Self-cruise fishery water quality sensing and intelligent regulation method and system

Abstract

The present application relates to a self-cruise fishery water quality sensing and intelligent control method and system, the system of the present application is highly intelligent, can control self-cruise automatic traversal water area, realizes whole process automatic feeding, can accurately feed the breeding object, only needs artificial regular shore maintenance base station and adds bait and micro-ecological preparation, greatly saves manpower and material; while feeding, the water quality is monitored, and the water quality condition is fed back to the base station to adjust the bait quantity, the water quality optimization and the breeding target form positive feedback, realizes the integration of protecting water environment and feeding. Meanwhile, the present application uses BP (back propagation) neural network and DQN (Deep Q-Network) algorithm deep learning neural network, the system can realize water quality judgment after receiving the monitored water quality physicochemical factor, determine the bait feeding quantity and the micro-ecological preparation feeding quantity, and further realize providing a good growth environment for the breeding organism.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of fishery culture equipment, and particularly relates to a self-cruise fishery water quality sensing and intelligent control method and system. BACKGROUND

[0002] With the development and progress of science and technology, the automation of the aquatic industry has become a topic of increasing concern. According to the survey, most places still use artificial throwing of bait and micro-ecological agents. The frequency and amount of bait thrown are based on the existing experience of fishermen, resulting in waste of manpower and resources. Inadequate consumption of bait can pollute the water environment. For example, bait can increase the ammonia nitrogen content in the culture area, thereby affecting the dissolved oxygen and pH physicochemical factors. With the improvement of people's nutritional requirements, people's demand for various aquatic products has also increased, and the addition of protein and other additives in the bait has also increased. In addition, the excretion of cultured organisms, the throwing of pesticides and fertilizers, etc. can cause eutrophication in the culture area, affecting the growth of cultured organisms. The sinking of bait to the bottom also causes great pollution to the sediment in the culture area. Therefore, real-time monitoring of water quality is a very important step.

[0003] The prior art does not have in-depth research on monitoring data after adding repair substances for monitoring. At the same time, in places with a relatively large culture area, fishermen can only throw bait at the shore or on fishing boats. The amount of bait carried for feeding at one time is limited, and when feeding a large water area, it is necessary to return to the shore to add materials, which is time-consuming and labor-intensive. Moreover, the throwing of bait is not uniform enough, resulting in a surplus of bait in some culture areas and a shortage of bait in some culture areas. Excess bait sinks to the bottom of the water, causing water pollution. The existing research only develops an automatic bait feeder, a water quality monitoring system, and a device for putting repair substances into water, and the function of the produced device is relatively single. SUMMARY

[0004] The present application aims to provide a self-cruise fishery water quality sensing and intelligent control method and system to solve the problems existing in the prior art.

[0005] A self-cruise fishery water quality sensing and intelligent control system, comprising a culture water quality sensing and intelligent control device and a shore-based control center. The culture water quality sensing and intelligent control device comprises an energy supply module, a power drive module, an automatic feeding module, a BDS positioning module, a water quality data acquisition module, a wireless transmission module, and a feeding device controller. The energy supply module is electrically connected to other modules. The shore-based main control center comprises a sending working mode module, an intelligent decision module, an intelligent control module, and a warning module. The sending working mode module comprises automatic feeding instructions, water quality monitoring instructions, and water quality control instructions. The water quality intelligent decision module comprises a water quality judgment module and an intelligent decision module. The intelligent control module comprises an oxygen supply machine and a supply station.

[0006] The other modules in the aquaculture water quality sensing and intelligent control device send and receive working signals of the shore-based main control center through the release device controller; the cruise instruction of the working mode module is decoded by the automatic release module, and the power driving module drives the automatic release module according to the decoded cruise instruction

[0007] The BDS positioning module locates to the specified area; after the water quality data acquisition module collects the water quality physicochemical factors, the signal is transmitted to the intelligent decision module, and the intelligent decision module analyzes the water quality physicochemical factors, and controls the automatic release module to release automatically through the sending working mode module and the intelligent control module.

[0008] The aquaculture water quality sensing and intelligent control device and the shore-based main control center are communicated through the wireless transmission module, which can be 4G, 5G or Wifi.

[0009] Further, the automatic release module is composed of a first antenna (3), a first positioning plate (4), a release bin (5), a controller (6), a first wireless transmission module (8), a first switch valve (10), an electric control cabinet (27) and a time relay, the first antenna (3) is welded on the top of the release bin (5), the first positioning plate (4) is welded on the top of the release bin (5), the controller (6) and the first wireless transmission module (8) are fixedly assembled in the electric control cabinet (27), and the first switch valve (10) is welded on the bottom of the release bin (5). The opening time of the first switch valve (10) is decoded by the controller (6). After the automatic release module receives the release signal and the water quality sensing and intelligent release device reaches the specified position, the electromagnetic switch valve is opened, the time relay is automatically disconnected after reaching the set time, the electromagnetic valve is automatically closed, and the release work is completed.

