Non-intrusive load monitoring method and device, computer equipment and readable storage medium
Through the aggregation of multiple load decomposition models and model parameter adjustment, the accuracy of non-invasive load monitoring is improved, and the problem of low load monitoring accuracy in the prior art is solved, thereby achieving more accurate identification of power load data.
Patent Information
- Application Number
- CN202411344290.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-25
- Publication Date
- 2024-12-27
AI Technical Summary
The existing non-invasive load monitoring methods are not very accurate and have poor electrical load recognition effect.
By inputting the target power load data into multiple pre-constructed power load decomposition models, and obtaining their respective weights, effectively aggregating multiple load decomposition results, and adjusting model parameters using random forest model, reinforcement learning model and deep learning model and weights to improve the accuracy of load decomposition.
It improves the accuracy of non-invasive load monitoring, avoids the shortcomings of the single power load decomposition model, and achieves more accurate power load data identification.
Smart Images
Figure CN119209566A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of load identification, and in particular to a non-invasive load monitoring method, apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Art
[0002] With the rapid development of information technology and big data analysis technology, non-intrusive load monitoring can identify the power load conditions of different equipment through data analysis.
[0003] Currently, non-intrusive load monitoring can be performed based on a single optimization model, pattern recognition model or deep learning model. However, the existing non-intrusive load monitoring methods are not accurate and have poor power load identification effects. Summary of the invention
[0004] Based on this, it is necessary to provide a non-invasive load monitoring method, device, computer equipment, computer-readable storage medium and computer program product to address the above technical problems.
[0005] In a first aspect, the present application provides a non-intrusive load monitoring method, comprising:
[0006] Inputting the target power load data into a plurality of pre-built power load decomposition models to obtain a plurality of corresponding first load decomposition results;
[0007] Obtaining respective weights of the plurality of pre-built power load decomposition models;
[0008] Obtaining a second load decomposition result according to the plurality of first load decomposition results and respective weights of the plurality of pre-built power load decomposition models;
[0009] The power load data of each device is obtained according to the second load decomposition result and the standard power load data of each device.
[0010] In one embodiment, before inputting the target power load data into a plurality of pre-built power load decomposition models to obtain a plurality of corresponding first load decomposition results, the method further includes:
[0011] According to a plurality of pre-built random forest models and the first load test set, a plurality of corresponding third load decomposition results are obtained;
[0012] Get pre-built reinforcement learning models;
[0013] According to the pre-constructed reinforcement learning model, the third load decomposition result output by each random forest model and the initial weight corresponding to each random forest model, the parameters and weights of each random forest model are adjusted to obtain the pre-constructed power load decomposition model and the weights corresponding to the power load decomposition model.
[0014] In one embodiment, before obtaining corresponding third load decomposition results according to the plurality of pre-built random forest models and the first load test set, the method further includes:
[0015] Get the second load test set;
[0016] Evenly dividing the second load test set into a plurality of load test subsets;
[0017] For any first parameter to be adjusted model, select any load test subset from the several load test subsets, train the any first parameter to be adjusted model, and obtain a pre-constructed random forest model.
[0018] In one embodiment, obtaining the second load test set includes:
[0019] According to various types of pre-built deep learning models and the first load test set, a corresponding plurality of fourth load decomposition results are obtained;
[0020] Obtaining a plurality of load generation results according to the plurality of fourth load decomposition results and the generative adversarial network model;
[0021] A second load test set is obtained according to the several load generation results and the first load test set.
[0022] In one embodiment, obtaining a pre-built reinforcement learning model includes:
[0023] According to various types of pre-built deep learning models and the first load test set, a corresponding plurality of fourth load decomposition results are obtained;
[0024] According to the third load decomposition result and the fourth load decomposition result, the second parameter to be adjusted model is trained to obtain a pre-constructed reinforcement learning model.
[0025] In one embodiment, before obtaining corresponding fourth load decomposition results according to the pre-built various types of deep learning models and the first load test set, the method further includes:
[0026] Obtain the third parameter of each type of model to be adjusted;
[0027] For any type of third parameter to be adjusted model, the third parameter to be adjusted model of any type is trained according to the first load training set to obtain the pre-constructed deep learning model of any type.
