The application requires in the right of priority that the application number that on August 27th, 2004 submitted to is 60/605,346, name is called the U.S. Provisional Patent Application of " MODEL ASSOCIATION IN FLEET MONITORING SYSTEM FOR LARGEPOWER PLANTS (model interaction in the computers group monitoring of big generating plant) ".
Embodiment
Fig. 1 is the exemplary environments 100 that can use system 102 according to an embodiment of the invention, as to be used to upgrade a plurality of monitoring models.Described environment 100 comprises a plurality of monitored system 104a, 104b 104c.Although show three so monitored systems clearly, it should be understood that environment 100 can comprise so monitored system more or still less in the embodiment that substitutes.More specifically, monitored system 104a, 104b, 104c can comprise electricity generation system, processing factory, multicompartment Medical Devices or other this system, it is characterized in that the synthetic operation of a plurality of or complex assemblies in the process that generates one or more measurable outputs in response to one or more measurable inputs.
Among monitored system 104a, 104b, the 104c each is all monitored by a plurality of sensor 106a, 106b, 106c respectively to illustrative.Each sensor 106a, 106b, 106c can comprise transducer, and this transducer response is in producing electric signal with the corresponding various physical phenomenons of the input and output of each system.For example, in the environment of generating, if each among monitored system 104a, 104b, the 104c all comprises the generating plant, then the output of sensor measurement not only comprises electric energy, and comprises it being other output of secondary product of inevitably generating electricity.For example other output can comprise temperature, pressure and the vibration of main electrification component (such as gas turbine, boiler, steam turbine and generator).In equivalent environment, for example the input of generating plant can comprise combustion gas, air and/or steam.
By a plurality of sensor 106a, 106b, the response signal that among the 106c each produced provides data or sensor vector, these data or sensor vector can be used to monitoring and detect monitored system 104a, 104b, the fault of 104c. the signal that can processes sensor produces is to produce quantifiable data. and for example, the signal that sensor produces can be digitized and be transformed to produce sensor vector by digital signal processor. comprise that other known technology of analog signal processing can be alternatively or additionally be used to produce corresponding to monitored system 104a, 104b, the quantifiable data of the operation of 104c.
Here, the sensor vector that obtains of the signal that is produced by sensor is that being illustrated property ground is described in the context that reasoning detects.Reasoning detects and makes being constructed to of estimation model essential, and this estimation model is carrying out modeling to the operation of monitored system 104a, 104b, 104c on the mathematics or on statistics.This estimation model provides the correlativity between each measured input and output of monitored system 104a, 104b, 104c.As one of ordinary skill will be understood, estimation model produces estimated value, and this estimated value can be compared with actual value to determine the tolerance interval of one or more residual errors and definite residual error.If the determined residual error of run duration drops on outside its tolerance interval in monitored system, represent fault so.
The model that can be used to this inference system comprise such as the standard regression model of least square and such as the different modification of nuclear regression model than new model with based on the model of neural network.From it is evident that this description, system 102 according to the present invention does not benefit from the characteristic limitations of the particular module of monitoring monitored system 104a, 104b, 104c.Regardless of employed particular module, the structure of this model is all finished during the training stage usually, in this training stage, raw data is used to " training " particular module so that produce the sensor estimator, as one of ordinary skill will be understood.During monitor stages subsequently, the sensing data that newly produces is input in the model of being trained like this or a plurality of model to detect the fault of the corresponding monitored system among a plurality of monitored system 104a, 104b, the 104c.
Be connected to system's 102 being illustrated property a plurality of sensors-system interface 108a, 108b108c, these a plurality of sensor-system interfaces are connected to monitoring a plurality of monitored system 104a, a plurality of sensor 106a of one of 104b, 104c, the special subset of 106b, 106c again separately.Illustrative ground, the signal that sensor produces offers among a plurality of monitored system 104a, 104b, the 104c corresponding one by each sensor among a plurality of sensor 106a, 106b, the 106c.It is the function of having described of quantifiable data that sensor- system interface 108a, 108b, 108c carry out conversion of signals.Therefore, sensor- system interface 108a, 108b, 108c can comprise one or more multiplexers, are used for the signal that multiplexed a plurality of sensor produces.According to another embodiment, sensor- system interface 108a, 108b 108c can comprise the digital signal processor that is used to handle the digitized signal that the signal that produced by sensor obtains.In the embodiment that substitutes, these signal processing functions are to carry out by being included in system's 102 interior elements own.In addition, alternatively, a plurality of sensor 106a, 106b, 106c can directly be connected to system 102.According among these various embodiment any one, the signal after the processing is used to construct aforesaid estimation model.