[0010] Further, the supply station of the shore-based control center includes bait supply bin (17) and micro-ecological preparation supply bin (18), second antenna (19) and second wireless transmission module (20), second switch valve (21), fixed support (23), stirring chamber (22), material conveying pipeline (24) and second positioning plate (25); the second switch valve (21), the stirring chamber (22) and the material conveying pipeline (24) are welded together through the stainless steel support; when the second positioning plate (25) and the first positioning plate (4) are connected together, the second switch valve (21) is opened, the bait is stirred in the stirring chamber (22), and then is transmitted to the release bin (5) through the material conveying pipeline.

[0011] The bait supply bin and the micro-ecological preparation supply bin provide sufficient materials, and the aquaculture water quality sensing and intelligent control device supplies bait and micro-ecological preparation according to the instruction on the shore.

[0012] The second antenna (19) and the second transmission module (20) are used for receiving and sending signals of the delivery device controller.

[0013] Further, the energy supply module is composed of a solar panel (1) and a storage battery (9), the solar panel (1) is welded at the top end of the water quality sensing and intelligent control device, the storage battery (9) is placed in the electric control cabinet (27), the solar panel (1) converts solar energy into electrical energy and stores it in the storage battery (9); and / or the power driving module is composed of a floating plate (7), a propeller (15) and a brushless motor (16), the propeller (15) is welded on both sides of the bottom bracket of the floating plate (7), and the brushless motor (16) drives the propeller (15) to work.

[0014] Further, the water quality data acquisition module includes an ammonia nitrogen sensor, a pH sensor and a dissolved oxygen sensor (12), and the ammonia nitrogen sensor and the pH sensor are collected in a collection probe (13). In the m locations of the breeding area, the ammonia nitrogen solubility of the water quality physicochemical factor is x i , the dissolved oxygen solubility is y i , and the pH value is z i , and finally the average solubility of the same environmental characteristic data in the same period is transmitted to the intelligent decision-making data.

[0015] Further, the automatic delivery module is welded with a filter cartridge (11) below, the filter cartridge (11) has holes with different diameters at the bottom end, which buffers the falling speed of the delivered bait or micro-ecological preparation, prevents the delivered material from sinking to the bottom due to gravity, for example, slows down the falling speed of the bait, and makes the bait more suitable for consumption by the cultured organisms after falling into the water, reducing bait waste. The micro-ecological preparation includes bacillus subtilis and photosynthetic bacteria.

[0016] Further, the warning module includes a buzzer, which sounds when the monitored water quality data is in an abnormal state, reminding the breeding personnel and the control center personnel.

[0017] The above-mentioned self-cruise fishery water quality sensing and intelligent control system control method comprises:

[0018] Step one: the power driving module sails according to the cruise channel set by the DBS positioning module;

[0019] Step two: the water quality acquisition module collects water quality factors in real time, and then transmits the water quality analysis module, and judges the water quality through the BP neural network;

[0020] Step three: the water quality judgment result is transmitted to the delivery amount decision-making module, and the delivery amount of bait and micro-ecological preparation is determined through the deep learning neural network of the DQN algorithm.

[0021] Step four: the intelligent decision module outputs the current amount of release, and then the sending mode module and the intelligent control module send instructions to the automatic release module for automatic release;

[0022] Step five: after the automatic release is completed, return to the original position and standby.

[0023] Further, the method for the automatic release module to automatically release is as follows: when the water quality sensing and intelligent control device first positioning plate (4) supply station pipe mouth is connected with the water quality sensing and control device loading port of the second positioning plate (25) of the supply station, the second on-off valve (21) of the supply station electromagnetic on-off valve is opened, when the set time is reached, the second supply station electromagnetic valve on-off valve is closed, and the bait or micro-ecological preparation material is thrown into the breeding area by the self-cruise.

