CN107860099A

CN107860099A - Frosting detection method, device, storage medium and the equipment of a kind of air-conditioning

Info

Publication number: CN107860099A
Application number: CN201710829628.6A
Authority: CN
Inventors: 刘佰兰; 黄辉; 宋德超; 陈翀; 田涛
Original assignee: Gree Electric Appliances Inc of Zhuhai
Current assignee: Gree Electric Appliances Inc of Zhuhai
Priority date: 2017-09-15
Filing date: 2017-09-15
Publication date: 2018-03-30
Anticipated expiration: 2037-09-15
Also published as: CN107860099B

Abstract

The invention discloses frosting detection method, device, storage medium and the equipment of a kind of air-conditioning, this method includes：Using the self-learning capability of artificial neural network, the corresponding relation between the operational factor of air-conditioning and the frosting state of condenser under heating mode is established；Obtain the current operating parameter of air-conditioning；By the corresponding relation, it is determined that current frosting state corresponding with the current operating parameter.The solution of the present invention, can overcome in the prior art that frosting judgment accuracy is poor, heating effect difference and the defects of poor user experience, realize the beneficial effect that frosting judgment accuracy is good, heating effect is good and Consumer's Experience is good.

Description

Frosting detection method, device, storage medium and the equipment of a kind of air-conditioning

Technical field

The invention belongs to defrost technical field, and in particular to the frosting detection method of air-conditioning a kind of, device, storage medium and Equipment, more particularly to a kind of method and this method that air-conditioning condenser frosting situation is identified using artificial neural network algorithm Corresponding device, it is stored with the computer-reader form storage medium of instruction corresponding to this method and can loads and perform The equipment of instruction corresponding to this method.

Background technology

Air-conditioning (i.e. air regulator), can refer to manually means, to the temperature of surrounding air in building/structures, The parameters such as humidity, cleanliness factor, speed are adjusted and controlled.After air-conditioning runs a period of time under heating state, outdoor unit cold The situation of frosting occurs in condenser, influences heating effect, therefore to carry out defrosting operation to condenser.

But the defrosting control strategy of the air-conditioning at present in industry, be all based on certain experience set in advance rule and when Between, run in air-conditioning heating when completely the temperature spot of a period of time and condenser tubes reaches to a certain degree and open defrosting.But This method is limited to the accuracy of judgement degree of frosting situation, does not obviously start frosting also sometimes and has begun to defrosting, and sometimes white Very thick still not actuated defrosting has been tied to obtain, has failed to obtain preferable defrosting effect.

In the prior art, have that frosting judgment accuracy is poor, heating effect difference and the defects of poor user experience.

The content of the invention

It is an object of the present invention to it is directed to drawbacks described above, there is provided a kind of frosting detection method of air-conditioning, device, storage are situated between Matter and equipment, it is limited to the accuracy of judgement degree of frosting situation in the prior art to solve the problems, such as, reach frosting judgment accuracy Good effect.

The present invention provides a kind of frosting detection method of air-conditioning, including：Using the self-learning capability of artificial neural network, build Corresponding relation under vertical heating mode between the operational factor of air-conditioning and the frosting state of condenser；Obtain the current operation of air-conditioning Parameter；By the corresponding relation, it is determined that current frosting state corresponding with the current operating parameter.

Alternatively, wherein, the operational factor, including：Ambient parameter and/or running parameter, and/or by by setting rule The array more than one-dimensional or bidimensional of the feature composition extracted from the ambient parameter, the running parameter；Wherein, the ring Border parameter, including：At least one of indoor and outdoor surroundingses temperature, user's shell temperature, indoor and outdoor surroundingses humidity；It is and/or described Running parameter, including：In design temperature, wind speed setting, setting wind shelves, condenser temperature, outdoor fan running current at least One of；And/or the corresponding relation, including：Functional relation；The operational factor be the functional relation input parameter, institute State the output parameter that frosting state is the functional relation；It is determined that current frosting state corresponding with the current operating parameter, Including：By frosting state corresponding with the current operating parameter identical operational factor in the corresponding relation, it is defined as working as Preceding frosting state；And/or when the corresponding relation includes functional relation, the current operating parameter is inputted into the function In relation, the output parameter for determining the functional relation is current frosting state.

Alternatively, the corresponding relation between the operational factor of air-conditioning and the frosting state of condenser under heating mode is established, Including：Obtain the sample data for establishing the corresponding relation between the operational factor and the frosting state；Described in analysis The characteristic and its rule of operational factor, according to the characteristic and its rule, determine the artificial neural network network structure and Its network parameter；Using the sample data, the network structure and the network parameter are trained and tested, determine institute State the corresponding relation of operational factor and the frosting state.

Alternatively, the sample number for establishing the corresponding relation between the operational factor and the frosting state is obtained According to, including：Collect service data and frosting state of the air-conditioning under different use environments；The service data is analyzed, And the expertise information to prestore is combined, the data related to the frosting state are chosen as the operational factor；By described in The data pair that frosting state and the operational factor chosen are formed, as sample data.

Alternatively, service data of the air-conditioning under different use environments is collected, including：Air-conditioning is obtained in laboratory simulation ring The operational factor under border, and/or, obtain the operation collected by technology of Internet of things when user actually uses air-conditioning and join Number；And/or the network structure, including：In BP neural network, CNN neutral nets, RNN neutral nets, residual error neutral net At least one of；And/or the network parameter, including：It is input number of nodes, output node number, hidden layer number, Hidden nodes, initial At least one of weights, bias.

Alternatively, the network structure and the network parameter are trained, including：Choose in the sample data The operational factor in the training sample is input to the network structure, passed through by a part of data as training sample The activation primitive of the network structure and the network parameter are trained, and obtain hands-on result；Determine the actual instruction Practice whether the hands-on error between result frosting state corresponding in the training sample meets to set training error；When When the hands-on error meets the setting training error, it is determined that to described in the network structure and the network parameter Training is completed；And/or the network structure and the network parameter are tested, including：Choose in the sample data The operational factor in the test sample is input to the institute for training and completing by another part data as test sample State in network structure, tested with the network parameter that the activation primitive and the training are completed, obtain actual test As a result；Determine whether is actual test error between actual test result frosting state corresponding in the test sample Meet setting test error；When the actual test error meets the setting test error, it is determined that to the network structure The test with the network parameter is completed.

Alternatively, the network structure and the network parameter are trained, in addition to：When the hands-on error When being unsatisfactory for the setting training error, the network parameter is updated by the error energy function of the network structure；Pass through The network parameter after the activation primitive of the network structure and renewal carries out re -training, until the re -training Hands-on error afterwards meets the setting training error；And/or the network structure and the network parameter are surveyed Examination, in addition to：When the actual test error is unsatisfactory for the setting test error, to the network structure and the network Parameter carries out re -training, until the actual test error setting test error at a slow speed after the re -training.

Alternatively, in addition to：Determine whether the current frosting state reaches the defrosting degree of setting；When the current knot When white state reaches the defrosting degree, into the defrosting mode of setting, and/or, initiate the current frosting state and reach institute State the prompting of defrosting degree；And/or according to the current frosting state, after needing to enter the defrosting mode of setting, determine institute State entry time and/or the post-set time of defrosting mode；And/or to the present mode of operation of the air-conditioning, the current operation Parameter, the current frosting state, the entry time of defrosting mode, at least one of post-set time of defrosting mode, carry out Display and/or output；And/or receive the result that the current frosting state is not inconsistent with actual frosting state, and/or When in the corresponding relation not with the current operating parameter identical operational factor, the corresponding relation is updated, Amendment, at least one of learn attended operation again.

Matching with the above method, another aspect of the present invention provides a kind of frosting detection device of air-conditioning, including：Establish single Member, for the self-learning capability using artificial neural network, establish the knot of the operational factor of air-conditioning and condenser under heating mode Corresponding relation between white state；

Acquiring unit, for obtaining the current operating parameter of air-conditioning；

Determining unit, for by the corresponding relation, it is determined that current frosting shape corresponding with the current operating parameter State.

