An electric power consumption anomaly identification method using a combined learning method

Abstract

The present application relates to the technical field of power detection, and particularly relates to a power abnormality identification method using a combination learning method, comprising the following steps: S1, obtaining household appliance power parameters according to preset requirements; S2, enumerating all possible power schemes of each household appliance through permutation and combination, and obtaining total load currents corresponding to each power scheme; the power scheme comprises the working number of each household appliance; S3, training a preset neural network model with each power scheme as input and the corresponding total load current as output; S4, obtaining the number of each household appliance actually in working state as an actual power scheme, inputting the trained neural network model, and obtaining a reference load current; S5, comparing the actual total load current with the reference load current, and if the deviation is greater than a preset degree, it is determined that there is an abnormality. The present application can greatly improve the monitoring efficiency of residential power abnormality and prevent possible power accidents.

Description

Technical Field

[0001] The present invention relates to the technical field of electricity consumption detection, and in particular to a method for identifying abnormal electricity consumption by adopting a combined learning method. Background Art

[0002] Energy conservation and emission reduction are key technical goals for smart grid development. While significant progress has been made in my country's power system reform, much remains to be done. In particular, improving the electricity market mechanism, increasing energy efficiency, and implementing a green development strategy remain challenging issues.

[0003] Current electricity consumption monitoring is primarily categorized into industrial and commercial (IC) and residential (residential) monitoring. While industrial and commercial (IC) use accounts for a significant portion of electricity, its usage patterns are relatively simple, and existing research is relatively extensive. In contrast, residential usage patterns are more complex, with a high proportion of flexible loads and greater demand elasticity. Studies have shown that understanding and improving residential electricity usage behavior can save over 27% of electricity consumption. Therefore, studying residential electricity usage is crucial for improving energy system efficiency, achieving energy conservation and emission reduction, and fostering green development. For residential electricity monitoring, the current mainstream technology is non-invasive, using load decomposition to analyze user behavior. However, this approach requires the computation of numerous electrical characteristics to identify abnormalities in electricity usage, resulting in relatively low efficiency.

[0004] Therefore, how to improve the efficiency of monitoring abnormal electricity consumption of residents while ensuring monitoring accuracy has become an urgent problem to be solved. Summary of the Invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for identifying abnormal electricity consumption using a combined learning method, which can significantly improve the efficiency of monitoring abnormal electricity consumption of residents while ensuring monitoring accuracy.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] A method for identifying abnormal power consumption using a combined learning method comprises the following steps:

[0008] S1. Obtaining electricity consumption parameters of household appliances according to preset requirements, wherein the preset requirements include the type and quantity of household appliances;

[0009] S2. Enumerate all possible power usage plans for each household appliance by permutation and combination, and obtain the total load current corresponding to each power usage plan; the power usage plan includes the number of working hours of each household appliance;

[0010] S3. Using each power consumption plan as input and the corresponding total load current as output, the preset neural network model is trained;

[0011] S4, the number of each household appliance actually in working state is obtained as an actual power consumption scheme, and is input into the trained neural network model to obtain a reference load current;

[0012] S5, the actual total load current is compared with the reference load current, and if the deviation is greater than a preset degree, it is determined that there is an anomaly.

[0013] Preferably, the neural network model is a three-layer neural network model.

[0014] Preferably, the structure of the neural network model comprises 15 input layer neurons, 9 hidden layer neurons and 15 output layer neurons.

[0015] Preferably, the activation function of the neural network is a sigmoid function.

[0016] Preferably, in S3, when training the neural network model, the number of iterations is set to 1000 times, the training variance is set to 1e-4, and the learning rate is set to 0.01.

[0017] Preferably, in S3, the total load current weighted by a preset multiple is taken as the output, and the neural network model is trained; in S5, the actual total load current weighted by a preset multiple is compared with the reference load current.

[0018] Preferably, the value of the weighting multiple is greater than or equal to 80.

[0019] Preferably, in S2, the total load current is:

[0020] i(t) = α1i1(t) + α2i2(t) + … α m i m (t) + … α n i n (t);

[0021] Wherein, i(t) is the total load current measured at t time at the home entrance; i m (t) is the working current of the mth appliance at t time, α m is the working number of the appliance, and the working number of 0 represents that the appliance is not turned on.

[0022] Preferably, in S5, when it is determined that there is an anomaly, a preset power consumption warning is also performed.

[0023] Compared with the prior art, the present application has the following beneficial effects:

[0024] 1. Unlike existing methods for monitoring residential electricity usage, which use load decomposition to analyze user electricity behavior, this present invention permutes and combines household appliances to obtain all possible electricity usage scenarios. The total load current for each scenario is then calculated and introduced into a neural network for learning. In other words, the present invention uses the acquired household appliances as variables, enumerates all possible electricity usage scenarios by permuting and combining the variables, and then trains the neural network model based on the total load current corresponding to each scenario. This, leveraging the self-learning nature of the neural network model, allows the relationship between the household appliance's electricity usage scenario and its total load current to be determined using limited training data. Subsequently, during actual monitoring, simply obtain the number of household appliances actually operating as the actual electricity usage scenario and input this into the trained neural network model to obtain a reference load current. By comparing the reference load current with the actual total load current, it is possible to determine whether there are any abnormalities in electricity usage. Compared to existing technologies, this monitoring method not only ensures the accuracy of monitoring results, but is also simple to implement, low-cost, and significantly improves monitoring efficiency.