[0024] Further, the release mode of the bait and the micro-ecological preparation in step three includes: when the monitored water quality physicochemical factors are normal, only the bait is released; when the monitored water quality physicochemical factors are abnormal, the signal is transmitted to the early warning module, and the bait and the micro-ecological preparation are released; after the micro-ecological preparation is released, 12-24 hours are waited, and it is judged by the intelligent decision release amount module that the micro-ecological preparation still needs to be released, so only the micro-ecological preparation needs to be released, until the monitored water quality physicochemical factors return to the normal range. That is, the working condition of the water quality monitoring mode in the application includes condition one: the water quality sensing and intelligent control device receives the working instruction first, monitors the water quality physicochemical factors, and intelligently releases according to the monitoring; condition two: the water quality monitoring mode is sent again by the shore-based controller after the micro-ecological preparation is released for 12-24 hours. The bait release working mode: when the water quality intelligent decision module judges that the water quality is normal, the water quality is not controlled, and the bait is released in a timely, quantitative and fixed-point manner according to the information set in advance by the system; the water quality control mode: when the water quality intelligent decision module makes a decision, the supply station and the water quality sensing and intelligent control device work simultaneously to throw the intelligent bait or the micro-ecological preparation into the breeding area.

[0025] Further, the release amount of the bait and the micro-ecological preparation decided by the release amount decision module in step three is determined according to the following method:

[0026] (1) Establish the brain of the intelligent decision module;

[0027] (2) Establish an experience database; set the state-action database to store N data, and evaluate the experience J[S i ,a i ,r i ,S′ iStored to experience database, when database is full of data, new learning experience replaces old data, satisfies database real-time update, wherein, S i Indicates the current water quality physicochemical factor state, a i Indicates the current bait and microecological agent quantity action, r i Indicates the current action reward value, S' i Indicates the water quality physicochemical factor state measured after bait and microecological agent action;

[0028] (3) the self-learning module of intelligent decision module is established;The reward value is determined according to the current state and the state after, the system can judge whether the current experience can make water quality control reach the best breeding state through reward value r and value Q, wherein reward value r represents the immediate reward after action, value Q represents the expected report that can be obtained in the future after state S i Take state a i , the expected report that can be obtained in the future after.

[0029] Further, the set time is calculated according to the material quantity-time relationship, which is determined by , m represents mass, V represents volume, ρ represents density, S represents the bottom area of on-off valve, v represents material falling speed, and t represents time.

[0030] Compared with the prior art, the present application has the following advantages:

[0031] (1) the self-cruise function of the present application can automatically traverse water area, realize full-process automatic bait feeding, can accurately feed the breeding object, and greatly save manpower and materials.

[0032] (2) the design of water quality perception and intelligent control device can meet the needs of most water areas, realize the monitoring of water quality while feeding, and feedback the water quality to the base station to adjust the bait quantity, so that the water quality optimization and breeding target form positive feedback, realizing the integration of water body environment protection and feeding.

[0033] (3) the system of the present application is highly intelligent, only needs artificial regular shore maintenance base station and adds bait and microecological agent, the water quality perception and intelligent control device will automatically supplement bait according to environmental factors, or add microecological agent, effectively monitor and control the water environment of breeding organisms, greatly liberate labor and reduce labor cost.

[0034] (4) The self-cruise fishery water quality sensing and intelligent regulation system uses a BP neural network and a deep learning neural network of a DQN algorithm, can realize water quality judgment after receiving monitored water quality physicochemical factors, determines bait feeding amount and micro-ecological preparation feeding amount, so that the aquaculture water quality meets the growth standard of the cultured object, and a good growth environment is provided for the cultured organism. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 is a self-cruise fishery water quality sensing and intelligent regulation system overall structure diagram of the present application;

[0036] Figure 2 is a structure diagram of the aquaculture water quality sensing and intelligent regulation device of the present application;

[0037] Figure 3 is a structure diagram of the supply station of the present application;

[0038] Figure 4 is a self-cruise fishery water quality sensing and intelligent regulation system module diagram of the present application;

[0039] Figure 5 is a self-cruise water quality sensing and regulation device system working diagram of the present application;

[0040] Figure 6 is an intelligent decision feeding material on-off valve control diagram of the present application;

[0041] Wherein 1 is a solar panel, 2 is a support, 3 is a first antenna, 4 is a first positioning plate, 5 is a feeding bin, 6 is a controller, 7 is a floating plate, 8 is a first wireless transmission module, 9 is a storage battery, 10 is a first on-off valve, 11 is a filter cartridge, 12 is a dissolved oxygen probe, 13 is a collection probe, 14 is a fixing frame, 15 is a propeller, 16 is a brushless motor, 17 is a bait supply bin, 18 is a micro-ecological preparation supply bin, 19 is a second antenna, 20 is a second antenna transmission module, 21 is a second electromagnetic on-off valve, 22 is a stirring chamber, 23 is a fixed support, 24 is a material conveying pipeline, 25 is a second positioning plate, and 26 is a shore-based control center;

[0042] Figure 7 is a feeding amount prediction and actual measurement broken line diagram of the bait feeding amount of example 2;

[0043] Figure 8 is a feeding amount prediction and actual measurement broken line diagram of the micro-ecological preparation of example 2. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings.