Alternatively, wherein, the operational factor, including：Ambient parameter and/or running parameter, and/or by by setting rule The array more than one-dimensional or bidimensional of the feature composition extracted from the ambient parameter, the running parameter；Wherein, the ring Border parameter, including：At least one of indoor and outdoor surroundingses temperature, user's shell temperature, indoor and outdoor surroundingses humidity；It is and/or described Running parameter, including：In design temperature, wind speed setting, setting wind shelves, condenser temperature, outdoor fan running current at least One of；And/or the corresponding relation, including：Functional relation；The operational factor be the functional relation input parameter, institute State the output parameter that frosting state is the functional relation；The determining unit determines work as corresponding with the current operating parameter Preceding frosting state, is specifically included：By knot corresponding with the current operating parameter identical operational factor in the corresponding relation White state, it is defined as current frosting state；And/or when the corresponding relation includes functional relation, by the current operation ginseng Number is inputted in the functional relation, and the output parameter for determining the functional relation is current frosting state.

Alternatively, the unit of establishing is established under heating mode between the operational factor of air-conditioning and the frosting state of condenser Corresponding relation, specifically include：Obtain the sample for establishing the corresponding relation between the operational factor and the frosting state Notebook data；The characteristic and its rule of the operational factor are analyzed, according to the characteristic and its rule, determines the ANN The network structure and its network parameter of network；Using the sample data, the network structure and the network parameter are instructed Practice and test, determine the corresponding relation of the operational factor and the frosting state.

Alternatively, the unit of establishing is obtained for establishing the corresponding pass between the operational factor and the frosting state The sample data of system, is specifically included：Collect service data and frosting state of the air-conditioning under different use environments；To the operation The expertise information to prestore is analyzed and combined to data, chooses the data related to the frosting state as the fortune Row parameter；The data pair that the frosting state and the operational factor chosen are formed, as sample data.

Alternatively, the unit of establishing collects service data of the air-conditioning under different use environments, specifically includes：Obtain empty The operational factor under laboratory simulation environment is adjusted, and/or, obtain and user's actual use sky is collected by technology of Internet of things The operational factor of timing；And/or the network structure, including：BP neural network, CNN neutral nets, RNN nerve nets At least one of network, residual error neutral net；And/or the network parameter, including：It is input number of nodes, output node number, hidden At least one of the number of plies, Hidden nodes, initial weight, bias.

Alternatively, the unit of establishing is trained to the network structure and the network parameter, is specifically included：Choose The operational factor in the training sample is input to institute by a part of data in the sample data as training sample Network structure is stated, is trained by the activation primitive and the network parameter of the network structure, obtains hands-on result； Determine whether the hands-on error between hands-on result frosting state corresponding in the training sample meets Set training error；When the hands-on error meets the setting training error, it is determined that to the network structure and institute The training for stating network parameter is completed；And/or the unit of establishing is surveyed to the network structure and the network parameter Examination, is specifically included：Another part data in the sample data are chosen as test sample, by the institute in the test sample State operational factor to be input in the network structure that the training is completed, the institute completed with the activation primitive and the training State network parameter to be tested, obtain actual test result；Determine the actual test result and the phase in the test sample Answer whether the actual test error between frosting state meets to set test error；Set described in meeting when the actual test error When determining test error, it is determined that the test to the network structure and the network parameter is completed.

Alternatively, the unit of establishing is trained to the network structure and the network parameter, is specifically also included：When When the hands-on error is unsatisfactory for the setting training error, institute is updated by the error energy function of the network structure State network parameter；Re -training is carried out by the network parameter after the activation primitive of the network structure and renewal, Until the hands-on error after the re -training meets the setting training error；And/or the unit of establishing is to described Network structure and the network parameter are tested, and are specifically also included：Surveyed when the actual test error is unsatisfactory for the setting When trying error, re -training is carried out to the network structure and the network parameter, until the actual survey after the re -training Try the error setting test error at a slow speed.

Alternatively, in addition to：The determining unit, it is additionally operable to determine whether the current frosting state reaches removing for setting White degree；When the current frosting state reaches the defrosting degree, into the defrosting mode of setting, and/or, described in initiation Current frosting state reaches the prompting of the defrosting degree；And/or the determining unit, it is additionally operable to according to the current frosting State, after needing to enter the defrosting mode of setting, determine entry time and/or the post-set time of the defrosting mode；With/ Or, the determining unit, it is additionally operable to the present mode of operation to the air-conditioning, the current operating parameter, the current frosting State, the entry time of defrosting mode, at least one of post-set time of defrosting mode, shown and/or exported；With/ Or, the determining unit, be additionally operable to receive the result that the current frosting state is not inconsistent with actual frosting state and/ Or when in the corresponding relation not with the current operating parameter identical operational factor, unit is established by described, to institute State corresponding relation and at least one of be updated, correct, learning again attended operation.

Matching with the above method, further aspect of the present invention provides a kind of storage medium, including：Deposited in the storage medium Contain a plurality of instruction；Wherein, a plurality of instruction, the frosting for being loaded by processor and being performed above-described air-conditioning detect Method.

Matching with the above method or device, further aspect of the present invention provides a kind of equipment, including：Processor, for holding The a plurality of instruction of row；Memory, for storing a plurality of instruction；Wherein, a plurality of instruction, for by the memory storage, and Loaded by processor and perform the frosting detection method of above-described air-conditioning；Or including：The frosting of above-described air-conditioning Detection means.

Alternatively, the equipment includes：For the air conditioner main body being controlled to itself, and/or, for entering to air conditioner main body The outside control terminal of row control；Wherein, the outside control terminal, including：In wireless communication module, router, server, terminal At least one of.

The solution of the present invention, by using the self-learning function of neutral net, establish the operational factor and condenser of air-conditioning Frosting state between corresponding relation；According to the current operating parameter of air-conditioning, can determine currently to tie by the corresponding relation White state, determination mode is reliable, determines that result accuracy is good.

Further, the solution of the present invention, by using the self-learning function of neutral net, by entering to the data collected Row training and study, grasp the corresponding relation letter between the operational factor and sensor parameters and condenser frosting state of air-conditioning Number, so as to which Real time identification goes out the condenser frosting state of air-conditioning, the accuracy of defrost detection is improved, it is whole to be beneficial to raising air-conditioning The heating effect of machine, heating efficiency is improved, improve the comfort of user.

Further, the solution of the present invention, the frosting of air-conditioning condenser is analyzed by using artificial neural network algorithm The running status rule of journey, condenser in air-conditioning heating operation is found by the self study of artificial neural network, adaptive characteristic Mapping principle between frosting situation and air conditioner operation parameters state, the current frosting situation of air-conditioning can be efficiently identified out, so as to Start for defrosting with terminating to provide accurate basis for estimation, and judged result accuracy is good.

Further, the solution of the present invention, by the corresponding relation between operational factor and frosting state, it is determined that current operation Current frosting state corresponding to parameter, mode of operation is easy, and operating result reliability is high.

Thus, the solution of the present invention, the frosting of condenser in air-conditioning heating operation is determined by using neural network algorithm Corresponding relation between state and the operational factor of air-conditioning, and then the current of condenser is determined according to the current operating parameter of air-conditioning Frosting state；Solve the problems, such as it is limited to the accuracy of judgement degree of frosting situation in the prior art, so as to, overcome and tie in the prior art White judgment accuracy is poor, heating effect difference and the defects of poor user experience, realize that frosting judgment accuracy is good, heating effect is good and The good beneficial effect of Consumer's Experience.

Other features and advantages of the present invention will be illustrated in the following description, also, partly becomes from specification Obtain it is clear that or being understood by implementing the present invention.

Below by drawings and examples, technical scheme is described in further detail.

Brief description of the drawings

Fig. 1 is the schematic flow sheet of an embodiment of the frosting detection method of the air-conditioning of the present invention；

Fig. 2 is to be established in the method for the present invention under heating mode between the operational factor of air-conditioning and the frosting state of condenser Corresponding relation an embodiment schematic flow sheet；

Fig. 3 is to be obtained in the method for the present invention for establishing the corresponding pass between the operational factor and the frosting state The schematic flow sheet of one embodiment of the sample data of system；

Fig. 4 is the stream for the embodiment being trained in the method for the present invention to the network structure and the network parameter Journey schematic diagram；

Fig. 5 is the stream for the embodiment tested in the method for the present invention the network structure and the network parameter Journey schematic diagram；

Fig. 6 is the embodiment for carrying out re -training in the method for the present invention to the network structure and the network parameter Schematic flow sheet；

Fig. 7 is to judge whether current frosting state reaches the flow of an embodiment of defrosting degree and show in the method for the present invention It is intended to；

Fig. 8 is that the implementation whether current frosting state is consistent with actual frosting state is verified in the method for the present invention The schematic flow sheet of example；

Fig. 9 is the structural representation of an embodiment of the frosting detection device of the air-conditioning of the present invention；

Figure 10 be the present invention method in artificial neural network algorithm structure one (such as：BP neural network) an embodiment Structural representation；

Figure 11 be the present invention method in artificial neural network algorithm network structure two (such as：CNN convolutional neural networks) An embodiment structural representation；

Figure 12 be the present invention method in artificial neural network algorithm network structure three (such as：Residual error neutral net) one The structural representation of embodiment；

Figure 13 is the structural representation of an embodiment of artificial neural network algorithm network structure four in method of the invention；

Figure 14 be the present invention equipment in run artificial neural network algorithm intelligent apparatus (such as：Terminal) an implementation The principle schematic of example.

