on remote communication integrated management, which is used to solve the technical problems of the existing underwater light control technology lacking adaptability to dynamic environments, lacking real-time monitoring and status adjustment capabilities, and limited intelligence level, thereby significantly improving the adaptability of underwater lights to dynamic environments, control efficiency and control accuracy.

[0014] To solve the above problems, the technical solution adopted by the present invention is as follows:

[0015] A method for controlling an LED underwater light based on remote communication integrated management comprises the following steps:

[0016] Collect a variety of underwater environmental data in real time, and compress them through a high-pressure method based on context relevance to obtain a variety of underwater environmental compressed data;

[0017] Collect the real-time working status data of the underwater light in real time, and compress it using the best sparse dictionary to obtain the real-time working status compressed data;

[0018] The various underwater environment compressed data and the real-time working status compressed data are nested and associated based on the subject, superset, header, support and confidence to form a nested underwater data packet;

[0019] Generate a real-time transmission strategy based on the nested underwater data packets using the control frame interval period and transmit;

[0020] Generate underwater data saliency map and feature matrix for nested underwater data packets, and improve The function is normalized to obtain the environment-state comprehensive dictionary;

[0021] Input the environment-state comprehensive dictionary into the normal operation state model of the underwater light to obtain the real-time operation state of the underwater light;

[0022] Based on the environment-state comprehensive dictionary, the environmental adaptation coupling degree and environmental adaptation coupling coordination degree are obtained, and the environmental adaptability of the underwater light is generated;

[0023] Construct an underwater light control model, input the real-time operating status and environmental adaptability of the underwater light into the underwater light control model, and generate underwater control instructions based on the state-action value function and transmit them to the underwater light.

[0024] As a preferred embodiment of the present invention, when using the optimal sparse dictionary for compression, it includes:

[0025] An observation matrix is ​​constructed for the real-time working state data measured signal, and the real-time working state data measured signal is mapped from the high-dimensional observation space to the low-dimensional observation space to obtain the real-time working state data observation signal;

[0026] According to the real-time working state data measured signal and the real-time working state data observation signal, the reconstruction error is obtained. ;

[0027] Perform atomic updates one by one until the reconstruction error Less than the set error threshold , output the best sparse dictionary, and use the best sparse dictionary to express the characteristics of the measured signal of the complete real-time working status data.

[0028] As a preferred embodiment of the present invention, when performing nested association, it includes:

[0029] The preset environment-state relationship Projecting onto the identification attribute and itemset attribute of the compressed data, and nesting the data tuples according to the identification attribute;

[0030] Group data tuples according to item set attributes, generate frequent item sets, and obtain the support of each frequent item set;

[0031] Extract association rules from frequent item sets, generate copies of frequent item sets, connect copies, and generate entities;

[0032] Identify the small item set that is the subject of the potential rule and the superset of the corresponding large item set;

[0033] Select the association rules whose confidence is greater than the confidence threshold, perform difference operation on the subject and the superset, and generate the head of the association rule;

[0034] The subject, superset, header, support and confidence are formed into a nested relationship, and the compressed data is formed into a nested underwater data packet according to the nested relationship.

[0035] As a preferred implementation mode of the present invention, when generating a real-time transmission strategy, it includes:

[0036] A control frame is transmitted for the subject, superset, and header on the control channel, and the data of the last successful scheduling is sent on the data channel;

[0037] Wait until the control frame returns, and obtain the value of the channel available time and the value of the target node receivable time respectively;

[0038] Get the data of the last unsuccessful scheduling, update the value of its data period using the control frame interval period, and update it with the data of this control frame according to The algorithm sorts the data and obtains a data queue;

[0039] Find the first available data channel for the data in the data queue, obtain the timeout judgment parameter and the first timeout judgment time according to the receivable time of the target node and the available time of the channel, and generate the second timeout judgment time;

[0040] The data transmission time is obtained according to the second timeout judgment time, the value of the target node receivable time and the value of the channel available time are updated, the successfully scheduled data is sent on the data channel, and the next round of control frames is sent on the control channel.

[0041] As a preferred embodiment of the present invention, when obtaining the environment-state comprehensive dictionary, it includes:

[0042] The nested underwater data packets are divided according to the frequency channels, and the underwater data of each frequency channel is convolved in any direction through multiple one-dimensional Gaussian smoothing filters, and down-sampled to obtain the representation of the underwater data of each frequency channel at multiple scales;

[0043] The center-surround difference algorithm is used to simulate the characteristics of local underwater environment suppression, and the underwater saliency of underwater data of each frequency channel at different scales is obtained;

[0044] The normalized underwater saliency is obtained by a global nonlinear method amplification algorithm;

[0045] Convolve all normalized underwater saliency in the frequency channel with a one-dimensional difference Gaussian filter, and obtain the difference between the result after convolution and the underwater saliency before normalization;

[0046] The difference is added to the normalized underwater saliency to obtain the cross-scale integrated underwater saliency of each frequency channel, and linear integration is performed to obtain the underwater data saliency map.