[0045] Example 1

[0046] The application discloses a self-cruise fishery water quality sensing and intelligent regulation system. Figure 3 As shown in the figure, the system comprises a breeding water quality sensing and intelligent regulation device and a shore-based control center. The breeding water quality sensing and intelligent regulation device comprises an energy supply module, a power driving module, an automatic feeding module, a BDS positioning module, a water quality data acquisition module, a wireless transmission module and a feeding device controller, wherein the energy supply module is electrically connected with the other modules; the shore-based main control center comprises a sending working mode module, an intelligent decision module, an intelligent regulation module and a warning module; the sending working mode module comprises automatic feeding instructions, water quality monitoring instructions and water quality regulation instructions; the water quality intelligent decision module comprises a water quality judgment module and an intelligent decision module; the intelligent regulation module comprises an oxygen supply machine and a supply station.

[0047] The feeding device controller sends and receives working signals of the power driving module, the automatic feeding module, the BDS positioning module and the water quality data acquisition module, stores the collected data in the memory of the controller, and also receives and sends signals from the shore-based control center through wireless transmission. The feeding device controller is sealed and welded into the water quality sensing and intelligent regulation device.

[0048] The cruise instructions of the working mode module are decoded by the automatic feeding module, and the power driving module reaches the designated area according to the decoded cruise instructions and is positioned by the BDS positioning module.

[0049] After the water quality data acquisition module collects water quality physicochemical factors, signals are transmitted to the intelligent decision module. After the intelligent decision module analyzes the water quality physicochemical factors, the automatic feeding module is controlled to automatically feed through the sending working mode module and the intelligent regulation module.

[0050] The breeding water quality sensing and intelligent regulation device and the shore-based main control center are communicated through a wireless network. The wireless communication mode can be 4G, 5G or Wifi.

[0051] The specific structure of the self-cruise fishery water quality sensing and intelligent regulation system is shown in the figure. Figures 1 to 3

[0052] ​The automatic feeding module is composed of a first antenna 3, a first positioning plate 4, a feeding bin 5, a controller 6, a first wireless transmission module 8, a first switch valve 10, an electric control cabinet 27 and a time relay. The first antenna 3 is welded at the top of the feeding bin 5. The first positioning plate 4 is welded at the top of the two sides of the feeding bin 5. The controller 6 and the first wireless transmission module 8 are fixedly assembled in the electric control cabinet 27. The first switch valve 10 is welded at the bottom of the feeding bin 5. The opening time of the first switch valve 10 is decoded by the controller 6. After the automatic feeding module receives the feeding signal and the water quality sensing and intelligent feeding device reaches the specified position, the electromagnetic switch valve is opened. The time relay is automatically disconnected after reaching the set time. The electromagnetic valve is automatically closed to complete a feeding work.

[0053] The supply station of the shore-based control center includes a bait supply bin 17 and a micro-ecological preparation supply bin 18, a second antenna 19 and a second wireless transmission module 20, a second switch valve 21, a fixed support 23, a stirring chamber 22, a material conveying pipeline 24 and a second positioning plate 25. The second switch valve 21, the stirring chamber 22 and the material conveying pipeline 24 are welded together through a stainless steel support. When the second positioning plate 25 and the first positioning plate 4 are connected together, the second switch valve 21 is opened. After the bait is stirred in the stirring chamber 22, it is transmitted to the feeding bin 5 through the material conveying pipeline. The second antenna 19 and the second transmission module 20 are used for receiving and sending the signal of the feeding device controller. The bait supply bin and the micro-ecological preparation supply bin provide sufficient materials. The water quality sensing and intelligent control device replenishes the bait and the micro-ecological preparation according to the instruction on the shore.

[0054] The energy supply module is composed of a solar panel 1 and a storage battery 9. The back of the solar panel 1 is fixed by an iron stand and welded at the top of the water quality sensing and intelligent control device. Two groups of storage batteries 9 are placed in the electric control cabinet 27. The solar panel 1 converts solar energy into electrical energy and stores it in the storage battery 9. The energy supply module is electrically connected with other modules to provide each module of the water quality sensing and intelligent control device.

[0055] The power driving module is composed of a floating plate 7, a propeller 15 and a brushless motor 16. The propeller 15 is welded on both sides of the bottom support of the floating plate 7. The brushless motor 16 drives the propeller 15 to work. The power driving module reaches the specified area according to the cruise area instruction sent by the main control center by the BDS positioning module.