With reference to accompanying drawing, reference is as follows in the embodiment of the present invention：

102- establishes unit；104- acquiring units；106- determining units.

Embodiment

To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with the specific embodiment of the invention and Technical solution of the present invention is clearly and completely described corresponding accompanying drawing.Obviously, described embodiment is only the present invention one Section Example, rather than whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art are not doing Go out under the premise of creative work the every other embodiment obtained, belong to the scope of protection of the invention.

According to an embodiment of the invention, there is provided the method for a kind of frosting detection method of air-conditioning, the as shown in Figure 1 present invention An embodiment schematic flow sheet.The frosting detection method of the air-conditioning can include：

At step S110, using the self-learning capability of artificial neural network, the operation for establishing air-conditioning under heating mode is joined Corresponding relation between number and the frosting state of condenser.

Such as：Using the self-learning function of neutral net, by the way that the data collected are trained and learnt, grasp Corresponding relation function between the operational factor and sensor parameters and condenser frosting state of air-conditioning.

Wherein, sensor parameters can refer to detect Temperature Humidity Sensor of external environment situation etc., as operation ginseng A several parts.

Such as：The running status rule of the Frost formation process of air-conditioning condenser is analyzed using artificial neural network algorithm, is led to Cross the self study of artificial neural network, adaptive characteristic finds condenser frosting situation and operation of air conditioner in air-conditioning heating operation and joined Mapping principle between number state.

Such as：Artificial neural network algorithm can be utilized, by (including but is not limited to such as under a large amount of different use environments Under one or more：Indoor and outdoor surroundingses temperature and humidity, design temperature etc.) collection of operation of air conditioner data summarization, choose some The running state parameter and frosting state parameter of air-conditioning are learnt and trained to neutral net, pass through tune as sample data Weights between whole network structure and network node, make the relation between neutral net fitting air conditioner operation parameters and frosting state, Finally neutral net is set accurately to fit the corresponding relation between the operational factor of air-conditioning and condenser frosting state.

In an optional example, the operational factor, it can include：Ambient parameter and/or running parameter, and/or by The array more than one-dimensional or bidimensional that the feature extracted by setting rule from the ambient parameter, the running parameter forms.

Alternatively, the ambient parameter, can include：Indoor and outdoor surroundingses temperature, user's shell temperature, indoor and outdoor surroundingses are wet At least one of degree.

Alternatively, the running parameter, can include：Design temperature, wind speed setting, setting wind shelves, condenser temperature, room At least one of outer blower fan running current.

Such as：Input parameter includes but is not limited to following one or more：Environment temperature, the air-conditioning setting temperature of indoor and outdoor Degree, wind speed setting (shelves), condenser temperature, outdoor fan running current etc..Input parameter is not only single parameter, includes pressing yet The input parameter one or more dimensions array of certain rule extraction feature composition.

Thus, by the operational factor of diversified forms, lifting is advantageous to corresponding between operational factor and frosting state The accuracy and reliability that relation determines.

In an optional example, the corresponding relation, it can include：Functional relation.

Alternatively, the operational factor is the input parameter of the functional relation, and the frosting state is closed for the function The output parameter of system.

Thus, by the corresponding relation of diversified forms, the flexibility that is determined to current frosting state and convenient can be lifted Property.

In an optional example, the fortune of air-conditioning under heating mode is established in method that can be of the invention with reference to shown in Fig. 2 The schematic flow sheet of one embodiment of the corresponding relation between row parameter and the frosting state of condenser, further illustrate step The detailed process of the corresponding relation under heating mode between the operational factor of air-conditioning and the frosting state of condenser is established in S110.

Step S210, obtain the sample for the corresponding relation that can be used for establishing between the operational factor and the frosting state Notebook data.

It is alternatively possible to obtained in method of the invention with reference to shown in Fig. 3 for establishing the operational factor and the knot The schematic flow sheet of one embodiment of the sample data of the corresponding relation between white state, further illustrate and obtained in step S210 It can be used for the detailed process of the sample data for the corresponding relation established between the operational factor and the frosting state.

Step S310, collect service data and frosting state of the air-conditioning under different use environments.

Service data of the air-conditioning under different use environments is collected in an optional specific example, in step S310, can With including：Obtain the operational factor of the air-conditioning under laboratory simulation environment.

Service data of the air-conditioning under different use environments is collected in an optional specific example, in step S310, also It can include：Obtain operational factor when user's actual use air-conditioning is collected by technology of Internet of things.

Such as：Data collection：Collect itself running state parameter and corresponding knot of the air-conditioning under different use environments White situation.Specific collection mode includes but is not limited to operational factor of the air-conditioning under laboratory simulation environment, by Internet of Things skill Art collects the modes such as air conditioner operation parameters when actual user uses.

Thus, service data is collected by number of ways, is advantageous to increase the amount of service data, lifts artificial neural network Learning ability, and then lifted determine corresponding relation accuracy and reliability.

Step S320, the expertise information to prestore is analyzed the service data and combined, chosen and the knot The related data of white state as the operational factor (such as：Choose on the influential parameter of air-conditioning frosting state as input Parameter, using frosting state as output parameter).

Such as：By the analysis to data and expertise knowledge is combined, is chosen on the influential ginseng of air-conditioning frosting state Number is used as input parameter, using frosting state as output parameter.

Step S330, the data pair that the frosting state and the operational factor chosen are formed, as sample number According to.

Such as：By obtained input, output parameter pair, a part is used as training book sample data, and a part is used as test specimens Notebook data.

Thus, by the way that the service data being collected into is analyzed and handled, operational factor is determined, and then obtain sample number According to operating process is simple, and operating result reliability is high.

Step S220, the characteristic and its rule of the operational factor are analyzed, according to the characteristic and its rule, it is determined that described The network structure and its network parameter of artificial neural network.

Such as：The data characteristic that is influenceed according to different design temperatures, environment temperature, shell temperature etc. on operation of air conditioner and Its rule contained, the basic structure of network, the input of network, output node number, network hidden layer number, hidden section can be primarily determined that Points, network initial weight etc..

Alternatively, the network structure, can include：BP neural network, CNN neutral nets, RNN neutral nets, residual error At least one of neutral net.

Alternatively, the network parameter, can include：Input number of nodes, output node number, hidden layer number, Hidden nodes, just At least one of beginning weights, bias.

Such as：BP neural network structural representation is as shown in Figure 10, during practical application, can adjust input as needed Layer, hidden layer, output layer nodes and hidden layers numbers.

Such as：Multilayer convolutional network, exactly constantly extract merging from the feature of low dimensional and obtain the feature of more higher-dimension, so as to It can be used for being classified or inter-related task.Such as：Multilayer convolutional network structural representation is as shown in figure 11, net during practical application Network structure can adjust according to actual conditions.

Such as：Residual error neural network structure schematic diagram is as shown in figure 12, and network structure can be according to actual feelings during practical application Condition adjusts.

Such as：When debugging CNN networks, deepen the network number of plies and change the method for convolution kernel size, net can not be caused Network performance gets a promotion.Front and rear data can preferably be connected by adding residual block, strengthen feature representation ability, so it can add The learning ability of strong convolutional network.As shown in figure 13, the input of certain section of neutral net is x, and desired output is H (x), input x Be passed to output as after initial configuration, it is necessary to learn target be just changed into F (x)=H (x)-x.