As schematically show, monitored system 104a, 104b, 104c are by system's 102 remote monitoring.Therefore, this system can be positioned at the position away from monitored system, for example, and the diagnostic center (not shown) place in the various systems that monitor away from it.Utilize remote monitoring, sensor vector (comprising the service data about monitored system 104a, 104b, 104c here) is sent to system 102 continuously, perhaps alternatively be sent to system 102 in the mode of sending in batch, wherein each batch comprises that such data, these data have been included in the performance of the monitored system of time durations since the last time of data sends in batch.Although system 102 is shown as the various monitored system 104a of remote monitoring, 104b, 104c, it should be understood that described system 102 alternatively can be included in a plurality of systems that independent monitored system place uses here.For example, the local system that uses can connect via data communication network, so that can work in coordination with local monitor.In addition, can be with the local a plurality of assemblies that use individual system to monitor monitored system of above-mentioned same way as.
Now additionally with reference to figure 2, schematically showing the specific embodiments of the system 102 that is used to upgrade a plurality of monitoring models. system comprises to 102 illustratives the model association module 202 that communicates each other, update module 204 and model modification module 206. are according to an embodiment, one or more modules in the described module realize with one or more dedicated, hardwired circuit that are used to carry out following each function. alternatively, one or more modules in the described module can realize to be arranged to the machine readable code of moving on general or specialized equipment. in another embodiment, one or more modules of described module realize with the combination of hard-wired circuit and machine readable code.
In the operation, among a plurality of monitored system 104a-c each, model association module 202 is determined monitored systems and the association between a plurality of estimation models of structure as mentioned above.Therefore, each special system among monitored system 104a, 104b, the 104c is associated with at least one this estimation model.Yet, one or more can being associated with more than one estimation model among monitored system 104a, 104b, the 104c by model association module.For example, a monitored system 104a can only be associated with recurrence pattern type.Another monitored system 104b can be with regression model, be associated based on the model of auto-associating neural network and/or nuclear regression model.Another monitored system 104c can only be associated with two this models.
In Fig. 3, schematically show the more general example of the model association scheme of carrying out by model association module 202 300.Described model association scheme 300 makes J monitored system S
1, S
2..., S
JWith K estimation model M
1, M
2..., M
KBe associated.As in this example describe first S of system
1Only by a model M
1Come modeling and therefore only with a model M
1Be associated.Second S of system
2With three different model M
1, M
2And M
3Be associated, although wherein be different models, these three models all are applicable to second S of system separately
2J the S of system
jWith second model M in K the model
2And K model be associated, and each model all provides the S of system
JDifferent modelings aspect.
It is evident that easily that from this example the model association scheme of being carried out by model association module 202 300 is enough general in to comprise various other possible combinations.Certainly, particular associative combination mainly is by the properties specify of monitored system and employed different models.
Further illustrate as the synoptic diagram of Fig. 3, each particular estimation model is based on a plurality of different estimation community set { x
1, x
2..., x
L}
T, { y
1, y
2..., y
K}
TAnd { z
1, z
2..., z
L}
TOne of.For example described estimation attribute can comprise sensor tabulation, sensor threshold value, cycle of training, estimation model algorithm and/or various algorithm parameter.These estimate that all or some combinations of the special estimation attribute in the attribute go for each in the different models.Therefore, though one or more can being suitable in the described estimation attribute to more than one model, but each estimates that community set is uniquely corresponding to particular estimation model.
Update module 204 is imported in response to the user and is upgraded one or more in the described estimation attribute.When estimating that attribute is updated, the estimation attribute that update module 204 then will be upgraded is sent to each estimation model corresponding to the different sets that comprises the estimation attribute that now has been updated.The estimation attribute that upgraded is replaced the preceding version of renewal in the set.
In case the estimation attribute of one or more renewals is sent to one or more estimation models by update module 204, this estimation model is corresponding at least one the unique estimation community set that comprises in the present estimation attribute that upgraded, and the estimation attribute after the renewal is sent to those estimation models that need be corrected or upgrade.Modification is carried out by model modification module 206, and it revises each estimation model corresponding to the different sets that comprises at least one the estimation attribute that was updated.As more specifically described as the following, in response to can " being trained again " to produce the sensor vector that new sensor produces subsequently to being comprised in the estimation model that was modified corresponding to the renewal of the one or more estimation attributes in the community set of estimation model.
In particular example, may think and be used to monitored system 104a, 104b, one or more estimation models that carry out modeling among the 104c are finished the work deficiently with respect to one of monitored system. and for example this may be because monitored system 104a, 104b, the variation of the foundation structure of one of 104c. on the contrary, the variation of system architecture or other environment may cause model to become being more suitable for inapplicable monitored system 104a before this model of monitoring, 104b, one of 104c. in addition, can develop new model for monitored system 104a, 104b, one or more uses among the 104c.