[0047] As a preferred embodiment of the present invention, after obtaining the underwater data saliency map, the method includes:

[0048] The nested underwater data packets and underwater data saliency maps are processed by time domain to obtain multiple groups of underwater data and underwater data saliency maps, and then input into the convolutional neural network model. layer, and obtain the corresponding one-dimensional features;

[0049] The one-dimensional feature is concatenated with underwater data having the same time dimension, and a convolutional neural network is used to extract features by combining the principal component analysis results of the underwater data and the underwater data saliency map, and a multi-channel feature of each group of underwater data after framing is output;

[0050] Linearly combine the multi-channel features of each group of underwater data in time order to generate a feature matrix;

[0051] Through the fully connected layer, the improvement The activation function of the function normalizes the feature matrix to obtain a normalized feature matrix;

[0052] Establish the mapping relationship between the target indicator ID and the normalized feature matrix to generate the environment-state comprehensive dictionary.

[0053] As a preferred embodiment of the present invention, when obtaining the real-time operating status of the underwater light, it includes:

[0054] according to Similarity obtains similar underwater light groups and their fusion datasets;

[0055] Divide the training data set and test data set of the target monitoring underwater lights from the fused data set and perform preprocessing;

[0056] The common feature information is extracted from the fusion data set through the spatiotemporal autoencoder, and the common feature information is transferred to the state monitoring model of the target underwater light through parameter migration. The normal operation state model of the underwater light is established in combination with the training data set.

[0057] based on The standard deviation of the statistic is combined Control chart to obtain the first threshold , the second threshold ;

[0058] The test data set is input into the normal operating state model of the underwater light to obtain the reconstruction residual and pass Control chart, first threshold and the second threshold Determine the operating status of the target monitoring underwater light.

[0059] As a preferred embodiment of the present invention, when extracting common feature information, it includes:

[0060] The fused data set is preprocessed and input into the spatiotemporal autoencoder. By adjusting the convolution kernel size and movement direction of the encoding network, common feature information extraction and regression analysis are completed.

[0061] When establishing the normal operating state model of underwater lights, include:

[0062] The pre-processed real-time working status time series data is input in a sliding window manner;

[0063] Select all monitoring variables of the last moment of the input real-time working status time series data as the target monitoring quantity;

[0064] The mapping relationship between the real-time working status time series data and the target monitoring quantity is obtained to obtain the normal operating status model of the underwater light.

[0065] As a preferred embodiment of the present invention, when obtaining the environmental adaptability of the underwater light, it includes:

[0066] Obtaining a target indicator set regarding underwater environmental sensitivity and underwater light intensity from the environment-state comprehensive dictionary;

[0067] The original value, maximum value and minimum value of each target indicator in the spatial pixel are obtained, and the range normalized value of underwater environmental sensitivity and the range normalized value of underwater light intensity are generated. The values ​​are resampled to a grid of a preset size, and a coupling degree model is introduced to generate the environmental adaptation coupling degree.

[0068] Introducing a coupling coordination model for the environmental adaptation coupling degree to generate an environmental adaptation coupling coordination degree;

[0069] The environmental adaptability of the underwater light is generated according to the environmental adaptability coupling degree and the environmental adaptability coupling coordination degree.

[0070] As a preferred embodiment of the present invention, when constructing an underwater light control model, it includes:

[0071] pass The algorithm combines deep neural networks with Algorithms are combined to make decisions network;

[0072] Initialization decision Network, through decision The network obtains the value of the state-action value function and stores the state, action, and reward through the experience replay cache;

[0073] Repeat the parameter update steps until the preset number of training times is reached or the algorithm converges, and the decision is completed. Update the network parameters to obtain the underwater light control model;

[0074] The parameter updating step includes: controlling action selection, obtaining state feedback, updating the experience playback cache, and updating network parameters;

[0075] Updating the network parameters involves sampling batches of data from the updated experience replay buffer, obtaining the value of the updated state-action value function, and updating the current decision Network parameters are copied to the target decision Network, decision making Update of network parameters.

[0076] Compared with the prior art, the present invention has the following beneficial effects:

[0077] (1) Efficient data compression and transmission

[0078] The underwater environment data and real-time working status data are compressed by using a context-related high-pressure method and an optimal sparse dictionary, which greatly reduces the data transmission volume, improves the transmission efficiency, adapts to the low bandwidth requirements under complex underwater communication conditions, and ensures the integrity and high accuracy of the data.

[0079] (2) Nested association and intelligent analysis

[0080] Through the nested association of multi-dimensional data and the combination of attributes such as subject, superset, head, support and confidence, in-depth mining and association analysis between underwater environment data and underwater light status data are achieved, so as to fully grasp the dynamic changing relationship between the operating environment and status of underwater lights.

[0081] (3) Integrated management of environment and status

[0082] Based on the environment-state comprehensive dictionary and underwater data saliency map, the multi-dimensional characteristics of the underwater environment and the working state of the LED underwater light are systematically integrated to form a unified analysis model, which supports the accurate prediction and optimal control of the operating state of the underwater light.

[0083] (4) Intelligent real-time control

[0084] An underwater light control model is constructed using an improved state-action value function and deep neural network. The real-time operating status and environmental adaptability evaluation results are combined to dynamically generate high-precision underwater control instructions, which effectively improves the intelligent operation capability of underwater lights, reduces manual intervention, and improves work efficiency.

