CN114977176B - Power load decomposition method, device, equipment and storage medium - Google Patents
Power load decomposition method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN114977176B CN114977176B CN202210845292.3A CN202210845292A CN114977176B CN 114977176 B CN114977176 B CN 114977176B CN 202210845292 A CN202210845292 A CN 202210845292A CN 114977176 B CN114977176 B CN 114977176B
- Authority
- CN
- China
- Prior art keywords
- power
- load
- load decomposition
- model
- data
- 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
Images
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2310/00—The network for supplying or distributing electric power characterised by its spatial reach or by the load
- H02J2310/70—Load identification
-
- 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
- Y04S20/00—Management or operation of end-user stationary applications or the last stages of power distribution; Controlling, monitoring or operating thereof
- Y04S20/20—End-user application control systems
- Y04S20/242—Home appliances
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Power Engineering (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention belongs to the technical field of power electronics, and discloses a power load decomposition method, a device, equipment and a storage medium. The method comprises the following steps: acquiring sampled power data; constructing a load decomposition sub-model set according to the sampled power data; aggregating each load decomposition sub-model set to obtain a load decomposition model; detecting current power data; and inputting the current power data into the load decomposition model to obtain a power load decomposition result. Through the mode, the plurality of load models are constructed according to different sampled power data, and then the load models are aggregated to obtain the load decomposition model capable of detecting various power loads, so that the load decomposition of the power data is realized, the power load information of a user is successfully collected and analyzed, and the accuracy of analyzing the power utilization condition of the user is improved.
Description
Technical Field
The present invention relates to the field of power electronics technologies, and in particular, to a power load decomposition method, apparatus, device, and storage medium.
Background
The smart grid is the inevitable direction of development of future power systems, and both a huge main power system and a tiny smart home system are developed rapidly in recent years. The smart grid relies on modern communication, edge calculation and artificial intelligence technologies, and realizes safe, efficient, economic, reliable and environment-friendly operation of a power system through efficient sensing, measuring and communication equipment and perfect edge perception and intelligent decision capability.
The current intelligent power grid can only determine the electricity utilization habits of users through information such as electricity utilization time of power utilization power and the like, but the method is not accurate, and the specific electric appliances used by the users and the service conditions of the electric appliances cannot be determined, so that the power grid monitoring requirements which grow day by day cannot be met.
The above is only for the purpose of assisting understanding of the technical aspects of the present invention, and does not represent an admission that the above is prior art.
Disclosure of Invention
The invention mainly aims to provide a power load decomposition method, a power load decomposition device, power load decomposition equipment and a storage medium, and aims to solve the technical problem of how to perform load decomposition in the prior art.
To achieve the above object, the present invention provides a power load splitting method, comprising the steps of:
acquiring sampled power data;
constructing a load decomposition sub-model set according to the sampled power data;
aggregating each load decomposition submodel set to obtain a load decomposition model;
detecting current power data;
and inputting the current power data into the load decomposition model to obtain a power load decomposition result.
Optionally, the constructing a load decomposition sub-model set according to the sampled power data includes:
determining the active power and the total active power of a target electrical appliance according to the sampled power data;
constructing an input characteristic according to the total active power;
determining a training sample according to the input characteristics and the active power of the target electric appliance;
and respectively training the training samples to obtain a load decomposition submodel set.
Optionally, the determining the active power and the total active power of the target electrical appliance according to the sampled power data includes:
determining power bus voltage, power bus current, target electrical appliance voltage and target electrical appliance current according to the sampled power data;
determining total active power according to the power bus voltage and the power bus current;
and determining the active power of the target electrical appliance according to the voltage and the current of the target electrical appliance.
Optionally, the training samples respectively to obtain a set of load decomposition submodels includes:
classifying the training samples to obtain target training samples corresponding to a plurality of target electrical appliances;
inputting the target training sample into a preset initial load decomposition submodel to obtain a load decomposition submodel;
and determining a load decomposition submodel set according to each load decomposition submodel.
Optionally, the aggregating each load decomposition submodel set to obtain a load decomposition model includes:
acquiring basic training samples corresponding to each load decomposition submodel in the load decomposition submodel set and corresponding model output parameters;
matching according to the basic training sample and the model output parameters to obtain aggregate sample data;
and training a preset aggregation model according to the aggregation sample data to obtain a load decomposition model.
Optionally, after aggregating each set of load decomposition submodels to obtain a load decomposition model, the method further includes:
acquiring cross-region sampling power data;
determining a transfer learning sample according to the cross-region sampling power data;
training the load decomposition model according to the cross-region migration learning sample to obtain a cross-region load decomposition model;
detecting cross-regional power data;
and inputting the trans-regional power data into a trans-regional load decomposition model to obtain a trans-regional power load decomposition result.
Optionally, the determining a transfer learning sample according to the cross-region sampling power data includes:
determining the active power of the trans-regional electric appliance and the total active power of the trans-regional electric appliance according to the trans-regional sampled power data;
constructing a transfer training input characteristic according to the trans-regional total active power;
and determining a transfer learning sample according to the transfer training input characteristics and the cross-region electric appliance active power.
