CN105592518A - Water quality monitoring system and method for aquaculture farm - Google Patents

Abstract

The invention discloses a water quality monitoring system for an aquaculture farm. The system comprises a main controller, a water sound communication module, a remote monitoring center terminal and a wireless sensor network. The main controller is connected with the water sound communication module. The main controller and the remote monitoring center terminal are in communication connection with each other via a wireless communication network. The wireless sensor network is composed of a plurality of respectively independent clusters and convergent nodes. Each cluster is composed of a cluster node and a plurality of common nodes. Each node is in the form of a sensor module, and is composed of a temperature sensor, a PH value sensor and a dissolved oxygen sensor, wherein the parameters of the water environment in the aquiculture area are acquired by the nodes. According to the technical scheme of the invention, the water quality of the large aquaculture water area can be monitored accurately in real time. Meanwhile, the parameters of the deep water environment of the net cage type aquaculture can be monitored. Moreover, the data transmission problem between an ashore remote monitoring center terminal and an underwater sensing network is solved. The requirements of the fishery production, such as the control informatization and the control networking, are also met.

Description

Aquatic farm water quality monitoring system and method

Technical field

The present invention relates to a kind of water quality monitoring system and method, relate in particular to a kind of aquatic farm water quality monitoring system andMethod, belongs to aquaculture technical field.

Background technology

Fish production is the important part in national product field, and the water quality parameter in cultivation waters directly affects aquaticThe existence of thing, such as temperature, water-soluble oxygen amount, pH value, ammonia-nitrogen content etc. all can have impact to different fingerling growths. But large faceDiversity, the polytropy of long-pending water area water-quality have brought difficulty to fishery cultivating technology, therefore, and to the prison of culture environment of aquatic products parameterSurvey and become a kind of problem of being badly in need of research. Because China's cultural technique is relatively backward, water quality still relies on people to a great extentWork realizes monitoring, and especially the detection in the deep water waters to water field of big area, cage type cultivation is wasted time and energy, if can not be real-timeAccurately monitoring water environment parameter, will inevitably cause heavy losses. Therefore, development one can Real-Time Monitoring aquatic farm water qualityMonitoring system has become exigence.

Underwater wireless sensor network (UWSN) can be used for the letter of object in perception, acquisition and processing network's coverage areaCease and send to recipient. Can be with it real-time dynamic monitoring that carries out to cultivation waters shallow-water environment, but only realized water ringThe parameter monitoring of border surf zone, can not monitor the environment in deep water waters. Due to the complexity of underwater environment, different depth waters ringBorder differs greatly, so, can be the cultivation of deep water box type stereo to real-time, the accurate monitoring of plant's different depth water environmentProvide safeguard.

Summary of the invention

The object of the present invention is to provide a kind of aquatic farm water quality monitoring system and method, to water field of big area, netThe water quality in the deep water waters of box cultivation is accurately monitoring in real time, and solving prior art cannot measure in real time to water quality parameter,Easily cause the problem of aquatic products loss.

Object of the present invention is achieved by the following technical programs:

A kind of aquatic farm water quality monitoring system, comprises master controller 1, underwater sound communication module 2, remote monitoring center eventuallyEnd 3, wireless sensor network 4, described master controller 1 is connected with underwater sound communication module 2, described master controller 1 and remote monitoringCenter terminal 3 communicates to connect by cordless communication network, and described wireless sensor network 4 comprises some bunch 42 Hes independently separatelyAggregation node 41; Comprise a leader cluster node 43 and several ordinary nodes 44 for described bunch 42, each node is sensor assembly,Formed by temperature, pH value and dissolved oxygen sensor, complete obtaining culture zone water environment parameter;

The composition of described wireless sensor network 4 and communication are carried out in the following manner,

1) initial phase

Step S101: network scenarios is set, and in network environment, random placement N has identical primary power, ID numberingFrom the ordinary node of 0～N-1; Aggregation node is deployed in certain of network surrounding, and each ordinary node can be directly and aggregation nodeCommunication, aggregation node Infinite Energy is large and can deal with data;

Step S102: netinit, aggregation node is broadcasted a hello message to all ordinary nodes in network,Each ordinary node is receiving after hello message the distance d that estimates itself and aggregation node according to received signal strength indicator(i, BS); Once, other ordinary nodes in certain ordinary node broadcasting area are it to each ordinary node broadcast simultaneouslyNeighbors, calculates each ordinary node and can reach power AMRP to the minimum average B configuration of its neighbors; Minimum average B configuration can reach power AMRP and useThe average signal strength of the neighbors that this ordinary node is received represents:

$AMRP=\underset{j=1}{\overset{n}{\Σ}}{\mathrm{Radio}}_j/n$

Wherein n is the neighbors number of this ordinary node, Radio_jThat this ordinary node receives that the signal of neighbors j is strongDegree;

2) the cluster stage

Step S201: cluster is prepared, first all ordinary nodes in network are set becomes the initial proportion of leader cluster nodeC_prob, then each ordinary node calculates it according to formula (1) becomes the probability CH of leader cluster node_prob, and initialize all commonThe alternative leader cluster node set of node is empty, and all ordinary nodes are not all candidate cluster head nodes, and its node state is commonNode;

CH_prob＝max(C_prob×E_residual/E_max,P_min)(1)

Wherein E_residualFor this ordinary node current remaining, E_maxFor the primary power of this ordinary node, regulationCH_probMinimum of a value is P_min, prevent that leader cluster node election iterative convergence speed is too slow;

Step S202: each ordinary node judges that whether the alternative leader cluster node set of oneself is as empty, and whether this nodeIn adding bunch, if so, enter step S203, otherwise go to step S204;

Step S203: each ordinary node judges that it becomes the probability CH of leader cluster node_probWhether be more than or equal to 1, ifBe, this ordinary node becomes final leader cluster node, and enters step S205, otherwise goes to step S206;

Step S204: alternative leader cluster node set does not judge oneself whether to be candidate cluster head node for empty ordinary node,If so, enter step S207, otherwise go to step S208;

Step S205: this ordinary node becomes final leader cluster node, upgrading node state is final leader cluster node, by oneselfAdd its alternative leader cluster node set, then calculate its bunch of radius R according to formula (2)_i, and with a bunch radius R_iTo common in its bunchNode broadcasts oneself is elected as the message of final leader cluster node, this message comprise No. ID of this final leader cluster node, node state,Minimum average B configuration can reach power AMRP and bunch radius; After information broadcast is complete, in its bunch of radius, all ordinary nodes upgrade its alternative bunchHead node set, adds this final leader cluster node in alternative leader cluster node set, and proceeds to step S212;