[0085] (5) Environmental adaptability and adaptive optimization

[0086] By calculating the environmental adaptation coupling degree and coupling coordination degree, the adaptability of underwater lights to the environment can be evaluated in real time, and adaptive optimization and adjustment can be performed according to different environmental conditions, thereby improving the applicability and stability of underwater lights in complex and changeable underwater environments.

[0087] (6) Multi-objective optimization and performance improvement

[0088] The system realizes multi-objective optimization analysis by extracting frequent item sets and mining rules. It improves the transmission stability and real-time performance of underwater data by controlling frame intervals and dynamic scheduling strategies, thereby further optimizing the overall system performance.

[0089] (7) Intelligent diagnosis and fault prediction

[0090] By using the spatiotemporal autoencoder to perform feature extraction and transfer learning on the fused data set, a normal operating status model of the underwater light was established, which can realize accurate monitoring of the real-time operating status and early warning of faults, thereby improving the safety and reliability of the system.

[0091] (8) Energy saving, high efficiency and economy

[0092] The present invention realizes dynamic optimization of energy consumption by precisely controlling the operating state of underwater lights and combining environmental adaptability analysis, thereby greatly reducing the energy consumption of underwater lights and extending the service life of equipment, thus having significant economic benefits and environmental protection value.

[0093] In summary, the present invention combines a variety of advanced algorithms and technical means to improve the intelligent operation, data transmission efficiency, environmental adaptability and system reliability of underwater lights, while realizing low-cost and high-efficiency comprehensive control and management of underwater lights. It has important technological innovation significance and broad application prospects.

[0094] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] Figure 1 It is a step diagram of the LED underwater light control method based on remote communication integrated management provided by the present invention;

[0096] Figure 2 It is a structural schematic diagram of the convolutional neural network model provided by the present invention.

[0097] Description of Figure Numbers: 1. Layer; 2, the first convolutional layer; 3, the first pooling layer; 4, the second convolutional layer; 5, the second pooling layer; 6, the fully connected layer. DETAILED DESCRIPTION

[0098] The LED underwater light control method based on remote communication integrated management provided by the present invention is as follows: Figure 1 As shown, the following steps are included:

[0099] Step S1: collecting a variety of underwater environment data in real time through the environment perception sensor built into the underwater light, and compressing them through a high-pressure method based on context relevance to obtain a variety of underwater environment compressed data;

[0100] Step S2: collecting the real-time working status data of the underwater light in real time through the device perception sensor built into the underwater light, and compressing it using the best sparse dictionary to obtain the real-time working status compressed data;

[0101] Step S3: nesting and associating the multiple underwater environment compressed data and the real-time working status compressed data based on the subject, superset, header, support, and confidence to form a nested underwater data packet;

[0102] Step S4: generating a real-time transmission strategy according to the nested underwater data packet using the control frame interval period through the remote communication module, and transmitting the nested underwater data packet to the remote control server based on the real-time transmission strategy;

[0103] Step S5: Generate underwater data saliency map and feature matrix for the nested underwater data packets through the remote control server, and The function is normalized to obtain the environment-state comprehensive dictionary;

[0104] Step S6: construct a normal operating state model of the underwater light, input the environment-state comprehensive dictionary into the normal operating state model of the underwater light, and obtain the real-time operating state of the underwater light;

[0105] Step S7: obtaining the environmental adaptation coupling degree and the environmental adaptation coupling coordination degree based on the environment-state comprehensive dictionary, and generating the environmental adaptability of the underwater light;

[0106] Step S8: construct an underwater light control model, input the real-time operating status and environmental adaptability of the underwater light into the underwater light control model, and generate underwater control instructions based on the state-action value function and transmit them to the underwater light.

[0107] Specifically, the environmental perception sensors include: temperature sensors, pressure sensors, light intensity sensors, turbidity sensors, and salinity sensors.

[0108] Temperature sensor, used to monitor the real-time temperature of the underwater environment.

[0109] The pressure sensor is used to measure the water depth and water pressure, and dynamically adjust the operation mode of the underwater light based on the measured pressure data.

[0110] Light intensity sensor, used to sense the natural light intensity of the underwater environment and assist in adjusting the brightness of the LED underwater light.

[0111] Turbidity sensors are used to monitor the turbidity of underwater environments to determine water transparency and adjust light output effects.

[0112] Salinity sensors are used to measure the salt concentration in water, especially to determine the conductivity and corrosiveness of water in seawater environments.

[0113] Device perception sensors include: temperature sensor, current and voltage sensor, power consumption sensor, operation time recording module, and humidity sensor.

[0114] Temperature sensors are used to monitor the temperature of the underwater light housing and interior (such as LED chip temperature, power module temperature) to prevent the equipment from being damaged due to overheating.

[0115] Current and voltage sensors are used to monitor the operating current and voltage of underwater lights to ensure that the equipment operates within the rated power range.

[0116] Power consumption sensor for real-time monitoring of total power consumption of underwater lights, helping to optimize energy efficiency and operating modes.