[0028] In one embodiment, obtaining the second load decomposition result according to the weights of the plurality of first load decomposition results and the plurality of pre-built power load decomposition models comprises:
[0029] sequentially taking the first load decomposition results as target load decomposition results;
[0030] According to the target load decomposition result, a target power load decomposition model is obtained;
[0031] Determining a target weight corresponding to the target power load decomposition model among the respective weights of the plurality of pre-constructed power load decomposition models;
[0032] According to the target load decomposition result and the target weight, a multiplication value corresponding to the target load decomposition result is obtained;
[0033] A second load decomposition result is obtained according to the multiplied values corresponding to the plurality of first load decomposition results.
[0034] In a second aspect, the present application also provides a non-intrusive load monitoring device, comprising:
[0035] A first decomposition result acquisition module inputs the target power load data into a plurality of pre-built power load decomposition models to obtain a plurality of corresponding first load decomposition results;
[0036] A weight acquisition module, used to acquire the weights of the plurality of pre-built power load decomposition models;
[0037] A second decomposition result acquisition module, used to obtain a second load decomposition result according to the weights of the plurality of first load decomposition results and the plurality of pre-built power load decomposition models;
[0038] The power load data acquisition module is used to obtain the power load data of each device according to the second load decomposition result and the standard power load data of each device.
[0039] In a third aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor executes the above method.
[0040] In a fourth aspect, the present application further provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and the computer program is executed by a processor to execute the above method.
[0041] In a fifth aspect, the present application further provides a computer program product, wherein the computer program product comprises a computer program, and the computer program is executed by a processor to execute the above method.
[0042] The above-mentioned non-intrusive load monitoring method, device, computer equipment, computer-readable storage medium and computer program product input the target power load data into several pre-built power load decomposition models to obtain several corresponding first load decomposition results; obtain the weights of several pre-built power load decomposition models; and obtain the second load decomposition result according to the several first load decomposition results and the weights of several pre-built power load decomposition models. The power load data of each device is obtained according to the second load decomposition result and the standard power load data of each device. The present application obtains the second load decomposition result according to the first load decomposition results and weights corresponding to several pre-built power load decomposition models, and obtains the power load data of each device according to the second load decomposition result and the standard power load data of each device, which can realize the effective aggregation of multiple first load decomposition results, avoid the insufficient load decomposition capability of a single power load decomposition model, and thus improve the accuracy of non-intrusive load monitoring. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.
[0044] Figure 1 is an application environment diagram of a non-intrusive load monitoring method in one embodiment;
[0045] Figure 2 is a flow chart of a non-intrusive load monitoring method in one embodiment;
[0046] Figure 3 A schematic diagram of a flow chart of a non-intrusive load monitoring method in another embodiment;
[0047] Figure 4 A schematic diagram of the difference between the predicted value of active power and the actual value of active power of cement mill transformer No. 1 in one embodiment;
[0048] Figure 5A schematic diagram of the difference between the predicted value of active power and the actual value of active power of cement mill transformer No. 2 in one embodiment;
[0049] Figure 6 A schematic diagram of the difference between the predicted value of active power and the actual value of active power of a packaging transformer in one embodiment;
[0050] Figure 7 A schematic diagram of the difference between the predicted value of active power and the actual value of active power of cement mill circulating fan No. 1 in one embodiment;
[0051] Figure 8 A schematic diagram of the difference between the predicted value of active power and the actual value of active power of cement mill circulating fan No. 2 in one embodiment;
[0052] Fig. 9 A schematic diagram of the difference between the predicted value of active power and the actual value of active power of the main motor No. 1 of the cement mill in one embodiment;
[0053] Fig.10 A schematic diagram of the difference between the predicted value of active power and the actual value of active power of the main motor No. 2 of the cement mill in one embodiment;
[0054] Fig.11 is a structural block diagram of a non-intrusive load monitoring device in one embodiment;
[0055] Fig.12 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION
[0056] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0057] It should be noted that this application conducts non-intrusive load monitoring with the user's authorization and permission. The electricity load data involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.
[0058] The present application embodiment provides a non-intrusive load monitoring method, which can be executed by a computer device, such as Figure 1 As shown, the computer device can obtain the target power load data, and then obtain the power load data of each device. It can be understood that the computer device can be implemented by a server, a terminal, or an interactive system between a terminal and a server. In this embodiment, the method includes Figure 2 The steps shown are:
[0059] Step S201: input target power load data into a plurality of pre-built power load decomposition models to obtain a plurality of corresponding first load decomposition results.