[0025] In summary, the present invention can significantly improve the efficiency of monitoring abnormalities in residents' electricity consumption while ensuring monitoring accuracy, and plays an auxiliary role in preventing possible electricity accidents.

[0026] 2. When the present invention detects abnormal electricity consumption of residents, it will send an electricity usage alert. After the user understands the situation, he can make subsequent confirmation and work with the user to determine whether the current electricity usage status is due to new load access or a fault, which can improve user participation.

[0027] 3. When the present invention obtains training data for the neural network model, due to the diversity of household appliances and the diversity of power consumption combinations, the difference between the total load currents corresponding to various household appliance solutions may be small. Directly using the original data for training may affect the accuracy of subsequent monitoring. In a further optimization of the present invention, after obtaining the total load current corresponding to each household appliance solution, it will be weighted by a preset multiple and used as training data. Correspondingly, during actual monitoring, the actual total load current will also be weighted by a preset multiple and then compared and judged. In this way, the differentiation of the total load current differences between various household appliance combinations is increased, which can further ensure the accuracy of monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to make the purpose, technical solutions and advantages of the invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings, in which:

[0029] Figure 1 This is a flow chart of a method for identifying abnormal power consumption using a combined learning method according to the present invention;

[0030] Figure 2 Network structure diagram of the neural network in the embodiment;

[0031] Figure 3 R value diagram of the neural network training result in the embodiment. DETAILED DESCRIPTION

[0032] The following is further described in detail through specific embodiments:

[0033] Embodiment:

[0034] As Figure 1 shown, the embodiment discloses an electricity abnormality identification method using a combination learning method, including the following steps:

[0035] S1, obtaining the electricity parameters of household appliances according to preset requirements, the preset requirements including the types and quantities of household appliances.

[0036] Household appliances, i.e. electrical appliances for residents, such as air conditioners, rice cookers, computers, televisions, fans, etc. In order to ensure the applicability of monitoring, the obtained household appliances can be as diverse as possible, and if there are different power types of the same type of household appliances, they can also be obtained separately, so as to enumerate as many electricity scheme data as possible for subsequent neural network model training.

[0037] S2, enumerating all possible electricity schemes of each household appliance through permutation and combination, and obtaining the total load current corresponding to each electricity scheme; the electricity scheme includes the working quantity of each household appliance.

[0038] In specific implementation, each household appliance can be set as a known variable:

[0039] For example, variable a = air conditioner, variable b = rice cooker, variable c = computer, variable d = television, variable e = fan, etc. In order to enumerate all electricity combination conditions, permutation and combination are performed, for example, a appliance is bundled, then there are {a, b}, {a, c}, {a, d}, {a, e}; if a appliance and b appliance are bundled, then there are {a, b, c}, {a, b, d}, {a, b, e}, and so on, enumerating all possible electricity schemes of each household appliance.

[0040] For ease of illustration, it is assumed that there are n types of electrical appliances, and the total load current composed of n types of electrical appliances can be represented as:

[0041] i(t) = α1i1(t) + α2i2(t) + … α m i m (t) + … α n i n (t);

[0042] Wherein, i(t) is the total load current measured at the household at t time; i m (t) is the working current of the mth electric appliance at t time; a m is the coefficient of the electric appliance, that is, the working number of each type of electric appliance, 0 represents that the electric appliance is not turned on.

[0043] S3, training the preset neural network model with each power consumption scheme as input and the corresponding total load current as output.

[0044] In specific implementation, the neural network model is a three-layer neural network model, and the structure of the neural network model includes 15 input layer neurons, 9 hidden layer neurons and 15 output layer neurons. The network level structure of the neural network is as shown in Figure 2 The weight coefficient ω i is randomly generated, and the back propagation iteration can be performed. In order to better learn and train, the number of iterations is set to 1000 times, the training variance is set to 1e-4, the learning rate is set to 0.01, and the activation function uses the sigmoid function. After adjustment, the result reaches the optimum. The neural network training result R value is as shown in Figure 3 .

[0045] It should be noted that due to the diversity of household appliances and the diversity of power consumption combinations, there may be a case that the difference between the total load currents corresponding to multiple household appliance schemes is small. Directly using the original data for training may affect the accuracy during subsequent monitoring. Therefore, in the further optimization of the present application, after obtaining the total load currents corresponding to each household appliance scheme, the total load currents are weighted by a preset multiple for use as training data. Correspondingly, during actual monitoring, the actual total load current is also weighted by a preset multiple before comparison and judgment. In this way, the total load current difference between various household appliance combinations is large, which can further ensure the accuracy of monitoring. At the same time, in order to ensure that the difference between each total load current in the training data is large enough, the weighted multiple should be greater than or equal to 80. In the present embodiment, the value of the weighted multiple is 100.