Therefore, according to another embodiment of the invention, described model association module 202 is arranged in response to the deletion of the interpolation of the variation of system architecture and other environment and new model or old model and upgrades association between monitored system and the associated estimation model.
Now additionally with reference to figure 4, in any given example, the association between M different system and N the estimation model can provide with fleet table 400 concisely.Described fleet table 400 may be implemented as M * N matrix, and wherein M is the integer corresponding to the number of monitored system greater than 1, and N is the integer that equals the number of the estimation model that is associated with different monitored systems.Association between i system and j the model is the i that is assigned with numerical value 1 by matrix, j unit's expression usually; When not having association, the i of matrix, j element is 0.Change 1 and 0 and easily revise fleet table 400 by upgrading association in response to environmental change (changing and/or the interpolation or the deletion of estimation model) such as system architecture along with model association module 202.
With reference now to Fig. 5,, the alternate embodiment that is used to upgrade the system 500 of a plurality of monitoring models comprises training module 508 in addition.Training module 508 is system's particular version of each monitored each estimation model of systematic training.Training is to utilize the sensing data that is produced by the sensor that communicates to connect special system to carry out, and wherein trains special system's particular version of estimation model for this special system.Communicate by letter to training module 508 illustratives with model association module 502, update module 504 and model modification module 506.
Described function before model association module 502, update module 504 and model modification module 506 are carried out separately.Therefore, In yet another embodiment, training module 508 can be arranged to other module in each system's particular version of each estimation model of being modified owing to the performed operation of other module with training again of each synthetic operation.
As Fig. 6 illustrative steps schematically showed, an alternative embodiment of the invention is to be used to upgrade the model association method 600 that the electronics of estimation model is realized.Each estimation model can comprise one of various models that are used for the detection of reasoning as has been described.Method 600 illustrative ground in step 602 comprises the association of determining between special monitored system and in a plurality of estimation model at least one.Each estimation model is based on one of a plurality of different estimation community sets, and each set is uniquely corresponding to particular estimation model.
Method 600 comprises in step 604 that additionally upgrading at least one estimates attribute.In step 606, method 600 comprises that further the estimation attribute that will upgrade is sent to each estimation model corresponding to the different sets that comprises at least one the estimation attribute that was updated.Method 600 comprises modification each estimation model corresponding to the different sets that comprises at least one the estimation attribute that was updated in step 608.Described method illustrative ground finishes in step 610.
According to another embodiment, method 600 can comprise at least one association of upgrading between monitored system and the associated estimation model.Additional step can be alternatively any point during being used for upgrading the process of a plurality of system monitoring models according to the step of having described carry out.
With reference now to the process flow diagram of Fig. 7,, shows the model association method 700 that realizes according to another embodiment electronics.Method 700 illustrative ground in step 702 comprises the association of determining between special monitored system and in a plurality of estimation model at least one.Each associated estimation is trained in step 704.Each model is trained for each system individually, and this model utilization comes modeling by the sensor vector that the sensor that communicates to connect special system provides.Each model definition of being trained like this is corresponding to system's particular version of the estimation model of system, and wherein this model is trained for this system.
Method 700 comprises also that in step 706 upgrading at least one estimates attribute.Method 700 comprises also that in step 708 the estimation attribute that will upgrade is sent to each estimation model corresponding to the different sets that comprises at least one the estimation attribute that was updated.Method 700 comprises modification each estimation model corresponding to the different sets that comprises at least one the estimation attribute that was updated in step 710.In step 712, method 700 comprises each system's particular version of each estimation model that training again has been modified.As training pattern, train described model based on the sensing data that produces by the sensor that communicates to connect special system again, wherein train special system's particular version of estimation model again for this special system.Described method illustrative ground finishes in step 712.
As described in the whole text, the present invention can realize with the combination of hardware, software or hardware and software.The present invention also can realize with centralized system in a computer system, and perhaps the distributed way that is distributed in the computer system of a plurality of interconnection with different elements realizes.The computer system or the miscellaneous equipment that are suitable for carrying out any kind of said method all are fit to.The typical combination of hardware and software can be the general-purpose computing system with computer program, and this computer program is controlled computer system when being loaded and be performed, so that it carries out method described herein.
The present invention can be embedded in the computer program, and this computer program comprises all features that can realize method described herein, and this computer program can be carried out these methods in being loaded into computer system the time.Computer program in current context means any expression of any language, code or the sign format of instruction set, described instruction set intention makes the system with information processing capability directly or in one of following operation or after both carry out special function, and described operation comprises: a) be converted to another kind of language, code or symbol; B) reproduce with the different materials form.
The present invention can realize with other form under the situation that does not break away from spirit of the present invention or base attribute.Therefore, should be with reference to the following claim of indication scope of the present invention, rather than with reference to above-mentioned instructions.