Non-intrusive residential user electric quantity decomposition method and system based on clustering
By combining the Kmeans clustering algorithm and the convolutional neural network model, it solves the problem of difficulty in accurately decomposing electrical appliances' power consumption in the complex residential power environment in the existing technology, achieving higher decomposition accuracy and more effective noise interference. reduce.
Patent Information
- Application Number
- CN202411159487.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-22
- Publication Date
- 2025-01-21
AI Technical Summary
The existing non-invasive load monitoring technology is difficult to accurately identify and decompose the electricity consumption of each electrical equipment in complex residential electricity environments, and the decomposition accuracy often decreases in scenarios where multiple electrical appliances operate simultaneously.
A sequence-to-point convolutional neural network model based on the Kmeans clustering algorithm is used, and the Kmeans clustering algorithm and the convolutional neural network are combined to decompose the electrical power of residential users through clustering and deep learning technology.
It significantly improves the accuracy of the power consumption decomposition of residential users' electrical equipment, reduces the interference of similar electrical appliance noise during model training, and can more effectively handle complex residential power consumption environments.
Smart Images
Figure CN119337078A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power load monitoring, and in particular to a cluster-based non-intrusive residential user power decomposition method and system. Background Art
[0002] With the continuous development of smart grids and the continuous promotion of smart sensor technology, power big data analysis has gradually become a hot topic for experts and scholars in the power industry. On the distribution side of the power grid, since residential user equipment has the characteristics of multiple load components, complex operating characteristics and various influencing factors, how to obtain the power consumption of load equipment from the load operation data of residential users, promote the benign interaction between power companies and users, optimize the operation of the power grid while protecting the interests of both parties is the focus of the power industry. Among them, non-intrusive load monitoring NILM technology refers to installing only one sensor at the user's entrance, monitoring the power consumption and working status of each or each type of equipment in the house by collecting and analyzing the user's total port data, so as to understand the power consumption of each device inside the user. The non-intrusive load monitoring (NILM) technology is of great significance for optimizing grid planning, operation and management and saving electricity and electricity bills for users. It is also one of the key technologies to promote refined management on the demand side and energy conservation and emission reduction. Therefore, the non-intrusive load monitoring (NILM) technology has received widespread attention since it was proposed. Currently, many methods have been applied to non-intrusive load decomposition, and the popular deep learning theory in recent years has become an important tool for realizing the non-intrusive load monitoring (NILM) technology. Among them, the convolutional neural network model can effectively mine the internal characteristics of various complex equipment, and obtain the window power sequence of a single target equipment based on the continuously sliding total power window sequence, showing good results in the field of load decomposition.
[0003] However, the current common solutions have many shortcomings, including: since the power information of residential users' electrical equipment is difficult to capture, and conventional deep learning models are easily interfered by other electrical noise during training, traditional non-intrusive load monitoring (NILM) technology often relies on device feature extraction, such as switching features and harmonic features. It is difficult to accurately identify and decompose the power consumption of various electrical equipment in the face of complex residential power consumption environments; due to mutual interference and superposition effects between different electrical appliances, non-intrusive load monitoring (NILM) technology often suffers from a decrease in decomposition accuracy when dealing with complex scenarios where multiple electrical appliances are running at the same time. Summary of the invention
[0004] In view of the fact that when decomposing the power consumption of residential electrical appliances, the power information of residential electrical equipment is difficult to capture in the existing technology, and the conventional deep learning model is easily interfered by the noise of other electrical appliances during the training process, the non-intrusive load monitoring (NILM) technology is difficult to accurately identify and decompose the power consumption of various electrical equipment in the face of complex residential power consumption environments, and the decomposition accuracy often decreases when dealing with complex scenarios where multiple electrical appliances are running at the same time. Therefore, the present invention is proposed.
[0005] Therefore, the problem to be solved by the present invention is how to provide a sequence-to-point convolutional neural network model based on the Kmeans clustering algorithm, combine the Kmeans clustering algorithm with the convolutional neural network, and solve the method of deploying and applying the non-intrusive load monitoring NILM technology at the residential user end.