[0060] The total electricity load data of an enterprise or household can be used as the target electricity load data.
[0061] Random forest is an ensemble learning method that generates the final prediction result by building a set of decision trees and combining their prediction results. It is widely used in classification and regression problems. The random forest model can be trained to obtain the power load decomposition model.
[0062] The power load decomposition model can perform load decomposition on the target power load data, so the target power load data can be input into a pre-built power load decomposition model, and the output result of the power load decomposition model can be used as the corresponding first load decomposition result.
[0063] Step S202: Obtain the weights of several pre-built power load decomposition models.
[0064] When the random forest model is trained according to the reinforcement learning model, the weights of several pre-built power load decomposition models can be obtained.
[0065] Step S203: obtaining a second load decomposition result according to the weights of the plurality of first load decomposition results and the plurality of pre-built power load decomposition models.
[0066] For each first load decomposition result, determine the weight corresponding to the first load decomposition result according to the pre-constructed power load decomposition model corresponding to the first load decomposition result, multiply the first load decomposition result by the weight corresponding to the first load decomposition result to obtain the multiplication value corresponding to each first load decomposition result, add the multiplication values corresponding to several first load decomposition results to obtain the second load decomposition result.
[0067] Step S204, obtaining the power load data of each device according to the second load decomposition result and the standard power load data of each device.
[0068] The standard power load data of each device can be obtained based on the actual power consumption data of each device.
[0069] Exemplarily, the second load decomposition results are 30W, 50W and 120W, the standard power load data of device A is 50W, the standard power load data of device B is 30W, the standard power load data of device C is 120W, the standard power load data of device D is 80W, …, then according to the second load decomposition result and the standard power load data of each device, the power load data of each device can be obtained. Specifically, the power load data of the above-mentioned devices are: the power load data of device A is 50W, the power load data of device B is 30W and the power load data of device C is 120W.
[0070] In the above-mentioned non-invasive load monitoring method, the second load decomposition result is obtained according to the first load decomposition results and weights corresponding to several pre-constructed power load decomposition models, and the power load data of each device is obtained according to the second load decomposition result and the standard power load data of each device. This can achieve effective aggregation of multiple first load decomposition results, avoid the insufficient load decomposition capability of a single power load decomposition model, and thus improve the accuracy of non-invasive load monitoring.
[0071] In one of the embodiments, before inputting the target power load data into several pre-built power load decomposition models respectively to obtain several corresponding first load decomposition results, the method provided in the present application also includes: obtaining several corresponding third load decomposition results based on several pre-built random forest models and a first load test set; obtaining a pre-built reinforcement learning model; adjusting the parameters and weights of each random forest model based on the pre-built reinforcement learning model, the third load decomposition results output by each random forest model, and the initial weights corresponding to each random forest model to obtain the pre-built power load decomposition model and the weights corresponding to the power load decomposition model.
[0072] Based on public residential activity learning datasets, such as the UK-DALE public dataset, the active power data on the bus is recorded at a sampling frequency of 1 Hz, and the active power data of each device is recorded at a sampling frequency of 1 / 6 Hz. The data is preprocessed, cubic spline interpolation is performed to supplement missing values, outliers are screened out, and all data are normalized to obtain the first load test set.
[0073] The first load test set can be enter Pre-built Random Forest models , get the corresponding The third load decomposition results , as shown in formula (1).
[0074] (1)
[0075] The reinforcement learning model consists of state space, action space, and reward function. These three parts are the key components for defining problems and training intelligent agents, and together determine how intelligent agents learn and optimize strategies.
[0076] The state space represents the set of all possible states currently in. Each state contains all information at a certain point in time. In this embodiment, the state space includes several third load decomposition results and parameters and weights of several current random forest models.
[0077] The action space represents the set of all possible actions that can be performed in each state. Each action is an action that can be taken in a specific state, such as adjusting the weights of the random forest model, changing the architecture of the random forest model, or redistributing the first load test set. The action space should cover all possible optimization strategies.
