Specific embodiment
To keep the technical problem to be solved in the present invention, technical solution and advantage clearer, below in conjunction with attached drawing and toolBody embodiment is described in detail.
The data processing method of earth's surface disaster alarm provided in an embodiment of the present invention can be applied to detect sensorMonitoring data carry out vector machine recurrence learning, and obtain prediction result, earth's surface disaster just can determine according to the prediction resultEarly warning prevents the hair of earth's surface disaster as a result, monitoring personnel can take adequate measures according to the earth's surface disaster alarm resultIt is raw, above-mentioned earth's surface disaster can be the movement of the rock as earth's surface, soil property etc. and caused by disaster, such as: mountain area mud-rock flowDisaster, building collapse disaster etc. use the multiple sensors row of acquisition respectively it is, of course, also possible to be Mine production disaster at this timeSurface displacement information, internal displacement information, rainfall information, soil pressure force information, the soil moisture content information, pore water pressure of Tu ChangForce information, temperature information and humidity information etc., and using above-mentioned earth's surface disaster alarm data processing method to these information intoRow processing predicts whether the refuse dump has and the danger such as mud-rock flow, collapsing occurs, and obtains prediction result.
Referring to Figure 1, Fig. 1 is a kind of process of the data processing method of earth's surface disaster alarm provided in an embodiment of the present inventionFigure, as shown in Figure 1, method includes the following steps:
Step 101, using feature vector selection algorithm, choose first eigenvector from n initial sample datas, it is describedFirst eigenvector includes m sample data, wherein n is positive integer, and m is the integer less than n, and the sample data is to pass throughThe monitoring data that sensor obtains.
Wherein, features described above vector selection algorithm can select m from n initial sample datas (x1, x2 ... ..., xn)A sample data (xs1, xs2 ... ..., xsv), wherein 1≤v≤m, above-mentioned m sample data (xs1, xs2 ... ..., xsv) areFeature vector FV after selection.
In this way, any vector in the m sample data can in the case where the set of eigenvectors of known sample dataWith by FV linear expression, convenient for using Support vector regression algorithm to be learnt in step 102 to step 104.
In addition, the temperature that above-mentioned monitoring data can be the pressure information of pressure sensor acquisition, temperature sensor obtainsAny one or more in humidity information that information and humidity sensor obtain etc., such as: it is being applied to refuse dump disasterIn monitoring process, above-mentioned monitoring data can be the surface displacement information of the refuse dump, internal displacement information, rainfall information,One of soil pressure force information, soil moisture content information, pore water pressure force information, temperature information and humidity information information or moreKind information.Therefore, monitoring data have the characteristic that data volume is big and dimension is high.
When learning using the Support vector regression algorithm monitoring data big to above-mentioned data volume, need to spend bigTo cause the overlong time for exporting prediction result the effect of early warning is not achieved, it is therefore desirable to above-mentioned monitoring number in the time of amountAccording to data volume reduced.
In this step, using features described above vector selection algorithm, to the above-mentioned initial sample comprising a large amount of sample dataData are reduced, and do not change the structure of initial sample data, can reduce the support in step 102 to step 104 in this wayVector machine regression algorithm is to operation time of the first eigenvector, so as to promote the data of the earth's surface disaster alarmThe efficiency of processing method.
Step 102, using the Support vector regression algorithm including insensitive loss function, respectively to the fisrt featureVector sum first increases sample data newly and is learnt, and obtains the first prediction result and the second prediction result, wherein described first is newIncreasing sample data is the sample data increased newly in first predetermined period.
Wherein, the value of above-mentioned insensitive loss function of ε can be to the prediction result of above-mentioned Support vector regression algorithmAccuracy has an impact.When the value of ε is excessive, the efficiency and accuracy of the Support vector regression algorithm will be reduced；When the value of εWhen too small, by generate overfitting the case where, the accuracy of prediction result will also decrease.
It should be noted that above-mentioned first prediction result is to be supported in first predetermined period to first eigenvectorPrediction result obtained from vector machine regression algorithm, above-mentioned second prediction result are in second predetermined period to the first newly-increased sampleNotebook data is supported prediction result obtained from vector machine regression algorithm, and second predetermined period is later than first predetermined period.