[0117] The operating time recording module is used to record the cumulative operating time of the equipment, evaluate the life of the lamps based on historical data, and prompt maintenance or replacement.

[0118] Humidity sensor to monitor the humidity inside the lamp to detect if there is a risk of seal failure or water ingress.

[0119] The remote communication module includes: a wired remote communication module and a wireless remote communication module. Preferably, a wireless remote communication module is used, and the wireless remote communication module supports remote communication modes such as Wi-Fi, 4G, 5G or LoRa.

[0120] Furthermore, the control method provided by the present invention also includes: the user accesses the remote control server through the terminal device, views the real-time working status data, operation history records and current underwater environment data of the underwater light in real time, and performs remote control operations.

[0121] In the above step S1, when compression is performed based on the context-related high-pressure method, it includes:

[0122] Read multiple underwater environment data and determine whether a reference value exists. If not, convert the first multiple underwater environment data into a reference value structure and use it as the reference value;

[0123] If it exists, perform difference calculation with the reference value and determine whether the difference exceeds the preset value. If not, save it as a difference structure;

[0124] Compressing various underwater environment data through a reference value structure and a difference value structure to obtain various underwater environment compressed data;

[0125] Among them, the last digit of the benchmark value structure is used to identify the number of difference structures carried by the benchmark value. Various underwater environmental data include: temperature, pressure, light intensity, turbidity, and salinity.

[0126] Specifically, when the correlation between underwater environmental data is high, the difference between the two is not large. At this time, when compressing these data, only a baseline value structure needs to be stored, and the underwater environmental data whose difference with the baseline value is within the preset value (that is, underwater environmental data with high correlation) is stored in a difference structure. Therefore, a baseline value structure plus several difference structures can accurately represent a series of underwater environmental data with high correlation, thereby achieving high-proportion compression. Therefore, it is called "context-relevance-based high-pressure method", where "high pressure" refers to high-proportion compression.

[0127] In the above step S2, when the optimal sparse dictionary is used for compression, it includes:

[0128] Obtain the sparse representation of the measured signal of real-time working status data in its sparse domain, as shown in Formula 1: (1);

[0129] In the formula, The measured signal is the real-time working status data. is a sparse dictionary, is the sparse expression coefficient;

[0130] An observation matrix is ​​constructed for the measured signal of the real-time working status data, and the measured signal of the real-time working status data is mapped from the high-dimensional observation space to the low-dimensional observation space to obtain the real-time working status data observation signal, as shown in Formula 2:

[0131] (2);

[0132] In the formula, Observe the signal for real-time working status data. is the observation matrix, is the perception matrix;

[0133] According to the real-time working state data measured signal and the real-time working state data observation signal, the reconstruction error is obtained. ;

[0134] Perform atomic updates one by one until the reconstruction error Less than the set error threshold , output the best sparse dictionary, and use the best sparse dictionary to express the characteristics of the measured signal of the complete real-time working status data;

[0135] Among them, when performing atomic updates one by one, it includes:

[0136] Provide an update loop matrix, and use the update loop matrix to make each element in the observation matrix independently obey the Bernoulli distribution, as shown in Formula 3:

[0137] (3);

[0138] In the formula, To update the matrix in the loop, For each element in the observation matrix, The size is The lateral size of the observation matrix.

[0139] In the above step S3, when performing nested association, it includes:

[0140] The preset environment-state relationship Projecting onto the identification attributes and item set attributes of various underwater environment compressed data and real-time working status compressed data, and nesting the data tuples according to the identification attributes;

[0141] Get the power set of each data tuple on the itemset attribute, de-nest the itemset attribute, group the data tuples according to the itemset attribute, and get the number of data tuples in each itemset;

[0142] Delete the items with less than the minimum support in the data tuple of the item set, generate frequent item sets, and obtain the support of each frequent item set;

[0143] Extracting association rules from frequent item sets and generating a first copy and a second copy of the frequent item sets;

[0144] Connect the first copy and the second copy, so that the frequent itemsets of the first copy are a proper subset of the frequent itemsets of the second copy, and generate the subject;

[0145] Rename the attributes and use the renamed attributes to identify the small item set as the subject of the potential rule and the superset of the corresponding large item set;

[0146] Get the confidence of the association rule, select the association rule whose confidence is greater than the confidence threshold, perform a difference operation on the subject and the superset, and generate the head of the association rule;

[0147] The subject, superset, head, support and confidence are formed into a nested relationship, and various underwater environment compressed data and real-time working status compressed data are formed into a nested underwater data packet according to the nested relationship.