[0046] S4, obtaining the number of each household appliance actually in working state as an actual power consumption scheme, inputting the trained neural network model to obtain a reference load current;

[0047] S5, comparing the actual total load current weighted by a preset multiple with the reference load current, if the deviation is greater than a preset degree, it is determined that there is an anomaly. In specific implementation, when abs{(I N (k)-I S )} / I S <D i , the power consumption behavior is normal; otherwise, the power consumption behavior is abnormal. In the formula, I SIt is the value after adding the weighted preset multiples of the actual total load current, I N (k) is the corresponding reference load current, D i = is a preset deviation value. If an abnormality is determined, a preset power usage warning is issued. The power usage warning can be in the form of an audio, visual, or textual prompt. Those skilled in the art can also choose a specific power usage warning method that they are familiar with, as long as the warning effect is achieved. This will not be repeated here.

[0048] Unlike existing methods for monitoring residential electricity usage, which use load decomposition to analyze user electricity behavior, the present invention permutes and combines household appliances to obtain all possible electricity usage scenarios. The total load current for each scenario is then calculated and introduced into a neural network for learning. In other words, the present invention uses the acquired household appliances as variables, enumerates all possible electricity usage scenarios by permuting and combining the variables, and then trains the neural network model based on the total load current corresponding to each scenario. This, leveraging the self-learning nature of the neural network model, allows the relationship between the household appliance's electricity usage scenario and its total load current to be determined using limited training data. Subsequently, during actual monitoring, simply obtain the number of household appliances actually operating as the actual electricity usage scenario and input this into the trained neural network model to obtain a reference load current. By comparing the reference load current with the actual total load current, it is possible to determine whether there are any abnormalities in electricity usage. Compared to existing technologies, this monitoring method not only ensures the accuracy of monitoring results, but is also simple to implement, low-cost, and significantly improves monitoring efficiency. In addition, when the present invention detects abnormal electricity consumption of residents, it will also send an electricity usage alert. After the user understands the situation, he can make subsequent confirmation and work with the user to determine whether the current electricity usage status is due to new load access or a fault, which can improve user participation.

[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the technical solutions. Those skilled in the art should understand that modifications or equivalent replacements of the technical solutions of the present invention that do not depart from the purpose and scope of the technical solutions of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A method for identifying abnormal power consumption using a combined learning method, characterized in that: The following steps are involved: S1. Obtaining electricity consumption parameters of household appliances according to preset requirements, wherein the preset requirements include the type and quantity of household appliances; S2. Enumerate all possible power usage plans for each household appliance by permutation and combination, and obtain the total load current corresponding to each power usage plan; the power usage plan includes the number of working hours of each household appliance; S3. Using each power consumption plan as input and the corresponding total load current as output, the preset neural network model is trained; S4. Obtain the number of household appliances actually in operation as the actual power consumption plan, input it into the trained neural network model, and obtain a reference load current; S5. Compare the actual total load current with the reference load current. If the deviation is greater than a preset degree, it is determined that an abnormality exists.

2. The method for identifying abnormal power consumption using a combined learning method according to claim 1, wherein: The neural network model is a three-layer neural network model.

3. The method for identifying abnormal power consumption using a combined learning method according to claim 2, wherein: The structure of the neural network model includes 15 input layer neurons, 9 hidden layer neurons and 15 output layer neurons.

4. The method for identifying abnormal power consumption using a combined learning method according to claim 3, wherein: The activation function of the neural network is a sigmoid function.

5. The method for identifying abnormal power consumption using a combined learning method according to claim 4, wherein: In S3, when training the neural network model, the number of iterations is set to 1000, the training variance is set to 1e-4, and the learning rate is set to 0.

01.

6. The method for identifying abnormal power consumption using a combined learning method according to claim 5, wherein: In S3, the total load current is weighted by a preset multiple as output to train the neural network model; in S5, the actual total load current is weighted by the preset multiple and compared with the reference load current.

7. The method for identifying abnormal power consumption using a combined learning method according to claim 6, wherein: The value of the weighted multiple is greater than or equal to 80.

8. The method for identifying abnormal power consumption using a combined learning method according to claim 7, wherein: In S2, the total load current is: i(t)=α1i1(t)+α2i2(t)+…α m i m (t)+…α n i n (t); Where i(t) is the total load current measured at the home entrance at time t; i m (t) is the working current of the mth appliance at time t, α m The working number of the appliance is 0, which means none of the appliances are turned on.

9. The method for identifying abnormal power consumption using a combined learning method according to claim 8, wherein: In S5, when it is determined that an abnormality exists, a preset power usage warning is also performed.

Images