Due to not verifying to above-mentioned first prediction result and the second prediction result, it not can confirm that above-mentioned first is pre-The accuracy of result and the second prediction result is surveyed, such as: since the fitness in the feature vector selection algorithm in step 101 takesNot be worthwhile and cause, the difference between the first eigenvector and initial sample data is excessive, will cause described first pre-Surveying has error between result and the second prediction result and actual conditions.
It, can be using Support vector regression algorithm to respectively to first eigenvector and the first newly-increased sample by this stepNotebook data is learnt, and obtains above-mentioned first prediction result and above-mentioned second prediction result respectively, is step 103 and step104 provide operating basis.
Step 103, according to difference between first prediction result and the first newly-increased sample data and describedDifference between second prediction result and the second newly-increased sample data adjusts the insensitive loss function, wherein described secondNewly-increased sample data is the sample data increased newly in second predetermined period, and described second predetermined period is later than first predictionPeriod.
Wherein, in the specific application process, the monitoring data of refuse dump are constantly newly-increased, are obtained in first predetermined periodThe prediction result obtained can be verified according to the newly-increased sample data obtained after first predetermined period.
In addition, if the above-mentioned prediction result obtained in first predetermined period increases newly with what is obtained after first predetermined periodHave between biggish difference and second prediction result and the second newly-increased sample data between sample data with largerDifference, then it represents that the value of the ε in above-mentioned Support vector regression algorithm may be improper, makes to prop up by adjusting the value of εThe accuracy for holding the prediction result of vector machine regression algorithm is higher.
Specifically, above-mentioned insensitive loss function can be adjusted using fuzzy control method.
For example, the fuzzy controller exports the increment Delta ε of ε as shown in Fig. 2, E1 and Δ E is inputted fuzzy controller, withAdjust the value of ε.The fuzzy controller by the fuzzy variable of E1 and Δ E be divided into 7 fuzzy subsets as shown in Figure 3 NL,NM, NS, ZE, PS, PM, PL }.Rule of thumb rule and sample calculate analysis, and E1 and Δ E respectively include as shown in Figure 3 be subordinate toRelationship.Δ ε is divided into 5 fuzzy subsets { NB, NS, ZE, PS, PB } as shown in Figure 4 by the fuzzy controller, and according to inputE1 and Δ E and the Δ ε of output between change relationship establish fuzzy rule as shown in Table 1:
Table 1
| (Δ E, U, E1) | NL | NM | NS | ZE | PS | PM | PL |
| NL | NB | NB | NB | ZE | NS | ZE | ZE |
| NM | NB | NB | NB | PS | ZE | ZE | NS |
| NS | NB | NB | NS | PB | ZE | NS | NS |
| ZE | NB | NS | ZE | PB | ZE | NS | NB |
| PS | NS | NS | ZE | PB | NS | NB | NB |
| PM | NS | ZE | ZE | PS | NB | NB | NB |
| PL | ZE | ZE | NS | ZE | NB | NB | NB |
Wherein, the first row indicates each fuzzy subset belonging to E1, and first row indicates each fuzzy subset belonging to Δ E, UIndicate fuzzy rule, the i.e. different fuzzy subsets that Δ ε is adapted to according to the fuzzy rule.For example, when E1 belongs to the fuzzy son of NLCollection, Δ E belong to PL fuzzy subset, then Δ ε will be adjusted according to fuzzy rule to the direction of ZE fuzzy subset.
In addition, when E1 is located in the section NL or the section PL, and Δ E indicates that the error amount of E2 is greater than the error amount of E1, thenThe output valve of Δ ε is adjusted to reduce the value of ε.
Certainly, the quantity of the fuzzy subset of above-mentioned E1, Δ E and Δ ε can also be other numbers such as 3,4, hereinWithout limitation.
It should be noted that offline SVR model described in Fig. 2 refers to, initial sample data is selected by feature vectorThe first eigenvector obtained after algorithm is reduced is selected, input Learning machine is supported vector machine regression algorithm (SupportVector Regression, SVR) off-line learning.
In this step, by comparing difference between first prediction result and the first newly-increased sample data, withAnd the difference between second prediction result and the second newly-increased sample data, can determine whether the value of ε is suitable, ε'sThe value of the ε is adjusted in the inappropriate situation of value, with promoted the Support vector regression algorithm prediction result it is accurateProperty, to reach the accuracy for promoting the prediction result of data processing method of the earth's surface disaster alarm.