In order to achieve the above object, the present invention also provides a power load splitting apparatus including:
the acquisition module is used for acquiring sampled power data;
the processing module is used for constructing a load decomposition sub-model set according to the sampled power data;
the processing module is further used for aggregating each load decomposition sub-model set to obtain a load decomposition model;
the acquisition module is also used for detecting current power data;
and the processing module is also used for inputting the current power data into the load decomposition model to obtain a power load decomposition result.
Further, to achieve the above object, the present invention also proposes a power load splitting apparatus including: a memory, a processor and a power load splitting program stored on the memory and executable on the processor, the power load splitting program configured to implement the steps of the power load splitting method as described above.
Furthermore, to achieve the above object, the present invention also proposes a storage medium having stored thereon a power load splitting program which, when executed by a processor, implements the steps of the power load splitting method as described above.
The method comprises the steps of acquiring sampled power data; constructing a load decomposition sub-model set according to the sampled power data; aggregating each load decomposition sub-model set to obtain a load decomposition model; detecting current power data; and inputting the current power data into the load decomposition model to obtain a power load decomposition result. Through the mode, the load models are constructed according to different sampled power data, and then the load models are aggregated to obtain the load decomposition model capable of detecting various power loads, so that the load decomposition of the power data is realized, the power load information of a user is successfully collected and analyzed, and the accuracy of analyzing the power utilization condition of the user is improved.
Drawings
FIG. 1 is a schematic diagram of a power load splitting device of a hardware operating environment according to an embodiment of the present invention;
FIG. 2 is a schematic flow chart illustrating a power load splitting method according to a first embodiment of the present invention;
FIG. 3 is a flow chart illustrating a power load splitting method according to a second embodiment of the present invention;
FIG. 4 is a schematic diagram illustrating a load splitting flow of an embodiment of the power load splitting method of the present invention;
fig. 5 is a block diagram showing the structure of the power load splitting apparatus according to the first embodiment of the present invention.
The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Referring to fig. 1, fig. 1 is a schematic structural diagram of a power load splitting apparatus in a hardware operating environment according to an embodiment of the present invention.
As shown in fig. 1, the power load splitting apparatus may include: a processor 1001, such as a Central Processing Unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to implement connection communication among these components. The user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface, a wireless interface. The network interface 1004 may optionally include a standard wired interface, a Wireless interface (e.g., a Wireless-Fidelity (Wi-Fi) interface). The Memory 1005 may be a Random Access Memory (RAM) Memory, or may be a Non-Volatile Memory (NVM), such as a disk Memory. The memory 1005 may alternatively be a storage device separate from the processor 1001 described previously.
Those skilled in the art will appreciate that the configuration shown in fig. 1 does not constitute a limitation of the electrical load splitting apparatus and may include more or fewer components than those shown, or some components may be combined, or a different arrangement of components.
As shown in fig. 1, the memory 1005, which is a storage medium, may include therein an operating system, a network communication module, a user interface module, and a power load decomposition program.
In the power load splitting apparatus shown in fig. 1, the network interface 1004 is mainly used for data communication with a network server; the user interface 1003 is mainly used for data interaction with a user; the processor 1001 and the memory 1005 in the power load splitting apparatus of the present invention may be provided in the power load splitting apparatus which calls the power load splitting program stored in the memory 1005 by the processor 1001 and executes the power load splitting method provided by the embodiment of the present invention.
An embodiment of the present invention provides a power load splitting method, and referring to fig. 2, fig. 2 is a schematic flow chart of a first embodiment of the power load splitting method according to the present invention.
In this embodiment, the power load splitting method includes the steps of:
step S10: sampled power data is acquired.
It should be noted that, an execution subject of this embodiment is a smart grid management system, and the smart grid management system may be a system composed of a non-invasive electrical signal acquisition device and a data processing device, or may be other devices with the same or similar functions, which is not limited in this embodiment.
It can be understood that, this embodiment is applied to the background of the smart grid, and under the scene that there is a demand to analyze the power consumption habits and power consumption information of the user, the artificial intelligence analysis technology is combined to decompose and recognize the power consumption load of the user, this embodiment is mainly applied to provide a data processing basis for non-invasive conformity recognition, the non-invasive load recognition means that the load monitoring device is installed outside the factory building and the family of the user, usually on the household electricity meter, and the load monitoring device detects and recognizes the electrical load inside the user on the basis that the user does not need to know and grasp the electrical appliance type owned by the user. Since the invasive load recognition has a high installation difficulty and requires the cooperation of users, the non-invasive load recognition method draws more attention and obtains a lot of research results. The non-invasive load monitoring is one of key technologies for smart grid fine management, and a power grid operator who recognizes the non-invasive load can master and know the operation state of a power system at any time, monitor the power consumption behavior of families or factories in a district, arrange reasonable power dispatching planning and predict the power consumption condition of users in the district. By monitoring the information such as the type and the running state of the electric equipment of a user in real time, the non-invasive load monitoring technology provides an important reference basis for efficient electric energy dispatching and power grid structure optimization. Meanwhile, the result of load decomposition can help to master the power utilization structure and power utilization habits of the user, and important reference is provided for reasonable power utilization of the user and avoiding power waste. Therefore, the present embodiment proposes a legal system for non-invasive load decomposition by training a plurality of models and aggregating the models.
It should be noted that the sampled power data is power data detected by the electrical appliance for test through a non-invasive detection means, where the power data may include one or more sets of power data such as current information, voltage information, phase angle information, and the like in the power bus used by the user.