$R_i=\mathrm{\ω}(1-\frac{d_{\max }-d(i,BS)}{d_{\max }-d_{\min }})R_0+(1-\mathrm{\ω})(1-c\frac{{\mathrm{CH}}_{prob\max }-{\mathrm{CH}}_{prob}\left(i\right)}{{\mathrm{CH}}_{prob\max }-{\mathrm{CH}}_{prob\min }})R_0---\left(2\right)$

Wherein d_maxAnd d_minBe respectively maximum and the minimum of a value of ordinary node to the distance of aggregation node, d (i, BS) isOrdinary node i is to the distance of aggregation node, CH_probmaxAnd CH_probminBe respectively ordinary node and become the maximum probability of leader cluster nodeAnd minimum probability, CH_prob(i) become the probability of leader cluster node for ordinary node i, R₀For ordinary node greatest irradiation radius, ω andC is the parameter for controlling span;

Step S206: each ordinary node judges that it becomes the probability CH of leader cluster node_probWhether be more than or equal to 0～1 itBetween the random number generating, if so, enter step S209, otherwise go to step S210;

Step S207: the ordinary node that becomes candidate cluster head node judges that it becomes the probability CH of leader cluster node_probWhetherBe more than or equal to 1, if so, enter step S205, otherwise go to step S209;

Step S208: for alternative leader cluster node set for empty and its oneself be not the common joint of candidate cluster head nodePoint, this ordinary node add minimum average B configuration in its alternative leader cluster node set can reach power AMRP minimum candidate cluster head node orUnder the final leader cluster node of person bunch in, and enter step S211;

Step S209: this ordinary node becomes candidate cluster head node, upgrading this ordinary node state is candidate cluster head node,To oneself add its alternative leader cluster node set, and calculate its bunch of radius R according to formula (2)_i, simultaneously to ordinary node in its bunchBroadcast the message of oneself being elected as candidate cluster head node, this message comprise No. ID of this candidate cluster head node self, node state,Minimum average B configuration can reach power AMRP and bunch radius; When after the complete message of candidate cluster head node broadcasts, all common joints in its bunch of radiusPoint upgrades its alternative leader cluster node set, this candidate cluster head node added in alternative leader cluster node set, and execution stepS210；

Step S210: ordinary node or candidate cluster head node are by the CH of self_probBe multiplied by 2 and enter step S211;

Step S211: judge iterations whether be greater than M time, M be maximum iteration time andAsFruit is end iteration and enter step S212, otherwise iterations to add 1, and go to step S202;

Step S212: cluster stops, and each node determines its end-state; If ordinary node has become in the time of iterative processFinal leader cluster node, it is elected as leader cluster node; If there is final bunch of cephalomere in the alternative leader cluster node set of ordinary nodePoint, it adds in minimum average B configuration in alternative leader cluster node set can reach under the final leader cluster node of power AMRP minimum bunch;If certain ordinary node is isolated node, in alternative leader cluster node set, there is not final leader cluster node, add its communication halfIn footpath under nearest cluster node bunch in; If this isolated node communication radius is interior without other nodes, isolated node listAlone become bunch;

3) stage of communication bunch

Adopt bunch in single-hop and bunch between multi-hop data transmission means, each leader cluster node need to be from the leader cluster node of adjacent clusterLeader cluster node of middle selection is as its via node, and forwarding data is to aggregation node; Distance for bunch head apart from aggregation nodeFrom bunch head that is less than T, all directly single-hop transmission is to aggregation node, and wherein T is a predefined value, and its value is always less than generalLogical node greatest irradiation radius R₀, be greater than bunch head of T apart from the distance of aggregation node for bunch head, from the set of adjacent cluster head nodeBunch head of a Least-cost of middle selection, as its via node, through multi-hop, completes transfer of data to aggregation node place; SpecificallyProcess is as follows:

Step S301: leader cluster node calculates its distance apart from aggregation node, and judge whether distance is less than T, if so,Enter step S302, otherwise go to step S303;

Step S302: leader cluster node is directly transported to aggregation node by data sheet jump set;

Step S303: leader cluster node c_iWith doubly bunch radius broadcast of its β, wherein between the general value 1～1.5 of β and itsBroadcast radius is less than ordinary node greatest irradiation radius R₀; The message of leader cluster node broadcast comprise leader cluster node ID, dump energy,The node cost of leader cluster node self and to the distance of aggregation node, node cost is calculated according to formula (3); Receive messageOther leader cluster nodes also send return messages to the leader cluster node c that sends message_i, return messages comprise its node cost andTo the distance of aggregation node; Leader cluster node c_iCalculate the set of adjacent cluster head node according to a bunch number of sending return messagesc_i.R_CH, a bunch c_iAdjacent cluster head node sets definition be: c_i.R_CH＝{c_j|d(c_i,c_j)≤βR_i,d(c_j,BS)＜d(c_i,BS) }; Wherein d (c_i,c_j) refer to leader cluster node c_iWith c_jDistance to each other, d (c_j, BS) and refer to leader cluster node c_jTo converging jointThe distance of point, d (c_i, BS) and refer to leader cluster node c_iTo the distance of aggregation node;

Node cost computing formula is: $\cos t\left(i\right)=\mathrm{\μ}\frac{\stackrel{\‾}{d_{c-BS}}}{d_{ci-BS}}+(1-\mathrm{\μ})\frac{E_{ci}}{\stackrel{\‾}{E_c}}---\left(3\right)$

WhereinFor the leader cluster node in this leader cluster node and its adjacent cluster head node set is to aggregation node distanceMean value, d_ci-BSFor leader cluster node c_iTo the distance of aggregation node,For in leader cluster node and its adjacent cluster head node setThe mean value of leader cluster node dump energy, E_ciFor leader cluster node c_iDump energy, μ is the parameter for controlling span;

Step S304: leader cluster node c_iFrom its adjacent cluster head node set, select the leader cluster node of Least-cost as itNext-hop node.

Object of the present invention can also further realize by following technical measures:

The method for supervising of aforementioned aquatic farm water quality monitoring system, comprises the following steps:

1) according to the record of water quality parameter to aquatic farm, count shallow waters temperature, water-soluble oxygen amount, pH value,Air mass flow, the data of the deep water waters temperature in corresponding moment, water-soluble oxygen amount, pH value; By shallow waters temperature, water-soluble oxygen amount,PH value, air mass flow, as input parameter, using deep water waters temperature, water-soluble oxygen amount, pH value as output parameter, are set up nerve netNetwork, according to existing Historical Monitoring data, uses BP neutral net, additional momentum learning rules, neural network training;

2) according to water biological species in the suitable temperature of deep water Watershed expand, water-soluble oxygen amount, pH value ideal value, calculated by populationMethod, the optimum that solves neutral net is inputted parameter, i.e. shallow waters temperature, water-soluble oxygen amount, pH value, air mass flow;

3) judge whether to need artificial sample according to a upper Recognition with Recurrent Neural Network evaluated error, as needs, by manually adoptingSample is off-line analysis then, and contrast draws the deep water that deep water waters temperature, water-soluble oxygen amount, pH value and the neutral net of actual measurement estimateThe error of waters temperature, water-soluble oxygen amount, pH value, then by deep water waters temperature, water-soluble oxygen amount, the pH value data of this group actual measurement,And neutral net estimation is together with the error information of surveying, and uses additional momentum learning rules, upgrades neural network training; AsDo not need artificial sample, return to step 2).