[0148] In the above step S4, when generating the real-time transmission strategy, it includes:

[0149] A control frame is transmitted for the subject, superset, and header on the control channel, and the last successfully scheduled data in the nested underwater data packet is sent on the data channel;

[0150] Wait until the control frame returns, obtain the value of the channel available time and the value of the target node receivable time, and save them to the available time table and receivable time table;

[0151] Obtain the data of the last unsuccessful scheduling in the nested underwater data packet, and update the value of its data deadline using the control frame interval period;

[0152] The data of the last unsuccessful scheduling and the data of this control frame are The algorithm sorts the data and obtains a data queue;

[0153] Find the first available data channel for the data in the data queue, and obtain the timeout judgment parameter according to the receivable time of the target node, as shown in Formula 4:

[0154] (4);

[0155] In the formula, For the timeout judgment parameter, is the receivable time of the target node, is the preset first time constant;

[0156] The first timeout judgment time is obtained according to the channel available time, as shown in Formula 5:

[0157] (5);

[0158] In the formula, is the first timeout judgment time, is the channel available time, To obtain a larger value;

[0159] The second timeout judgment time is obtained according to the first timeout judgment time and the timeout judgment parameter, as shown in Formula 6:

[0160] (6);

[0161] In the formula, is the second timeout judgment time, is a preset second time constant;

[0162] Obtain the data transmission time according to the second timeout judgment time , and look for potential slow data based on the data’s emission time;

[0163] If there is delayed data, ,in, is the transmission time of the data, is the length of the data, is the preset second time constant, is the value of the data period, then use The scheduling method discards data. If there is no delayed data, the value of the target node's receivable time and the value of the channel's available time are updated, and the data that is not successfully scheduled is added to the unsuccessful data queue;

[0164] The successfully scheduled data is sent on the data channel, and the next round of control frames is sent on the control channel.

[0165] In the above step S5, when obtaining the environment-state comprehensive dictionary, it includes:

[0166] The nested underwater data packets are divided according to the frequency channels, and the underwater data of each frequency channel is convolved in any direction through 8 one-dimensional Gaussian smoothing filters, and then down-sampled to obtain the representation of the underwater data of each frequency channel at 8 scales;

[0167] The center-surround difference algorithm is used to simulate the characteristics of local underwater environment suppression, and the underwater saliency of underwater data of each frequency channel at different scales is obtained;

[0168] All underwater saliency is normalized to a set range through a global nonlinear amplification algorithm to obtain the normalized underwater saliency, so as to enhance the salient features, suppress the smooth features, and eliminate the cross-modal amplitude differences caused by different feature extraction mechanisms.

[0169] All normalized underwater saliency in the frequency channel are convolved with the one-dimensional difference Gaussian filter, and the difference between the result after convolution and the underwater saliency before normalization is obtained, and the difference is added to the normalized underwater saliency to obtain the cross-scale integrated underwater saliency of each frequency channel, as shown in Formula 7:

[0170] (7);

[0171] In the formula, To integrate underwater saliency across scales, is the normalized underwater saliency, is the convolution operation, is a one-dimensional difference Gaussian filter, is the underwater saliency before normalization;

[0172] The cross-scale integrated underwater saliency of each frequency channel is linearly integrated to obtain the underwater data saliency map, as shown in Formula 8:

[0173] (8);

[0174] In the formula, is the underwater data saliency map, is the number of frequency channels, For the Cross-scale integrated underwater saliency of frequency channels;

[0175] Among them, the center-surround difference algorithm includes: performing cross-scale subtraction between fine scale and coarse scale, and correcting the result.

[0176] Furthermore, after obtaining the underwater data saliency map, the following steps are included:

[0177] The nested underwater data packets and underwater data saliency maps are frame-processed in the time domain to obtain multiple groups of underwater data and underwater data saliency maps with the same time interval;

[0178] Input multiple sets of underwater data and underwater data saliency maps into the convolutional neural network model Layer 1, obtain the corresponding one-dimensional features;

[0179] The one-dimensional features are concatenated with the underwater data with the same time dimension, and the principal component analysis results of the underwater data and the underwater data saliency map are combined to extract features using a convolutional neural network after two convolutions, two poolings, and one full connection operation, and the multi-channel features of each group of underwater data after framing are output;

[0180] The multi-channel features of each group of underwater data are linearly combined in time order to generate the final feature matrix that can well highlight the features of the underwater data;

[0181] Through the fully connected layer, the improvement The activation function of the function normalizes the feature matrix to obtain a normalized feature matrix;

[0182] Among them, improvement Function, as shown in Formula 9:

[0183] (9);

[0184] In the formula, is the element in the feature matrix, is a natural constant;

[0185] A mapping relationship between the target indicator ID and the normalized feature matrix is ​​established to generate an environment-state comprehensive dictionary containing the mapping relationship between multiple target indicators and their corresponding feature matrices.

[0186] Specifically, Figure 2 As shown, the convolutional neural network model includes: Layer 1, first convolutional layer 2, first pooling layer 3, second convolutional layer 4, second pooling layer 5 and fully connected layer 6.

[0187] In the above step S6, when obtaining the real-time operating status of the underwater light, it includes:

[0188] according to Similarity is used to obtain underwater lights with similar operating states to the target monitored underwater lights to form a similar underwater light group, and a fusion data set of the similar underwater light group is obtained;

[0189] Divide the training data set and the test data set of the target monitoring underwater lights from the fused data set, and preprocess the training data set and the test data set;

[0190] The common feature information is extracted from the fusion data set through the spatiotemporal autoencoder, and the common feature information is transferred to the state monitoring model of the target underwater light through parameter migration. The normal operation state model of the underwater light based on group similarity and parameter migration is established in combination with the training data set.