[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:
[0007] In the first aspect, an embodiment of the present invention provides a clustering-based non-intrusive residential user electricity decomposition method, which includes collecting the total electricity data of original residential users and the total electricity signal data of target residential users and aggregating them to construct an original user set; using the original user set to train the residential user's electrical appliance electricity decomposition structure to construct an electrical appliance electricity decomposition model; based on the electrical appliance electricity decomposition model, decomposing the total electricity signal data of the target residential user to obtain an electricity decomposition sequence of the target electrical equipment; using a confidence evaluation model to evaluate the electricity decomposition sequence of the target electrical equipment to obtain a confidence score and optimize the electricity decomposition sequence of the target electrical equipment to achieve residential user electricity decomposition.
[0008] As a preferred solution of the clustering-based non-intrusive residential user electricity decomposition method described in the present invention, wherein: the original user set is composed of labeled total electricity data of residential users; the construction of the electrical appliance electricity decomposition model includes the following steps: the labeled total electricity data of residential users is segmented based on the ratio of B1:B2:B3 to obtain training data, verification data and test data; in the supervised training and verification process, the network parameters are updated in a supervised manner to construct the electrical appliance electricity decomposition model.
[0009] As a preferred solution of the clustering-based non-intrusive residential user electricity decomposition method of the present invention, the calculation formula of the electrical appliance electricity decomposition model is as follows:
[0010]
[0011]
[0012] in, is the corresponding power value in the power sequence window of the mth load device; F p is the parameter set of the model; Y t:t+W-1 is the sliding window of the total power sequence; t is the time; W is the length of the sliding window of the total power sequence; ε is the W-dimensional Gaussian random noise; τ is the middle moment of the sequence window.
[0013] As a preferred solution of the clustering-based non-intrusive residential user electricity decomposition method of the present invention, the calculation formula of the loss function in the supervised training process is as follows:
[0014]
[0015] Among them, L p is the loss obtained from model training; W is the length of the sliding window of the total power sequence; T is the total length of the input total power sequence; p is the probability of the output of the electrical appliance power decomposition model and the input of the electrical appliance power decomposition model; is the power sequence window of the mth load device; Y t:t+W-1 is the sliding window of the total power sequence; t is the time; θ p is the network parameter.
[0016] As a preferred solution of the clustering-based non-intrusive residential user electricity decomposition method described in the present invention, the method of obtaining the electricity decomposition sequence of the target electrical equipment includes the following steps: inputting the window sequence of the target user's total electricity signal data into the electrical equipment electricity decomposition model to obtain the electricity value of a single electrical equipment; as the input window of the electrical equipment electricity decomposition model continues to slide, the electricity value of the target electrical equipment window midpoint is combined into the electricity decomposition sequence of the target electrical equipment.
[0017] As a preferred solution of the clustering-based non-intrusive residential user electricity decomposition method described in the present invention, the aggregation includes the following steps: using the original residential user total electricity data and the target residential user total electricity signal data to calculate, to obtain the average hourly electricity value and the average daily electricity value of all residential users; normalizing the average hourly electricity value of hour A1 and the average daily electricity value of day A2 to construct a feature set; clustering the feature set using the Kmeans clustering algorithm to obtain multiple classes composed of similar users, and constructing an original user set similar to the target user.
[0018] As a preferred solution of the clustering-based non-intrusive residential user electricity decomposition method of the present invention, the calculation formula of the normalization processing is as follows:
[0019]
[0020] Among them, x tis the actual measured value of the electric quantity series; x is the average value of the electric quantity series; σ is the standard deviation of the electric quantity series.
[0021] In the second aspect, in order to further solve the safety problems existing in power load monitoring, the present invention provides a clustering-based non-intrusive residential user power decomposition system, which includes: a user clustering module, which is used to collect the total power data of original residential users and the total power signal data of target residential users, and perform calculations, normalization and clustering to construct a set of original users similar to the target users; a model construction module, which is used to divide the original user set into training data, verification data and test data according to a proportion, and update the network parameters in a supervised manner during the supervised training and verification process to construct an electrical appliance power decomposition model; a data decomposition module, which is used to input the window sequence of the total power signal data of the target user into the electrical appliance power decomposition model and combine the power values to obtain the power decomposition sequence of the target electrical equipment; a data optimization module, which is used to evaluate the power decomposition sequence of the target electrical equipment using a confidence evaluation model to obtain a confidence score, and optimize the power decomposition sequence of the target electrical equipment using the confidence score.