Step 104, using the Support vector regression algorithm including insensitive loss function adjusted, to described firstFeature vector is learnt, and obtains target prediction result.
It should be noted that in actual application, since above-mentioned monitoring data continual may update, becauseThis, above-mentioned target prediction result can be using the Support vector regression algorithm for including insensitive loss function adjusted,To the first eigenvector, and/or, the first newly-increased sample data, and/or, the second newly-increased sample data is learnt and is obtainedThe name of prediction result out, above-mentioned first eigenvector, the first newly-increased sample data and the second newly-increased sample data is answered hereinCan not be identical for distinguishing its three, over time, above-mentioned first eigenvector, the first newly-increased sample data andThe role of two newly-increased sample datas can change, such as: if can be 08:01 when current, in the inspection that the 08:00 moment obtainsMeasured data is that newly-increased sample data still if current time is 09:00, becomes in the detection data that the 08:00 moment obtainsHistorical data.
Certainly, above-mentioned target prediction is as a result, can also be according to the sample number increased newly after described second predetermined periodAccording to being supported prediction result obtained from vector machine regression algorithm.
In this step, by using the Support vector regression algorithm including ε adjusted, to the first eigenvectorLearnt again with the described first newly-increased sample data, the accuracy of the target prediction result made is higher.
In the embodiment of the present invention, by using feature vector selection algorithm, the first spy is chosen from n initial sample datasVector is levied, the first eigenvector includes m sample data, wherein n is positive integer, and m is the integer less than n, the sampleData are the monitoring data obtained by sensor；Using the Support vector regression algorithm including insensitive loss function, divideIt is other that the first eigenvector and the first newly-increased sample data are learnt, obtain the first prediction result and the second prediction knotFruit, wherein the first newly-increased sample data is the sample data increased newly in first predetermined period；According to first predictionAs a result the difference between the described first newly-increased sample data and second prediction result and the second newly-increased sample data itBetween difference, adjust the insensitive loss function, wherein the second newly-increased sample data is new in second predetermined periodThe sample data of increasing, described second predetermined period are later than described first predetermined period；Using including insensitive loss adjustedThe Support vector regression algorithm of function, learns the first eigenvector, obtains target prediction result.In this way, canTo verify according to the sample data increased newly after prediction result to the prediction result, and insensitive damage is adjusted according to verification resultThe value for losing function, to promote the accuracy of the prediction result of Support vector regression algorithm, so as to promote the earth's surface calamityThe accuracy of the prediction result of the data processing method of evil early warning.
Fig. 5 is referred to, Fig. 5 is the stream of the data processing method of another earth's surface disaster alarm provided in an embodiment of the present inventionCheng Tu, as shown in figure 5, method includes the following steps:
Step 501, using feature vector selection algorithm, choose first eigenvector from n initial sample datas, it is describedFirst eigenvector includes m sample data, wherein n is positive integer, and m is the integer less than n, and the sample data is to pass throughThe monitoring data that sensor obtains.
Step 502, using the Support vector regression algorithm including insensitive loss function, respectively to the fisrt featureVector sum first increases sample data newly and is learnt, and obtains the first prediction result and the second prediction result, wherein described first is newIncreasing sample data is the sample data increased newly in first predetermined period.
Optionally, described using the Support vector regression algorithm for including insensitive loss function, to the first newly-increased sampleBefore the step of data are learnt, the method also includes:
In the case that the quantity for the sample data for including in the described first newly-increased sample data is greater than preset value, using spyVector selection algorithm is levied, second feature vector is chosen from the described first newly-increased sample data, wherein the second feature vectorIn include sample data quantity be less than the described first newly-increased sample data in include sample data quantity；
It is described to use the Support vector regression algorithm including insensitive loss function, the first newly-increased sample data is carried outThe step of study, comprising:
Using the Support vector regression algorithm including insensitive loss function, the second feature vector is carried out onlineStudy, obtains the second prediction result.
Wherein, above-mentioned preset value, which can according to need, is configured, such as: it is greater than in the quantity of the first newly-increased sample dataIn the biggish situations of any numbers such as 200,500,1000, using feature vector selection algorithm from the first newly-increased sample numberAccording to middle selection second feature vector, wherein the sample size in the second feature vector is less than in the first newly-increased sample dataSample size.