[0078] The reward function defines the reward obtained after performing an action, which is a direct feedback on its behavior. The reward function is used to guide the selection of actions that can maximize long-term benefits. In this application, the reward function comprehensively considers the accuracy (such as minimizing the mean absolute error) and stability (such as minimizing the variance) of load decomposition, and achieves multi-objective optimization by balancing these two objectives, as shown in formula (2).
[0079] (2)
[0080] In the formula, MAE represents the mean absolute error, Variance represents the variance, and is the weight coefficient used to balance the mean absolute error and variance.
[0081] The third load decomposition result output by each random forest model can be input into a pre-constructed reinforcement learning model. The pre-constructed reinforcement learning model can adjust the parameters and weights of each random forest model based on the initial weights corresponding to each random forest model and the initial parameters corresponding to each random forest model. Each time an adjustment operation is performed, a reward function can be obtained. When the reward function meets the conditions, the random forest model at this time is used as the pre-constructed power load decomposition model, and the weights corresponding to the random forest model at this time are used as the weights corresponding to the power load decomposition model.
[0082] In this embodiment, the parameters and weights of each random forest model are adjusted according to the pre-constructed reinforcement learning model to obtain an electricity load decomposition model with better load decomposition capability and weights corresponding to a more appropriate electricity load decomposition model, thereby improving the accuracy of non-invasive load monitoring.
[0083] In one of the embodiments, before obtaining a corresponding number of third load decomposition results based on a number of pre-built random forest models and a first load test set, the method provided in the present application also includes: obtaining a second load test set; evenly dividing the second load test set into a number of load test subsets; for any first parameter to be adjusted model, selecting any load test subset from the number of load test subsets, training any first parameter to be adjusted model, and obtaining a pre-built random forest model.
[0084] The second load test set can be Divide evenly into load test subsets, each load test subset is represented by ,in , each load test subset The number of samples is the same and the consistency of load data distribution is maintained as much as possible to ensure that each random forest model can fully learn the load data characteristics.
[0085] For any model to be adjusted for the first parameter, select one load test subset from several load test subsets , train any first parameter to be adjusted model to obtain a pre-built random forest model , as shown in formula (3).
[0086] (3)
[0087] In this embodiment, the second load test set is evenly divided into several load test subsets, and one load test subset is selected from the several load test subsets, and any first parameter to be adjusted model is trained to obtain a pre-constructed random forest model, which can improve the load decomposition capability of the random forest model, thereby improving the accuracy of non-invasive load monitoring.
[0088] In one of the embodiments, a second load test set is obtained, and the specific steps are as follows: based on the pre-built types of deep learning models and the first load test set, a corresponding number of fourth load decomposition results are obtained; based on the number of fourth load decomposition results and the generative adversarial network model, a number of load generation results are obtained; based on the number of load generation results and the first load test set, a second load test set is obtained.
[0089] The first load test set can be input into various types of deep learning models to obtain several fourth load decomposition results corresponding to each type of deep learning model. .
[0090] Several fourth load decomposition results can be Input the generative adversarial network model to obtain several load generation results , as shown in formula (4).
[0091] (4)
[0092] in, is the number of samples used to generate data.
[0093] The Generative Adversarial Networks (GAN) model consists of a generator and a discriminator. The generator network accepts a random noise vector And the fourth load decomposition results As input, generate realistic load data ,in Represents the parameters of the generator. The discriminator network accepts the real load data and generate data As input, output the judgment result and ,in Represents the parameters of the discriminator.
[0094] The training process of the generative adversarial network model includes the discriminator loss function, the generator loss function and the optimization process.
[0095] The discriminator loss function is shown in formula (5):
[0096] (5)
[0097] in, is the discriminator loss function, is the real load data, is a random noise vector, is the fourth load decomposition result, To generate load data, is the batch size.
[0098] The generator loss function is shown in formula (6):
[0099] (6)
[0100] in, is the generator loss function, is a random noise vector, is the fourth load decomposition result,
[0101] To generate load data, is the batch size.
[0102] Optimization process: Alternately update the parameters of the generator and the discriminator, and minimize the loss function by gradient descent, as shown in equations (7) and (8):
[0103] (7)
[0104] (8)
[0105] in, is the learning rate, represents the parameters of the discriminator, represents the parameters of the generator, represents the gradient of the discriminator loss function, represents the gradient of the generator loss function.