[0022] In a third aspect, an embodiment of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the clustering-based non-intrusive residential user electricity decomposition method as described in the first aspect of the present invention is implemented.
[0023] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the clustering-based non-intrusive residential user electricity decomposition method as described in the first aspect of the present invention is implemented.
[0024] The beneficial effects of the present invention are as follows: in view of the problem that the existing non-intrusive load monitoring NILM model is difficult to accurately decompose the power consumption of residential users' electrical appliances, the present invention designs a clustering-based non-intrusive residential user power decomposition method, combines the Kmeans clustering algorithm with the convolutional neural network, greatly reduces the interference degree of similar electrical appliance noise during the model training process, provides effective support for the electrical appliance power decomposition of residential user loads on the distribution network side, and improves the accuracy of the power consumption decomposition of residential users' electrical equipment. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:
[0026] Figure 1 This is a flow chart of the clustering-based non-intrusive residential user electricity decomposition method in Example 1.
[0027] Figure 2 Schematic diagram of the neural network structure of the electrical appliance power decomposition model in Example 1. DETAILED DESCRIPTION
[0028] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.
[0029] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0030] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.
[0031] Example 1
[0032] Reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a non-intrusive residential user electricity decomposition method based on clustering, including the following steps:
[0033] S1: Collect the total electricity data of original residential users and the total electricity signal data of target residential users and aggregate them to construct the original user set.
[0034] Specifically, Figure 1The figure shows the process of the clustering-based non-invasive residential user electricity decomposition method. The total electricity data of the residential users is obtained by combining the original user hourly and daily average electricity values with the target user hourly and daily average electricity values. The model is then supervised trained and verified based on the total electricity data of the residential users and the target electrical equipment label electricity data to build an electricity decomposition model for the target equipment. The total electricity signal of the residential users is then used to obtain the target equipment electricity window sequence, and finally the decomposed target equipment electricity value is obtained.
[0035] Preferably, the aggregation includes the following steps: using the original residential users' total electricity data and the target residential users' total electricity signal data to perform calculations to obtain average hourly electricity values and average daily electricity values for all residential users.
[0036] The average hourly electricity value of 24 hours and the average daily electricity value of 7 days are normalized to construct a feature set, where the normalization is the electricity consumption minus the average value and divided by the standard deviation.
[0037] The Kmeans clustering algorithm is used to cluster the feature set to obtain multiple clusters composed of similar users, and to construct an original user set similar to the target user.
[0038] Specifically, the calculation formula for normalization is as follows:
[0039]
[0040] Among them, x t is the actual measured value of the power sequence. The power sequence includes the total power sequence and the power sequence of the target electrical equipment. The sequence value of the power sequence is obtained by measuring the equipment. is the mean value of the electric quantity series; σ is the standard deviation of the electric quantity series.
[0041] Preferably, by calculating the average hourly electricity value and the average daily electricity value and normalizing them, a more comprehensive feature set is constructed, which effectively captures the time characteristics of the user's electricity consumption pattern, eliminates the differences in electricity consumption levels between different users, and makes the clustering results more accurate; the Kmeans clustering algorithm is used to further enhance the grouping effect of similar users, provide a high-quality data foundation for subsequent model training, improve the generalization ability of the model, effectively reduce the complexity of electricity decomposition calculation, and improve the efficiency and scope of application of the algorithm.
[0042] S2: Use the original user set to train the household electrical appliance power decomposition structure and build an electrical appliance power decomposition model.
[0043] Preferably, the original user set is composed of tagged total electricity data of residential users.
[0044] Specifically, building an electrical appliance power decomposition model includes the following steps: segmenting the labeled total power data of residential users based on a ratio of 3:1:1 to obtain training data, verification data, and test data.