The method for supervising of aforementioned aquatic farm water quality monitoring system, wherein particle cluster algorithm, step is as follows:

1) initialize population: determine population size NP, particle cluster algorithm iterations NG, initializes particle position,Calculate the fitness of each particle and initialize globally optimal solution and individual optimal solution;

The function that calculates particle fitness is:

$Fitness=\underset{i}{\Σ}{(O_i-O_i^{\′})}^2$

Wherein, O_iRepresent i element of neutral net output vector, O_i' be the i of the output vector of theoretical expectationElement;

2) upgrade population: the equation of motion of population is as follows:

v(t)＝ω·v(t-1)+c₁·(lbest-x(t))+c₂·(gbest-x(t))

x(t+1)＝x(t)+c₃·v(t)

Wherein ω is taken asI is this iterations of particle cluster algorithm, c₁,c₂,c₃For constant, c₁,c₂GetValue is 2.8, c₃Value is that 0.3, lbest is the individual optimal solution that each particle search is crossed, and gbest is that all particle search are crossedGlobally optimal solution;

3) calculate the particle fitness of this iteration, upgrade individual optimal solution and globally optimal solution: to each particle, willThe fitness that this iteration produces, compared with current individual optimal solution, get fitness less be individual optimal solution, with all grainsThe globally optimal solution search for of son is compared, get fitness less be globally optimal solution;

4) judge whether to reach iteration NG time, if so, export globally optimal solution, if not, return to step 2).

The method for supervising of aforementioned aquatic farm water quality monitoring system, wherein additional momentum learning method, update rule is as followsFormula:

$\mathrm{\ω}(t+1)=\mathrm{\ω}\left(t\right)-(1-a)\mathrm{\η}\frac{\∂E_T}{\∂\mathrm{\ω}\left(t\right)}+a\mathrm{\Δ}\mathrm{\ω}\left(t\right)$

Wherein Δ ω (t)=ω (t)-ω (t-1), E_TFor the training error of neutral net, η is weight, a be momentum because ofSon, gets 0.9.

The method for supervising of aforementioned aquatic farm water quality monitoring system can also be achieved by another kind of technical scheme:

A method for supervising for aquatic farm water quality monitoring system, comprises the following steps:

1) according to the record of water quality parameter to aquatic farm, count shallow waters temperature, water-soluble oxygen amount, pH value,Air mass flow, the data of the deep water waters temperature in corresponding moment, water-soluble oxygen amount, pH value; By shallow waters temperature, water-soluble oxygen amount,PH value, air mass flow, as input parameter, using deep water waters temperature, water-soluble oxygen amount, pH value as output parameter, are set up nerve netNetwork, according to existing Historical Monitoring data, uses BP neutral net, additional momentum learning rules, neural network training;

2) according to water biological species in the suitable temperature of deep water Watershed expand, water-soluble oxygen amount, pH value ideal value, calculated by heredityMethod, the optimum that solves neutral net is inputted parameter, i.e. shallow waters temperature, water-soluble oxygen amount, pH value, air mass flow;

Described genetic algorithm comprises the following steps:

1. adopt real coding, initialize chromosome, form initial population;

2. utilize fitness function to evaluate the each chromosome in each generation;

3. carry out genetic manipulation;

4. recalculate the adaptive value of each individuality;

5. choose after new population, the optimum individual in new population is retained, use the previous generation's optimum individual to replace thisThe poorest individuality in generation;

6. judge whether to reach evolutionary generation, if do not have, return to the 2. step, otherwise finish;

7. using the value of the optimum individual in new population as with, remain unchanged, adopt BP algorithm to learn, until meetPerformance indications;

3) judge whether to need artificial sample according to a upper Recognition with Recurrent Neural Network evaluated error, as needs, by manually adoptingSample is off-line analysis then, and contrast draws the deep water that deep water waters temperature, water-soluble oxygen amount, pH value and the neutral net of actual measurement estimateThe error of waters temperature, water-soluble oxygen amount, pH value, then by deep water waters temperature, water-soluble oxygen amount, the pH value data of this group actual measurement,And neutral net estimation is together with the error information of surveying, and uses additional momentum learning rules, upgrades neural network training; AsDo not need artificial sample, return to step 2).

Compared with prior art, the invention has the beneficial effects as follows: the present invention realizes real-time to cultivating large area water area water-qualityAccurately monitoring, the present invention, except realizing the ambient parameter monitoring of waters surf zone, can also realize dark that cage type is cultivatedWater water environment parameter monitoring; The present invention, by wireless network communication technique is combined with UWSN technology, has well solvedData transmission problems between remote monitoring terminal and underwater sensing network on the bank, to meet fish production control information, netThe requirement of network.

Brief description of the drawings

Fig. 1 is aquatic farm water quality monitoring system structure chart of the present invention;

Fig. 2 is aquatic farm water quality monitoring system wireless sensor network structure chart of the present invention.

Detailed description of the invention

Below in conjunction with the drawings and specific embodiments, the invention will be further described.

In aquatic farm, waters temperature, water-soluble oxygen amount, pH value are very important to the growth of aquatic organism, in order to improveThen aquatic products quality and output, need to measure and regulate cultivation waters physical parameter, controls if realize automaticallyConvenient, efficient and be conducive to aquatic products growth. In order to realize the detection to waters water environment, the present invention adopts wireless senserNetwork detects water field of big area. The self-organizing network of normally a kind of foundation-free facility of wireless sensor network, sensingDevice node uses powered battery, and disposes the general unmanned in waters, easily because depleted of energy lost efficacy, finally causes wholeNetwork paralysis. If the energy consumption of energy active balance whole system, can prolonging network survival time. As shown in Figure 1, in order to realizeThis purpose, aquatic farm water quality monitoring system of the present invention comprises master controller 1, underwater sound communication module 2, remote monitoring centerTerminal 3, wireless sensor network 4, described master controller 1 is connected with underwater sound communication module 2, described master controller 1 and remote monitoringControl center terminal 3 communicates to connect by cordless communication network, and described wireless sensor network 4 comprises some separately independently bunches 42With aggregation node 41; Comprise a leader cluster node 43 and several ordinary nodes 44 for described bunch 42, each node is sensor diePiece, is made up of temperature, pH value and dissolved oxygen sensor, completes obtaining culture zone water environment parameter;