In addition, the load label corresponding to the sampled power data is the type of the electrical appliance for testing corresponding to the sampled power data, and the load label is associated with the sampled power data to provide a data sample for training of the load identification model.
Step S20: and constructing a load decomposition sub-model set according to the sampled power data.
It can be understood that the building of the load decomposition submodel set is mainly used for building a power load decomposition teacher model for subsequent knowledge distillation and transfer learning. The main function of the part is to construct a separate power load decomposition model for each electric appliance based on the real power consumption data of each available electric appliance and the total power data of the household electric meter. The sampled power data are classified according to the type of the electric appliance of the data source, the sampled power data are divided into different groups, and then a load decomposition sub-model for identifying the electric appliance is trained according to each group of the sampled power data. And summarizing the load decomposition submodels corresponding to the electrical appliances to form a sub-model conforming to decomposition. It should be noted that the training process may be trained by different sample data, for example: and training is directly carried out according to the sampled voltage and current, so that a load decomposition sub-model with complete functions can be generated.
In the embodiment, the active power and the total active power of the target electrical appliance are determined according to the sampled power data; constructing an input characteristic according to the total active power; determining a training sample according to the input characteristics and the active power of the target electrical appliance; and respectively training the training samples to obtain a load decomposition sub-model set.
In specific implementation, the total active power and the active power of a target electrical appliance can be calculated in an early stage, so that more ideal characteristics are extracted, and because the correlation of the load relation when the active power is in accordance with the electrical appliance is stronger, the power is used as the input characteristic of a model, compared with direct training through direct input of voltage and current, the training cost is reduced to a greater extent, and the accuracy of the training is improved.
In the present embodiment, a power bus voltage, a power bus current, a target appliance voltage, and a target appliance current are determined from the sampled power data; determining total active power according to the power bus voltage and the power bus current; and determining the active power of the target electrical appliance according to the voltage and the current of the target electrical appliance.
It should be noted that the active power calculation may be calculated by determining the power bus voltage, the power bus current, the target appliance voltage and the target appliance current according to the sampled power data, and only the voltage V, the current I and the phase angle for the training user of the load splitting model need to be obtained from the power block chain
The active power of the specific electric appliance and the total active power consumption on the power bus are calculated, and the calculation formula is as follows:
wherein, the first and the second end of the pipe are connected with each other,
and
the voltage and current values at the moment t on the power bus are respectively;
and
the current and voltage values of the appliance app.
The main task of this section is to construct a plurality of load decomposition submodels, which are power load decomposition teacher models for subsequent knowledge distillation and transfer learning. The main function of the part is to construct a separate power load decomposition model for each electric power appliance based on the real power consumption data of each available electric appliance and the total power data of the household electric meter. The preferred scheme for training the load decomposition submodel set is as follows:
subsequently, a power consumption decomposition model is built for the appliance app. The power consumption decomposition deep learning model takes a one-dimensional convolution kernel as a core to construct 5 layers of convolution layers, a full connection layer with 1024 hidden nodes is connected behind the convolution layers, and the output layer of the model is a model layer only containing one hidden node. For the sake of clarity, we document the power consumption decomposition model built for the appliance app. The loss function of the model is:
in the formula (I), the compound is shown in the specification,
in order to obtain the value of the loss,
and (4) predicting the app power at the time t for a load decomposition model built for the appliance app. Minimizing the loss function by using a standard deep learning model training process to finish the model
And (4) training.
And finally, repeating the process according to the types of the electric appliances needing to be detected, and constructing a load decomposition sub-model for each type of electric appliances to obtain
And a series of models are formed into a load decomposition sub-model set.
In this embodiment, the training samples are classified to obtain target training samples corresponding to a plurality of target electrical appliances; inputting the target training sample into a preset initial load decomposition submodel to obtain a load decomposition submodel; and determining a load decomposition submodel set according to each load decomposition submodel.
It should be noted that, in the process of load decomposition, the load condition of each target electrical appliance needs to be decomposed from the power data in the bus, so that data acquisition needs to be performed on multiple different electrical appliances, load recognition models for different electrical appliances are trained, and each electrical appliance has a corresponding load decomposition submodel, so that only training samples need to be classified according to types of the electrical appliances, and a preset initial load decomposition submodel needs to be trained respectively to obtain a plurality of final trained load decomposition submodels.
Step S30: and aggregating all the load decomposition submodels to obtain a load decomposition model.
It should be noted that, in the actual generation, the process of load decomposition can be realized by forming a plurality of load decomposition submodels for identifying each electrical appliance, and only the power data in the bus detected in real time needs to be sequentially input into each load decomposition submodel, but this method has two significant drawbacks. Firstly, the method needs to train a load decomposition model for each type of electric appliance, which causes the method to consume a large amount of time and computing resources in the training and testing stage, and the detection time and the occupied resources of the system are improved and the real-time performance of the system is greatly reduced due to the fact that a plurality of models need to be input during real-time detection. Secondly, the load recognition algorithm is trained and tested on the same data set, namely the load recognition algorithm is trained on data collected by the region A and is tested on the data of the region A, so that when the model is migrated across regions, a plurality of models need to be migrated and trained, and the model migration cost is greatly improved. Therefore, the load decomposition submodel set is aggregated and integrated into a complete load decomposition model which can be used for detecting the load conditions of various electrical appliances.