[0106] Through repeated iterative training, the data generated by the generator becomes more and more realistic, while the discriminator's discrimination ability continues to improve.
[0107] Generate results for several loads and the first load test set Combine to get the second load test set , as shown in formula (9).
[0108] (9)
[0109] In this embodiment, a second load test set with a richer data volume is obtained based on a number of load generation results and the first load test set.
[0110] In one of the embodiments, a pre-built reinforcement learning model is obtained, and the specific steps are as follows: according to the pre-built various types of deep learning models and the first load test set, a corresponding number of fourth load decomposition results are obtained; according to the third load decomposition result and the fourth load decomposition result, the second parameter model to be adjusted is trained to obtain the pre-built reinforcement learning model.
[0111] You can use Deep Q-Network (DQN) as the reinforcement learning model and randomly initialize the policy network parameters and value network parameters of the reinforcement learning model. The experience replay mechanism can be used to store samples obtained during training in a buffer, and randomly extract samples for network updates to improve training stability and efficiency.
[0112] The formula for updating the value network parameters in the deep Q network is shown in formula (10).
[0113] (10)
[0114] in, is the current state, For the current action, For reward, For the next state, is the learning rate, is the discount factor, is the target network parameter, Indicates the next state An action from the set of all possible actions.
[0115] The strategy update formula is shown in formula (11):
[0116] (11)
[0117] The policy and value network are updated by continuously conducting experiments in the simulation environment. Each trial consists of selecting an action from the current state, executing the action, and observing the new state and reward. The policy is updated using reward feedback so that future action selections can maximize the cumulative reward.
[0118] According to various types of pre-built deep learning models and the first load test set, a corresponding number of fourth load decomposition results can be obtained.
[0119] The second parameter to be adjusted model may be trained according to the third load decomposition result and the fourth load decomposition result to obtain a pre-constructed reinforcement learning model.
[0120] That is, in this embodiment, the state space includes several third load decomposition results and the parameters and weights of several current random forest models, several fourth load decomposition results and the parameters and weights of each type of current deep learning model. The action space may include adjusting the weight of the random forest model, changing the parameters of the random forest model, reallocating the first load test set, adjusting the weights of each type of deep learning model, and changing the parameters of each type of deep learning model. The operation of adjusting the parameters and weights of each random forest model and the operation of the parameters and weights of each type of deep learning model can be performed. Each time an adjustment operation is performed, a reward function can be obtained. When the reward function meets the conditions, a pre-constructed reinforcement learning model is obtained.
[0121] In this embodiment, the second parameter to be adjusted model is trained according to the third load decomposition result and the fourth load decomposition result to obtain a pre-constructed reinforcement learning model, and a reinforcement learning model with better performance is obtained through relatively rich training data.
[0122] In one embodiment, before obtaining a corresponding number of fourth load decomposition results based on pre-constructed deep learning models of various types and a first load test set, the method provided in the present application also includes: obtaining each type of third parameter model to be adjusted; for any type of third parameter model to be adjusted, training any type of third parameter model to be adjusted according to the first load training set to obtain a pre-constructed deep learning model of any type.
[0123] Various types of models to be adjusted for the third parameters may include a denoising autoencoder deep learning model, a sequence-sequence deep learning model, and a sequence-point deep learning model.
[0124] Based on public residential activity learning datasets, such as the UK-DALE public dataset, the active power data on the bus is recorded at a sampling frequency of 1 Hz, and the active power data of each device is recorded at a sampling frequency of 1 / 6 Hz. The data is preprocessed, missing values are supplemented by cubic spline interpolation, outliers are screened out, and all data are normalized to obtain the first load training set.
[0125] For any type of the third parameter to be adjusted model, the third parameter to be adjusted model is trained according to the first load training set to capture the characteristics of the load data, and the parameters of the third parameter to be adjusted model are adjusted to optimize the performance of the third parameter to be adjusted model. The loss function L is shown in formula (12).
[0126] (12)
[0127] Where b represents the total number of load samples, u represents the load sample index, represents the predicted output of the model with the third parameter to be adjusted, Represents the true value of the load sample.
[0128] When the loss function L satisfies the condition, the third parameter model to be adjusted at this time is used as any type of pre-built deep learning model.