Such as：Artificial neural network used in this programme, a certain network structure is not limited to, can be classical Artificial neural network such as BP neural network or high-level manual's neutral net, or deep learning network such as CNN (convolution Neutral net) and RNN (Recognition with Recurrent Neural Network) etc., concrete scheme can select according to practical application scene.

Thus, the artificial neural network structure by diversified forms and network parameter, it can be lifted and network structure is selected Flexibility, it is also possible that between operational factor and frosting state corresponding relation determine convenience and reliability.

Step S230, using the sample data, the network structure and the network parameter are trained and tested, Determine the corresponding relation of the operational factor and the frosting state.

Such as：By the way that the sample data is trained and tested, training result and test result are obtained.Described in selection Training result and the test result reach the sample data of sets requirement, according to determining the sample data of selection The corresponding relation of operational factor and the frosting state.

Such as：After the completion of network design, training sample data need to be used, network is trained.Training method can be according to reality The problem of being found in the network structure on border and training is adjusted.

Thus, by collecting service data, selection sample data, and it is trained and tests based on sample data, it is determined that Corresponding relation between operational factor and frosting state, be advantageous to convenience and accuracy of the lifting to frosting condition adjudgement.

It is alternatively possible to the network structure and the network parameter are carried out in method of the invention with reference to shown in Fig. 4 The schematic flow sheet of one embodiment of training, further illustrate and the network structure and the network parameter are entered in step S230 The detailed process of row training.

Step S410, a part of data in the sample data is chosen as training sample, by the training sample The operational factor be input to the network structure, pass through the activation primitive of the network structure and the network parameter and carry out Training, obtains hands-on result.

Such as：(1) input data x is imported, according to activation primitive, the weights of initialization and biasing, calculates the reality of network Export a (x), i.e. a (x)=1/ (1+e^-z), wherein Z=W_k*x+b_l。

Step S420, determine the reality between hands-on result frosting state corresponding in the training sample Whether training error meets to set training error.Such as：Expectation is used as using the corresponding frosting state in the sample data Training result.

Such as：Judge whether the desired output y (x) and reality output a (x) of network meet output accuracy requirement, i.e.,：‖y (x)-a(x)‖<∈, ∈ are target minimal error.

Step S430, when the hands-on error meets the setting training error, it is determined that to the network structure The training with the network parameter is completed.

Such as：Terminate to train if the desired output y (x) of network meets required precision with reality output a (x).

Thus, by the way that training sample is trained for selected network structure and network parameter, can obtain more Reliable network structure and network parameter, be advantageous to the essence that lifting determines to the corresponding relation between operational factor and frosting state Parasexuality and reliability.

More it is alternatively possible to enter in method of the invention with reference to shown in Fig. 6 to the network structure and the network parameter The schematic flow sheet of one embodiment of row re -training, further illustrate in step S230 to the network structure and the network The detailed process of re -training when parameter is trained.

Step S610, when the hands-on error is unsatisfactory for the setting training error, pass through the network structure Error energy function update the network parameter.

Step S620, carried out again by the network parameter after the activation primitive of the network structure and renewal Training, until the hands-on error after the re -training meets the setting training error.

Such as：Network is updated according in the following manner if the desired output y (x) and reality output a (x) of network are unsatisfactory for Weights W_k, bias b_l：

C (w, b) is error energy function (by taking standard variance function as an example), and n is the total quantity of training sample, summation be Carried out on total training sample x：

Update each layer weights：

Update each layer biasing：

Wherein：

W_kFor initial weight,For the partial derivative of error energy function pair weights, b_lFor initial bias,For error energy Partial derivative of the flow function to biasing；Value can be obtained by chain type Rule for derivation, until network output error it is small Untill ∈ (target minimal error).

Thus, by re -training after being modified network parameter when training error is larger, be advantageous to obtain more Accurate and reliable network structure, and then obtain more accurate and reliable corresponding relation.

It is alternatively possible to the network structure and the network parameter are carried out in method of the invention with reference to shown in Fig. 5 The schematic flow sheet of one embodiment of test, further illustrate and the network structure and the network parameter are entered in step S230 The detailed process of row test.

Step S510, another part data in the sample data is chosen as test sample, by the test sample In the operational factor be input in the network structure that the training is completed, with the activation primitive and described trained Into the network parameter tested, obtain actual test result.

Such as：After the completion of network training, then with test sample positive test network.

Step S520, determine actual test result frosting state corresponding in the test sample (i.e. with described The frosting state in sample data is as desired output parameter) between actual test error whether meet setting test miss Difference.

Step S530, when the actual test error meets the setting test error, it is determined that to the network structure The test with the network parameter is completed.

Such as：If test error meets to require, network training test is completed.

Thus, by being used to test sample train obtained network structure and network parameter to be tested, with further Verify network structure and the reliability of network parameter.

More alternatively, in being tested in step S230 the network structure and the network parameter, can also include： When the actual test error is unsatisfactory for the setting test error, weight is carried out to the network structure and the network parameter New training, until actual test error after the re -training setting test error at a slow speed.

Such as：When test error is unsatisfactory for requiring, then repeatedly above step, re -training network.

Thus, by carrying out re -training to network structure to retest when test error is larger, be advantageous to More accurate and reliable network structure is obtained, and then lifts the accuracy determined to frosting state.

At step S120, the current operating parameter of air-conditioning is obtained.

At step S130, by the corresponding relation, it is determined that current frosting shape corresponding with the current operating parameter State.

Such as：Real time identification goes out the condenser frosting state of air-conditioning.

Wherein, the application be control air-conditioning state of a control (such as：Setup parameter etc.) to realize comfortableness, this Shen It please be that air-conditioning state is identified, identify the situation of frosting, which dictates that the difference of the parameter of input and output.

Thus, by the way that based on corresponding relation, the current frosting situation of air-conditioning is efficiently identified out according to current operating parameter, So as to start for defrosting with terminating to provide accurate basis for estimation, and judged result accuracy is good.

Current frosting state corresponding with the current operating parameter is determined in an optional example, in step S130, It can include：By frosting state corresponding with the current operating parameter identical operational factor in the corresponding relation, it is determined that For current frosting state.

Current frosting state corresponding with the current operating parameter is determined in an optional example, in step S130, It can also include：When the corresponding relation can include functional relation, the current operating parameter is inputted into the function and closed In system, the output parameter for determining the functional relation is current frosting state.

Thus, by the way that based on corresponding relation or functional relation, current frosting state is determined according to current operating parameter, it is determined that Mode is easy, determines result reliability height.

In an optional embodiment, it can also include：Judge whether current frosting state reaches the mistake of defrosting degree Journey.

It is alternatively possible to judge whether current frosting state reaches defrosting degree in method of the invention with reference to shown in Fig. 7 An embodiment schematic flow sheet, further explanation judge whether current frosting state reaches the detailed process of defrosting degree.

Step S710, determines whether the current frosting state reaches the defrosting degree of setting.

Step S720, when the current frosting state reaches the defrosting degree, into the defrosting mode of setting；With/ Or, initiate the prompting that the current frosting state reaches the defrosting degree.

Wherein, when the current frosting state is not up to the defrosting degree, current operation is maintained.

Thus, by whether needing to defrost based on current frosting condition adjudgement, the accuracy of defrost detection is improved, it is beneficial In the heating effect for improving air conditioner, heating efficiency is improved, improves the comfort of user.

In an optional embodiment, it can also include：Determine entry time, the mistake of post-set time of defrosting mode Journey.

It is alternatively possible to according to the current frosting state, after needing to enter the defrosting mode of setting, it is determined that described remove The entry time of white pattern and/or post-set time (such as：It is determined that defrosting time started and/or defrosting end time).

Such as：It can be preset according to existing experience, e.g., when thickness reaches F, carry out defrosting operation；It is thick during defrosting When degree is reduced to A, defrosting operation is exited.

Thus, entry time, post-set time by determining defrosting mode etc., can be with preferably to grasp defrosting process Lifting defrosting reliability, it is also possible that user is apparent to Air conditioners running mode, and then convenient use, Consumer's Experience are good.

In an optional embodiment, it can also include：The displaying operation such as shown and/or exported to relevant parameter Process.

It is alternatively possible to present mode of operation, the current operating parameter, the current frosting shape to the air-conditioning State, the entry time of defrosting mode, at least one of post-set time of defrosting mode, shown and/or exported.

