CN116308304B - New energy intelligent operation and maintenance method and system based on meta learning concept drift detection - Google Patents
New energy intelligent operation and maintenance method and system based on meta learning concept drift detection Download PDFInfo
- Publication number
- CN116308304B CN116308304B CN202310586561.3A CN202310586561A CN116308304B CN 116308304 B CN116308304 B CN 116308304B CN 202310586561 A CN202310586561 A CN 202310586561A CN 116308304 B CN116308304 B CN 116308304B
- Authority
- CN
- China
- Prior art keywords
- drift detection
- concept drift
- data
- meta
- learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 146
- 238000000034 method Methods 0.000 title claims abstract description 49
- 238000012423 maintenance Methods 0.000 title claims abstract description 45
- 238000012544 monitoring process Methods 0.000 claims abstract description 62
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 60
- 238000013507 mapping Methods 0.000 claims abstract description 17
- 230000002159 abnormal effect Effects 0.000 claims abstract description 9
- 238000012549 training Methods 0.000 claims description 35
- 239000011159 matrix material Substances 0.000 claims description 19
- 238000012545 processing Methods 0.000 claims description 17
- 230000006870 function Effects 0.000 claims description 14
- 101100455978 Arabidopsis thaliana MAM1 gene Proteins 0.000 claims description 10
- 238000007781 pre-processing Methods 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 7
- 230000000295 complement effect Effects 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 230000005856 abnormality Effects 0.000 claims description 5
- 238000010276 construction Methods 0.000 claims description 5
- 238000013528 artificial neural network Methods 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000013145 classification model Methods 0.000 claims description 2
- 238000000638 solvent extraction Methods 0.000 claims 1
- 230000008901 benefit Effects 0.000 abstract description 2
- 230000008859 change Effects 0.000 description 3
- 238000010248 power generation Methods 0.000 description 3
- 238000003745 diagnosis Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/20—Administration of product repair or maintenance
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- Marketing (AREA)
- General Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Water Supply & Treatment (AREA)
- Biophysics (AREA)
- Public Health (AREA)
- Operations Research (AREA)
- Entrepreneurship & Innovation (AREA)
- Primary Health Care (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Quality & Reliability (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention belongs to the field of intelligent operation and maintenance, and provides a new energy intelligent operation and maintenance method and system based on meta learning concept drift detection, which introduces a meta learning method, regards fault detection of each device as an independent meta learning task, and generalizes the concept drift detection to different devices; adopting a convolutional neural network model, and introducing a concept drift detection method to obtain a prediction result; and calculating the loss between the predicted result and the real result, judging whether the corresponding monitoring data are still in the same distribution according to the loss, if the corresponding monitoring data are still in the same distribution, the mapping relation between the current data and the influencing factors is not changed, the equipment does not have faults, and if the corresponding monitoring data are not in the same distribution, the mapping relation between the data and the influencing factors is changed, and the equipment has faults or anomalies. The method has the advantages that by detecting whether the concept drift phenomenon exists in the known data, the abnormal detection and fault prediction method of the equipment can be more accurate and reasonable.
Description
Technical Field
The invention belongs to the field of intelligent operation and maintenance, and particularly relates to a new energy intelligent operation and maintenance method and system based on meta learning concept drift detection.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
The existing operation and maintenance management of new energy power stations and solar photovoltaic power generation equipment mainly discovers equipment problems through daily manual inspection, operation and maintenance management and optimization are carried out through operation and maintenance decision making carried out manually, the management mode is relatively dependent on the skill level of operation and maintenance personnel, the current new energy belongs to an emerging industry, technical expert resources are deficient, the quality of the operation and maintenance decision making is uneven, a standard unified operation and maintenance management mode of the solar photovoltaic power generation equipment cannot be formed, and the manual operation and maintenance cost is high.
Most intelligent operation and maintenance schemes used in the current industry are based on operation data of a power station and a solar photovoltaic power generation equipment set in the form of flow data such as generated energy, current, voltage and temperature, diagnosis or early warning is carried out on a certain equipment or a certain type of defect fault, but the reasons such as data deficiency, insufficient sample size and insufficient talents of technical experts exist, experience and an operation and maintenance system cannot be organically combined and timeliness is poor, and most operation and maintenance use an early warning diagnosis model obtained through offline learning, performance is reduced along with time, the operation and maintenance system cannot keep accuracy constantly, and operation and maintenance quality cannot be guaranteed. In addition, when the solar photovoltaic equipment fails or the operation and maintenance system is abnormal, the mapping relation between the operation data and the influencing factors thereof can fluctuate, for example, when the solar photovoltaic equipment fails, the generating capacity of the generator set can be reduced for a long time and even interrupted, at the moment, the abnormal state of the equipment and the model accuracy for predicting the equipment failure can be outdated with time, the model is misjudged to be still normal after the operation state indexes of the photovoltaic equipment set such as generating capacity, current, voltage and temperature are abnormal for a long time, the phenomenon is called concept drift, and in sum, the concept drift is a difficult problem in the operation and maintenance of the new energy equipment.