[0129] In this embodiment, for any type of third parameter to be adjusted model, any type of third parameter to be adjusted model is trained according to the first load training set to obtain any type of deep learning model with better performance.
[0130] In one embodiment, a second load decomposition result is obtained based on several first load decomposition results and the respective weights of several pre-constructed electricity load decomposition models, and the specific steps are as follows: the first load decomposition results are taken as target load decomposition results in sequence; a target electricity load decomposition model is obtained based on the target load decomposition result; among the respective weights of several pre-constructed electricity load decomposition models, the target weight corresponding to the target electricity load decomposition model is determined; based on the target load decomposition result and the target weight, a multiplication value corresponding to the target load decomposition result is obtained; based on the multiplication values corresponding to the several first load decomposition results, a second load decomposition result is obtained.
[0131] The first load decomposition results can be used as target load decomposition results in turn; according to the target load decomposition results, a target power load decomposition model is obtained; among the weights of the respective pre-constructed power load decomposition models, the target weight corresponding to the target power load decomposition model is determined; according to the target load decomposition result and the target weight, a multiplication value corresponding to the target load decomposition result is obtained; according to the multiplication values corresponding to the several first load decomposition results, a second load decomposition result is obtained, as shown in formula (13).
[0132] (13)
[0133] in, is the second load decomposition result, For the The weight of the target electricity load decomposition model, For the The target load decomposition results.
[0134] In this embodiment, the first load decomposition results can be used as target load decomposition results in turn, and the multiplication value corresponding to the target load decomposition result can be obtained according to the target load decomposition result and the target weight; and a more accurate second load decomposition result can be obtained according to the multiplication values corresponding to several first load decomposition results.
[0135] In order to better understand the above method, an application example of the non-intrusive load monitoring method of the present application is described in detail below. Figure 3 shown.
[0136] With the authorization of the cement plant, training data can be obtained based on the actual industrial load data of the cement plant. Specifically, the active power data on the cement plant bus and the active power data of each device are recorded at a sampling frequency of 3 minutes in the cement mill production line as the actual industrial load data of the cement plant. The equipment includes cement mill transformer No. 1, cement mill transformer No. 2, packaging transformer, cement mill circulating fan No. 1, cement mill circulating fan No. 2, cement mill main motor No. 1 and cement mill main motor No. 2. The actual industrial load data of the cement plant from August 19, 2024 to September 1, 2024 is used as the first load training set, and the actual industrial load data of the cement plant from September 2, 2024 to September 8, 2024 is used as the first load test set.
[0137] Obtain each type of third parameter model to be adjusted; for any type of third parameter model to be adjusted, train any type of third parameter model to be adjusted according to the first load training set to obtain a pre-constructed deep learning model of any type.
[0138] According to the pre-built various types of deep learning models and the first load test set, several corresponding fourth load decomposition results are obtained; according to several fourth load decomposition results and the generative adversarial network model, several load generation results are obtained; according to several load generation results and the first load test set, the second load test set is obtained.
[0139] The second load test set is evenly divided into n load test subsets; for any first parameter to be adjusted model, any load test subset is selected from the n load test subsets, and any first parameter to be adjusted model is trained to obtain n pre-constructed random forest models. According to several pre-constructed random forest models and the first load test set, corresponding n third load decomposition results are obtained.
[0140] According to the third load decomposition result and the fourth load decomposition result, the second parameter to be adjusted model is trained to obtain a pre-constructed reinforcement learning model. The third load decomposition result and the fourth load decomposition result are the predicted active power values of each device. Therefore, according to the difference between the predicted active power value and the actual active power value of each device, the second parameter to be adjusted model can be trained to obtain a pre-constructed reinforcement learning model. The difference between the predicted active power value and the actual active power value of each device is as follows: Figure 4-Figure 10 shown. Specifically, Figure 4 Schematic diagram of the difference between the predicted active power value and the actual active power value of cement mill transformer No. 1. Figure 5 Schematic diagram of the difference between the predicted value and the actual value of active power of cement mill transformer No. 2. Figure 6Schematic diagram of the difference between the predicted active power value and the actual active power value of the packaging transformer. Figure 7 Schematic diagram of the difference between the predicted value and the actual value of active power of cement mill circulating fan No. 1. Figure 8 Schematic diagram of the difference between the predicted value and the actual value of active power of cement mill circulating fan No. 2. Fig. 9 Schematic diagram of the difference between the predicted value and the actual value of active power of the main motor No. 1 of the cement mill. Fig.10 Schematic diagram of the difference between the predicted value of active power and the actual value of active power of cement mill main motor No. 2.