In addition, as shown in Fig. 2, the quantity for the sample data for being included in the first newly-increased sample data be less than or equal to it is pre-If can be learnt point by point to each of the described first newly-increased sample data sample data in the case where value, in this way,It, can be with the accuracy of hoisting machine study in the case where the negligible amounts for the sample data that first newly-increased sample data is included.
In present embodiment, when the quantity of the first newly-increased sample data is too many, feature vector selection algorithm can be usedThe sample size of the first newly-increased sample data is reduced, to reduce in subsequent step using Support vector regression algorithmThe first newly-increased sample data is learnt and is verified and consumes a large amount of time.
Step 503, according to difference between first prediction result and the first newly-increased sample data and describedDifference between second prediction result and the second newly-increased sample data adjusts the insensitive loss function, wherein described secondNewly-increased sample data is the sample data increased newly in second predetermined period, and described second predetermined period is later than first predictionPeriod.
Optionally, above-mentioned steps 503 can with comprising the following specific steps
It obtains absolute between the sample data generated at first in first prediction result and the first newly-increased sampleError E 1；
It obtains absolute between the sample data generated at first in second prediction result and the second newly-increased sampleError E 2；
According to the difference DELTA E between E1 and E2, the insensitive loss function is adjusted, wherein if E1 is more than or equal toPreset error value, and Δ E indicates that E2 is greater than E1, then reduces the insensitive loss function.
Wherein, during obtaining newly-increased sample data, it may occur however that the failures such as network signal broken string, server crashSituation and the case where cause newly-increased sample data not obtain timely and cause newly-increased data accumulation.Above-mentioned will be caused in this wayOne newly-increased sample data and the second newly-increased sample data include a large amount of sample data.
In this way, being carried out to above-mentioned comprising the first newly-increased sample data of a large amount of sample data and the second newly-increased sample dataThe process that Support vector regression algorithm is learnt will devote a tremendous amount of time, to cause the warning data Processing AlgorithmHave big difference between the time that the time of the target prediction result obtained and actual conditions occur, and the effect of early warning is not achieved.
In addition, when the generation time of the sample data increased newly in first predetermined period is later than the generation of initial sample dataBetween.And above-mentioned first prediction result is the prediction result that the initial sample data according to first eigenvector, that is, after reducing obtains,And the first newly-increased sample is the sample increased newly in first predetermined period.In this way, including multiple sample numbers in the first newly-increased sampleIn the case where, comparison is carried out by the sample data that will be generated at first in the first newly-increased sample and the first prediction result,It may insure using immediate sample data is compared with first prediction result in time with the first prediction result, it canTo ensure the accuracy of verification result.
In addition, since the above-mentioned second newly-increased sample is the data increased newly in second predetermined period, and second predetermined periodIt is later than first predetermined period.And second prediction result be the prediction result obtained according to the first newly-increased sample data two.In this way,In the case that second newly-increased sample includes multiple sample datas, by the sample data that will be generated at first in the second newly-increased sample withSecond prediction result carries out comparison, it can be ensured that use and the second prediction result in time immediate sample data andSecond prediction result is compared, it can be ensured that the accuracy of verification result.
It include a large amount of sample data in the first newly-increased sample data or the second newly-increased sample data in present embodimentIn the case where, the reduction of sample size is carried out, to the first newly-increased sample data or the second newly-increased sample data to be promotedThe processing time for stating the data processing method of earth's surface disaster alarm, promote the efficiency of early warning.
Step 504, using the Support vector regression algorithm including insensitive loss function adjusted, to described firstFeature vector is learnt, and obtains target prediction result.
Step 505, according to the size relation between the target prediction result and preset threshold, determine disaster alarm knotFruit.
Wherein, above-mentioned preset threshold may include multiple, above-mentioned disaster alarm result also may include with it is above-mentioned multiple pre-If the one-to-one warning grade of threshold value, such as: when target prediction the result is that when the soil pressure force value of refuse dump, when the soil pressure force valueIn the case where greater than the first preset threshold, 1 grade of early warning of disaster alarm result is obtained；When the soil pressure force value is greater than the second default thresholdIn the case where value, 2 grades of early warning of disaster alarm is obtained as a result, the wherein value of the first preset threshold and the second preset threshold not phaseTogether.