Disclosure of Invention
In order to solve at least one technical problem in the background art, the invention provides a new energy intelligent operation and maintenance method and system based on meta learning concept drift detection, which enable the results of an abnormality detection and fault prediction method of equipment to be more accurate and reasonable by detecting whether the concept drift phenomenon exists in known data.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
the first aspect of the invention provides a new energy intelligent operation and maintenance method based on meta learning concept drift detection, which comprises the following steps:
dividing all the monitoring data of the new energy equipment to obtain a support set and a query set;
training to obtain an independent concept drift detection model of each device based on the support set, the query set and the concept drift detection model;
performing concept drift detection according to real-time new energy equipment monitoring data and the trained concept drift detection model, and if the concept drift exists, the equipment fails and overhauls;
the construction process of the concept drift detection model comprises the following steps:
introducing a meta learning method, regarding fault detection of each device as a single meta learning task, and generalizing concept drift detection to each different device; for each element learning task, adopting a convolutional neural network model, and introducing a concept drift detection method to obtain a prediction result and a concept drift detection threshold parameter at each moment;
and calculating the loss between the predicted result and the real result at each moment, judging whether the monitoring data of the corresponding equipment are still in the same distribution according to the loss and the conceptual drift detection threshold parameter, if so, not changing the mapping relation between the current monitoring data and the influencing factors of the current monitoring data, and if not, changing the mapping relation between the current monitoring data and the influencing factors of the current monitoring data, and if so, and if the equipment is in the failure or abnormality.
Further, for each element learning task, a convolutional neural network model is adopted, and a concept drift detection method is introduced to obtain a prediction result and a concept drift detection threshold parameter at each moment, and the method specifically comprises the following steps:
regarding the data set of each device as an independent meta-learning task, firstly performing internal circulation training, for each meta-learning task, performing one or more iterative training on a convolutional neural network by using a support set, using KL divergence with a sliding window as a concept drift detection method, and learning a concept drift detection threshold parameter and the convolutional neural network by using a meta-learning method MAML to obtain a loss function of a minimized support set; after the internal circulation training is completed, the convolutional neural network is evaluated by adopting a query set of the element learning task;
and carrying out iterative training on all the element learning tasks, updating the network parameters of the convolutional neural network and the conceptual drift detection threshold parameters in each iteration, and obtaining the network parameters of the convolutional neural network which are suitable for the distribution of each element learning task through multiple iterations.
Further, for each element learning task, training the convolutional neural network and the concept drift detection by using the support set for one or more iterations to obtain a loss function of the minimized support set, which specifically includes:
firstly, forward propagation is carried out on a convolutional neural network by using randomly initialized network parameters and threshold parameters, and a support set is transmitted to obtain a device fault prediction result;
and calculating loss based on the equipment failure prediction result and the real label, calculating gradient by using back propagation, and updating network parameters and conceptual drift detection threshold parameters of the convolutional neural network by using gradient descent so as to minimize a loss function of the support set.
Further, dividing all the monitoring data of the new energy equipment into a support set and a query set, including:
and storing the data in a time sequence form according to a minute or hour unit, generating a two-dimensional matrix of monitoring data of each device, preprocessing the data, taking the data in a period of time as one-time input by adopting a sliding window, taking each new time sequence two-dimensional matrix as a data sample of different devices, and dividing each new time sequence two-dimensional matrix into a support set and a query set.
Further, the preprocessing includes missing data complement processing and normalization processing.
Further, the length of the sliding window is smaller than the length of the support set and the query set.
Further, the monitoring data of the equipment which is on line again after maintenance is regarded as new meta-tasks to be re-transmitted into the concept drift detection model for training.
A second aspect of the present invention provides a new energy intelligent operation and maintenance system based on meta learning concept drift detection, comprising:
the data dividing module is used for dividing all the monitoring data of the new energy equipment into a support set and a query set;
the concept drift detection module is used for training to obtain independent concept drift detection models of all the devices based on the support set, the query set and the concept drift detection models;
performing concept drift detection according to real-time new energy equipment monitoring data and the trained concept drift detection model, and if the concept drift exists, the equipment fails and overhauls;
the construction process of the concept drift detection model comprises the following steps:
introducing a meta learning method, regarding fault detection of each device as a single meta learning task, and generalizing concept drift detection to each different device; for each element learning task, adopting a convolutional neural network model, and introducing a concept drift detection method to obtain a prediction result and a concept drift detection threshold parameter at each moment;
and calculating the loss between the predicted result and the real result at each moment, judging whether the monitoring data of the corresponding equipment are still in the same distribution according to the loss and the conceptual drift detection threshold parameter, if so, not changing the mapping relation between the current monitoring data and the influencing factors of the current monitoring data, and if not, changing the mapping relation between the current monitoring data and the influencing factors of the current monitoring data, and if so, and if the equipment is in the failure or abnormality.