[0141] According to the pre-constructed reinforcement learning model, the third load decomposition result output by each random forest model and the initial weight corresponding to each random forest model, the parameters and weights of each random forest model are adjusted to obtain the pre-constructed power load decomposition model and the weights corresponding to the power load decomposition model.
[0142] The target power load data is input into several pre-built power load decomposition models to obtain several corresponding first load decomposition results; the second load decomposition result is obtained according to the weights of the several first load decomposition results and the several pre-built power load decomposition models; the power load data of each device is obtained according to the second load decomposition result and the standard power load data of each device.
[0143] In this embodiment, a second load decomposition result is obtained based on the first load decomposition results and weights corresponding to several pre-constructed power load decomposition models, and the power load data of each device is obtained based on the second load decomposition result and the standard power load data of each device. This can achieve effective aggregation of multiple first load decomposition results, avoid the insufficient load decomposition capability of a single power load decomposition model, and thus improve the accuracy of non-invasive load monitoring.
[0144] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.
[0145] Based on the same inventive concept, the embodiment of the present application also provides a non-intrusive load monitoring device for implementing the non-intrusive load monitoring method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more non-intrusive load monitoring device embodiments provided below can refer to the limitations of the non-intrusive load monitoring method above, and will not be repeated here.
[0146] In an exemplary embodiment, Fig.11 As shown, a non-intrusive load monitoring device is provided, wherein:
[0147] The first decomposition result acquisition module 1101 inputs the target power load data into a plurality of pre-built power load decomposition models to obtain a plurality of corresponding first load decomposition results;
[0148] The weight acquisition module 1102 is used to acquire the weights of the plurality of pre-built power load decomposition models;
[0149] A second decomposition result acquisition module 1103 is used to obtain a second load decomposition result according to the plurality of first load decomposition results and the respective weights of the plurality of pre-built power load decomposition models;
[0150] The power load data acquisition module 1104 is used to obtain the power load data of each device according to the second load decomposition result and the standard power load data of each device.
[0151] In one embodiment, the device also includes a model training module, which is used to: obtain a corresponding number of third load decomposition results based on a number of pre-constructed random forest models and a first load test set; obtain a pre-constructed reinforcement learning model; adjust the parameters and weights of each random forest model based on the pre-constructed reinforcement learning model, the third load decomposition results output by each random forest model, and the initial weights corresponding to each random forest model, to obtain a pre-constructed power load decomposition model and the weights corresponding to the power load decomposition model.
[0152] In one of the embodiments, the model training module is also used to: obtain a second load test set; evenly divide the second load test set into a number of load test subsets; for any first parameter model to be adjusted, select any one load test subset from the several load test subsets, train any first parameter model to be adjusted, and obtain a pre-constructed random forest model.
[0153] In one embodiment, the model training module is also used to: obtain a corresponding number of fourth load decomposition results based on pre-built types of deep learning models and a first load test set; obtain a number of load generation results based on the number of fourth load decomposition results and a generative adversarial network model; obtain a second load test set based on the number of load generation results and the first load test set.
[0154] In one embodiment, the model training module is also used to: obtain a corresponding number of fourth load decomposition results based on pre-constructed deep learning models of various types and a first load test set; train the second parameter model to be adjusted based on the third load decomposition result and the fourth load decomposition result to obtain a pre-constructed reinforcement learning model.
[0155] In one of the embodiments, the model training module is also used to: obtain various types of third parameter models to be adjusted; for any type of third parameter model to be adjusted, train the third parameter model to be adjusted of any type according to the first load training set to obtain a pre-built deep learning model of any type.
[0156] In one embodiment, the second decomposition result acquisition module 1103 is also used to: sequentially use the first load decomposition results as target load decomposition results; obtain a target power load decomposition model based on the target load decomposition results; determine the target weight corresponding to the target power load decomposition model among the respective weights of the several pre-built power load decomposition models; obtain a multiplication value corresponding to the target load decomposition result based on the target load decomposition result and the target weight; and obtain a second load decomposition result based on the multiplication values corresponding to several first load decomposition results.