The composition of described wireless sensor network 4 and communication are carried out in the following manner,

1) initial phase

Step S101: network scenarios is set, and in network environment, random placement N has identical primary power, ID numberingFrom the ordinary node of 0～N-1; Aggregation node is deployed in certain of network surrounding, and each ordinary node can be directly and aggregation nodeCommunication, aggregation node Infinite Energy is large and can deal with data;

Step S102: netinit, aggregation node is broadcasted a hello message to all ordinary nodes in network,Each ordinary node is receiving after hello message the distance d that estimates itself and aggregation node according to received signal strength indicator(i, BS); Once, other ordinary nodes in certain ordinary node broadcasting area are it to each ordinary node broadcast simultaneouslyNeighbors, calculates each ordinary node and can reach power AMRP to the minimum average B configuration of its neighbors; Minimum average B configuration can reach power AMRP and useThe average signal strength of the neighbors that this ordinary node is received represents:

$AMRP=\underset{j=1}{\overset{n}{\Σ}}{\mathrm{Radio}}_j/n$

Wherein n is the neighbors number of this ordinary node, Radio_jThat this ordinary node receives that the signal of neighbors j is strongDegree;

2) the cluster stage

Step S201: cluster is prepared, first all ordinary nodes in network are set becomes the initial proportion of leader cluster nodeC_prob, then each ordinary node calculates it according to formula (1) becomes the probability CH of leader cluster node_prob, and initialize all commonThe alternative leader cluster node set of node is empty, and all ordinary nodes are not all candidate cluster head nodes, and its node state is commonNode;

CH_prob＝max(C_prob×E_residual/E_max,P_min)(1)

Wherein E_residualFor this ordinary node current remaining, E_maxFor the primary power of this ordinary node, regulationCH_probMinimum of a value is P_min, prevent that leader cluster node election iterative convergence speed is too slow;

Step S202: each ordinary node judges that whether the alternative leader cluster node set of oneself is as empty, and whether this nodeIn adding bunch, if so, enter step S203, otherwise go to step S204;

Step S203: each ordinary node judges that it becomes the probability CH of leader cluster node_probWhether be more than or equal to 1, ifBe, this ordinary node becomes final leader cluster node, and enters step S205, otherwise goes to step S206;

Step S204: alternative leader cluster node set does not judge oneself whether to be candidate cluster head node for empty ordinary node,If so, enter step S207, otherwise go to step S208;

Step S205: this ordinary node becomes final leader cluster node, upgrading node state is final leader cluster node, by oneselfAdd its alternative leader cluster node set, then calculate its bunch of radius R according to formula (2)_i, and with a bunch radius R_iTo common in its bunchNode broadcasts oneself is elected as the message of final leader cluster node, this message comprise No. ID of this final leader cluster node, node state,Minimum average B configuration can reach power AMRP and bunch radius; After information broadcast is complete, in its bunch of radius, all ordinary nodes upgrade its alternative bunchHead node set, adds this final leader cluster node in alternative leader cluster node set, and proceeds to step S212;

$R_i=\mathrm{\ω}(1-\frac{d_{\max }-d(i,BS)}{d_{\max }-d_{\min }})R_0+(1-\mathrm{\ω})(1-c\frac{{\mathrm{CH}}_{prob\max }-{\mathrm{CH}}_{prob}\left(i\right)}{{\mathrm{CH}}_{prob\max }-{\mathrm{CH}}_{prob\min }})R_0---\left(2\right)$

Wherein d_maxAnd d_minBe respectively maximum and the minimum of a value of ordinary node to the distance of aggregation node, d (i, BS) isOrdinary node i is to the distance of aggregation node, CH_probmaxAnd CH_probminBe respectively ordinary node and become the maximum probability of leader cluster nodeAnd minimum probability, CH_prob(i) become the probability of leader cluster node for ordinary node i, R₀For ordinary node greatest irradiation radius, ω andC is the parameter for controlling span;

Step S206: each ordinary node judges that it becomes the probability CH of leader cluster node_probWhether be more than or equal to 0～1 itBetween the random number generating, if so, enter step S209, otherwise go to step S210;

Step S207: the ordinary node that becomes candidate cluster head node judges that it becomes the probability CH of leader cluster node_probWhetherBe more than or equal to 1, if so, enter step S205, otherwise go to step S209;

Step S208: for alternative leader cluster node set for empty and its oneself be not the common joint of candidate cluster head nodePoint, this ordinary node add minimum average B configuration in its alternative leader cluster node set can reach power AMRP minimum candidate cluster head node orUnder the final leader cluster node of person bunch in, and enter step S211;

Step S209: this ordinary node becomes candidate cluster head node, upgrading this ordinary node state is candidate cluster head node,To oneself add its alternative leader cluster node set, and calculate its bunch of radius R according to formula (2)_i, simultaneously to ordinary node in its bunchBroadcast the message of oneself being elected as candidate cluster head node, this message comprise No. ID of this candidate cluster head node self, node state,Minimum average B configuration can reach power AMRP and bunch radius; When after the complete message of candidate cluster head node broadcasts, all common joints in its bunch of radiusPoint upgrades its alternative leader cluster node set, this candidate cluster head node added in alternative leader cluster node set, and execution stepS210；

Step S210: ordinary node or candidate cluster head node are by the CH of self_probBe multiplied by 2 and enter step S211;

Step S211: judge iterations whether be greater than M time, M be maximum iteration time andAsFruit is end iteration and enter step S212, otherwise iterations to add 1, and go to step S202;

Step S212: cluster stops, and each node determines its end-state; If ordinary node has become in the time of iterative processFinal leader cluster node, it is elected as leader cluster node; If there is final bunch of cephalomere in the alternative leader cluster node set of ordinary nodePoint, it adds in minimum average B configuration in alternative leader cluster node set can reach under the final leader cluster node of power AMRP minimum bunch;If certain ordinary node is isolated node, in alternative leader cluster node set, there is not final leader cluster node, add its communication halfIn footpath under nearest cluster node bunch in; If this isolated node communication radius is interior without other nodes, isolated node listAlone become bunch;