Further, in the concept drift detection module, for each element learning task, a convolutional neural network model is adopted, and a concept drift detection method is introduced to obtain a prediction result and a concept drift detection threshold parameter at each moment, which specifically includes:
regarding the data set of each device as an independent meta-learning task, performing internal circulation training, for each meta-learning task, performing one or more iterative training on the convolutional neural network by using a support set, using KL divergence with a sliding window as a concept drift detection method, and learning the concept drift detection threshold parameters and the convolutional neural network by using a meta-learning method MAML to obtain a loss function and a concept drift detection threshold parameter of a minimized support set; after the internal circulation training is completed, the convolutional neural network is evaluated by adopting a query set of the element learning task;
and carrying out iterative training on all the element learning tasks, updating the network parameters of the convolutional neural network and the conceptual drift detection threshold parameters in each iteration, and obtaining the network parameters of the convolutional neural network which are suitable for the distribution of each element learning task through multiple iterations.
Further, in the data dividing module, the dividing the monitoring data of all the new energy devices into a support set and a query set includes:
and storing the data in a time sequence form according to a minute or hour unit, generating a two-dimensional matrix of monitoring data of each device, preprocessing the data, taking the data in a period of time as one-time input by adopting a sliding window, taking each new time sequence two-dimensional matrix as a data sample of different devices, and dividing each new time sequence two-dimensional matrix into a support set and a query set.
Compared with the prior art, the invention has the beneficial effects that:
the invention generalizes concept drift detection to various different devices for fault prediction by introducing a meta learning method MAML; the existence of the concept drift phenomenon gradually leads to the change of the data distribution through the form of the change of the data value, the data distribution is not identical to the old data, the input monitoring data is output with a fault prediction result by introducing a concept drift detection method and a convolution neural network model, the loss is calculated with a real result to judge whether the monitoring data is still identical to the distribution, if the monitoring data is still identical to the distribution through detection, the fact that the mapping relation between the current data and the influencing factors is unchanged is shown, the equipment does not fail, if the data does not belong to the identical distribution, the fact that the mapping relation between the data and the influencing factors is changed is shown, the equipment fails or is abnormal, and at the moment, an alarm is sent to prompt technicians to check and overhaul or replace components, so that the accuracy and the rationality of the fault prediction of the operation and maintenance system are improved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.
FIG. 1 is a flow chart of a method according to an embodiment of the present invention;
FIG. 2 is a flow chart of an operation and maintenance system with a conceptual drift detection model according to an embodiment of the present invention;
FIG. 3 is a schematic view of a sliding window according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a convolutional neural network according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of a conceptual drift detection model using MAML as employed in an embodiment of the present invention.
Detailed Description
The invention will be further described with reference to the drawings and examples.
It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present invention. As used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof.
Example 1
As shown in fig. 1-2, the present embodiment provides a new energy intelligent operation and maintenance method based on meta learning concept drift detection, which includes the following steps:
step 1: acquiring monitoring data of new energy equipment;
in step 1, the monitoring data of the new energy device is that the individual current, voltage, temperature, device switch state and the like of each device are obtained through the sensors of different devices such as a solar photovoltaic device assembly, an inverter, a combiner, a transformer and the like.
Step 2: processing the acquired monitoring data of the new energy equipment;
the processing process specifically comprises the following steps:
the monitoring data of the new energy equipment are stored in a time sequence mode according to a minute or hour unit, a two-dimensional matrix of each monitoring data is generated, the data are preprocessed, the data in a period of time are used as one-time input by adopting a sliding window, each new time sequence two-dimensional matrix is obtained as a data sample of different equipment, and a support set (containing manual marking information) and a query set (not containing manual marking information) are respectively divided.
In connection with fig. 4, for the used time series data, the convolutional neural network performs convolutional operation on the time series data by using two-dimensional convolution, and a full connection layer and a softmax layer are added to convert the output of the convolutional neural network for predicting and classifying normal and fault data into probability distribution so as to detect concept drift.
The convolution operation of the one-dimensional convolution layer can be expressed as the following formula:
wherein ,is the output->Personal characteristic value->Is convolution kernel->Weight value->Is +.f. in the sliding window input sequence>Data sample,/->Is a bias item->Is the size of the convolution kernel, +.>Is a ReLU activation function, and is output as an input value +.>And 0.
Wherein the manual marking information is fault and normal data determined according to historical data and professional knowledge of technicians.
As shown in fig. 3, a fixed length sliding Window length window_size is selected from the initial timeThe data samples of Window_Size moments are input backwards as the time sequence input of Window_Size moments, and the length Window_Size is selected: and setting the length Window_Size of the time Window for sliding and intercepting the data samples, and ensuring that the Window_Size is smaller than the lengths of the support set and the query set.