In the embodiment, basic training samples corresponding to each load decomposition submodel in the load decomposition submodel set and corresponding model output parameters are obtained; matching according to the basic training sample and the model output parameters to obtain aggregate sample data; and training a preset aggregation model according to the aggregation sample data to obtain a load decomposition model.
In a specific implementation, this embodiment proposes a model aggregation scheme, which is as follows: the main task of the polymerization is to polymerize the multiple load decomposition submodels obtained in the previous part into a unified load decomposition model by using an improved knowledge distillation technology. Firstly, the prediction outputs of a plurality of load decomposition submodels need to be integrated, and then a plurality of power consumption decomposition models are aggregated to construct a unified load decomposition model, wherein the model can realize the function of performing load decomposition on a plurality of electrical appliances by one model. Specifically, this section mainly includes the following specific operations and steps: training samples constructed in the sample construction process
Are respectively input into
In a series of models, get
Wherein, in the step (A),
and (4) outputting the prediction of the power of the corresponding appliance at the time t for the power consumption decomposition model constructed for the appliance type in the nth.
Secondly, will
As a prediction output of the aggregation model, will
As an input to the model, the aggregate model is trained. It is emphasized that the network size of the aggregate model F is consistent with the model structure constructed for a single appliance, the only difference being that the output of F is the power score of n types of appliancesSolve the predicted result, and
only the power split prediction results for the appliance app are output. The loss function for the polymerization model F is shown below:
the process is then trained using a standard deep learning model to minimize the loss function L F To target, the aggregation model F is trained.
In this embodiment, the aggregation sample data is input into a preset aggregation model to obtain a prediction result
And substituting the prediction output into a preset loss function to calculate a loss value, and adjusting a preset aggregation model according to the loss value until the preset aggregation model is converged to obtain a load decomposition model.
Step S40: current power data is detected.
It can be understood that after the training and aggregation of the model are completed, the power data of the target to be measured needs to be acquired in real time in the use stage. The acquisition mode can be directly collected from a user electric meter through non-invasive detection equipment, and the main task of the acquisition mode is to measure and collect the electric power data of a user. The main function of the part is to sample the current I and the voltage V on the power bus of the monitored user at low frequency by utilizing a measuring device arranged at a household electric meter; then, preprocessing the acquired power data; and finally, sending the processed data to the power block chain.
Furthermore, the collected power data can be preprocessed to improve the data accuracy, and the power operator collects the power data of the users in real time in a low-frequency sampling (the sampling frequency is 1Hz in this embodiment) mode through the measuring device installed on the household electricity meter of each user, wherein the power data comprises current I, voltage V, phase angle V and power data of the users
. Due to the large randomness and instability of the power system operation, the measured current I and voltage V data will jitter. For this purpose, the acquired data is processed in a moving average mode in the step, and data jitter is eliminated. The specific calculation method is as follows:
wherein Vt and It are processing results after eliminating jitter at time t; vi and Ii are original sampling results at the moment i; t is a parameter of the moving average algorithm for specifying how long a span of time data is used for moving average.
In addition, due to unstable working state of the measuring equipment, occasionally, the situation that data is not collected occurs. Aiming at the problem, two effective data before and after the moment of lacking numerical values are used for averaging in the step, and the data which are not collected are supplemented. And finally, removing invalid data and error data, and supplementing the removed data by using the difference method, so that the data quality is improved, and the data confusion are avoided. And uploading the processed voltage and current data to an electric power block chain, and performing chain winding operation on the data to realize persistent storage of the acquired data.
Step S50: and inputting the current power data into the load decomposition model to obtain a power load decomposition result.
It should be noted that, when the current power data is input into the load decomposition model, a power load decomposition result may be obtained, and the form of the power load decomposition result may be a plurality of types of electrical appliances and their corresponding load capacities (wattages), for example: the use condition of a kettle, a microwave oven, a refrigerator, a dish washing machine and a washing machine.
The embodiment acquires sampled power data; constructing a load decomposition sub-model set according to the sampled power data; aggregating each load decomposition sub-model set to obtain a load decomposition model; detecting current power data; and inputting the current power data into the load decomposition model to obtain a power load decomposition result. Through the mode, the plurality of load models are constructed according to different sampled power data, and then the load models are aggregated to obtain the load decomposition model capable of detecting various power loads, so that the load decomposition of the power data is realized, the power load information of a user is successfully collected and analyzed, and the accuracy of analyzing the power utilization condition of the user is improved.
Referring to fig. 3, fig. 3 is a schematic flow chart of a power load splitting method according to a second embodiment of the present invention.
Based on the first embodiment, before the step S10, the power load splitting method of the present embodiment further includes:
step S301: and acquiring cross-region sampling power data.
It should be noted that the data content of the cross-region sampling power data is consistent with the sampling power data, where the cross-region sampling power data is only the sampling power data of the target region obtained during model migration, and the model is optimized by the sampling power data of the target region, so that higher accuracy can be ensured when the model is used in the target region.
Step S302: and determining a transfer learning sample according to the cross-region sampling power data.