3) stage of communication bunch

Adopt bunch in single-hop and bunch between multi-hop data transmission means, each leader cluster node need to be from the leader cluster node of adjacent clusterLeader cluster node of middle selection is as its via node, and forwarding data is to aggregation node; Distance for bunch head apart from aggregation nodeFrom bunch head that is less than T, all directly single-hop transmission is to aggregation node, and wherein T is a predefined value, and its value is always less than generalLogical node greatest irradiation radius R₀, be greater than bunch head of T apart from the distance of aggregation node for bunch head, from the set of adjacent cluster head nodeBunch head of a Least-cost of middle selection, as its via node, through multi-hop, completes transfer of data to aggregation node place; SpecificallyProcess is as follows:

Step S301: leader cluster node calculates its distance apart from aggregation node, and judge whether distance is less than T, if so,Enter step S302, otherwise go to step S303;

Step S302: leader cluster node is directly transported to aggregation node by data sheet jump set;

Step S303: leader cluster node c_iWith doubly bunch radius broadcast of its β, wherein between the general value 1～1.5 of β and itsBroadcast radius is less than ordinary node greatest irradiation radius R₀; The message of leader cluster node broadcast comprise leader cluster node ID, dump energy,The node cost of leader cluster node self and to the distance of aggregation node, node cost is calculated according to formula (3); Receive messageOther leader cluster nodes also send return messages to the leader cluster node c that sends message_i, return messages comprise its node cost andTo the distance of aggregation node; Leader cluster node c_iCalculate the set of adjacent cluster head node according to a bunch number of sending return messagesc_i.R_CH, a bunch c_iAdjacent cluster head node sets definition be: c_i.R_CH＝{c_j|d(c_i,c_j)≤βR_i,d(c_j,BS)＜d(c_i,BS) }; Wherein d (c_i,c_j) refer to leader cluster node c_iWith c_jDistance to each other, d (c_j, BS) and refer to leader cluster node c_jTo converging jointThe distance of point, d (c_i, BS) and refer to leader cluster node c_iTo the distance of aggregation node;

Node cost computing formula is: $\cos t\left(i\right)=\mathrm{\μ}\frac{\stackrel{\‾}{d_{c-BS}}}{d_{ci-BS}}+(1-\mathrm{\μ})\frac{E_{ci}}{\stackrel{\‾}{E_c}}---\left(3\right)$

WhereinFor the leader cluster node in this leader cluster node and its adjacent cluster head node set is to aggregation node distanceMean value, d_ci-BSFor leader cluster node c_iTo the distance of aggregation node,For in leader cluster node and its adjacent cluster head node setThe mean value of leader cluster node dump energy, E_ciFor leader cluster node c_iDump energy, μ is the parameter for controlling span;

Step S304: leader cluster node c_iFrom its adjacent cluster head node set, select the leader cluster node of Least-cost as itNext-hop node.

The building form energy active balance whole system of the radio sensing network of above-mentioned aquatic farm water quality monitoring systemEnergy consumption, extend network time of surviving in the unattended situation of water field of big area.

Above-mentioned aquatic farm water quality monitoring system carries out real-time dynamic monitoring to cultivation waters shallow-water environment, but only realizesThe parameter monitoring of water environment surf zone, for adopting in the situation of deep-water net cage culture mode, due to underwater environment complexityChangeable, need to monitor the water quality parameter in deep water waters, to ensure quality and the output of deep water culture. The present invention sets up shallow watersMapping between the deep water waters temperature in temperature, water-soluble oxygen amount, pH value, air mass flow and corresponding moment, water-soluble oxygen amount, pH value is closedSystem, sets up neutral net also after training, by ginsengs such as the shallow waters temperature that can survey, water-soluble oxygen amount, pH value, air mass flowsNumber estimates the parameter that is difficult to the deep water waters of directly measuring in real time.

For realizing this purpose, should first serviceability temperature meter, the sensor such as mass air flow sensor, water-soluble oxygen amount detector is right respectivelyThe water quality parameter in shallow waters, deep water waters is measured, and sets up this aquatic farm water quality physical quantity database, thenUtilize neutral net that deep water parameter is indirectly measured and controlled. Specifically comprise the following steps:

1) according to the record of water quality parameter to aquatic farm, count shallow waters temperature, water-soluble oxygen amount, pH value,Air mass flow, the data of the deep water waters temperature in corresponding moment, water-soluble oxygen amount, pH value; By shallow waters temperature, water-soluble oxygen amount,PH value, air mass flow, as input parameter, using deep water waters temperature, water-soluble oxygen amount, pH value as output parameter, are set up nerve netNetwork, according to existing Historical Monitoring data, uses BP neutral net, additional momentum learning rules, neural network training;

2) according to water biological species in the suitable temperature of deep water Watershed expand, water-soluble oxygen amount, pH value ideal value, calculated by populationMethod, the optimum that solves neutral net is inputted parameter, i.e. shallow waters temperature, water-soluble oxygen amount, pH value, air mass flow; Wherein particleGroup's algorithm, step is as follows:

(1) initialize population: determine population size NP, particle cluster algorithm iterations NG, initializes particle position,Calculate the fitness of each particle and initialize globally optimal solution and individual optimal solution;

The function that calculates particle fitness is:

$Fitness=\underset{i}{\Σ}{(O_i-O_i^{\′})}^2$

Wherein, O_iRepresent i element of neutral net output vector, O_i' be the i of the output vector of theoretical expectationElement;

(2) upgrade population: the equation of motion of population is as follows:

v(t)＝ω·v(t-1)+c₁·(lbest-x(t))+c₂·(gbest-x(t))

x(t+1)＝x(t)+c₃·v(t)

Wherein ω is taken asI is this iterations of particle cluster algorithm, c₁,c₂,c₃For constant, c₁,c₂GetValue is 2.8, c₃Value is that 0.3, lbest is the individual optimal solution that each particle search is crossed, and gbest is that all particle search are crossedGlobally optimal solution;

(3) calculate the particle fitness of this iteration, upgrade individual optimal solution and globally optimal solution: to each particle,The fitness that this iteration is produced, compared with current individual optimal solution, get fitness less be individual optimal solution, with allThe globally optimal solution that particle search is crossed is compared, get fitness less for globally optimal solution;

(4) judge whether to reach iteration NG time, if so, export globally optimal solution, if not, return to step 2).

3) due to neural network shallow waters temperature, water-soluble oxygen amount, pH value, air mass flow, corresponding moment darkMapping relations between water waters temperature, water-soluble oxygen amount, pH value, neutral net, can be with the shallow water that can survey after trainingParameter Estimation goes out the deep water parameter that can not directly measure in real time. In order to obtain better measurement effect, dwindle neutral net and estimateError, need to upgrade neural network training.