In this embodiment, the sliding time Window length window_size has a value between 50 and 100.
Wherein the preprocessing comprises missing data complement processing and normalization processing.
Specifically, the missing data complement processing specifically adopts average adjacent load to fill a default position;
the normalization processing uses a Min-Max method to normalize the data set after the missing data completion processing, and the calculation formula is as follows:
wherein the data is converted to [0,1 ]],For normalizing the processed data, +.>For raw data such as temperature, voltage, current, etc, +.>For maximum raw data, +.>Is the smallest raw data.
Step 3: initializing a meta learning concept drift detection model:
using a convolutional neural network as a concept drift detection basic prediction classification model to obtain KL divergence of the concept drift detection model, and randomly initializing network parameters comprising the convolutional neural network and element learning parameters of a concept drift detection model threshold value;
Specifically, the calculation formula of the conceptual drift detection model KL divergence is as follows:
wherein ,for the loss of all variables in the sliding window after classification by convolutional neural network, +.>Probability distribution for loss of sliding window at the previous moment,/->A probability distribution of the loss of the sliding window for the subsequent moment;
wherein, the meta-learning parametersThe convolution kernel size of the initialized convolution neural network can be randomly selected>Convolution kernel weight->Neural network learning rate, network structure, concept drift detection threshold, etc.
Step 4: introducing a meta learning method MAML, training a concept drift detection model by using the MAML, introducing the meta learning method, regarding fault detection of each device as a single meta learning task, and generalizing the concept drift detection to each different device; adopting a convolutional neural network model, and introducing a concept drift detection method to obtain a prediction result in a sliding window at each moment and a threshold value parameter obtained by a meta learning method;
and calculating the loss between the predicted result and the real result in the sliding window at each moment, judging whether the monitoring data of the corresponding equipment are still in the same distribution according to the loss and the threshold value parameter obtained by the meta-learning method, if so, the mapping relation between the current data and the influencing factors is not changed, the equipment is not in fault, and if not in the same distribution, the mapping relation between the data and the influencing factors is changed, and the equipment is in fault or abnormal.
The method for introducing meta learning regards the fault detection of each device as a single meta learning task and generalizes the concept drift detection to each different device; adopting a convolutional neural network model, and introducing a concept drift detection method to obtain a prediction result in a sliding window at each moment, wherein the method specifically comprises the following steps:
treating the data sets of the individual devices as separate meta-learning tasks, for each learning meta-task, usingSliding window, updating from the support set of the task through forward propagation, element learning parameters including network parameters of convolutional neural network and concept drift detection thresholdObtaining a device fault prediction classification result; then calculate the loss between the predicted result and the real label, calculate the gradient using the back propagation, update the meta-learning parameter using the gradient descent +.>。
Meta-learning parameters from support setsAnd carrying out classified prediction on the query set to obtain a device fault prediction result, then calculating loss between the prediction result and a real label, obtaining a KL divergence value through a conceptual drift detection model, and evaluating the performance of the convolutional neural network on the element learning task by using the loss and the KL divergence value. Computing total meta-learning loss using the performance of query sets of all meta-learning tasks and updating meta-learning parameters therewith>And obtaining a final concept drift detection training model.
As shown in fig. 5, the method specifically comprises the following steps:
step 4.1: performing inner loop training, for each metatask, performing one or more iterations of training using its support set, during which randomly initialized parameters are used firstForward propagation is carried out on a convolutional neural network and a concept drift detection model, a fault prediction result is obtained by introducing a support set, loss is calculated with a real label, a KL divergence value is obtained by using the concept drift detection model, a gradient is calculated by using backward propagation, and a meta learning parameter is calculated by using gradient descent>Updated to->To minimize the loss function of the support set;
the updated expression is:
wherein ,is a meta learning parameter, < >>The update step size can be fixed, can be obtained through meta-learning,is->Personal task->Uses meta-learning parameters on support set +.>Is>Is a gradient of (a).
Step 4.2: after the inner loop training is completed, the set of queries for the meta-task is used to evaluate the performance of the convolutional neural network. Meta-learning parameters using internal circulationForward propagation is carried out on the query set to obtain a result of fault prediction, and loss is calculated with the real result, and the loss function on the query set is used for calculating the lossTo evaluate the performance of convolutional neural networks and then update +.>;
The update expression of (2) is:
wherein ,for updating the step size, it can be fixed or can be obtained by meta-learning,/for updating the step size>Is from the total->The +.o. selected from the individual tasks>Personal task->,/>Is shown in the task->Network parameters used on the query set of (2)>Is a loss function of->Gradient that is the sum of the loss functions of all selected meta-tasks.
Step 4.3: iterative training of all meta-tasks, each timeIn iteration, meta-learning parametersAre updated, and finally, meta learning parameters obtained after a plurality of iterations are +.>The method is suitable for the distribution of each meta-task, and can be used for carrying out fault prediction on a new task to be tested.