[0056] The water quality data acquisition module includes an ammonia nitrogen sensor, a pH sensor and a dissolved oxygen sensor 12. The ammonia nitrogen sensor and the pH sensor are collected in the collection probe 13. In the m positions of the breeding area, the ammonia nitrogen solubility is x i , the dissolved oxygen solubility is y i , and the pH value is zi The data finally transmitted to the intelligent decision-making is the average solubility of the same environmental characteristic data in the same period.

[0057] The filter cartridge 11 is welded below the automatic feeding module, and the filter cartridge 11 is provided with holes with different diameters at the bottom end, which buffers the falling speed of the bait or the micro-ecological preparation, prevents the bait from sinking to the bottom due to gravity, and reduces the waste of bait. The micro-ecological preparation includes bacillus subtilis and photosynthetic bacteria.

[0058] Example 2

[0059] A self-cruise fishery water quality sensing and intelligent control method, as shown in Figure 5 and Figure 6 , comprising:

[0060] Step 1: The power driving module sails according to the cruise channel set by the DBS positioning module.

[0061] Step 2: The water quality collection module collects the ammonia nitrogen concentration value, the dissolved oxygen concentration value and the pH value in real time, and then transmits them to the water quality analysis module of the shore-based controller through Wifi, and judges the water quality through the BP neural network.

[0062] Step 3: The water quality judgment analysis result is transmitted to the feeding amount decision-making module, and the feeding amount of bait and micro-ecological preparation is determined through the deep learning neural network of the DQN algorithm; when the monitored water quality physicochemical factors are normal, only the bait is fed;

[0063] When the monitored water quality physicochemical factors are abnormal, the bait and the micro-ecological preparation are fed at the same time, and the signal is transmitted to the early warning module, the buzzer sounds, reminding the breeder that the water quality is in an abnormal state; then after feeding the micro-ecological preparation, after waiting for 24 hours, if the intelligent decision-making feeding amount module still needs to feed the micro-ecological preparation, only the micro-ecological preparation needs to be fed, until the monitored water quality physicochemical factors return to the normal range.

[0064] Step 4: After the intelligent decision-making module of the shore-based controller outputs the current feeding amount, the automatic feeding module feeds according to the instructions sent by the sending mode module and the intelligent control module, when the feeding station pipe is connected with the water quality sensing and intelligent control device loading port, the feeding station electromagnetic valve is opened, when the set time is reached, the feeding station electromagnetic valve is closed, and the water quality sensing and intelligent control device self-cruise throws the material into the breeding area.

[0065] Step 5: After completing a throwing task, the water quality sensing and intelligent control device returns to the designated position and waits, and then performs the throwing work again after receiving the working instruction sent by the shore-based controller.

[0066] Example 3

[0067] The water quality judgment module in this embodiment is based on BP neural network to judge water quality. The normalized data is imported into the trained neural network to pre-judge water quality. When the water quality is normal, the output result is in the interval [0-0.5), which represents normal water quality. When the water quality is abnormal, the output result is in the interval [0.5-1], which represents abnormal water quality.

[0068] After the water quality analysis module judges the water quality by the BP neural network, the water quality judgment analysis result is transmitted to the feeding amount decision module to determine the feeding amount of bait and micro-ecological preparation. The specific method is as follows:

[0069] 1. Obtain historical data, i.e. obtain 300 groups of physicochemical factor data which have important influence on aquaculture water quality, i.e. ammonia nitrogen concentration value, dissolved oxygen concentration value and pH value and water quality grade relationship. This relationship is divided according to "Surface Water Environmental Quality Standard" (GB3838.2002);

[0070] Table 1 Surface water environmental quality standard limit unit mg / L

[0071]

[0072] The above table is divided according to the different standards of water function. The present application is to study the surface water for aquaculture. After water quality detection and judgment, the water quality index contains and is above the standard data of standard III, which indicates that the water quality is normal, and the system output result is [0-0.5). The data of other water quality standards indicates that the water quality is abnormal, and the output is [0.5-1].

[0073] The sample data for learning and testing used in this embodiment comes from the actual monitoring data of a breeding company from January to March 2023. The breeding area is 2000m 2 , the water depth is 1.4 meters, and part of the data is shown in Table 2:

[0074] Table 2 Water quality monitoring table report

[0075]

[0076] 2. Divide the historical data into training sample set and test sample set and perform data normalization. The first 250 data is the training sample set, and the last 50 data is the test sample set. The Max-Min planning method is selected to normalize the data, and each factor is mapped to the range of [0, 1]. The expression formula is:

[0077] X ij represents the jth data of the ith factor, X' ijRepresents the data after maximum and minimum normalization preprocessing, X mini represents the minimum value of the factor in the i-th order, X maxi is the maximum value of the factor in the i-th order. Some normalized data are shown in Table 3:

[0078] Table 3 Normalization of water quality data

[0079]

[0080] 3. Construct aquaculture water quality prediction model - based on BP neural network model.