[0045] like Figure 2 The figure shows the neural network structure of the electrical appliance power decomposition model. During the supervised training and verification process, the network parameters are updated in a supervised manner to construct the electrical appliance power decomposition model. The input of the electrical appliance power decomposition model is the total power signal data window sequence of the target user, and the output of the electrical appliance power decomposition model is the power value of the corresponding electrical device at the midpoint of the corresponding window sequence.
[0046] Specifically, the calculation formula of the electrical appliance power decomposition model is as follows:
[0047]
[0048] in, is the corresponding power value in the mth load device power sequence window, that is, the output of the electrical appliance power decomposition model; F p is the parameter set of the model; Y t:t+W-1 is the sliding window of the total power sequence, i.e., the input of the electrical appliance power decomposition model; t is the time; W is the length of the sliding window of the total power sequence; ε is the W-dimensional Gaussian random noise; τ is the middle moment of the sequence window.
[0049] Furthermore, the loss function in the supervised training process is calculated as follows:
[0050]
[0051] Among them, L p is the loss obtained from model training; W is the length of the sliding window of the total power sequence; T is the total length of the input total power sequence, and the total power sequence value Y t:t+W-1 When inputting into the electrical appliance power decomposition model, zero values with a length of [(W-1) / 2] are added at both ends of the total power sequence; p is the probability of the output of the electrical appliance power decomposition model and the input of the electrical appliance power decomposition model; is the mth load equipment power sequence window, that is, the output of the electrical equipment power decomposition model; Y t:t+W-1 is the sliding window of the total power sequence, i.e., the input of the electrical appliance power decomposition model; t is the time; θ p is the network parameter.
[0052] Specifically, since the present invention adopts a sequence-to-point convolutional neural network, the input of the network is a window sequence of total power signal data of the target residential user load node, and the output of the network is the power value corresponding to the midpoint of the corresponding window, that is, the label. Therefore, in the supervised training and verification process, the network parameters are updated in a supervised manner.
[0053] Preferably, unlike the fully connected network or simple convolutional network used in the prior art, which has difficulty in effectively capturing the long-term dependencies and local features of the electricity data, the present invention adopts a sequence-to-point convolutional neural network structure, which can effectively capture the temporal features and contextual information of the electricity data, and by introducing Gaussian random noise, improves the robustness and generalization ability of the model, and can better cope with data fluctuations in practical applications; the design of the loss function takes multiple aspects into consideration, effectively balances the accuracy and generalization ability of the model, and ensures the physical rationality of the decomposition results.
[0054] S3: Decompose the total power signal data of the target residential users based on the electrical appliance power decomposition model to obtain the power decomposition sequence of the target electrical equipment.
[0055] Preferably, obtaining the power decomposition sequence of the target electrical device includes the following steps: inputting the window sequence of the total power signal data of the target user into the electrical power decomposition model to obtain the power value of a single electrical device.
[0056] As the input window of the electrical appliance power decomposition model continues to slide, the power values at the midpoints of the target electrical appliance windows are combined into a power decomposition sequence of the target electrical appliance.
[0057] Preferably, by combining the sliding window technology with the sequence-to-point convolutional neural network, it can not only process variable-length sequence data, but also effectively utilize contextual information to improve the prediction accuracy of the model. Compared with the traditional fixed window method, by continuously sliding the input window and combining the power values, the continuous prediction of the power of the target electrical equipment is achieved, which ensures the temporal consistency and accuracy of the prediction results and provides more reliable data support for energy management and power consumption behavior analysis.
[0058] S4: Use the confidence assessment model to evaluate the power decomposition sequence of the target electrical equipment, obtain the confidence score and optimize the power decomposition sequence of the target electrical equipment to achieve power decomposition of residential users.
[0059] Preferably, the confidence assessment model includes the following steps: performing difference calculation based on the power decomposition sequence of the target electrical equipment, analyzing the difference between the output of the electrical power decomposition model and the actual value in a single load device, and the specific formula is as follows:
[0060]
[0061] Where D is the difference between the output of the electrical power decomposition model and the actual value in a single load device; is the corresponding power value in the power sequence window of the mth load device, that is, the output of the electrical power decomposition model; is the actual power value of the mth load device.