[0191] based on The standard deviation of the statistic is combined Control chart to obtain the first threshold , the second threshold , as shown in Formula 10 and Formula 11:

[0192] (10);

[0193] (11);

[0194] In the formula, is the mean of the forecast residuals during historical normal operating time, is the threshold coefficient, for The standard deviation of the statistic, for The weighting coefficients in the control chart, for historical uptime;

[0195] The test data set is input into the normal operating state model of the underwater light to obtain the reconstruction residual and pass Control chart, first threshold and the second threshold Determine the operating status of the target monitoring underwater light;

[0196] If the reconstruction error The value continues at the first threshold , the second threshold If the target monitoring underwater light is within , it is determined that it is in normal operation;

[0197] If the reconstruction error The value continues to exceed the first threshold , the second threshold , it is determined that the target monitoring underwater light is in a faulty operating state.

[0198] Furthermore, when extracting the common feature information, it includes:

[0199] The fused data set is preprocessed and input into the spatiotemporal autoencoder. By adjusting the convolution kernel size and moving direction of the encoding network, the common feature information extraction and regression analysis of the preprocessed fused data set are completed. The convolution process of the encoding network is shown in Formula 12:

[0200] (12);

[0201] In the formula, For the Tier Output feature map, for Tier Input feature maps, For the The weight matrix of the convolution kernel of the layer, is the convolution operation, For the The number of feature maps at the layer input, For the The bias matrix of the layer, is the activation function;

[0202] Among them, the spatiotemporal autoencoder includes: an encoding network and a decoding network. The encoding network consists of cascade Network composition, the decoding network uses a fully connected network;

[0203] When establishing the normal operating state model of underwater lights based on group similarity and parameter migration, it includes:

[0204] The pre-processed real-time working status time series data , using a sliding window for input, where Indicates the sliding window Time The observed The value of the monitoring variable. is the sliding window size, is the number of dimensions of the monitoring variable;

[0205] Select all monitoring variables of the last moment of the input real-time working status time series data as the target monitoring variables ,in, The last time in the sliding window The observed Dimension monitoring variable value;

[0206] The normal operating state model of the underwater light is as shown in Formula 13, Formula 14 and Formula 15:

[0207] (13);

[0208] (14);

[0209] (15);

[0210] In the formula, It is the model function mapping relationship.

[0211] In the above step S7, when obtaining the environmental adaptability of the underwater light, it includes:

[0212] Obtaining a first target indicator set and a second target indicator set regarding underwater environment sensitivity and underwater light intensity from an environment-state comprehensive dictionary;

[0213] The original value, maximum value and minimum value of each first target indicator in the first target indicator set in the spatial pixel are obtained to generate the range normalized value of the underwater environmental sensitivity, as shown in Formula 16:

[0214] (16);

[0215] In the formula, is the normalized value of the underwater environmental sensitivity, For each first target indicator in the spatial pixel The original value of For each first target indicator in the spatial pixel The maximum value of For each first target indicator in the spatial pixel The minimum value of

[0216] The original value, maximum value and minimum value of each second target indicator in the second target indicator set in the spatial pixel are obtained to generate the range normalized value of the underwater light intensity, as shown in Formula 17:

[0217] (17);

[0218] In the formula, is the normalized value of the underwater light intensity, For each second target indicator in the spatial pixel The original value of For each second target indicator in the spatial pixel The maximum value of For each second target indicator in the spatial pixel The minimum value of

[0219] The range normalized values ​​of underwater environmental sensitivity and underwater light intensity are resampled to a grid of preset size, and the coupling degree model is introduced to generate the environmental adaptation coupling degree, as shown in Formula 18:

[0220] (18);

[0221] In the formula, To adapt the coupling degree to the environment, The normalized value of the extreme difference of underwater environmental sensitivity in the grid The pixel value on The range normalized value of underwater light intensity in the grid The pixel value on ;

[0222] A coupling coordination model is introduced for the environmental adaptation coupling degree to generate the environmental adaptation coupling coordination degree, as shown in Formula 19:

[0223] (19);

[0224] In the formula, The coupling coordination degree for environmental adaptation is is the first coordination coefficient, is the second coordination coefficient;

[0225] The environmental adaptability of the underwater light is generated according to the environmental adaptability coupling degree and the environmental adaptability coupling coordination degree.

[0226] Specifically, the value of the first coordination coefficient is 0.4-0.6, and the value of the second coordination coefficient is 0.4-0.6.

[0227] The smaller the environmental adaptation coupling degree is, the more coupled the underwater environmental sensitivity and the underwater light intensity are. The larger the environmental adaptation coupling coordination degree is, the more coordinated the underwater environmental sensitivity and the underwater light intensity are.