[0081] (1) Determination of the number of neurons

[0082] The neural network is divided into an input layer, a hidden layer, and an output layer. Since the present invention uses the normalized ammonia nitrogen concentration value, dissolved oxygen concentration value, and pH as the input of the input layer, it is determined that there are three unit nodes in the input layer, and the normalized water quality level is used as the output of the output layer, so the unit node of the output layer is one. The present invention uses the more widely used formula for hidden layer nodes to determine the number of hidden layer nodes, that is, the formula is:

[0083]

[0084] i represents the number of hidden layers, m represents the number of input layers, n represents the number of output layers, and the value range of a is [1, 10]. In the present invention, i has three values, namely the BP neural network 3-3-1 structure for judging the level of water quality.

[0085] (2) Parameter determination.

[0086] The activation function uses the Sigmoid function. The number of iterations is 100 and the root mean square error is 0.01;

[0087] (3) Training BP neural network

[0088] The water quality data value s(x i ,y i ,z i ) is input into the BP neural network, and the output water quality prediction value is o(x i ,y i ,z i )=f(w ij s-θ i ), o(s) represents the predicted value of water quality. The network is trained based on the error of the measured water quality value to determine whether it meets the error. The root mean square error (MSE) formula is:

[0089]

[0090] Where N represents the amount of test sample set, o i represents the true value, o' i represents the predicted value, when the trained network does not meet the root mean square error requirement, the back propagation algorithm is used to correct the network until the error requirement is met or the number of learning is reached.

[0091] 4. Build a model of intelligent decision-making release amount module - deep reinforcement learning model based on DQN

[0092] Regarding the deep reinforcement learning intelligent decision-making module based on DQN algorithm, first, the brain of the decision-making module is established, then the experience database is established, and then the self-learning module is established.

[0093] (1) Establish the brain of the intelligent decision-making module

[0094] ① Get the historical monitoring water quality physicochemical factor ammonia nitrogen concentration value, dissolved oxygen concentration value and pH value and the data of bait and microecological preparation to form the current environment E(S i ,a i ), wherein the ammonia nitrogen concentration value x i , the dissolved oxygen concentration value y i and the pH value z i represent the current state S(x i ,y i ,z i ), the bait amount m and the microecological preparation amount n represent the current action a(m,n);

[0095] ② Load evaluation-action neural network, the evaluation-action neural network model uses DQN algorithm, and the evaluation-action neural network has the same structure as the target neural network.

[0096] ③ Determine the bait amount and microecological preparation release amount strategy, input the pretreated environmental data into the convolutional neural network, that is, input the current state S(x i ,y i ,z i ), input the ammonia nitrogen concentration value x i , the dissolved oxygen concentration value y i and the pH value z i , and output the function value Q(s; θ), the bait amount m and the microecological preparation amount n are determined by ε-greedy, and the formula is as follows:

[0097]

[0098] This formula represents the next action explored with a probability of ε, and the remaining 1-ε probability selects the optimal action value, and ε is set to 0.1;

[0099] IV. Constructing the reward value formula, obtaining the data of the measured water quality physicochemical factors before and after the bait and microecological preparation are put in, and the data relationship of the put-in, calculating the reward value r of the evaluation-action pair neural network, the formula is as follows:

[0100]

[0101] wherein s i represents the water quality physicochemical factor data of the monitored ammonia nitrogen concentration value, dissolved oxygen concentration value and pH value at i moment, s i ' represents the water quality physicochemical factor data of the monitored ammonia nitrogen concentration value, dissolved oxygen concentration value and pH value after the bait and microecological preparation are put in, S' i indicates the state of the measured water quality physicochemical factor after the action of putting in the bait and microecological preparation, a i indicates the action of the current put-in bait and microecological preparation quantity, r represents the immediate reward after the action is executed, that is, the reward value, and whether the standard of fish farming is reached is evaluated;

[0102] V. Training the evaluation-action neural network, another neural network with the same structure as the evaluation-action neural network is set, that is, the target neural network, which is used as a label to assist in training the evaluation-action neural network, so as to meet the stability of the network; the current state s and the current action a are input into the evaluation-action neural network, and the value Q of the current action is output, that is, the function is Q(s, a; ω), ω represents the current network weight value; the target value of the target network with the same structure as the evaluation network is y i =r i +γmax a Q(s t+1 ,a; ω'), ω' represents the target network weight value, and γ is the decay factor, γ is 0.8; the evaluation network assigns the weight value ω to ω' after the Nth iteration; the loss evaluation function of this network is that is, the deep reinforcement network is trained, the function weight value of the action network is updated by the gradient descent method, and the gradient descent formula is:

[0103]

[0104] Table 4 is a test example of the evaluation-action model

[0105]

[0106] Table 4 is a deep reinforcement learning model based on the constructed DQN, which can output the microecological preparation and bait put-in quantity by inputting the monitored water quality physicochemical factor data, and provide the put-in strategy. The predicted put-in quantity and the actual put-in quantity are compared and analyzed in the early training, so as to ensure that the put-in strategy of the microecological preparation and the bait for different water quality grades is optimized and adjusted.