It should be noted that the sample run of the migration learning samples may be consistent with the training samples in the load split submodel. In this embodiment, the active power of the trans-regional electrical appliance and the total active power of the trans-regional electrical appliance are determined according to the trans-regional sampled power data; constructing a transfer training input characteristic according to the trans-regional total active power; and determining a transfer learning sample according to the transfer training input features and the active power of the cross-region electric appliance.
In a specific implementation, voltage V, current I and phase angle are sampled across a region of power data
The active power of the trans-regional electric appliance and the total active power of the trans-regional electric appliance are obtained through calculation, and the calculation formula is as follows:
step S303: and training the load decomposition model according to the cross-region migration learning sample to obtain a cross-region load decomposition model.
It should be noted that, the consistent decomposition model obtained after aggregating the models may already be put into formal use, but since the training samples are all from one region, there may be a certain bias in the trained model, for example: training on region a data, and testing on region B data, a significant degradation in algorithm performance occurs. Therefore, it is necessary to build a power load decomposition model that can cross-domain using migration learning. In this embodiment, the following preferred model migration scheme is proposed:
step S304: cross-region power data is detected.
It should be noted that the cross-region power data is power data corresponding to a target region where the smart grid management system is deployed in the use phase, and is used for performing power load decomposition on the target region.
Step S305: and inputting the cross-region power data into a cross-region load decomposition model to obtain a cross-region power load decomposition result.
When the power consumption information of each type of electric appliance is acquired from the total power consumption data of the user to be tested by using the non-invasive load decomposition model G. For the user needing to be detected, data on the power bus of the detected user is firstly obtained through the block chain. The data is processed according to the processing steps designed in the first part of data acquisition; then, an acquisition device externally connected to a user electric meter uploads the electric power data of the user to an electric power block chain; finally, the method provided by the embodiment obtains the to-be-detected data preprocessed on the power bus through the data search function of the power blockchainVoltage measurement
And current
And (6) data. On the data to be detected, a data processing method for feature extraction is used for constructing and obtaining the data to be detected
(ii) a As a modelled feature, the same approach as in training sample data, namely x, can be used t And inputting the data to be tested into the aggregated load decomposition model G for the active power at the time t to obtain the final power consumption decomposition result aiming at the N-type electrical appliances:
wherein the content of the first and second substances,
. Finally, training of the load decomposition submodel (classroom model), model aggregation and model migration are integrated, so that the effect that the model with multiple load decomposition recognition has high accuracy and convenient deployment can be achieved, and the total implementation flow is shown in fig. 4.
Firstly, selecting power consumers from other regions (different from the region where the consumers in the first part are located), repeating the operation in the sample construction corresponding to the load decomposition submodel, and constructing and obtaining the training sample for transfer learning
. And performing transfer learning on the aggregation model F obtained in the previous part by using the transfer learning training sample constructed in the previous step. The process of transfer learning comprises the following steps: using the polymerization model F vs. x t TL Predicting to obtain a prediction result P _ TL _ t of each power decomposition of the N electric appliances; calculating the true value from P _ TL _ t to electric appliance power consumption decomposition
The error loss of (2); carrying out gradient derivation on the error loss obtained by the last step of calculation, and using a standard deep learning optimization algorithm to carry out data analysis on the model F
Training is carried out; when training is over, the final model is denoted as G.
Wherein the content of the first and second substances,
and
the active power of the trans-regional electric appliance and the total active power of the trans-regional electric appliance are respectively, namely power data of a region where the model needs to be deployed are needed, the data type of the region is completely consistent with that of the load decomposition submodel, and only the regions for data sampling are different;
and
the current and voltage values of the appliance app of the target area. For app power consumption value of electric appliance at time t
The method includes the steps that input features xt of an electric appliance app during load decomposition at t moment are constructed to obtain a migration learning sample
。
Firstly, sample construction of training samples is needed, and app power consumption value of electric appliance at t moment
The prediction is performed using the sampled data 10minutes before time t. Based on this setting, we are directed to the load split of the appliance app at time tAnd constructing an input feature xt. In particular, the present invention relates to a method for producing,
wherein T =10minutes. Subsequently, we can get samples of appliance app at power consumption t instant for model training
. The feature construction part is repeated until all data are used for construction of the training sample.
The embodiment acquires cross-region sampling power data; determining a transfer learning sample according to the cross-region sampling power data; training the load decomposition model according to the cross-region migration learning sample to obtain a cross-region load decomposition model; detecting cross-region power data; and inputting the cross-region power data into a cross-region load decomposition model to obtain a cross-region power load decomposition result. By the mode, model migration of the model is achieved, the existing model is optimized according to the power data of the target area, accuracy of the model in cross-region load decomposition is improved, and the application range of the model is widened.
Furthermore, an embodiment of the present invention also provides a storage medium, on which a power load decomposition program is stored, which when executed by a processor implements the steps of the power load decomposition method as described above.
Referring to fig. 5, fig. 5 is a block diagram showing the structure of the first embodiment of the power load splitting apparatus according to the present invention.
As shown in fig. 5, the power load splitting apparatus according to the embodiment of the present invention includes:
and an obtaining module 10, configured to obtain the sampled power data.
And the processing module 20 is used for constructing a load decomposition sub-model set according to the sampled power data.
The processing module 20 is further configured to aggregate the load decomposition submodels to obtain a load decomposition model.
The obtaining module 10 is further configured to detect current power data.