Therefore,, in a upper circulation, the error of the data of artificial sample gained and neutral net data estimator judgesWhether this needs again to carry out artificial sample measurement, and its judgment rule is: if artificial sample last time data and neutral net numberLess according to error, extend the interval time of next artificial sample and this sampling, if this artificial sample sampled data and godLarger through network data error, to dwindle next artificial sample and this sampling interval time; The concrete time in sampling intervalNeed to determine according to the requirement of producing. If neutral net is constantly updated to training, estimated value constantly approaches measured value, instituteThe number of times of the off-line artificial sample Measurement and analysis needing can reduce, and the sampling interval can strengthen, and has alleviated manual working burden.

Specific practice is for to judge whether to need artificial sample according to a upper Recognition with Recurrent Neural Network evaluated error, as needs, logicalCross then off-line analysis of artificial sample, contrast draws errors actual measurement and data estimation, then by this group measured data, withAnd together with error information between neutral net estimated value and measured value, use additional momentum learning rules, upgrade training neuralNetwork; If do not needed artificial sample, return to step 2).

Described additional momentum learning method, update rule as shown in the formula:

$\mathrm{\ω}(t+1)=\mathrm{\ω}\left(t\right)-(1-a)\mathrm{\η}\frac{\∂E_T}{\∂\mathrm{\ω}\left(t\right)}+a\mathrm{\Δ}\mathrm{\ω}\left(t\right)$

Wherein Δ ω (t)=ω (t)-ω (t-1), E_TFor the training error of neutral net, η is weight, a be momentum because ofSon, gets 0.9.

Described step 2) solve the optimum input parameter of neutral net, required in order to obtain the growth of deep water aquatic products speciesTemperature, water-soluble oxygen amount, the ideal value of pH value, solve corresponding with it shallow water temperature, water-soluble oxygen by method of the present inventionAmount, pH value, air mass flow, then pass through firing equipment, oxygenerator, oxygenation equipment, blower fan, input soda acid material etc. to plantShallow waters water quality parameter is controlled, thereby indirectly controls deepwater environment, finally reaches the growth of deep water aquatic products species requiredThe object of the ideal value of temperature, water-soluble oxygen amount, pH value.

Object of the present invention can also be achieved by another kind of technical scheme, equally based on neutral net, stillUse genetic algorithm, solve the optimum input parameter of neutral net.

Specifically comprise the following steps:

1) according to the record of water quality parameter to aquatic farm, count shallow waters temperature, water-soluble oxygen amount, pH value,Air mass flow, the data of the deep water waters temperature in corresponding moment, water-soluble oxygen amount, pH value; By shallow waters temperature, water-soluble oxygen amount,PH value, air mass flow, as input parameter, using deep water waters temperature, water-soluble oxygen amount, pH value as output parameter, are set up nerve netNetwork, according to existing Historical Monitoring data, uses BP neutral net, additional momentum learning rules, neural network training;

2) according to water biological species in the suitable temperature of deep water Watershed expand, water-soluble oxygen amount, pH value ideal value, calculated by heredityMethod, the optimum that solves neutral net is inputted parameter, i.e. shallow waters temperature, water-soluble oxygen amount, pH value, air mass flow;

Described genetic algorithm comprises the following steps:

1. adopt real coding, initialize chromosome, form initial population;

2. utilize fitness function to evaluate the each chromosome in each generation;

3. carry out genetic manipulation;

4. recalculate the adaptive value of each individuality;

5. choose after new population, the optimum individual in new population is retained, use the previous generation's optimum individual to replace thisThe poorest individuality in generation;

6. judge whether to reach evolutionary generation, if do not have, return to the 2. step, otherwise finish;

7. using the value of the optimum individual in new population as with, remain unchanged, adopt BP algorithm to learn, until meetPerformance indications;

3) judge whether to need artificial sample according to a upper Recognition with Recurrent Neural Network evaluated error, as needs, by manually adoptingSample is off-line analysis then, and contrast draws the deep water that deep water waters temperature, water-soluble oxygen amount, pH value and the neutral net of actual measurement estimateThe error of waters temperature, water-soluble oxygen amount, pH value, then by deep water waters temperature, water-soluble oxygen amount, the pH value data of this group actual measurement,And neutral net estimation is together with the error information of surveying, and uses additional momentum learning rules, upgrades neural network training; AsDo not need artificial sample, return to step 2).

Genetic algorithm is a kind of global optimization method of the random search based on biological evolution process, and it is by intersecting and becomingThe different impact that greatly reduces original state, makes search obtain optimal result, and does not rest on local minimum place. Therefore, forBring into play genetic algorithm and BP algorithm strong point separately, there is the parameter of locality with the adjusting of BP algorithm and optimization, use genetic algorithmOptimize and there is parameter of overall importance.

What no matter which kind of method realized can will be in real time by wired or wireless network to the measured value in plant watersParameter transfers to long-range monitoring server, to realize informationization, the networking to large area fish production control.

In addition to the implementation, the present invention can also have other embodiments, and all employings are equal to replaces or equivalent transformation shapeThe technical scheme becoming, all drops in the protection domain of requirement of the present invention.

Claims (5)

1. an aquatic farm water quality monitoring system, comprises master controller (1), underwater sound communication module (2), remote monitoring centerTerminal (3), wireless sensor network (4), described master controller (1) is connected with underwater sound communication module (2), described master controller(1) communicate to connect by cordless communication network with remote monitoring center terminal (3), described wireless sensor network (4) is if compriseDry independently bunch (42) and aggregation node (41) separately; Described bunch (42) comprise a leader cluster node (43) and several common jointsPoint (44), each node is sensor assembly, comprises temperature, pH value and dissolved oxygen sensor, completes culture zone water environment is joinedObtaining of number; It is characterized in that, the composition of described wireless sensor network (4) and communication are carried out as follows:

1) initial phase

Step S101: network scenarios is set, and in network environment, random placement N has identical primary power, ID numbers from 0The ordinary node of～N-1; Aggregation node is deployed in certain of network surrounding, each ordinary node can be directly and aggregation node logicalLetter, aggregation node Infinite Energy is large and can deal with data;

Step S102: netinit, a hello message of aggregation node broadcast is to all ordinary nodes in network, eachOrdinary node receive after hello message according to received signal strength indicator estimate itself and aggregation node distance d (i,BS); Once, other ordinary nodes in certain ordinary node broadcasting area are its adjacent joint to each ordinary node broadcast simultaneouslyPoint, calculates each ordinary node and can reach power AMRP to the minimum average B configuration of its neighbors; It is general with this that minimum average B configuration can reach power AMRPThe average signal strength of the neighbors that logical node is received represents:

$AMRP=\underset{j=1}{\overset{n}{\Σ}}{\mathrm{Radio}}_j/n$

Wherein n is the neighbors number of this ordinary node, Radio_jIt is the signal strength signal intensity that this ordinary node is received neighbors j;