[0107] After the training is completed, when the ammonia nitrogen, dissolved oxygen and pH data are collected, and enter the intelligent decision module. At this time, the deep reinforcement model will determine the amount of bait and microbial preparation according to the current water quality and historical data, and output the result to the instruction of sending working mode. Play a role in improving water quality, and provide a good water environment for the cultured organisms. Next, a specific case is explained, the monitored ammonia nitrogen, dissolved oxygen and pH value is 0.3mg / L, 7.21mg / , 8.39. After prediction, the water quality grade evaluation result of the culture area is good. However, the historical data shows that the water area also needs to be regularly adjusted bacteria and algae, and under the condition of normal bait feeding, it is also necessary to put in microbial preparation to maintain good water quality state, so put in bait 22.5Kg, microbial preparation 1.125Kg.

[0108] (2) Establish an experience database

[0109] The state-action database can store N data, and the evaluation-action network evaluates the network after evaluation. The experience J[S i ,a i ,r i ,S i+1 ] is stored in the experience database. When the database is full of data, the new learning experience J replaces the old data, which satisfies the real-time update of the database, and determines that the experience database data is 2000;

[0110] (3) Self-learning module of intelligent decision module

[0111] After a period of time, M experiences are extracted from the experience database and input into the evaluation-action neural network for value evaluation. The evaluation-action neural network is trained and the weight value is updated, which satisfies the network self-updating. It is determined that M is 20; Since the experience data stored in the database is related to the previous state, the experience is learned by random extraction, which reduces the relationship between the input states, so that the brain can more accurately predict the action of the current state, that is, according to the measured ammonia nitrogen concentration value x i , dissolved oxygen concentration value y i

[0112] and pH value z i , the bait amount m and the microbial preparation amount n value are estimated; This module can learn from past experience and also obtain new experience values from the current state-action, that is, this module can determine the reward value according to the current state and the subsequent state. The system can judge whether the current experience can make the water quality control reach the best breeding state through the reward value r and the value Q. The whole process is a cycle of state-action value judgment without human intervention.

[0113] 5、According to the material quantity-time relation, the time of opening the electromagnetic switch valve determines the quantity of material, and the above relation is determined by the formula , where m represents mass, V represents volume, p represents density, S represents the area of the bottom of the switch valve, v represents the speed of the material falling, and t represents time. Then, the single-chip microcomputer in the controller of the supply station decodes the time needed to be opened. When the timing time arrives, the electromagnetic switch valve is closed. The water quality sensing and intelligent control device performs material throwing in the set area.

Claims

1. A self-cruise fishery water quality sensing and intelligent control system, characterized in that, The application relates to an aquaculture water quality sensing and intelligent control device and a shore-based control center; the aquaculture water quality sensing and intelligent control device comprises an energy supply module, a power driving module, an automatic feeding module, a BDS positioning module, a water quality data acquisition module, a wireless transmission module and a feeding device controller, the energy supply module is electrically connected with other modules; the shore-based main control center comprises a sending working mode module, an intelligent decision module, an intelligent control module and a warning module; the sending working mode module comprises automatic feeding instructions, water quality monitoring instructions and water quality control instructions; the water quality intelligent decision module comprises a water quality judgment module and an intelligent decision module; the intelligent control module comprises an oxygen supply machine and a supply station; The working signals of the shore-based main control center are sent and received by the other modules in the aquaculture water quality sensing and intelligent control device through the feeding device controller; the cruise instructions of the working mode module are decoded by the automatic feeding module, the power driving module reaches the designated area for work according to the decoded cruise instructions, the water quality data acquisition module collects water quality physicochemical factors and transmits signals to the intelligent decision module, the intelligent decision module analyzes the water quality physicochemical factors through the deep learning neural network of the BP neural network and the DQN algorithm, and controls the automatic feeding module to automatically feed through the sending working mode module and the intelligent control module; The automatic feeding module is composed of a first antenna (3), a first positioning plate (4), a feeding bin (5), a controller (6), a first wireless transmission module (8), a first switch valve (10), an electric control cabinet (27) and a time relay; the first antenna (3) is welded at the top of the feeding bin (5), the first positioning plate (4) is welded at the top of the feeding bin (5), the controller (6) and the first wireless transmission module (8) are fixedly assembled in the electric control cabinet (27), and the first switch valve (10) is welded at the bottom of the feeding bin (5); the opening time of the first switch valve (10) is decoded by the controller (6); The supply station of the shore-based control center comprises a bait supply bin (17) and a micro-ecological preparation supply bin (18), a second antenna (19) and a second wireless transmission module (20), a second switch valve (21), a fixed support (23), a stirring chamber (22), a material conveying pipeline (24) and a second positioning plate (25); the second switch valve (21), the stirring chamber (22) and the material conveying pipeline (24) are welded together through a stainless steel support; when the second positioning plate (25) and the first positioning plate (4) are connected together, the second switch valve (21) is opened, the bait is stirred in the stirring chamber (22) and then transmitted to the feeding bin (5) through the material conveying pipeline.