[0062] The Gaussian distribution is used to calculate the probability of the difference between the output of the electrical appliance power decomposition model and the actual value in a single load device, and the probabilities of all load devices are summed to obtain the probability of the difference between the output of the electrical appliance power decomposition model and the actual value in all load devices. The specific formula is as follows:
[0063]
[0064] Where P is the probability of the difference between the output of the electrical power decomposition model and the actual value in all load devices; N is the total number of load devices; σ m is the standard deviation of the power decomposition sequence of the mth target electrical equipment; is the corresponding power value in the power sequence window of the mth load device, that is, the output of the electrical power decomposition model; is the actual power value of the mth load device.
[0065] Based on the time interval [0, T], the probability of the difference between the output of the electrical power decomposition model and the actual value in all load devices is integrated to obtain the confidence score.
[0066] Specifically, the calculation formula of the confidence assessment model is as follows:
[0067]
[0068] Among them, C m is the confidence score, i.e., the confidence of the corresponding power value in the power sequence window of the mth load device; T is the total time; N is the total number of load devices; σ m is the standard deviation of the power decomposition sequence of the mth target electrical equipment; is the corresponding power value in the power sequence window of the mth load device, that is, the output of the electrical power decomposition model; is the actual power value of the mth load device; t is the time variable.
[0069] Preferably, optimizing the power decomposition sequence of the target electrical equipment includes: introducing the average power value based on the confidence score and the power decomposition sequence of the target electrical equipment to provide a stable benchmark for the uncertainty of the model prediction, thereby ensuring the stability of the data reliability; at the same time, the introduction of the confidence score can enhance the reliability of the calculation results, avoid errors caused by abnormal individual data points, and obtain the optimized power decomposition sequence of the target electrical equipment.
[0070] Specifically, the calculation formula of the optimized power decomposition sequence of the target electrical equipment is as follows:
[0071]
[0072] Among them, Y m The power decomposition sequence of the optimized target electrical equipment; is the corresponding power value in the mth load device power sequence window, that is, the output of the electrical appliance power decomposition model; C m is the confidence score, i.e., the confidence of the corresponding power value in the power sequence window of the mth load device; It is the average power value corresponding to the power sequence window of the mth load device.
[0073] If the confidence score C m < the first threshold, it is determined that the credibility of the power decomposition sequence of the target electrical equipment is low, the weight of the average power value is increased, and the power decomposition sequence of the target electrical equipment is optimized using the confidence score to obtain the optimized power decomposition sequence of the target electrical equipment.
[0074] If the confidence score C m ≥ the first threshold and ≤ the second threshold, the credibility of the power decomposition sequence of the target electrical equipment is determined to be medium, the weight of the output of the electrical power decomposition model is increased, and the power decomposition sequence of the target electrical equipment is optimized using the confidence score to obtain the optimized power decomposition sequence of the target electrical equipment.
[0075] If the confidence score C m > the second threshold, it is determined that the reliability of the power decomposition sequence of the target electrical equipment is high, and there is no need to optimize the power decomposition sequence of the target electrical equipment.
[0076] Specifically, for determining whether the first and second thresholds for determining whether the power decomposition sequence of the target electrical equipment is optimized, power data of different users and different time periods are collected and tested, and the obtained confidence scores are statistically analyzed and a distribution graph is drawn. Based on the distribution characteristics, appropriate data are selected as the initial first and second thresholds, and the thresholds are continuously adjusted according to the results in subsequent tests, with the goal of retaining data with higher confidence and better processing data with low confidence, so as to obtain the final first and second thresholds.
[0077] Preferably, by introducing a confidence assessment model to calculate the difference between the model output and the actual value, and using Gaussian distribution for probability calculation, not only can the reliability of the decomposition result be evaluated, but the optimization strategy can also be dynamically adjusted according to the confidence level. By introducing the average power value as a stable benchmark, the problem of insufficient uncertainty prediction in the existing technology is effectively solved, the credibility and accuracy of the power decomposition results are improved, and a more reliable basis is provided for energy management decisions.