2) the cluster stage

Step S201: cluster is prepared, first all ordinary nodes in network are set becomes the initial proportion C of leader cluster node_prob, soRear each ordinary node calculates it according to formula (1) becomes the probability CH of leader cluster node_prob, and initialize all ordinary nodesAlternative leader cluster node set is empty, and all ordinary nodes are not all candidate cluster head nodes, and its node state is ordinary node;

CH_prob＝max(C_prob×E_residual/E_max,P_min)(1)

Wherein E_residualFor this ordinary node current remaining, E_maxFor the primary power of this ordinary node, regulation CH_prob?Little value is P_min, prevent that leader cluster node election iterative convergence speed is too slow;

Step S202: each ordinary node judges that whether the alternative leader cluster node set of oneself is as empty, and whether this node addsIn bunch, if so, enter step S203, otherwise go to step S204;

Step S203: each ordinary node judges that it becomes the probability CH of leader cluster node_probWhether be more than or equal to 1, if so,This ordinary node becomes final leader cluster node, and enters step S205, otherwise goes to step S206;

Step S204: alternative leader cluster node set does not judge oneself whether to be candidate cluster head node for empty ordinary node, ifBe, enter step S207, otherwise go to step S208;

Step S205: this ordinary node becomes final leader cluster node, upgrading node state is final leader cluster node, and oneself is addedIts alternative leader cluster node set, then calculates its bunch of radius R according to formula (2)_i, and with a bunch radius R_iTo ordinary node in its bunchBroadcast the message of oneself being elected as final leader cluster node, this message comprises No. ID of this final leader cluster node, node state, minimumOn average can reach power AMRP and bunch radius; After information broadcast is complete, in its bunch of radius, all ordinary nodes upgrade its alternative bunch of cephalomerePoint set, adds this final leader cluster node in alternative leader cluster node set, and proceeds to step S212;

$R_i=\mathrm{\ω}\left(1-c\frac{d_{\max }-d\left(i,BS\right)}{d_{\max }-d_{\min }}\right)R_0+\left(1-\mathrm{\ω}\right)\left(1-c\frac{{\mathrm{CH}}_{prob\max }-{\mathrm{CH}}_{prob}\left(i\right)}{{\mathrm{CH}}_{prob\max }-{\mathrm{CH}}_{prob\min }}\right)R_0---\left(2\right)$

Wherein d_maxAnd d_minBe respectively maximum and the minimum of a value of ordinary node to the distance of aggregation node, d (i, BS) is commonNode i is to the distance of aggregation node, CH_probmaxAnd CH_probminBe respectively ordinary node become leader cluster node maximum probability andMinimum probability, CH_prob(i) become the probability of leader cluster node for ordinary node i, R₀For ordinary node greatest irradiation radius, ω and cIt is the parameter for controlling span;

Step S206: each ordinary node judges that it becomes the probability CH of leader cluster node_probRandom between whether being more than or equal to 0～1The number generating, if so, enters step S209, otherwise goes to step S210;

Step S207: the ordinary node that becomes candidate cluster head node judges that it becomes the probability CH of leader cluster node_probWhether be greater than etc.In 1, if so, enter step S205, otherwise go to step S209;

Step S208: for alternative leader cluster node set for empty and its oneself be not the ordinary node of candidate cluster head node, shouldOrdinary node adds in its alternative leader cluster node set minimum average B configuration can reach the candidate cluster head node of power AMRP minimum orUnder whole leader cluster node bunch in, and enter step S211;

Step S209: this ordinary node becomes candidate cluster head node, upgrading this ordinary node state is candidate cluster head node, will be fromOneself adds its alternative leader cluster node set, and calculates its bunch of radius R according to formula (2)_i, simultaneously to ordinary node broadcast in its bunchOneself be elected as the message of candidate cluster head node, this message comprises No. ID of this candidate cluster head node self, node state, minimumOn average can reach power AMRP and bunch radius; When after the complete message of candidate cluster head node broadcasts, in its bunch of radius, all ordinary nodes moreNew its alternative leader cluster node set, adds this candidate cluster head node in alternative leader cluster node set, and performs step S210;

Step S210: ordinary node or candidate cluster head node are by the CH of self_probBe multiplied by 2 and enter step S211;

Step S211: judge iterations whether be greater than M time, M be maximum iteration time andIfEnd iteration and enter step S212, otherwise iterations to add 1, and go to step S202;

Step S212: cluster stops, and each node determines its end-state; If ordinary node has become finally in the time of iterative processLeader cluster node, it is elected as leader cluster node; If there is final leader cluster node in the alternative leader cluster node set of ordinary node,It adds in minimum average B configuration in alternative leader cluster node set can reach under the final leader cluster node of power AMRP minimum bunch; If certain is generalLogical node is isolated node, in alternative leader cluster node set, does not have final leader cluster node, adds distance in its communication radiusIn under nearest cluster node bunch; If this isolated node communication radius is interior without other nodes, the independent cluster of isolated node;

3) stage of communication bunch

Adopt bunch in single-hop and bunch between multi-hop data transmission means, each leader cluster node need to select from the leader cluster node of adjacent clusterSelect a leader cluster node as its via node, forwarding data is to aggregation node; Little apart from the distance of aggregation node for bunch headIn bunch head of T, all directly single-hop transmission is to aggregation node, and wherein T is a predefined value, and its value is always less than common jointPoint greatest irradiation radius R₀, be greater than bunch head of T apart from the distance of aggregation node for bunch head, from the set of adjacent cluster head node, selectSelect bunch head of a Least-cost as its via node, through multi-hop, complete transfer of data to aggregation node place; Detailed processAs follows:

Step S301: leader cluster node calculates its distance apart from aggregation node, and judge that whether distance is less than T, if so, entersStep S302, otherwise go to step S303;

Step S302: leader cluster node is directly transported to aggregation node by data sheet jump set;

Step S303: leader cluster node c_iWith doubly bunch radius broadcast of its β, wherein between the general value 1～1.5 of β and its broadcast halfFootpath is less than ordinary node greatest irradiation radius R₀; The message of leader cluster node broadcast comprises leader cluster node ID, dump energy, bunch cephalomerePut the node cost of self and the distance to aggregation node, node cost is calculated according to formula (3); Receive other bunches of messageHead node also sends return messages to the leader cluster node c that sends message_i, return messages comprise its node cost and to convergingThe distance of node; Leader cluster node c_iCalculate adjacent cluster head node set c according to a bunch number of sending return messages_i.R_CH, bunch headc_iAdjacent cluster head node sets definition be: c_i.R_CH＝{c_j|d(c_i,c_j)≤βR_i,d(c_j,BS)＜d(c_i, BS) }; Wherein d(c_i,c_j) refer to leader cluster node c_iWith c_jDistance to each other, d (c_j, BS) and refer to leader cluster node c_jTo the distance of aggregation node, d(c_i, BS) and refer to leader cluster node c_iTo the distance of aggregation node;