2. The self-cruise fishery water quality sensing and intelligent control system according to claim 1, characterized in that, The energy supply module is composed of a solar panel (1) and a storage battery (9), the solar panel (1) is welded at the top end of the water quality sensing and intelligent control device, the storage battery (9) is placed in the electric control cabinet (27), the solar panel (1) converts solar energy into electrical energy and stores it in the storage battery (9); and / or the power driving module is composed of a floating plate (7), a propeller (15) and a brushless motor (16), the propeller (15) is welded on both sides of the bottom bracket of the floating plate (7), and the brushless motor (16) drives the propeller (15) to work.

3. The self-cruise fishery water quality sensing and intelligent control system according to claim 1, characterized in that, The water quality data acquisition module includes ammonia nitrogen sensor, pH sensor and dissolved oxygen sensor (12).

4. The self-cruise fishery water quality sensing and intelligent control system according to claim 1, characterized in that, The lower part of the automatic feeding module is welded with a filter cartridge (11), and the bottom end of the filter cartridge (11) is provided with holes with different diameters.

5. The method of claim 1, wherein the method further comprises: Comprise: Step one: the power driving module sails according to the cruise channel set by the DBS positioning module; Step two: the water quality acquisition module collects water quality factors in real time, and then transmits them to the water quality analysis module, and judges the water quality through the BP neural network; Step three: the water quality judgment result is transmitted to the feeding amount decision module, and the feeding amount of bait and micro-ecological preparation is determined through the deep learning neural network of DQN algorithm; Step four: after the intelligent decision module outputs the current feeding amount, the sending working mode module and the intelligent control module send instructions to the automatic feeding module for automatic feeding; Step five: after automatic feeding is completed, return to the original position and standby; The feeding amount decision module determines the feeding amount of bait and micro-ecological preparation according to the following method in step three: (1) establish the brain of intelligent decision module; (2) Establish experience database; Set state-action database can store N data, action-evaluation network after the evaluation of the network will experience J[S i ,a i ,r i ,S i '] to experience database, when the database is full of data, new learning experience J replaces the old data, to meet the real-time update of the database, wherein, S i represents the state of the current water quality physicochemical factors, a i represents the current amount of bait and microecological preparation action, r i represents the reward value after the current action, S i v represents the state of water quality physicochemical factors measured after the action of feeding bait and microecological preparation; (3) The self-learning module of the intelligent decision-making module is established; the reward value is determined according to the current state and the subsequent state, and the system can judge whether the current experience can make the water quality regulation reach the best breeding state through the reward value r and the value Q, wherein the reward value r represents the immediate reward after the action is performed, and the value Q represents the expected report that can be obtained in the future i Take state a i , The formula for calculating the reward value r is as follows: The current state S i and the current action a i are input into an evaluation-action neural network, and an output value Q, i.e., a function Q(S i ,a i ; ω) is output, where ω represents the current network weight value; the evaluation-action neural network model adopts a DQN algorithm.

6. The control method according to claim 5, characterized in that: The feeding mode of bait and micro-ecological preparation in step three includes: when the monitored water quality physicochemical factors are normal, only bait is fed; when the monitored water quality physicochemical factors are abnormal, the signal is transmitted to the warning module, and bait and micro-ecological preparation are fed; after feeding micro-ecological preparation, wait for 12-24 hours, and if the intelligent decision feeding amount module still needs to feed micro-ecological preparation after judging, only micro-ecological preparation needs to be fed, until the monitored water quality physicochemical factors return to the normal range.

7. The method of claim 5, wherein the step of regulating is performed by a processor. The set time is calculated from the material amount-time relationship, and is determined by where m represents mass, V represents volume, p represents density, S represents the bottom area of the on-off valve, v represents the speed of the material falling, and t represents time.

Images