Step 5: an operation and maintenance system with a conceptual drift detection model: and transmitting each monitoring data acquired from the sensor in real time into a concept drift detection model by using a sliding window and taking a data sample as a query set to obtain an independent concept drift detection model of each equipment component and outputting a respective KL divergence value and a concept drift threshold value, if the KL divergence value is larger than the concept drift threshold value, indicating that the concept drift occurs, namely, if the equipment is faulty or abnormal, sending an early warning signal to inform operation and maintenance personnel to carry out fault investigation and shutdown maintenance, and taking the monitoring data acquired by restarting the on-line equipment after the maintenance as a new meta-task to reuse the MAML to train the concept drift detection model of the task.
The technical advantage of the scheme is that when equipment fails or an operation and maintenance system is abnormal, the concept drift detection technology can detect the change of the state of the equipment and forecast the equipment failure so as to take operation and maintenance measures in time.
Example two
The embodiment provides a new energy intelligent operation and maintenance system based on meta learning concept drift detection, which comprises:
the data dividing module is used for dividing all the monitoring data of the new energy equipment into a support set and a query set;
the concept drift detection module is used for training to obtain independent concept drift detection models of all the devices based on the support set, the query set and the concept drift detection models;
performing concept drift detection according to real-time new energy equipment monitoring data and the trained concept drift detection model, and if the concept drift exists, the equipment fails and overhauls;
the construction process of the concept drift detection model comprises the following steps:
introducing a meta learning method, regarding fault detection of each device as an independent meta learning task, generalizing concept drift detection to different devices, adopting a convolutional neural network model for each meta learning task, and introducing a concept drift detection method to obtain a prediction result and a concept drift detection threshold parameter at each moment;
and calculating the loss between the predicted result and the real result at each moment, judging whether the monitoring data of the corresponding equipment are still in the same distribution according to the loss and the conceptual drift detection threshold parameter, if so, not changing the mapping relation between the current monitoring data and the influencing factors of the current monitoring data, and if not, changing the mapping relation between the current monitoring data and the influencing factors of the current monitoring data, and if so, and if the equipment is in the failure or abnormality.
In the concept drift detection module, for each element learning task, a convolutional neural network model is adopted, and a concept drift detection method is introduced to obtain a prediction result, which specifically comprises the following steps:
regarding the data set of each device as an independent meta-learning task, performing internal circulation training, for each meta-learning task, performing one or more iterative training on the convolutional neural network by using a support set, using KL divergence with a sliding window as a concept drift detection method, and learning the concept drift detection threshold parameters and the convolutional neural network by using a meta-learning method MAML to obtain a loss function and a concept drift detection threshold parameter of a minimized support set; after the internal circulation training is completed, the convolutional neural network is evaluated by adopting a query set of the element learning task;
and carrying out iterative training on all the element learning tasks, updating the network parameters of the convolutional neural network and the conceptual drift detection threshold parameters in each iteration, and obtaining the network parameters of the convolutional neural network which are suitable for the distribution of each element learning task through multiple iterations.
In the data dividing module, the dividing the monitoring data of all the new energy devices into a support set and a query set includes:
and storing the data in a time sequence form according to a minute or hour unit, generating a two-dimensional matrix of monitoring data of each device, preprocessing the data, taking the data in a period of time as one-time input by adopting a sliding window, taking each new time sequence two-dimensional matrix as a data sample of different devices, and dividing each new time sequence two-dimensional matrix into a support set and a query set.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (9)
1. The new energy intelligent operation and maintenance method based on meta learning concept drift detection is characterized by comprising the following steps of:
dividing all the monitoring data of the new energy equipment to obtain a support set and a query set;
training to obtain an independent concept drift detection model of each device based on the support set, the query set and the concept drift detection model;
performing concept drift detection according to real-time new energy equipment monitoring data and the trained concept drift detection model, and if the concept drift exists, the equipment fails and overhauls;
the construction process of the concept drift detection model comprises the following steps:
introducing a meta learning method, regarding fault detection of each device as an independent meta learning task, generalizing concept drift detection to different devices, adopting a convolutional neural network model for each meta learning task, and introducing a concept drift detection method to obtain a prediction result and a concept drift detection threshold parameter at each moment;
calculating the loss between the predicted result and the real result at each moment, judging whether the monitoring data of the corresponding equipment are still in the same distribution according to the loss and the conceptual drift detection threshold parameter, if so, the mapping relation between the current monitoring data and the influencing factors is not changed, the equipment is not in fault, and if not, the mapping relation between the current monitoring data and the influencing factors is changed, and the equipment is in fault or abnormal;
for each element learning task, a convolutional neural network model is adopted, and a concept drift detection method is introduced to obtain a prediction result and a concept drift detection threshold parameter at each moment, and the method specifically comprises the following steps:
regarding the data set of each device as an independent meta-learning task, performing internal circulation training, for each meta-learning task, performing one or more iterative training on the convolutional neural network by using a support set, using KL divergence with a sliding window as a concept drift detection method, and learning the concept drift detection threshold parameters and the convolutional neural network by using a meta-learning method MAML to obtain a loss function and a concept drift detection threshold parameter of a minimized support set; after the internal circulation training is completed, the convolutional neural network is evaluated by adopting a query set of the element learning task;
performing iterative training on all the element learning tasks, updating the network parameters of the convolutional neural network and the conceptual drift detection threshold parameters in each iteration, and obtaining the network parameters of the convolutional neural network which are suitable for the distribution of each element learning task through multiple iterations;
the obtained monitoring data of the new energy equipment are processed; the processing process specifically comprises the following steps:
storing monitoring data of new energy equipment in a time sequence form according to a minute or hour unit, generating a two-dimensional matrix of each monitoring data, preprocessing the data, taking data in a period of time as one-time input by adopting a sliding window, taking each obtained new time sequence two-dimensional matrix as a data sample of different equipment, and dividing a support set (containing manual marking information) and a query set (not containing manual marking information) respectively;
for the used time sequence data, the convolutional neural network carries out convolutional operation on the time sequence data by using two-dimensional convolution, and a full-connection layer and a softmax layer are added to convert the output of the convolutional neural network for predicting and classifying normal and fault data into probability distribution so as to detect concept drift;
the convolution operation of the one-dimensional convolution layer can be expressed as the following formula:
wherein ,is the output->Personal characteristic value->Is convolution kernel->Weight value->Is +.f. in the sliding window input sequence>Data sample,/->Is a bias item->Is the size of the convolution kernel, +.>Is a ReLU laserLiving function, output as input value +.>And 0;
wherein, the manual marking information is fault and normal data determined according to historical data and professional knowledge of technicians;
wherein, a sliding Window length Window_Size with a fixed length is selected from the initial momentThe data samples of Window_Size moments are input backwards as the time sequence input of Window_Size moments, and the length Window_Size is selected: setting the length Window_Size of a time Window for sliding interception of the data sample, and ensuring that the Window_Size is smaller than the lengths of the support set and the query set;
the sliding time Window length Window_Size has a value of 50 to 100;
the preprocessing comprises missing data complement processing and normalization processing;
specifically, the missing data complement processing specifically adopts average adjacent load to fill a default position;
the normalization processing uses a Min-Max method to normalize the data set after the missing data completion processing, and the calculation formula is as follows:
wherein the data is converted to [0,1 ]],For normalizing the processed data, +.>For raw data such as air temperature, voltage, current,for maximum raw data, +.>Is the smallest raw data;
initializing a meta learning concept drift detection model:
using a convolutional neural network as a concept drift detection basic prediction classification model to obtain KL divergence of the concept drift detection model, and randomly initializing network parameters comprising the convolutional neural network and element learning parameters of a concept drift detection model threshold value;
Specifically, the calculation formula of the conceptual drift detection model KL divergence is as follows:
wherein ,for the loss of all variables in the sliding window after classification by convolutional neural network, +.>Probability distribution for loss of sliding window at the previous moment,/->A probability distribution of the loss of the sliding window for the subsequent moment;
wherein, the meta-learning parametersThe convolution kernel size of the initialized convolution neural network can be randomly selected>Convolution kernel weight->Neural network learning rate, network structure, and concept drift detection threshold.
2. The new energy intelligent operation and maintenance method based on the meta learning concept drift detection according to claim 1, wherein for each meta learning task, training of one or more iterations of the convolutional neural network is performed by using a support set, specifically comprising:
firstly, forward propagation is carried out on a convolutional neural network by using randomly initialized network parameters, and a support set is transmitted to obtain a device fault prediction result;
based on the equipment failure prediction result and the real label, calculating the loss, calculating the gradient by using back propagation, and updating the network parameters of the convolutional neural network by using gradient descent so as to minimize the loss function of the support set.
3. The method for intelligent operation and maintenance of new energy based on meta learning concept drift detection of claim 1, wherein dividing all new energy device monitoring data into a support set and a query set comprises:
and storing the data in a time sequence form according to a minute or hour unit, generating a two-dimensional matrix of monitoring data of each device, preprocessing the data, taking the data in a period of time as one-time input by adopting a sliding window, taking each new time sequence two-dimensional matrix as a data sample of different devices, and dividing each new time sequence two-dimensional matrix into a support set and a query set.
4. The new energy intelligent operation and maintenance method based on meta learning concept drift detection as claimed in claim 3, wherein the preprocessing includes missing data complement processing and normalization processing.
5. The new energy intelligent operation and maintenance method based on meta learning concept drift detection of claim 3, wherein the length of the sliding window is smaller than the lengths of the support set and the query set.