Node cost computing formula is: $\cos t\left(i\right)=\mathrm{\μ}\frac{\stackrel{\‾}{d_{c-BS}}}{d_{ci-BS}}+(1-\mathrm{\μ})\frac{E_{ci}}{\stackrel{\‾}{E_c}}---\left(3\right)$

WhereinFor the leader cluster node in this leader cluster node and its adjacent cluster head node set to aggregation node distance on averageValue, d_ci-BSFor leader cluster node c_iTo the distance of aggregation node,For bunch head in leader cluster node and its adjacent cluster head node setThe mean value of residue energy of node, E_ciFor leader cluster node c_iDump energy, μ is the parameter for controlling span;

Step S304: leader cluster node c_iThe leader cluster node of selecting Least-cost from its adjacent cluster head node set as it nextHop node.

2. a method for supervising for aquatic farm water quality monitoring system as claimed in claim 1, is characterized in that, comprise withLower step:

1) according to the record of the water quality parameter to aquatic farm, count shallow waters temperature, water-soluble oxygen amount, pH value, airFlow, the data of the deep water waters temperature in corresponding moment, water-soluble oxygen amount, pH value; By shallow waters temperature, water-soluble oxygen amount, pH value,Air mass flow, as input parameter, using deep water waters temperature, water-soluble oxygen amount, pH value as output parameter, is set up neutral net, rootAccording to existing Historical Monitoring data, use BP neutral net, additional momentum learning rules, neural network training;

2) according to water biological species in the suitable temperature of deep water Watershed expand, water-soluble oxygen amount, pH value ideal value, by particle cluster algorithm,Solve optimum input parameter, i.e. shallow waters temperature, water-soluble oxygen amount, pH value, the air mass flow of neutral net;

3) judge whether to need artificial sample according to a upper Recognition with Recurrent Neural Network evaluated error, as needs, right by artificial sampleRear off-line analysis, contrast draws the deep water waters that deep water waters temperature, water-soluble oxygen amount, pH value and the neutral net of actual measurement estimateThe error of temperature, water-soluble oxygen amount, pH value, then by deep water waters temperature, water-soluble oxygen amount, the pH value data of this group actual measurement, andNeutral net is estimated, together with the error information of surveying, to use additional momentum learning rules, upgrades neural network training; If do not neededWant artificial sample, return to step 2).

3. the method for supervising of aquatic farm water quality monitoring system as claimed in claim 2, is characterized in that, described populationAlgorithm, step is as follows:

1) initialize population: determine population size NP, particle cluster algorithm iterations NG, initializes particle position, calculatesThe fitness of each particle also initializes globally optimal solution and individual optimal solution;

The function that calculates particle fitness is:

$Fitness=\underset{i}{\Σ}{(O_i-O_i^{\′})}^2$

Wherein, O_iRepresent i element of neutral net output vector, O '_iFor i element of the output vector of theoretical expectation;

2) upgrade population: the equation of motion of population is as follows:

v(t)＝ω·v(t-1)+c₁·(lbest-x(t))+c₂·(gbest-x(t))

x(t+1)＝x(t)+c₃·v(t)

Wherein ω is taken asI is this iterations of particle cluster algorithm, c₁,c₂,c₃For constant, c₁,c₂Value is2.8，c₃Value is that 0.3, lbest is the individual optimal solution that each particle search is crossed, and gbest is the overall situation that all particle search are crossedOptimal solution;

3) calculate the particle fitness of this iteration, upgrade individual optimal solution and globally optimal solution: to each particle, by thisThe fitness that iteration produces, compared with current individual optimal solution, get fitness less for individual optimal solution, search with all particlesThe globally optimal solution that rope is crossed is compared, get fitness less for globally optimal solution;

4) judge whether to reach iteration NG time, if so, export globally optimal solution, if not, return to step 2).

4. the method for supervising of aquatic farm water quality monitoring system as claimed in claim 2, is characterized in that, described additional movingAmount learning method, update rule as shown in the formula:

$\mathrm{\ω}(t+1)=\mathrm{\ω}\left(t\right)-(1-a)\mathrm{\η}\frac{\∂E_T}{\∂\mathrm{\ω}\left(t\right)}+a\mathrm{\Δ}\mathrm{\ω}\left(t\right)$

Wherein Δ ω (t)=ω (t)-ω (t-1), E_TFor the training error of neutral net, η is weight, and a is factor of momentum, gets0.9。

5. a method for supervising for aquatic farm water quality monitoring system as claimed in claim 1, is characterized in that, comprise withLower step:

1) according to the record of the water quality parameter to aquatic farm, count shallow waters temperature, water-soluble oxygen amount, pH value, airFlow, the data of the deep water waters temperature in corresponding moment, water-soluble oxygen amount, pH value; By shallow waters temperature, water-soluble oxygen amount, pH value,Air mass flow, as input parameter, using deep water waters temperature, water-soluble oxygen amount, pH value as output parameter, is set up neutral net, rootAccording to existing Historical Monitoring data, use BP neutral net, additional momentum learning rules, neural network training;

2) according to water biological species in the suitable temperature of deep water Watershed expand, water-soluble oxygen amount, pH value ideal value, by genetic algorithm, askSeparate optimum input parameter, i.e. shallow waters temperature, water-soluble oxygen amount, pH value, the air mass flow of neutral net;

Described genetic algorithm comprises the following steps:

1. adopt real coding, initialize chromosome, form initial population;

2. utilize fitness function to evaluate the each chromosome in each generation;

3. carry out genetic manipulation;

4. recalculate the adaptive value of each individuality;

5. choose after new population, the optimum individual in new population is retained, replace this generation with the previous generation's optimum individualThe poorest individuality;

6. judge whether to reach evolutionary generation, if do not have, return to the 2. step, otherwise finish;

7. using the value of the optimum individual in new population as with, remain unchanged, adopt BP algorithm to learn, until meet performanceIndex;

3) judge whether to need artificial sample according to a upper Recognition with Recurrent Neural Network evaluated error, as needs, right by artificial sampleRear off-line analysis, contrast draws the deep water waters that deep water waters temperature, water-soluble oxygen amount, pH value and the neutral net of actual measurement estimateThe error of temperature, water-soluble oxygen amount, pH value, then by deep water waters temperature, water-soluble oxygen amount, the pH value data of this group actual measurement, andNeutral net is estimated, together with the error information of surveying, to use additional momentum learning rules, upgrades neural network training; If do not neededWant artificial sample, return to step 2).