Such as：Can also be to the operational mode of the air-conditioning, the current operating parameter, the current frosting state, institute State the defrosting time started, described at least one of end time of defrosting is transmitted.Such as：If in air-conditioning side, can send To client.If in end side, can send to air-conditioning or other clients.

Thus, operated by the displaying such as being shown, being exported to relevant parameter, user can be lifted to running state of air conditioner Understanding and check, intuitive is strong, and hommization is good.

In an optional embodiment, it can also include：Verify the current frosting state is with actual frosting state The no process being consistent.

It is alternatively possible to receive the result that the current frosting state is not inconsistent with actual frosting state, and/or really When in the fixed corresponding relation not with the current operating parameter identical operational factor, the corresponding relation is carried out more Newly, correct, at least one of learn attended operation again.

Such as：Air-conditioning can not learn actual frosting state, it is necessary to have the feedback operation of user in itself, i.e., if empty Intelligent decision is adjusted to go out frosting state, its state with reality is not inconsistent user by operational feedbacks such as remote controls, and air-conditioning can just be known.

Such as：It can be verified by remote control.Wherein, can be of the invention with reference to shown in Fig. 8 when remote control is verified The schematic flow sheet for the embodiment whether current frosting state is consistent with actual frosting state is verified in method, further Illustrate the detailed process for verifying whether the current frosting state is consistent with actual frosting state.

Step S810, verify whether the current frosting state is consistent with actual frosting state (such as：It can be shown by AR Show that module is shown to actual frosting state, with the current frosting state that checking determines and actual frosting state whether phase Symbol).

Step S820, when the current frosting state and actual frosting state be not inconsistent, and/or the corresponding relation in do not have During with the current operating parameter identical operational factor, the corresponding relation is updated, corrected, learn again at least A kind of attended operation.

Such as：Current frosting state can be determined according to the current operating parameter according to the corresponding relation after maintenance.Example Such as：By frosting state corresponding with the current operating parameter identical operational factor in the corresponding relation after maintenance, really It is set to current frosting state.

Thus, by the maintenance of the corresponding relation between pair operational factor and frosting state of determination, lifting pair is advantageous to The accuracy and reliability that frosting state determines.

Through substantial amounts of verification experimental verification, using the technical scheme of the present embodiment, by using the self-learning function of neutral net, The corresponding relation established between the operational factor of air-conditioning and the frosting state of condenser；According to the current operating parameter of air-conditioning, lead to Current frosting state can be determined by crossing the corresponding relation, and determination mode is reliable, determine that result accuracy is good.

According to an embodiment of the invention, a kind of frosting inspection of air-conditioning of the frosting detection method corresponding to air-conditioning is additionally provided Survey device (such as：Condenser frosting situation detecting system).The structure of one embodiment of device of the invention shown in Figure 9 is shown It is intended to.The frosting detection device of the air-conditioning can include：Establish unit 102, acquiring unit 104 and determining unit 106.

In an optional example, unit 102 is established, can be used for the self-learning capability using artificial neural network, build Corresponding relation under vertical heating mode between the operational factor of air-conditioning and the frosting state of condenser.This establishes the tool of unit 102 Body function and processing are referring to step S110.

Alternatively, the operational factor, can include：Ambient parameter and/or running parameter, and/or by by setting rule The array more than one-dimensional or bidimensional of the feature composition extracted from the ambient parameter, the running parameter.

In an optional specific example, the ambient parameter, it can include：Indoor and outdoor surroundingses temperature, user's body surface temperature At least one of degree, indoor and outdoor surroundingses humidity.

In an optional specific example, the running parameter, it can include：Design temperature, wind speed setting, setting wind At least one of shelves, condenser temperature, outdoor fan running current.

Alternatively, the corresponding relation, can include：Functional relation.

In an optional specific example, the operational factor be the functional relation input parameter, the frosting shape State is the output parameter of the functional relation.

Alternatively, the unit 102 of establishing establishes the operational factor of air-conditioning and the frosting state of condenser under heating mode Between corresponding relation, can specifically include：Acquisition can be used for establishing between the operational factor and the frosting state The sample data of corresponding relation.This establishes the concrete function of unit 102 and processing referring further to step S210.

More alternatively, it is described establish unit 102 obtain can be used for establishing the operational factor and the frosting state it Between corresponding relation sample data, can specifically include：Collect service data and frosting of the air-conditioning under different use environments State.This establishes the concrete function of unit 102 and processing referring further to step S310.

In a more optional specific example, the unit 102 of establishing collects operation of the air-conditioning under different use environments Data, it can specifically include：Obtain the operational factor of the air-conditioning under laboratory simulation environment.

In a more optional specific example, the unit 102 of establishing collects operation of the air-conditioning under different use environments Data, it can also specifically include：Obtain operational factor when user's actual use air-conditioning is collected by technology of Internet of things.

More alternatively, it is described establish unit 102 obtain can be used for establishing the operational factor and the frosting state it Between corresponding relation sample data, can also specifically include：Analyzed the service data and combined the expert to prestore Posterior infromation, choose data related to the frosting state as the operational factor (such as：Choose to air-conditioning frosting state Influential parameter is as input parameter, using frosting state as output parameter).This establishes the concrete function of unit 102 and place Reason is referring further to step S320.

More alternatively, it is described establish unit 102 obtain can be used for establishing the operational factor and the frosting state it Between corresponding relation sample data, can also specifically include：By the frosting state and the operational factor structure chosen Into data pair, as sample data.This establishes the concrete function of unit 102 and processing referring further to step S330.

Alternatively, the unit 102 of establishing establishes the operational factor of air-conditioning and the frosting state of condenser under heating mode Between corresponding relation, can also specifically include：Analyze the characteristic and its rule of the operational factor, according to the characteristic and its Rule, determine the network structure and its network parameter of the artificial neural network.This establishes the concrete function of unit 102 and processing Referring further to step S220.

More alternatively, the network structure, can include：It is BP neural network, CNN neutral nets, RNN neutral nets, residual At least one of poor neutral net.

More alternatively, the network parameter, can include：Input number of nodes, output node number, hidden layer number, Hidden nodes, At least one of initial weight, bias.

Alternatively, the unit 102 of establishing establishes the operational factor of air-conditioning and the frosting state of condenser under heating mode Between corresponding relation, can also specifically include：Using the sample data, the network structure and the network parameter are entered Row training and test, determine the corresponding relation of the operational factor and the frosting state.This establishes the specific of unit 102 Function and processing are referring further to step S230.

More alternatively, the unit 102 of establishing is trained to the network structure and the network parameter, specifically can be with Including：A part of data in the sample data are chosen as training sample, the operation in the training sample is joined Number is input to the network structure, is trained by the activation primitive and the network parameter of the network structure, obtains reality Border training result.This establishes the concrete function of unit 102 and processing referring further to step S410.

In a more optional specific example, the unit 102 of establishing enters to the network structure and the network parameter Row training, can also specifically include：Determine between hands-on result frosting state corresponding in the training sample Hands-on error whether meet set training error.This establishes the concrete function of unit 102 and processing referring further to step S420.Such as：Expectation training result is used as using the corresponding frosting state in the sample data.

In a more optional specific example, the unit 102 of establishing enters to the network structure and the network parameter Row training, can also specifically include：When the hands-on error meets the setting training error, it is determined that to the network The training of structure and the network parameter is completed.This establishes the concrete function of unit 102 and processing referring further to step S430.

More alternatively, the unit 102 of establishing is trained to the network structure and the network parameter, specifically may be used also With including：When the hands-on error is unsatisfactory for the setting training error, pass through the error energy of the network structure Function updates the network parameter.This establishes the concrete function of unit 102 and processing referring further to step S610.

In a more optional specific example, the unit 102 of establishing enters to the network structure and the network parameter Row training, can also specifically include：Entered by the network parameter after the activation primitive of the network structure and renewal Row re -training, until the hands-on error after the re -training meets the setting training error.This establishes unit 102 Concrete function and processing referring further to step S620.

Update each layer weights：

Update each layer biasing：

Wherein：

W_kFor initial weight,For the partial derivative of error energy function pair weights, b_lFor initial bias,For error Partial derivative of the energy function to biasing；Value can be obtained by chain type Rule for derivation, until network output error Untill ∈ (target minimal error).