6. The new energy intelligent operation and maintenance method based on meta-learning concept drift detection according to claim 1, wherein the monitoring data of the equipment which is on line again after maintenance is regarded as new meta-tasks to be re-transmitted into the concept drift detection model for training.
7. The new energy intelligent operation and maintenance system based on the meta learning concept drift detection is realized by the new energy intelligent operation and maintenance method based on the meta learning concept drift detection as set forth in claim 1, and is characterized by comprising the following steps:
the data dividing module is used for dividing all the monitoring data of the new energy equipment into a support set and a query set;
the concept drift detection module is used for training to obtain independent concept drift detection models of all the devices based on the support set, the query set and the concept drift detection models;
performing concept drift detection according to real-time new energy equipment monitoring data and the trained concept drift detection model, and if the concept drift exists, the equipment fails and overhauls;
the construction process of the concept drift detection model comprises the following steps:
introducing a meta learning method, regarding fault detection of each device as an independent meta learning task, generalizing concept drift detection to different devices, adopting a convolutional neural network model for each meta learning task, and introducing a concept drift detection method to obtain a prediction result and a concept drift detection threshold parameter at each moment;
and calculating the loss between the predicted result and the real result at each moment, judging whether the monitoring data of the corresponding equipment are still in the same distribution according to the loss and the conceptual drift detection threshold parameter, if so, not changing the mapping relation between the current monitoring data and the influencing factors of the current monitoring data, and if not, changing the mapping relation between the current monitoring data and the influencing factors of the current monitoring data, and if so, and if the equipment is in the failure or abnormality.
8. The intelligent operation and maintenance system for new energy based on meta-learning concept drift detection as set forth in claim 7, wherein in the concept drift detection module, for each meta-learning task, a convolutional neural network model is adopted, and a concept drift detection method is introduced to obtain a prediction result and a concept drift detection threshold parameter at each moment, specifically including:
regarding the data set of each device as an independent meta-learning task, performing internal circulation training, for each meta-learning task, performing one or more iterative training on the convolutional neural network by using a support set, using KL divergence with a sliding window as a concept drift detection method, and learning the concept drift detection threshold parameters and the convolutional neural network by using a meta-learning method MAML to obtain a loss function and a concept drift detection threshold parameter of a minimized support set; after the internal circulation training is completed, the convolutional neural network is evaluated by adopting a query set of the element learning task;
and carrying out iterative training on all the element learning tasks, updating the network parameters of the convolutional neural network and the conceptual drift detection threshold parameters in each iteration, and obtaining the network parameters of the convolutional neural network which are suitable for the distribution of each element learning task through multiple iterations.
9. The smart operation and maintenance system for new energy based on meta-learning concept drift detection of claim 7, wherein the data partitioning module divides all new energy device monitoring data into a support set and a query set, comprising:
and storing the data in a time sequence form according to a minute or hour unit, generating a two-dimensional matrix of monitoring data of each device, preprocessing the data, taking the data in a period of time as one-time input by adopting a sliding window, taking each new time sequence two-dimensional matrix as a data sample of different devices, and dividing each new time sequence two-dimensional matrix into a support set and a query set.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310586561.3A CN116308304B (en) | 2023-05-24 | 2023-05-24 | New energy intelligent operation and maintenance method and system based on meta learning concept drift detection |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310586561.3A CN116308304B (en) | 2023-05-24 | 2023-05-24 | New energy intelligent operation and maintenance method and system based on meta learning concept drift detection |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116308304A CN116308304A (en) | 2023-06-23 |
CN116308304B true CN116308304B (en) | 2023-08-25 |
Family
ID=86829128
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310586561.3A Active CN116308304B (en) | 2023-05-24 | 2023-05-24 | New energy intelligent operation and maintenance method and system based on meta learning concept drift detection |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116308304B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116991137B (en) * | 2023-07-04 | 2024-03-26 | 山东省计算中心(国家超级计算济南中心) | Concept drift-oriented industrial control system abnormality detection method capable of adapting to interpretation |
CN117249873B (en) * | 2023-11-20 | 2024-01-30 | 精智未来(广州)智能科技有限公司 | Quality monitoring method and equipment for gas molecular analysis |