The processing module 20 is further configured to input the current power data into the load decomposition model to obtain a power load decomposition result.
It should be understood that the above is only an example, and the technical solution of the present invention is not limited in any way, and in a specific application, a person skilled in the art may set the technical solution as needed, and the present invention is not limited thereto.
The embodiment obtains the module 10 and obtains the sampled power data; the processing module 20 constructs a load decomposition submodel set according to the sampled power data; the processing module 20 aggregates the load decomposition submodels to obtain a load decomposition model; the acquisition module 10 detects current power data; the processing module 20 inputs the current power data into the load decomposition model to obtain a power load decomposition result. Through the mode, the load models are constructed according to different sampled power data, and then the load models are aggregated to obtain the load decomposition model capable of detecting various power loads, so that the load decomposition of the power data is realized, the power load information of a user is successfully collected and analyzed, and the accuracy of analyzing the power utilization condition of the user is improved.
It should be noted that the above-mentioned work flows are only illustrative and do not limit the scope of the present invention, and in practical applications, those skilled in the art may select some or all of them according to actual needs to implement the purpose of the solution of the present embodiment, and the present invention is not limited herein.
In addition, the technical details that are not elaborated in this embodiment may refer to the power load splitting method provided in any embodiment of the present invention, and are not described herein again.
In this embodiment, the processing module is further configured to determine a target electrical appliance active power and a total active power according to the sampled power data;
constructing an input characteristic according to the total active power;
determining a training sample according to the input characteristics and the active power of the target electrical appliance;
and respectively training the training samples to obtain a load decomposition sub-model set.
In this embodiment, the processing module is further configured to
Determining power bus voltage, power bus current, target electrical appliance voltage and target electrical appliance current according to the sampled power data;
determining total active power according to the power bus voltage and the power bus current;
and determining the active power of the target electrical appliance according to the voltage and the current of the target electrical appliance.
In this embodiment, the processing module is further configured to classify the training samples to obtain target training samples corresponding to a plurality of target electrical appliances;
inputting the target training sample into a preset initial load decomposition submodel to obtain a load decomposition submodel;
and determining a load decomposition submodel set according to each load decomposition submodel.
In this embodiment, the processing module is further configured to obtain a basic training sample and a corresponding model output parameter corresponding to each load decomposition submodel in the load decomposition submodel set;
matching according to the basic training sample and the model output parameters to obtain aggregate sample data;
and training a preset aggregation model according to the aggregation sample data to obtain a load decomposition model.
In this embodiment, the processing module is further configured to acquire cross-region sampling power data;
determining a transfer learning sample according to the cross-region sampling power data;
training the load decomposition model according to the cross-region migration learning sample to obtain a cross-region load decomposition model;
detecting cross-regional power data;
and inputting the cross-region power data into a cross-region load decomposition model to obtain a cross-region power load decomposition result.
In this embodiment, the processing module is further configured to determine active power of a trans-regional electrical appliance and total active power of the trans-regional electrical appliance according to the trans-regional sampled power data;
constructing a transfer training input characteristic according to the trans-regional total active power;
and determining a transfer learning sample according to the transfer training input features and the active power of the cross-region electric appliance.
Further, it is to be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrases "comprising one of 8230; \8230;" 8230; "does not exclude the presence of additional like elements in a process, method, article, or system that comprises the element.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present invention or portions thereof that contribute to the prior art may be embodied in the form of a software product, where the computer software product is stored in a storage medium (e.g. a Read Only Memory (ROM)/RAM, a magnetic disk, and an optical disk), and includes several instructions for enabling a terminal device (which may be a mobile phone, a computer, a server, or a network device, etc.) to execute the method according to the embodiments of the present invention.
The above description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention, and all equivalent structures or equivalent processes performed by the present invention or directly or indirectly applied to other related technical fields are also included in the scope of the present invention.
Claims (8)
1. A power load splitting method, characterized by comprising:
acquiring sampled power data;
constructing a load decomposition sub-model set according to the sampled power data;
aggregating each load decomposition submodel set to obtain a load decomposition model;
detecting current power data;
inputting the current power data into the load decomposition model to obtain a power load decomposition result;
wherein, the constructing of the load decomposition submodel set according to the sampled power data comprises:
determining the active power and the total active power of a target electrical appliance according to the sampled power data;
constructing an input characteristic according to the total active power;
determining a training sample according to the input characteristics and the active power of the target electric appliance;
respectively training the training samples to obtain a load decomposition sub-model set;
wherein, the training samples respectively to obtain a set of load decomposition submodels comprises:
classifying the training samples to obtain target training samples corresponding to a plurality of target electrical appliances;
inputting the target training sample into a preset initial load decomposition submodel to obtain a load decomposition submodel, wherein the loss function of the training process is
In the formula (I), wherein,
in order to obtain the value of the loss,
the prediction result of the load decomposition submodel constructed aiming at a plurality of target electrical appliances to the power of each target electrical appliance at the time T, wherein T is a parameter of a moving average algorithm and is used for specifying the data in how long a time span is used for carrying out moving average, T is a certain time in a target training sample,
，
and
respectively the voltage and current values of the target appliance app at time t,
a phase angle for the target appliance app;
and determining a load decomposition submodel set according to each load decomposition submodel.