In the embodiment of the present invention, by the way that target prediction result to be compared with preset threshold, to determine the target predictionAs a result the risk of generation disaster is indicated whether, to obtain more intuitive disaster alarm as a result, more convenient for monitoring personnelEasy knows the disaster alarm result.
Optionally, the data processing method of earth's surface disaster alarm provided in an embodiment of the present invention can be applied to refuse dump calamityEvil monitoring, specifically, the refuse dump disaster monitoring may comprise steps of:
Obtain the geographical monitoring figure of refuse dump；
Show disaster alarm interface, wherein the disaster alarm interface includes geographical distribution window and is shown in describedManage the warning information window in distribution window；
Wherein, the data processing method embodiment using the earth's surface disaster alarm is shown in the warning information windowIn obtained target prediction as a result, the geographical distribution window shows the geographical monitoring figure of the refuse dump, the geographyThe monitoring being arranged in a one-to-one correspondence with the installation site for the multiple monitoring sensors being arranged in the refuse dump is shown in monitoring figurePoint identification.
Wherein, above-mentioned geographical monitoring figure can be using monitoring camera, GPS satellite view, unmanned plane aerial view etc.Arbitrarily to the monitoring view of refuse dump.
The step of geographical monitoring figure of above-mentioned acquisition refuse dump, can be in such a way that webpage transmits, from being stored with the geographyThe server of monitoring figure obtains the implementation geography monitoring figure of refuse dump.
Wherein, above-mentioned warning information window can also be referred to as " newest warning information " window, can show in the windowUsing in the data processing method embodiment of earth's surface disaster alarm as described above obtained target prediction result and/or disaster it is pre-Alert result.
For example, above-mentioned target prediction result can be the calculating knot in newest warning information window 602 as shown in Figure 6Fruit, above-mentioned disaster alarm result can be comprehensive pre-warning grade in newest warning information window 602 as shown in Figure 6.
In addition, the multiple monitoring sensors being arranged in above-mentioned refuse dump may be respectively used for the surface displacement of detection refuse dumpInformation, internal displacement information, rainfall information, soil pressure force information, soil moisture content information, pore water pressure force information, temperature letterBreath and humidity information etc., detected data can be used for differentiating that whether there is or not the danger such as mud-rock flow, collapsing occur for the refuse dump.
In addition, above-mentioned monitoring point identification can use the figure of different appearances according to the type of corresponding monitoring sensorShape or text, in order to identify.
In the case where with above-mentioned danger, just it can be incited somebody to action using the data processing method of earth's surface disaster alarm as described aboveThe risky target prediction of tool is predicted as the result is shown in above-mentioned warning information window, is taken timely measure convenient for monitoring personnel pre-The generation to take precautions against calamities.
In present embodiment, by obtaining the geographical monitoring figure of refuse dump, monitoring personnel can be made to check refuse dump in timePresence states, convenient in time discovery calamity danger.In addition, pass through while showing geographical distribution window and warning information window,Be conducive to monitoring personnel while viewing the monitoring figure of warning information and scene, in addition, due to using the earth's surface disaster alarmThe obtained target prediction result of data processing method have the advantages that calculate that the time is short and accuracy is high, therefore, in this stepThe target prediction result being shown in above-mentioned warning information window equally has the advantages that fast response time and accuracy is high.
Optionally, as shown in fig. 6, above-mentioned geographical distribution window 601 is shown in bottom, above-mentioned warning information window 602 is aobviousIt is shown on above-mentioned geographical distribution window, and also shows multiple monitoring point identifications on above-mentioned geographical distribution window 601.It is detectingIn the case where the first operation for target monitoring point identification 604, mesh corresponding with the target monitoring point identification 604 is shownThe messagewindow 603 of mark monitoring sensor；
Wherein, the messagewindow 603 of the target monitoring sensor is shown on the geographical distribution window 601, describedThe messagewindow 603 of target monitoring sensor show the target monitoring sensor title, model and monitoring data andAt least one of the monitoring figure of the target monitoring sensor.
Wherein, above-mentioned monitoring data can be specific value detected by target monitoring sensor, such as: soil pressure prisonSurvey pressure value detected by sensor etc..