More alternatively, the unit 102 of establishing is tested the network structure and the network parameter, specifically can be with Including：Another part data in the sample data are chosen as test sample, by the operation in the test sample Parameter is input in the network structure that the training is completed, the network completed with the activation primitive and the training Parameter is tested, and obtains actual test result.This establishes the concrete function of unit 102 and processing referring further to step S510.

In a more optional specific example, the unit 102 of establishing enters to the network structure and the network parameter Row test, can also specifically include：Determine actual test result frosting state corresponding in the test sample (i.e. with The frosting state in the sample data is as desired output parameter) between actual test error whether meet setting survey Try error.This establishes the concrete function of unit 102 and processing referring further to step S520.

In a more optional specific example, the unit 102 of establishing enters to the network structure and the network parameter Row test, can also specifically include：When the actual test error meets the setting test error, it is determined that to the network The test of structure and the network parameter is completed.This establishes the concrete function of unit 102 and processing referring further to step S530.

Such as：If test error meets to require, network training test is completed.

More alternatively, the unit 102 of establishing is tested the network structure and the network parameter, specifically may be used also With including：When the actual test error is unsatisfactory for the setting test error, the network structure and the network are joined Number carries out re -trainings, until the actual test error setting test error at a slow speed after the re -training.

Thus, by carrying out re -training to network structure to retest when test error is larger, be advantageous to More accurate and reliable network structure is obtained, and then lifts the accuracy determined to frosting state

In an optional example, acquiring unit 104, it can be used for the current operating parameter for obtaining air-conditioning.The acquisition list The concrete function of member 104 and processing are referring to step S120.

In an optional example, determining unit 106, can be used for by the corresponding relation, it is determined that with it is described current Current frosting state corresponding to operational factor.The concrete function of the determining unit 106 and processing are referring to step S130.

Alternatively, the determining unit 106 determines current frosting state corresponding with the current operating parameter, specifically may be used With including：By frosting state corresponding with the current operating parameter identical operational factor in the corresponding relation, it is defined as Current frosting state.

Alternatively, the determining unit 106 determines current frosting state corresponding with the current operating parameter, specific to go back It can include：When the corresponding relation can include functional relation, the current operating parameter is inputted into the functional relation In, the output parameter for determining the functional relation is current frosting state.

In an optional example, the determining unit 106, it can be also used for determining whether the current frosting state reaches To the defrosting degree of setting.This establishes the concrete function of unit 102 and processing referring further to step S710.

In an optional example, the determining unit 106, it can be also used for when the current frosting state reaches described During defrosting degree, into the defrosting mode of setting, and/or, initiate the current frosting state and reach carrying for the defrosting degree Show.This establishes the concrete function of unit 102 and processing referring further to step S720.

Wherein, or when the current frosting state is not up to the defrosting degree, current operation is maintained.

In an optional example, the determining unit 106, it can be also used for according to the current frosting state, when need Will enter setting defrosting mode after, determine the defrosting mode entry time and/or post-set time (such as：It is determined that defrosting Time started and/or defrosting end time).

In an optional example, the determining unit 106, can be also used for the present mode of operation to the air-conditioning, The current operating parameter, the current frosting state, the entry time of defrosting mode, defrosting mode post-set time in extremely It is one of few, shown and/or exported.

In an optional example, the determining unit 106, it can be also used for receiving the current frosting state and reality The result that border frosting state is not inconsistent, and/or determine in the corresponding relation not with the current operating parameter identical During operational factor, attended operation at least one of is updated, corrected, learning again to the corresponding relation.

Such as：It can be verified by remote control.Wherein, when remote control is verified, following operation can be performed：

Such as：The determining unit 106, it can be also used for verifying whether are the current frosting state and actual frosting state Be consistent (such as：Actual frosting state can be shown by AR display modules, the current frosting shape determined with checking Whether state is consistent with actual frosting state).The concrete function of the determining unit 106 and processing are referring further to step S810.

Such as：The determining unit 106, can be also used for when the current frosting state is not inconsistent with actual frosting state, And/or when in the corresponding relation not with the current operating parameter identical operational factor, unit is established by described 102, attended operation at least one of is updated, corrected, learning again to the corresponding relation.The tool of the determining unit 106 Body function and processing are referring further to step S820.

The processing and function realized by the device of the present embodiment essentially correspond to earlier figures 1 to the method shown in Fig. 8 Embodiment, principle and example, therefore not detailed part in the description of the present embodiment may refer to mutually speaking on somebody's behalf in previous embodiment It is bright, it will not be described here.

Through substantial amounts of verification experimental verification, using technical scheme, by using the self-learning function of neutral net, lead to Cross and the data collected are trained and learnt, grasp operational factor and sensor parameters and the condenser frosting of air-conditioning Corresponding relation function between state, so as to which Real time identification goes out the condenser frosting state of air-conditioning, improve the accurate of defrost detection Property, it is beneficial to the heating effect for improving air conditioner, improves heating efficiency, improves the comfort of user.

According to an embodiment of the invention, a kind of storage medium of the frosting detection method corresponding to air-conditioning is additionally provided.Should Storage medium can include：A plurality of instruction is stored with the storage medium；Wherein, a plurality of instruction, for by processor Load and perform the frosting detection method of above-described air-conditioning.

The processing and function realized by the storage medium of the present embodiment essentially correspond to earlier figures 1 to shown in Fig. 8 Embodiment, principle and the example of method, therefore not detailed part in the description of the present embodiment, may refer to the phase in previous embodiment Speak on somebody's behalf bright, will not be described here.

Through substantial amounts of verification experimental verification, using technical scheme, analyzed by using artificial neural network algorithm The running status rule of the Frost formation process of air-conditioning condenser, sky is found by the self study of artificial neural network, adaptive characteristic The mapping principle between condenser frosting situation and air conditioner operation parameters state in heating operation is adjusted, air-conditioning can be efficiently identified out and worked as Preceding frosting situation, so as to start for defrosting with terminating to provide accurate basis for estimation, and judged result accuracy is good.

According to an embodiment of the invention, the frosting detection side of the frosting detection method or air-conditioning corresponding to air-conditioning is additionally provided A kind of equipment of method.The equipment can include：Processor, for performing a plurality of instruction；Memory, for storing a plurality of instruction； Wherein, a plurality of instruction, for by the memory storage, and loaded by processor and perform the knot of above-described air-conditioning White detection method.Or the equipment can include：The frosting detection device of above-described air-conditioning.

Such as：Provide a kind of method realization device based on neural network recognization air-conditioning condenser frosting situation.Such as figure Shown in 14, when the air conditioner with wireless telecommunications is run, the operational factor of air-conditioning is uploaded to intelligent apparatus.

Such as：Operational factor is input in the network algorithm trained by intelligent apparatus, after judging frosting situation, to sky Readjust the distribution and send frosting situation either to directly transmit defrosting or terminate the order of defrosting.

Alternatively, the equipment can include：For the air conditioner main body being controlled to itself, and/or, for air-conditioning sheet The outside control terminal that body is controlled.

Such as：Algorithm can also be directly integrated in the controller of air-conditioning in itself, not connect intelligent apparatus additionally.

Wherein, the outside control terminal, can include：In wireless communication module, router, server, terminal at least One of.

Such as：Intelligent apparatus includes but is not limited to wireless communication module, router, server, smart mobile phone etc..Such as： Intelligent apparatus, it can include：At least one of wireless communication module, router, server, smart mobile phone.

In an optional example, the equipment, artificial neural network algorithm can be utilized, by using ring to a large amount of differences (include but is not limited to following one or more under border：Indoor and outdoor surroundingses temperature and humidity, design temperature etc.) operation of air conditioner number According to collect collect, choose some air-conditionings running state parameter and frosting state parameter as sample data, neutral net is entered Row study and training, by adjusting the weights between network structure and network node, make neutral net fitting air conditioner operation parameters and Relation between frosting state, neutral net is finally set accurately to fit between the operational factor of air-conditioning and condenser frosting state Corresponding relation.

Such as：Network structure, network node, all it is the term of field of neural networks, node is neuron, represents a kind of spy Fixed output function, structure are the composition situation of whole neutral net, the summary of connection.

Such as：Relation between network structure and network node, it can include：Network structure, can be by each network node Between connection situation determine that different nodes, the connecting mode of node and node, the number of plies etc. of node can show difference Network structure.

Such as：The weights between network structure and network node are adjusted, can be included：Overall network structure can be pre- in advance Good if (several similar structures can also be preset, the later stage selects most suitable according to the situation of network training), the adjustment of weights It may refer to the network training in following step 4 once and test.