2. The method of claim 1, wherein determining a target appliance active power and a total active power from the sampled power data comprises:
determining power bus voltage, power bus current, target electrical appliance voltage and target electrical appliance current according to the sampled power data;
determining total active power according to the power bus voltage and the power bus current;
and determining the active power of the target electrical appliance according to the voltage and the current of the target electrical appliance.
3. The method of claim 1, wherein aggregating each set of load split sub-models to obtain a load split model comprises:
acquiring basic training samples corresponding to each load decomposition submodel in the load decomposition submodel set and corresponding model output parameters;
matching according to the basic training sample and the model output parameters to obtain aggregate sample data;
and training a preset aggregation model according to the aggregation sample data to obtain a load decomposition model.
4. The method of claim 1, wherein after aggregating each set of load split sub-models to obtain a load split model, further comprising:
acquiring cross-region sampling power data;
determining a transfer learning sample according to the cross-region sampling power data;
training the load decomposition model according to the cross-region migration learning sample to obtain a cross-region load decomposition model;
detecting cross-region power data;
and inputting the cross-region power data into a cross-region load decomposition model to obtain a cross-region power load decomposition result.
5. The method of claim 4, wherein determining a transfer learning sample from the cross-region sampled power data comprises:
determining the active power of a trans-regional electric appliance and the total active power of the trans-regional electric appliance according to the trans-regional sampled power data;
constructing a transfer training input characteristic according to the trans-regional total active power;
and determining a transfer learning sample according to the transfer training input characteristics and the cross-region electric appliance active power.
6. An electric load splitting apparatus, characterized by comprising:
the acquisition module is used for acquiring sampled power data;
the processing module is used for constructing a load decomposition submodel set according to the sampled power data;
the processing module is further used for aggregating each load decomposition sub-model set to obtain a load decomposition model;
the acquisition module is also used for detecting current power data;
the processing module is further configured to input the current power data into the load decomposition model to obtain a power load decomposition result;
the processing module is further used for determining the active power and the total active power of the target electrical appliance according to the sampled power data;
constructing an input characteristic according to the total active power;
determining a training sample according to the input characteristics and the active power of the target electrical appliance;
respectively training the training samples to obtain a load decomposition sub-model set;
the processing module is further used for classifying the training samples to obtain target training samples corresponding to a plurality of target electrical appliances;
inputting the target training sample into a preset initial load decomposition submodel to obtain a load decomposition submodel, wherein the loss function of the training process is
In the formula (I), wherein,
in order to obtain the value of the loss,
the prediction result of the app power of each target electrical appliance by the load decomposition submodel constructed aiming at the apps of the target electrical appliances at the time T, wherein T is a parameter of a moving average algorithm and is used for specifying the data in how long a time span is used for moving average, T is a certain time in a target training sample,
，
and
respectively the voltage and current values of the target appliance app at time t,
a phase angle for the target appliance app;
and determining a load decomposition submodel set according to each load decomposition submodel.
7. An electrical load splitting apparatus, characterized in that the apparatus comprises: a memory, a processor and a power load splitting program stored on the memory and executable on the processor, the power load splitting program being configured to implement the steps of the power load splitting method of any of claims 1 to 5.
8. A storage medium having stored thereon a power load splitting program which, when executed by a processor, implements the steps of the power load splitting method according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210845292.3A CN114977176B (en) | 2022-07-19 | 2022-07-19 | Power load decomposition method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210845292.3A CN114977176B (en) | 2022-07-19 | 2022-07-19 | Power load decomposition method, device, equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114977176A CN114977176A (en) | 2022-08-30 |
| CN114977176B true CN114977176B (en) | 2022-11-18 |
Family
ID=82969099
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210845292.3A Active CN114977176B (en) | 2022-07-19 | 2022-07-19 | Power load decomposition method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114977176B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116051910A (en) * | 2023-03-10 | 2023-05-02 | 深圳曼顿科技有限公司 | Non-invasive load identification method, device, terminal equipment and storage medium |
| CN116859169B (en) * | 2023-09-01 | 2023-11-17 | 国网浙江省电力有限公司余姚市供电公司 | Non-invasive power load decomposition method and system |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103577878A (en) * | 2013-08-30 | 2014-02-12 | 国家电网公司 | Load model parameter identification method based on clone selection algorithm |
| CN108879702A (en) * | 2018-06-21 | 2018-11-23 | 浙江大学 | A kind of power consumption control system based on household load decomposition |