In the present embodiment, checks the data of the monitoring sensor of each monitoring point setting at any time convenient for monitoring personnel and showField monitoring view, is conducive to find faulty monitoring sensor in time, to debug in time, ensures refuse dump monitoring dataThe accuracy of processing method.
Fig. 7 is referred to, Fig. 7 is a kind of structure chart of warning data processing system provided in an embodiment of the present invention, the systemSystem 700 includes:
First chooses module 701, for using feature vector selection algorithm, chooses first from n initial sample datasFeature vector, the first eigenvector include m sample data, wherein n is positive integer, and m is the integer less than n, the sampleNotebook data is the monitoring data obtained by sensor；
First study module 702, it is right respectively for using the Support vector regression algorithm including insensitive loss functionThe first eigenvector and the first newly-increased sample data are learnt, and the first prediction result and the second prediction result are obtained,In, the first newly-increased sample data is the sample data increased newly in first predetermined period；
Module 703 is adjusted, for according to the difference between first prediction result and the first newly-increased sample data,And the difference between second prediction result and the second newly-increased sample data, adjust the insensitive loss function, whereinThe second newly-increased sample data is the sample data increased newly in second predetermined period, and described second predetermined period is later than describedFirst predetermined period；
Second study module 704, for being calculated using the Support vector regression for including insensitive loss function adjustedMethod learns the first eigenvector, obtains target prediction result.
Optionally, the system also includes:
Second chooses module, and the quantity of the sample data for including in the described first newly-increased sample data is greater than defaultIn the case where value, using feature vector selection algorithm, second feature vector is chosen from the described first newly-increased sample data,In, the quantity for the sample data for including in the second feature vector is less than the sample in the described first newly-increased sample data includedThe quantity of data；
First study module 702 is also used to using the Support vector regression algorithm including insensitive loss function,On-line study is carried out to the second feature vector, obtains the second prediction result.
Optionally, the adjustment module 703 includes:
First acquisition unit, for obtaining the sample generated at first in first prediction result and the described first newly-increased sampleAbsolute error E1 between notebook data；
Second acquisition unit, for obtaining the sample generated at first in second prediction result and the described second newly-increased sampleAbsolute error E2 between notebook data；
Adjustment unit, for adjusting the insensitive loss function, wherein if E1 according to the difference DELTA E between E1 and E2More than or equal to preset error value, and Δ E indicates that E2 is greater than E1, then reduces the insensitive loss function.
Optionally, the system also includes:
Determining module, for determining that disaster is pre- according to the size relation between the target prediction result and preset thresholdAlert result.
Each step in the data processing method embodiment of the earth's surface disaster alarm may be implemented in the embodiment of the present inventionSuddenly, and identical beneficial effect can be obtained, to avoid repeating, does not do extra repeat herein.
In several embodiments provided herein, it should be understood that disclosed method and system, it can be by otherMode realize.For example, system embodiment described above is only schematical, for example, the division of the unit, onlyFor a kind of logical function partition, there may be another division manner in actual implementation, such as multiple units or components can combineOr it is desirably integrated into another system, or some features can be ignored or not executed.Another point, shown or discussed phaseCoupling, direct-coupling or communication connection between mutually can be through some interfaces, the INDIRECT COUPLING or communication of device or unitConnection can be electrical property, mechanical or other forms.
It, can also be in addition, the functional units in various embodiments of the present invention may be integrated into one processing unitIt is that the independent physics of each unit includes, can also be integrated in one unit with two or more units.Above-mentioned integrated listMember both can take the form of hardware realization, can also realize in the form of hardware adds SFU software functional unit.
The above-mentioned integrated unit being realized in the form of SFU software functional unit can store and computer-readable deposit at oneIn storage media.Above-mentioned SFU software functional unit is stored in a storage medium, including some instructions are used so that a computerEquipment (can be personal computer, server or the network equipment etc.) executes information data described in each embodiment of the present inventionThe part steps of the processing method of block.And storage medium above-mentioned includes: USB flash disk, mobile hard disk, read-only memory (Read-OnlyMemory, abbreviation ROM), random access memory (Random Access Memory, abbreviation RAM), magnetic or disk etc. it is eachKind can store the medium of program code.
The above is a preferred embodiment of the present invention, it is noted that for those skilled in the artFor, without departing from the principles of the present invention, it can also make several improvements and retouch, these improvements and modificationsIt should be regarded as protection scope of the present invention.