[0157] Each module in the above non-intrusive load monitoring device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.
[0158] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Fig.12As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. 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, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data of a non-invasive load monitoring method. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a non-invasive load monitoring method is implemented.
[0159] Those skilled in the art will understand that Fig.12 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0160] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.
[0161] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.
[0162] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.
[0163] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.
[0164] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.
[0165] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
Claims
1. A non-intrusive load monitoring method, characterized in that: The method further comprises: Inputting the target power load data into a plurality of pre-built power load decomposition models to obtain a plurality of corresponding first load decomposition results; Obtaining respective weights of the plurality of pre-built power load decomposition models; Obtaining a second load decomposition result according to the plurality of first load decomposition results and respective weights of the plurality of pre-built power load decomposition models; The power load data of each device is obtained according to the second load decomposition result and the standard power load data of each device.
2. The method according to claim 1, characterized in that Before inputting the target power load data into a plurality of pre-built power load decomposition models to obtain a plurality of corresponding first load decomposition results, the method further includes: According to a plurality of pre-built random forest models and the first load test set, a plurality of corresponding third load decomposition results are obtained; Get pre-built reinforcement learning models; According to the pre-constructed reinforcement learning model, the third load decomposition result output by each random forest model and the initial weight corresponding to each random forest model, the parameters and weights of each random forest model are adjusted to obtain the pre-constructed power load decomposition model and the weights corresponding to the power load decomposition model.
3. The method according to claim 2, characterized in that Before obtaining corresponding third load decomposition results according to the plurality of pre-built random forest models and the first load test set, the method further includes: Get the second load test set; Evenly dividing the second load test set into a plurality of load test subsets; For any first parameter to be adjusted model, select any load test subset from the several load test subsets, train the any first parameter to be adjusted model, and obtain a pre-constructed random forest model.
4. The method according to claim 3, characterized in that The obtaining of the second load test set includes: According to various types of pre-built deep learning models and the first load test set, a corresponding plurality of fourth load decomposition results are obtained; Obtaining a plurality of load generation results according to the plurality of fourth load decomposition results and the generative adversarial network model; A second load test set is obtained according to the several load generation results and the first load test set.
5. The method according to claim 2, characterized in that: The obtaining of a pre-built reinforcement learning model comprises: According to various types of pre-built deep learning models and the first load test set, a corresponding plurality of fourth load decomposition results are obtained; According to the third load decomposition result and the fourth load decomposition result, the second parameter to be adjusted model is trained to obtain a pre-constructed reinforcement learning model.
6. The method according to claim 5, characterized in that Before obtaining corresponding fourth load decomposition results according to the pre-built various types of deep learning models and the first load test set, the method further includes: Obtain the third parameter of each type of model to be adjusted; For any type of third parameter to be adjusted model, the third parameter to be adjusted model of any type is trained according to the first load training set to obtain the pre-constructed deep learning model of any type.
7. The method according to claim 1, characterized in that The obtaining of a second load decomposition result according to the plurality of first load decomposition results and the respective weights of the plurality of pre-built power load decomposition models comprises: sequentially taking the first load decomposition results as target load decomposition results; According to the target load decomposition result, a target power load decomposition model is obtained; Determining a target weight corresponding to the target power load decomposition model among the respective weights of the plurality of pre-constructed power load decomposition models; According to the target load decomposition result and the target weight, a multiplication value corresponding to the target load decomposition result is obtained; A second load decomposition result is obtained according to the multiplied values corresponding to the plurality of first load decomposition results.
8. A non-intrusive load monitoring device, characterized in that: The device comprises: A first decomposition result acquisition module inputs the target power load data into a plurality of pre-built power load decomposition models to obtain a plurality of corresponding first load decomposition results; A weight acquisition module, used to acquire the weights of the plurality of pre-built power load decomposition models; A second decomposition result acquisition module, used to obtain a second load decomposition result according to the weights of the plurality of first load decomposition results and the plurality of pre-built power load decomposition models; The power load data acquisition module is used to obtain the power load data of each device according to the second load decomposition result and the standard power load data of each device.
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 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 method according to any one of claims 1 to 7 are implemented.