[0228] In the above step S8, when constructing the underwater light control model, it includes:

[0229] pass The algorithm combines deep neural networks with Algorithms are combined to make decisions network;

[0230] Initialization decision Network, through decision The network obtains the value of the state-action value function and stores the state, action, and reward through the experience replay cache;

[0231] Control action selection: At each time step, the agent selects the action based on the real-time operating status of the underwater light. The strategy selects an underwater light control action;

[0232] Obtaining status feedback: After the underwater light control action is executed, the next real-time operating state and reward value of the underwater light are received through the agent;

[0233] Update the experience replay cache: store the current underwater light's environmental adaptability, underwater light control action, reward value, and the next real-time operating state of the underwater light in the experience replay cache;

[0234] Update network parameters: Sample batches of data from the updated experience replay buffer, obtain the updated value of the state-action value function, and update the current decision Network parameters are copied to the target decision Network, decision making The update of network parameters is shown in formula 20:

[0235] (20);

[0236] In the formula, For updated decision Network parameters, For current decision Network parameters; is a discount factor used to balance the impact of the current immediate reward value and the subsequent reward value; For current decision The value of the state-action value function obtained by taking the current underwater light control action under the network parameters; is the learning rate, used to control the current decision Network parameters and previous decisions The impact of network parameters on the state-action value function; For current decision The instant reward value obtained by taking the current underwater light control action under the network parameters; For the next decision Among all possible underwater light control actions under the network parameters, the value of the state-action value function corresponding to the underwater light control action with the maximum state-action value function can be obtained. For the next decision The value of the state-action value function for all possible underwater light control actions under the network parameters;

[0237] Repeat the control action selection, obtain state feedback, update the experience playback cache, and update the network parameters until the preset number of training times is reached or the algorithm converges, and the decision is completed. Update the network parameters to obtain the underwater light control model;

[0238] Among them, the underwater light control action is selected according to the state-action value function of the underwater light control model to generate the corresponding underwater control instruction.

[0239] The above-mentioned embodiments are only preferred embodiments of the present invention and cannot be used to limit the scope of protection of the present invention. Any non-substantial changes and substitutions made by technicians in this field on the basis of the present invention shall fall within the scope of protection required by the present invention.

Claims

1. A method for controlling LED underwater lights based on remote communication integrated management, characterized in that: The following steps are involved: Collect a variety of underwater environment data in real time, and compress them through a high-ratio compression method based on context relevance to obtain a variety of underwater environment compressed data; Collect the real-time working status data of the underwater light in real time, and compress it using the best sparse dictionary to obtain the real-time working status compressed data; Nesting and associating a variety of underwater environment compressed data and real-time working status compressed data to form a nested underwater data packet; Generate a real-time transmission strategy based on the nested underwater data packets using the control frame interval period and transmit; Generate underwater data saliency map and feature matrix for nested underwater data packets, and improve The function normalizes the feature matrix to obtain the environment-state comprehensive dictionary; Input the environment-state comprehensive dictionary into the normal operation state model of the underwater light to obtain the real-time operation state of the underwater light; Based on the environment-state comprehensive dictionary, the environmental adaptation coupling degree and environmental adaptation coupling coordination degree are obtained, and the environmental adaptability of the underwater light is generated; Construct an underwater light control model, input the real-time operating status and environmental adaptability of the underwater light into the underwater light control model, generate underwater control instructions based on the state-action value function and transmit them to the underwater light; Among them, when performing nested associations, it includes: Get frequent itemsets and their support of compressed data; Extract association rules from frequent item sets, generate copies of frequent item sets, connect copies, and generate entities; Identify the small item set that is the subject of the potential rule and the superset of the corresponding large item set; Select the association rules whose confidence is greater than the confidence threshold, perform difference operation on the subject and the superset, and generate the head of the association rule; The subject, superset, header, support and confidence are formed into a nested relationship, and the compressed data is formed into a nested underwater data packet according to the nested relationship.

2. The LED underwater light control method based on remote communication integrated management according to claim 1 is characterized in that: When compressing with the best sparse dictionary, this includes: An observation matrix is ​​constructed for the real-time working state data measured signal, and the real-time working state data measured signal is mapped from the high-dimensional observation space to the low-dimensional observation space to obtain the real-time working state data observation signal; According to the real-time working state data measured signal and the real-time working state data observation signal, the reconstruction error is obtained. ; Perform atomic updates one by one until the reconstruction error Less than the set error threshold , output the best sparse dictionary, and use the best sparse dictionary to express the characteristics of the measured signal of the complete real-time working status data.

3. The LED underwater light control method based on remote communication integrated management according to claim 2 is characterized in that: When obtaining the frequent itemsets and their support of compressed data, it includes: The preset environment-state relationship Projecting onto the identification attributes and item set attributes of various underwater environment compressed data and real-time working status compressed data, and nesting the data tuples according to the identification attributes; The data tuples are grouped according to the item set attributes, frequent item sets are generated, and the support of each frequent item set is obtained.

4. The LED underwater light control method based on remote communication integrated management according to claim 3 is characterized in that: When generating a real-time transmission policy, include: A control frame is transmitted for the subject, superset, and header on the control channel, and the data of the last successful scheduling is sent on the data channel; Wait until the control frame returns, and obtain the value of the channel available time and the value of the target node receivable time respectively; Get the data of the last unsuccessful scheduling, update the value of its data period using the control frame interval period, and update it with the data of this control frame according to The algorithm sorts the data and obtains a data queue; Find the first available data channel for the data in the data queue, obtain the timeout judgment parameter and the first timeout judgment time according to the receivable time of the target node and the available time of the channel, and generate the second timeout judgment time; The data transmission time is obtained according to the second timeout judgment time, the value of the target node receivable time and the value of the channel available time are updated, the successfully scheduled data is sent on the data channel, and the next round of control frames is sent on the control channel.