[0078] The present embodiment also provides a clustering-based non-intrusive residential user electricity decomposition system, including: a user clustering module, which is used to collect the total electricity data of original residential users and the total electricity signal data of target residential users and perform calculations, normalization processing and clustering to construct a set of original users similar to the target users; a model construction module, which is used to divide the original user set into training data, verification data and test data according to a proportion, and update network parameters in a supervised manner during supervised training and verification to construct an electrical appliance electricity decomposition model; a data decomposition module, which is used to input the window sequence of the total electricity signal data of the target user into the electrical appliance electricity decomposition model and combine the electricity values to obtain the electricity decomposition sequence of the target electrical equipment; a data optimization module, which is used to evaluate the electricity decomposition sequence of the target electrical equipment using a confidence evaluation model to obtain a confidence score, and optimize the electricity decomposition sequence of the target electrical equipment using the confidence score.
[0079] This embodiment also provides a computer device, which is applicable to the case of a clustering-based non-intrusive residential user electricity decomposition method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the clustering-based non-intrusive residential user electricity decomposition method proposed in the above embodiment.
[0080] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.
[0081] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the non-intrusive residential user electricity decomposition method based on clustering as proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, magnetic disk or optical disk.
[0082] In summary, in order to address the problem that the existing non-intrusive load monitoring NILM model is difficult to accurately decompose the power consumption of residential users' electrical appliances, the present invention designs a clustering-based non-intrusive residential user power decomposition method, which combines the Kmeans clustering algorithm and the convolutional neural network, greatly reduces the interference of similar electrical appliance noise during the model training process, provides effective support for the electrical appliance power decomposition of residential user loads on the distribution network side, and improves the accuracy of the power consumption decomposition of residential users' electrical equipment.
[0083] Example 2
[0084] Referring to Table 1, which is a second embodiment of the present invention, this embodiment is different from the first embodiment in that, in order to verify its beneficial effects, operating data and related instructions of the present invention in an actual environment are provided.
[0085] This example uses 9 residential users in a community in northern China as samples, deploys 9 measuring devices to collect the total user and internal electrical appliance power data, divides all data into training data, verification data and test data in a ratio of 3:1:1, and selects 4 target electrical appliances, including washing machines, refrigerators, dishwashers and televisions, and uses the electrical appliance power decomposition model proposed in the present invention to perform power decomposition; first, the parameters of the electrical appliance power decomposition model are initialized, and the normalized data is input into the network structure for training and the electrical appliance power decomposition model is obtained; secondly, the trained electrical appliance power decomposition model is tested, and the total power sequence of the test data is input into the trained electrical appliance power decomposition model, so as to perform power decomposition on the target electrical appliances in the test data.
[0086] For the above-obtained power decomposition results, the mean absolute error MAE and signal total error SAE are used to evaluate the power decomposition results. The specific formula is as follows:
[0087]
[0088] Among them, MAE is the power prediction value The actual measured value x t The average value of the absolute error between ; T is the total duration of the test period; is the power prediction value, i.e., the power decomposition value obtained by the method of the present invention; x t is the actual measured value; SAE is the predicted value of power during the test period The error ratio between the true value r; is the predicted value of power; r is the actual value of power.
[0089] Table 1 shows the evaluation index data of mean absolute error MAE and signal total error SAE.
[0090] Table 1 Evaluation index table of mean absolute error MAE and signal total error SAE
[0091] Evaluation indicators washing machine refrigerator dishwasher TV set Mean absolute error MAE(kWh) 0.64 0.7 0.3 0.36 Signal total error SAE (%) 26.2 18.8 19.7 25.9
[0092] It can be seen from the above table that the present invention reduces the interference degree of similar electrical noise in the model training process by combining the Kmeans clustering algorithm and the convolutional neural network, provides effective support for the electrical power decomposition of the residential user load on the distribution network side, and can accurately decompose the power consumption of the residential user's electrical equipment.
[0093] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.