Wherein, specific implementation step is as follows：

Step 1, data collection：

Collect itself running state parameter and corresponding frosting situation of the air-conditioning under different use environments.It is specific to collect Mode, which includes but is not limited to operational factor of the air-conditioning under laboratory simulation environment, actual user is collected by technology of Internet of things makes The modes such as the air conditioner operation parameters of used time.

Step 2, sample data selection：

By the analysis to data and expertise knowledge is combined, is chosen on the influential parameter conduct of air-conditioning frosting state Input parameter, using frosting state as output parameter.In this programme, input parameter includes but is not limited to following one kind or more Kind：The environment temperature of indoor and outdoor, air-conditioning design temperature, wind speed setting (shelves), condenser temperature, outdoor fan running current etc.. Input parameter is not only single parameter, includes the input parameter one or more dimensions array for extracting feature composition according to certain rules yet.

By obtained input, output parameter pair, a part is used as training book sample data, and a part is used as test sample number According to.

Step 3, network structure design：

The data characteristic influenceed according to different design temperatures, environment temperature, shell temperature etc. on operation of air conditioner and its institute The rule contained, the basic structure of network, the input of network, output node number, network hidden layer number, hidden node can be primarily determined that Number, network initial weight etc..

Such as：The data influenceed on operation of air conditioner, can come from all service datas collected in step 1, also may be used To be a few items therein.

Such as：The characteristic of the data influenceed on operation of air conditioner and its rule contained, such as frosting here, frosting Intuitively there is relation with condenser temperature very, temperature is lower, and the possibility and thickness of frosting can be bigger, therefore first can be Condenser temperature is as input parameter.

Wherein, neural network algorithm schematic diagram such as Fig. 1 of the invention, specific artificial neural network structure include but is not limited to Three kinds of network structures below：

(1) BP neural network

BP neural network structural representation is as shown in Figure 10, during practical application, can adjust input layer, hidden as needed Layer, output layer nodes and hidden layers numbers.

Such as：Such as 8 output parameters are determined, and the frosting situation exported has 10 kinds, then can tentatively choose defeated It is 8 to enter node layer number, and output layer nodes are 10；And intermediate layer rule of thumb chooses 5, behind training result is bad to carry out Adjustment.

(2) CNN convolutional neural networks

Multilayer convolutional network, exactly constantly extract merging from the feature of low dimensional and obtain the feature of more higher-dimension, so as to For being classified or inter-related task.

Multilayer convolutional network structural representation is as shown in figure 11, and network structure can be adjusted according to actual conditions during practical application It is whole.

(3) residual error neutral net

When debugging CNN networks, deepen the network number of plies and change the method for convolution kernel size, network can not be showed Get a promotion.Front and rear data can preferably be connected by adding residual block, strengthen feature representation ability, so it can strengthen convolution The learning ability of network.As shown in figure 13, the input of certain section of neutral net is x, and desired output is H (x), and input x is passed to Output as after initial configuration, it is necessary to learn target be just changed into F (x)=H (x)-x.

Residual error neural network structure schematic diagram is as shown in figure 12, and network structure can be adjusted according to actual conditions during practical application It is whole.

That is, the artificial neural network used in this programme, is not limited to a certain network structure, can be through The artificial neural network of allusion quotation such as BP neural network or high-level manual's neutral net, or deep learning network such as CNN (convolutional neural networks) and RNN (Recognition with Recurrent Neural Network) etc., concrete scheme can select according to practical application scene.

Step 4, network training and test：

After the completion of network design, training sample data need to be used, network is trained.

Training method can be adjusted according to the problem of discovery in the network structure of reality and training.Herein only for this hair Bright one of which method is illustrated below：

(1) input data x is imported, according to activation primitive, the weights of initialization and biasing, calculates the reality output a of network (x), i.e. a (x)=1/ (1+e^-z), wherein Z=W_k*x+b_l。

(2) judge whether the desired output y (x) and reality output a (x) of network meet output accuracy requirement, i.e.,：

‖y(x)-a(x)‖<∈, ∈ are target minimal error.

(3) terminate to train if required precision is met, update the weights W of network according in the following manner if being unsatisfactory for_k, Bias b_l：

Update each layer weights：

Update each layer biasing：

Wherein：

After the completion of network training, then with test sample positive test network.When test error is unsatisfactory for requiring, then repeat Above step, re -training network；If test error meets to require, network training test is completed.

The processing and function realized by the equipment of the present embodiment essentially correspond to earlier figures 1 to the method shown in Fig. 8 Or embodiment, principle and the example of the device shown in Fig. 9, therefore not detailed part in the description of the present embodiment, it may refer to foregoing Related description in embodiment, will not be described here.

Through substantial amounts of verification experimental verification, using technical scheme, pass through pair between operational factor and frosting state It should be related to, determine current frosting state corresponding to current operating parameter, mode of operation is easy, and operating result reliability is high.

To sum up, it will be readily appreciated by those skilled in the art that on the premise of not conflicting, above-mentioned each advantageous manner can be certainly Combined, be superimposed by ground.

Embodiments of the invention are the foregoing is only, are not intended to limit the invention, for those skilled in the art For member, the present invention can have various modifications and variations.Any modification within the spirit and principles of the invention, being made, Equivalent substitution, improvement etc., should be included within scope of the presently claimed invention.

Claims

A kind of 1. frosting detection method of air-conditioning, it is characterised in that including：

Using the self-learning capability of artificial neural network, the frosting shape of the operational factor of air-conditioning and condenser under heating mode is established Corresponding relation between state；

Obtain the current operating parameter of air-conditioning；

By the corresponding relation, it is determined that current frosting state corresponding with the current operating parameter.
2. according to the method for claim 1, it is characterised in that wherein,

The operational factor, including：Ambient parameter and/or running parameter, and/or by by setting rule from the ambient parameter, The array more than one-dimensional or bidimensional of the feature composition extracted in the running parameter；Wherein,

The ambient parameter, including：At least one of indoor and outdoor surroundingses temperature, user's shell temperature, indoor and outdoor surroundingses humidity； And/or

The running parameter, including：Design temperature, wind speed setting, setting wind shelves, condenser temperature, outdoor fan running current At least one of；

And/or

The corresponding relation, including：Functional relation；

The operational factor is the input parameter of the functional relation, and the frosting state is joined for the output of the functional relation Number；

It is determined that current frosting state corresponding with the current operating parameter, including：

By frosting state corresponding with the current operating parameter identical operational factor in the corresponding relation, it is defined as current Frosting state；And/or

When the corresponding relation includes functional relation, the current operating parameter is inputted in the functional relation, determines institute The output parameter for stating functional relation is current frosting state.
3. method according to claim 1 or 2, it is characterised in that establish under heating mode the operational factor of air-conditioning with it is cold Corresponding relation between the frosting state of condenser, including：

Obtain the sample data for establishing the corresponding relation between the operational factor and the frosting state；

The characteristic and its rule of the operational factor are analyzed, according to the characteristic and its rule, determines the artificial neural network Network structure and its network parameter；

Using the sample data, the network structure and the network parameter are trained and tested, determine the operation The corresponding relation of parameter and the frosting state.
4. according to the method for claim 3, it is characterised in that obtain for establishing the operational factor and the frosting shape The sample data of corresponding relation between state, including：

Collect service data and frosting state of the air-conditioning under different use environments；

The expertise information to prestore is analyzed the service data and combined, is chosen related to the frosting state Data are as the operational factor；

The data pair that the frosting state and the operational factor chosen are formed, as sample data.
5. according to the method for claim 4, it is characterised in that

Service data of the air-conditioning under different use environments is collected, including：

The operational factor of the air-conditioning under laboratory simulation environment is obtained, and/or,

Obtain operational factor when user's actual use air-conditioning is collected by technology of Internet of things；

And/or

The network structure, including：In BP neural network, CNN neutral nets, RNN neutral nets, residual error neutral net at least One of；And/or

The network parameter, including：In input number of nodes, output node number, hidden layer number, Hidden nodes, initial weight, bias At least one of.
6. according to the method described in one of claim 3-5, it is characterised in that

The network structure and the network parameter are trained, including：

A part of data in the sample data are chosen as training sample, by the operational factor in the training sample The network structure is input to, is trained by the activation primitive and the network parameter of the network structure, obtains reality Training result；