5. The LED underwater light control method based on remote communication integrated management according to claim 4 is characterized in that: When obtaining the environment-state comprehensive dictionary, it includes: The nested underwater data packets are divided according to the frequency channels, and the underwater data of each frequency channel is convolved in any direction through multiple one-dimensional Gaussian smoothing filters, and down-sampled to obtain the representation of the underwater data of each frequency channel at multiple scales; The center-surround difference algorithm is used to simulate the characteristics of local underwater environment suppression, and the underwater saliency of underwater data of each frequency channel at different scales is obtained; All underwater saliency is normalized by a global nonlinear method amplification algorithm to obtain the normalized underwater saliency; Convolve all normalized underwater saliency in the frequency channel with a one-dimensional difference Gaussian filter, and obtain the difference between the result after convolution and the underwater saliency before normalization; The difference is added to the normalized underwater saliency to obtain the cross-scale integrated underwater saliency of each frequency channel, and linear integration is performed to obtain the underwater data saliency map.

6. The LED underwater light control method based on remote communication integrated management according to claim 5 is characterized in that: After obtaining the underwater data saliency map, including: The nested underwater data packets and underwater data saliency maps are processed by time domain to obtain multiple groups of underwater data and underwater data saliency maps, and then input into the convolutional neural network model. layer, and obtain the corresponding one-dimensional features; The one-dimensional feature is concatenated with underwater data having the same time dimension, and a convolutional neural network is used to extract features by combining the principal component analysis results of the underwater data and the underwater data saliency map, and a multi-channel feature of each group of underwater data after framing is output; Linearly combine the multi-channel features of each group of underwater data in time order to generate a feature matrix; Through the fully connected layer, the improvement The activation function of the function normalizes the feature matrix to obtain a normalized feature matrix; Establish the mapping relationship between the target indicator ID and the normalized feature matrix to generate the environment-state comprehensive dictionary; The target indicator ID includes: a first target indicator ID regarding underwater environment sensitivity and a second target indicator ID regarding underwater light intensity.

7. The LED underwater light control method based on remote communication integrated management according to any one of claims 1 to 6, characterized in that: When getting the real-time operating status of the underwater light, it includes: according to Similarity obtains similar underwater light groups and their fusion datasets; Divide the training data set and test data set of the target monitoring underwater lights from the fused data set and perform preprocessing; The common feature information is extracted from the fusion data set through the spatiotemporal autoencoder, and the common feature information is transferred to the state monitoring model of the target underwater light through parameter migration. The normal operation state model of the underwater light is established in combination with the training data set. based on The standard deviation of the statistic is combined Control chart to obtain the first threshold , the second threshold ; The test data set is input into the normal operating state model of the underwater light to obtain the reconstruction residual and pass Control chart, first threshold and the second threshold Determine the operating status of the target monitoring underwater light.

8. The LED underwater light control method based on remote communication integrated management according to claim 7, characterized in that: When extracting common feature information, include: The fused data set is preprocessed and input into the spatiotemporal autoencoder. By adjusting the convolution kernel size and movement direction of the encoding network, common feature information extraction and regression analysis are completed. When establishing the normal operating state model of underwater lights, include: The pre-processed real-time working status time series data is input in a sliding window manner; Select all monitoring variables of the last moment of the input real-time working status time series data as the target monitoring quantity; The mapping relationship between the real-time working status time series data and the target monitoring quantity is obtained to obtain the normal operating status model of the underwater light.

9. The LED underwater light control method based on remote communication integrated management according to claim 8, characterized in that: When obtaining the environmental adaptability of underwater lights, include: Obtaining a target indicator set regarding underwater environmental sensitivity and underwater light intensity from the environment-state comprehensive dictionary; The original value, maximum value and minimum value of each target indicator in the spatial pixel are obtained, and the range normalized value of underwater environmental sensitivity and the range normalized value of underwater light intensity are generated. The values ​​are resampled to a grid of a preset size, and a coupling degree model is introduced to generate the environmental adaptation coupling degree. Introducing a coupling coordination model for the environmental adaptation coupling degree to generate an environmental adaptation coupling coordination degree; The environmental adaptability of the underwater light is generated according to the environmental adaptability coupling degree and the environmental adaptability coupling coordination degree.

10. The LED underwater light control method based on remote communication integrated management according to claim 9, characterized in that: When building an underwater light control model, include: pass The algorithm combines deep neural networks with Algorithms are combined to make decisions network; Initialization decision Network, through decision The network obtains the value of the state-action value function and stores the state, action, and reward through the experience replay cache; Repeat the parameter update steps until the preset number of training times is reached or the algorithm converges, and the decision is completed. Update the network parameters to obtain the underwater light control model; The parameter updating step includes: controlling action selection, obtaining state feedback, updating the experience playback cache, and updating network parameters; Updating the network parameters involves sampling batches of data from the updated experience replay buffer, obtaining the value of the updated state-action value function, and updating the current decision Network parameters are copied to the target decision Network, decision making Update of network parameters.

Images