| CN110309984A (en) * | 2019-07-09 | 2019-10-08 | 上海上塔软件开发有限公司 | A kind of identification of non-intrusion type load and Short-term user behavior prediction method |
| CN111193256A (en) * | 2020-01-07 | 2020-05-22 | 上海电力大学 | Power load prediction method based on variational modal decomposition and gated cyclic unit |
| CN111242161A (en) * | 2019-12-27 | 2020-06-05 | 国网山西省电力公司大同供电公司 | Non-invasive non-resident user load identification method based on intelligent learning |
| CN113537556A (en) * | 2021-06-03 | 2021-10-22 | 杭州电子科技大学 | Household short-term load prediction method based on state frequency memory network |
| CN113629701A (en) * | 2021-06-21 | 2021-11-09 | 国网山东省电力公司枣庄供电公司 | Power load decomposition method and system based on sequence-to-point model and transfer learning |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106124852A (en) * | 2016-06-13 | 2016-11-16 | 国网天津宝坻供电有限公司 | Non-intrusion type load monitor system |
| CN111242391B (en) * | 2020-03-06 | 2023-10-31 | 云南电网有限责任公司电力科学研究院 | Machine learning model training method and system for power load identification |
| CN111582610A (en) * | 2020-07-13 | 2020-08-25 | 清华四川能源互联网研究院 | Prediction method for family energy decomposition based on convolutional neural network |
| CN113033076A (en) * | 2021-03-02 | 2021-06-25 | 苏州纽克斯电源技术股份有限公司 | Non-invasive load monitoring method and device based on multi-chain decomposition method |
| CN113962314A (en) * | 2021-10-27 | 2022-01-21 | 南京富尔登科技发展有限公司 | Non-invasive enterprise load decomposition method based on federal learning |
| CN114545066A (en) * | 2022-01-11 | 2022-05-27 | 国网江苏省电力有限公司信息通信分公司 | Non-invasive load monitoring model polymerization method and system |
| CN114676883A (en) * | 2022-03-02 | 2022-06-28 | 深圳江行联加智能科技有限公司 | Power grid operation management method, device and equipment based on big data and storage medium |
-
2022
- 2022-07-19 CN CN202210845292.3A patent/CN114977176B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103577878A (en) * | 2013-08-30 | 2014-02-12 | 国家电网公司 | Load model parameter identification method based on clone selection algorithm |
| CN108879702A (en) * | 2018-06-21 | 2018-11-23 | 浙江大学 | A kind of power consumption control system based on household load decomposition |
| CN110309984A (en) * | 2019-07-09 | 2019-10-08 | 上海上塔软件开发有限公司 | A kind of identification of non-intrusion type load and Short-term user behavior prediction method |
| CN111242161A (en) * | 2019-12-27 | 2020-06-05 | 国网山西省电力公司大同供电公司 | Non-invasive non-resident user load identification method based on intelligent learning |
| CN111193256A (en) * | 2020-01-07 | 2020-05-22 | 上海电力大学 | Power load prediction method based on variational modal decomposition and gated cyclic unit |
| CN113537556A (en) * | 2021-06-03 | 2021-10-22 | 杭州电子科技大学 | Household short-term load prediction method based on state frequency memory network |
| CN113629701A (en) * | 2021-06-21 | 2021-11-09 | 国网山东省电力公司枣庄供电公司 | Power load decomposition method and system based on sequence-to-point model and transfer learning |
Non-Patent Citations (2)
| Title |
|---|
| Active/reactive power decomposition approaches to the AC optimal power flow problem;ByungKwon Park等;《2014 North American Power Symposium (NAPS)》;20140924;1-6页 * |
| 基于深度学习和迁移学习的公共楼宇非侵入式负荷分解;杨秀等;《电网技术》;20220331;第46卷(第3期);1160-1167页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114977176A (en) | 2022-08-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114977176B (en) | Power load decomposition method, device, equipment and storage medium | |
| Gopinath et al. | Energy management using non-intrusive load monitoring techniques–State-of-the-art and future research directions | |
| CN103020459B (en) | A kind of cognitive method of various dimensions electricity consumption behavior and system | |
| CN110335168B (en) | Method and system for optimizing power utilization information acquisition terminal fault prediction model based on GRU | |
| CN105184886A (en) | Cloud data center intelligence inspection system and cloud data center intelligence inspection method | |
| Humala et al. | Universalnilm: A semi-supervised energy disaggregation framework using general appliance models | |
| CN115330357A (en) | Intelligent stereo garage data management method, device, equipment and storage medium | |
| Ghosh et al. | Non-intrusive identification of harmonic polluting loads in a smart residential system | |
| CN113656461A (en) | Data processing method, device, electronic equipment and medium | |
| CN116681186B (en) | Power quality analysis method and device based on intelligent terminal | |
| Schirmer et al. | Improving energy disaggregation performance using appliance-driven sampling rates | |
| CN116861316B (en) | Electrical appliance monitoring method and device | |
| KR102285374B1 (en) | Method of automatically recognizing power pattern consumed for work based on artificial intelligence and system for the same | |
| CN116720983A (en) | Power supply equipment abnormality detection method and system based on big data analysis | |
| CN110807014A (en) | Cross validation based station data anomaly discrimination method and device | |
| CN111178679A (en) | Phase identification method based on clustering algorithm and network search | |
| CN116108376A (en) | Monitoring system and method for preventing electricity stealing, electronic equipment and medium | |
| CN113627621B (en) | Active learning method for optical network transmission quality regression estimation | |
| CN115018011B (en) | Power load type identification method, device, equipment and storage medium | |
| Kommey et al. | An artificial intelligence‐based non‐intrusive load monitoring of energy consumption in an electrical energy system using a modified K‐Nearest Neighbour algorithm | |
| CN113033076A (en) | Non-invasive load monitoring method and device based on multi-chain decomposition method | |
| US20200151554A1 (en) | Machine learning based model for spectral scan and analysis | |
| He et al. | Post-processing for event-based non-intrusive load monitoring | |
| CN112884201B (en) | Non-intrusive load monitoring method, device and system based on artificial intelligence algorithm | |
| CN117878925B (en) | Power transmission data control method and system of smart power grid |
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 |