Determine whether is hands-on error between hands-on result frosting state corresponding in the training sample Meet setting training error；

When the hands-on error meets the setting training error, it is determined that to the network structure and the network parameter The training complete；

And/or

The network structure and the network parameter are tested, including：

Another part data in the sample data are chosen as test sample, the operation in the test sample is joined Number is input in the network structure that the training is completed, and is joined with the network that the activation primitive and the training are completed Number is tested, and obtains actual test result；

Determine whether is actual test error between actual test result frosting state corresponding in the test sample Meet setting test error；

When the actual test error meets the setting test error, it is determined that to the network structure and the network parameter The test complete.
7. according to the method for claim 6, it is characterised in that

The network structure and the network parameter are trained, in addition to：

When the hands-on error is unsatisfactory for the setting training error, pass through the error energy function of the network structure Update the network parameter；

Re -training is carried out by the network parameter after the activation primitive of the network structure and renewal, until described Hands-on error after re -training meets the setting training error；

And/or

The network structure and the network parameter are tested, in addition to：

When the actual test error is unsatisfactory for the setting test error, the network structure and the network parameter are entered Row re -training, until the actual test error setting test error at a slow speed after the re -training.
8. according to the method described in one of claim 1-7, it is characterised in that also include：

Determine whether the current frosting state reaches the defrosting degree of setting；

When the current frosting state reaches the defrosting degree, into the defrosting mode of setting, and/or, initiate described work as Preceding frosting state reaches the prompting of the defrosting degree；

And/or

According to the current frosting state, after needing to enter the defrosting mode of setting, determine that entering for the defrosting mode is fashionable Between and/or post-set time；

And/or

The entrance of present mode of operation, the current operating parameter, current the frosting state, defrosting mode to the air-conditioning Time, at least one of post-set time of defrosting mode, shown and/or exported；

And/or

Receive the result that frosting state is not inconsistent with actual frosting state described currently, and/or do not have in the corresponding relation When having with the current operating parameter identical operational factor, the corresponding relation is updated, corrected, learn again in extremely A kind of few attended operation.
A kind of 9. frosting detection device of air-conditioning, it is characterised in that including：

Establish unit, for the self-learning capability using artificial neural network, establish under heating mode the operational factor of air-conditioning with Corresponding relation between the frosting state of condenser；

Acquiring unit, for obtaining the current operating parameter of air-conditioning；

Determining unit, for by the corresponding relation, it is determined that current frosting state corresponding with the current operating parameter.
10. device according to claim 9, it is characterised in that wherein,

The operational factor, including：Ambient parameter and/or running parameter, and/or by by setting rule from the ambient parameter, The array more than one-dimensional or bidimensional of the feature composition extracted in the running parameter；Wherein,

The ambient parameter, including：At least one of indoor and outdoor surroundingses temperature, user's shell temperature, indoor and outdoor surroundingses humidity； And/or

The running parameter, including：Design temperature, wind speed setting, setting wind shelves, condenser temperature, outdoor fan running current At least one of；

And/or

The corresponding relation, including：Functional relation；

The operational factor is the input parameter of the functional relation, and the frosting state is joined for the output of the functional relation Number；

The determining unit determines current frosting state corresponding with the current operating parameter, specifically includes：

By frosting state corresponding with the current operating parameter identical operational factor in the corresponding relation, it is defined as current Frosting state；And/or

When the corresponding relation includes functional relation, the current operating parameter is inputted in the functional relation, determines institute The output parameter for stating functional relation is current frosting state.
11. the device according to claim 9 or 10, it is characterised in that the unit of establishing establishes air-conditioning under heating mode Operational factor and condenser frosting state between corresponding relation, specifically include：

Obtain the sample data for establishing the corresponding relation between the operational factor and the frosting state；

The characteristic and its rule of the operational factor are analyzed, according to the characteristic and its rule, determines the artificial neural network Network structure and its network parameter；

Using the sample data, the network structure and the network parameter are trained and tested, determine the operation The corresponding relation of parameter and the frosting state.
12. device according to claim 11, it is characterised in that the unit of establishing is obtained for establishing the operation ginseng The sample data of several corresponding relations between the frosting state, is specifically included：

Collect service data and frosting state of the air-conditioning under different use environments；

The expertise information to prestore is analyzed the service data and combined, is chosen related to the frosting state Data are as the operational factor；

The data pair that the frosting state and the operational factor chosen are formed, as sample data.
13. device according to claim 12, it is characterised in that

The unit of establishing collects service data of the air-conditioning under different use environments, specifically includes：

The operational factor of the air-conditioning under laboratory simulation environment is obtained, and/or,

Obtain operational factor when user's actual use air-conditioning is collected by technology of Internet of things；

And/or

The network structure, including：In BP neural network, CNN neutral nets, RNN neutral nets, residual error neutral net at least One of；And/or

The network parameter, including：In input number of nodes, output node number, hidden layer number, Hidden nodes, initial weight, bias At least one of.
14. according to the device described in one of claim 11-13, it is characterised in that

The unit of establishing is trained to the network structure and the network parameter, is specifically included：

A part of data in the sample data are chosen as training sample, by the operational factor in the training sample The network structure is input to, is trained by the activation primitive and the network parameter of the network structure, obtains reality Training result；

Determine whether is hands-on error between hands-on result frosting state corresponding in the training sample Meet setting training error；

When the hands-on error meets the setting training error, it is determined that to the network structure and the network parameter The training complete；

And/or

The unit of establishing is tested the network structure and the network parameter, is specifically included：

Another part data in the sample data are chosen as test sample, the operation in the test sample is joined Number is input in the network structure that the training is completed, and is joined with the network that the activation primitive and the training are completed Number is tested, and obtains actual test result；

Determine whether is actual test error between actual test result frosting state corresponding in the test sample Meet setting test error；

When the actual test error meets the setting test error, it is determined that to the network structure and the network parameter The test complete.
15. device according to claim 14, it is characterised in that

The unit of establishing is trained to the network structure and the network parameter, is specifically also included：

When the hands-on error is unsatisfactory for the setting training error, pass through the error energy function of the network structure Update the network parameter；

Re -training is carried out by the network parameter after the activation primitive of the network structure and renewal, until described Hands-on error after re -training meets the setting training error；

And/or

The unit of establishing is tested the network structure and the network parameter, is specifically also included：

When the actual test error is unsatisfactory for the setting test error, the network structure and the network parameter are entered Row re -training, until the actual test error setting test error at a slow speed after the re -training.
16. according to the device described in one of claim 9-15, it is characterised in that also include：

The determining unit, it is additionally operable to determine the defrosting the degree whether current frosting state reaches setting；

When the current frosting state reaches the defrosting degree, into the defrosting mode of setting, and/or, initiate described work as Preceding frosting state reaches the prompting of the defrosting degree；

And/or

The determining unit, it is additionally operable to, according to the current frosting state, after needing to enter the defrosting mode of setting, determine institute State entry time and/or the post-set time of defrosting mode；

And/or

The determining unit, it is additionally operable to the present mode of operation to the air-conditioning, the current operating parameter, the current frosting State, the entry time of defrosting mode, at least one of post-set time of defrosting mode, shown and/or exported；

And/or

The determining unit, be additionally operable to receive the result that the current frosting state is not inconsistent with actual frosting state and/ Or when in the corresponding relation not with the current operating parameter identical operational factor, unit is established by described, to institute State corresponding relation and at least one of be updated, correct, learning again attended operation.
A kind of 17. storage medium, it is characterised in that including：A plurality of instruction is stored with the storage medium；

Wherein, a plurality of instruction, for being loaded by processor and performing the frosting of the air-conditioning as described in claim 1-8 is any Detection method.
A kind of 18. equipment, it is characterised in that including：

Processor, for performing a plurality of instruction；

Memory, for storing a plurality of instruction；

Wherein, a plurality of instruction, for by the memory storage, and loaded by processor and performed such as claim 1-8 The frosting detection method of any described air-conditioning；

Or

The frosting detection device of air-conditioning as described in claim 9-16 is any.
19. equipment according to claim 18, it is characterised in that the equipment includes：For the sky being controlled to itself Body is adjusted, and/or, for the outside control terminal being controlled to air conditioner main body；Wherein,

The outside control terminal, including：At least one of wireless communication module, router, server, terminal.