Claims
1. A non-intrusive residential user electricity decomposition method based on clustering, characterized by: include: Collect the total electricity data of original residential users and the total electricity signal data of target residential users and aggregate them to construct the original user set; Using the original user set to train the household electrical appliance power decomposition structure, and construct an electrical appliance power decomposition model; Decomposing the total power signal data of the target residential user based on the electrical appliance power decomposition model to obtain a power decomposition sequence of the target electrical appliance; The power decomposition sequence of the target electrical equipment is evaluated using a confidence evaluation model to obtain a confidence score and optimize the power decomposition sequence of the target electrical equipment to achieve power decomposition of residential users.
2. The clustering-based non-intrusive residential user electricity decomposition method according to claim 1, characterized in that: The original user set is composed of the labeled total power data of residential users; The construction of the electrical appliance power decomposition model comprises the following steps: Based on the ratio of B1:B2:B3, the labeled total electricity data of residential users is divided to obtain training data, verification data and test data; During the supervised training and verification process, the network parameters are updated in a supervised manner to build an appliance power decomposition model.
3. The clustering-based non-intrusive residential user electricity decomposition method according to claim 2, characterized in that: The calculation formula of the electrical appliance power decomposition model is as follows: in, is the corresponding power value in the power sequence window of the mth load device; F p is the parameter set of the model; Y t:t+W-1 is the sliding window of the total power sequence; t is the time; W is the length of the sliding window of the total power sequence; ε is the W-dimensional Gaussian random noise; τ is the middle moment of the sequence window.
4. The clustering-based non-intrusive residential user electricity decomposition method according to claim 3, characterized in that: The loss function in the supervised training process is calculated as follows: Among them, L p is the loss obtained from model training; W is the length of the sliding window of the total power sequence; T is the total length of the input total power sequence; p is the probability of the output of the electrical appliance power decomposition model and the input of the electrical appliance power decomposition model; is the power sequence window of the mth load device; Y t:t+W-1 is the sliding window of the total power sequence; t is the time; θ p is the network parameter.
5. The clustering-based non-intrusive residential user electricity decomposition method according to claim 4, characterized in that: The method of obtaining the power decomposition sequence of the target electrical equipment comprises the following steps: Input the window sequence of the total power signal data of the target user into the electrical appliance power decomposition model to obtain the power value of a single electrical device; As the input window of the electrical appliance power decomposition model continues to slide, the power values at the midpoints of the target electrical appliance windows are combined into a power decomposition sequence of the target electrical appliance.
6. The clustering-based non-intrusive residential user electricity decomposition method according to claim 5, characterized in that: The polymerization comprises the following steps: Calculate the total power data of the original residential users and the total power signal data of the target residential users to obtain the average hourly power value and the average daily power value of all residential users; Normalize the average hourly electricity value of hour A1 and the average daily electricity value of day A2 to construct a feature set; The Kmeans clustering algorithm is used to cluster the feature set to obtain multiple clusters composed of similar users, and to construct an original user set similar to the target user.
7. The clustering-based non-intrusive residential user electricity decomposition method according to claim 6, characterized in that: The calculation formula for the normalization process is as follows: Among them, x t is the actual measured value of the electric quantity series; is the mean value of the electric quantity series; σ is the standard deviation of the electric quantity series.
8. A clustering-based non-intrusive residential user electricity decomposition system, based on the clustering-based non-intrusive residential user electricity decomposition method according to any one of claims 1 to 7, characterized in that: include, The user clustering module is used to collect the total power data of original residential users and the total power signal data of target residential users and perform calculation, normalization and clustering to construct a set of original users similar to the target users; A model building module is used to divide the original user set into training data, verification data, and test data according to the proportion, and to update the network parameters in a supervised manner during the supervised training and verification process to build an appliance power decomposition model; A data decomposition module is used to input the window sequence of the total power signal data of the target user into the electrical appliance power decomposition model and combine the power values to obtain the power decomposition sequence of the target electrical equipment; The data optimization module is used to evaluate the power decomposition sequence of the target electrical equipment using the confidence evaluation model to obtain a confidence score, and optimize the power decomposition sequence of the target electrical equipment using the confidence score.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the clustering-based non-intrusive residential user electricity decomposition method according to any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the clustering-based non-intrusive residential user electricity decomposition method according to any one of claims 1 to 7 are implemented.