CN117556365B (en) * | 2024-01-12 | 2024-04-09 | 山东建筑大学 | Method and device for training process data drift detection model and drift detection method and device |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111325112A (en) * | 2020-01-31 | 2020-06-23 | 贵州大学 | Cutter wear state monitoring method based on depth gate control circulation unit neural network |
CN114065613A (en) * | 2021-10-27 | 2022-02-18 | 中国华能集团清洁能源技术研究院有限公司 | Multi-working-condition process industrial fault detection and diagnosis method based on deep migration learning |
CN114422450A (en) * | 2022-01-21 | 2022-04-29 | 中国人民解放军国防科技大学 | Network flow analysis method and device based on multi-source network flow data |
CN114895656A (en) * | 2022-06-20 | 2022-08-12 | 河海大学常州校区 | Industrial Internet of things equipment fault diagnosis system capable of adaptively triggering incremental learning |
CN115169704A (en) * | 2022-07-13 | 2022-10-11 | 广东电网有限责任公司 | CVT error state prediction method and device based on increment integrated learning model |
CN115409154A (en) * | 2022-08-08 | 2022-11-29 | 上海大学 | Concept drift active detection method integrating meta-learning and pre-training |
CN115513951A (en) * | 2022-11-17 | 2022-12-23 | 山东建筑大学 | Power load prediction method and system based on concept drift detection |
CN115577237A (en) * | 2022-09-08 | 2023-01-06 | 沈阳工业大学 | Elevator fault diagnosis method based on one-dimensional convolutional neural network and meta-learning |
CN115758156A (en) * | 2022-11-25 | 2023-03-07 | 桂林电子科技大学 | Business process anomaly detection method based on concept drift discovery |
-
2023
- 2023-05-24 CN CN202310586561.3A patent/CN116308304B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111325112A (en) * | 2020-01-31 | 2020-06-23 | 贵州大学 | Cutter wear state monitoring method based on depth gate control circulation unit neural network |
CN114065613A (en) * | 2021-10-27 | 2022-02-18 | 中国华能集团清洁能源技术研究院有限公司 | Multi-working-condition process industrial fault detection and diagnosis method based on deep migration learning |
CN114422450A (en) * | 2022-01-21 | 2022-04-29 | 中国人民解放军国防科技大学 | Network flow analysis method and device based on multi-source network flow data |
CN114895656A (en) * | 2022-06-20 | 2022-08-12 | 河海大学常州校区 | Industrial Internet of things equipment fault diagnosis system capable of adaptively triggering incremental learning |
CN115169704A (en) * | 2022-07-13 | 2022-10-11 | 广东电网有限责任公司 | CVT error state prediction method and device based on increment integrated learning model |
CN115409154A (en) * | 2022-08-08 | 2022-11-29 | 上海大学 | Concept drift active detection method integrating meta-learning and pre-training |
CN115577237A (en) * | 2022-09-08 | 2023-01-06 | 沈阳工业大学 | Elevator fault diagnosis method based on one-dimensional convolutional neural network and meta-learning |
CN115513951A (en) * | 2022-11-17 | 2022-12-23 | 山东建筑大学 | Power load prediction method and system based on concept drift detection |
CN115758156A (en) * | 2022-11-25 | 2023-03-07 | 桂林电子科技大学 | Business process anomaly detection method based on concept drift discovery |
Non-Patent Citations (1)
Title |
---|
一种基于原型学习的自适应概念漂移分类方法;苏静;裘晓峰;李书芳;刘道伟;张春红;;北京邮电大学学报(第03期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN116308304A (en) | 2023-06-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN116308304B (en) | New energy intelligent operation and maintenance method and system based on meta learning concept drift detection | |
CN113156917B (en) | Power grid equipment fault diagnosis method and system based on artificial intelligence | |
CN112817280A (en) | Implementation method for intelligent monitoring alarm system of thermal power plant | |
CN117176560B (en) | Monitoring equipment supervision system and method based on Internet of things | |
CN107807860B (en) | Power failure analysis method and system based on matrix decomposition | |
CN116937575A (en) | Energy monitoring management system for grid system | |
CN116614177B (en) | Optical fiber state multidimensional parameter monitoring system | |
CN117764422B (en) | Intelligent energy-saving operation and maintenance management cloud platform | |
CN117408162B (en) | Power grid fault control method based on digital twin | |
CN114579659A (en) | System and method for estimating and sorting utilization potential of power battery by gradient utilization | |
CN117113157B (en) | Platform district power consumption fault detection system based on artificial intelligence | |
CN115600695B (en) | Fault diagnosis method for metering equipment | |
CN112801815B (en) | Power communication network fault early warning method based on federal learning | |
KR102411915B1 (en) | System and method for froviding real time monitering and ai diagnosing abnormality sign for facilities and equipments | |
Shan et al. | Root Cause Analysis of Failures for Power Communication Network Based on CNN | |
CN118114101B (en) | Fuel cell fault diagnosis method and system | |
CN114414938B (en) | Dynamic response method and system for power distribution network faults | |
CN111832832B (en) | District self-inspection system based on thing networking | |
CN118096131B (en) | Operation and maintenance inspection method based on electric power scene model | |
CN117216673B (en) | Current transformer monitoring evaluation overhauls platform | |
CN118101429B (en) | Power distribution network communication fault diagnosis method and system | |
CN118673456A (en) | Abnormality detection method based on real-time data of distributed photovoltaic power station | |
CN118656272A (en) | Equipment operation process monitoring system | |
CN118568439A (en) | Equipment state diagnosis method, system and medium based on multi-source data mining | |
CN117272042A (en) | Photovoltaic power station monitoring method and device, electronic equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |