Voltage sag category identification method and device based on non-intrusive load monitoring

Abstract

The application relates to a voltage sag type identification method and device based on non-intrusive load monitoring, which comprises the following steps: obtaining user monitoring data, extracting trajectory features from the user monitoring data, comparing the trajectory features with a pre-constructed sensitive load loss rate evaluation feature library, obtaining corresponding sensitive load loss rates, and obtaining the user voltage sag type based on the sensitive load loss rates, wherein the voltage sag type reflects the severity of the voltage sag suffered by the user. Compared with the prior art, the application has the advantages of high reliability, convenient identification and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power quality, in particular to a voltage sag type identification method and device based on non-intrusive load monitoring. BACKGROUND

[0002] The power quality of a power grid is mainly measured by voltage quality, frequency quality and power supply reliability. The ideal power should be stable voltage, 50Hz frequency and no harmonics, and the power grid should be continuously and reliably powered. Among many power quality problems such as frequency deviation, three-phase imbalance, harmonics and voltage sag, the harm caused by voltage sag to the production and life of the public is more serious.

[0003] At present, the evaluation methods of voltage sag severity mainly include investigation and statistical method, probability evaluation method, etc. The investigation and statistical method is to statistically evaluate the direct cost and indirect cost of users caused by voltage sag, and to reflect the degree of influence of users affected by voltage sag. The probability evaluation method is to construct a voltage sag severity evaluation model, including sensitive load trip probability evaluation, production process interruption probability evaluation, and voltage sag loss evaluation model, to reflect the severity of voltage sag suffered by users. The above methods have high dependence on the structure parameters of the production process of users, and the related information data is difficult to obtain in engineering practice.

[0004] A large number of power quality monitoring devices are installed on the grid side, which record the voltage and current waveforms during the occurrence of voltage sag events, which can be used to extract key features and find the corresponding relationship with the production process of users, providing a data basis for non-intrusive identification of sensitive loads. However, in actual production, due to the structural characteristics of the production process, the stop of some sensitive loads does not affect the actual production. If only the non-intrusive method is used to identify the sensitive load trip rate to reflect the severity of the influence of voltage sag on users, there will be deviation from the actual results. SUMMARY

[0005] The purpose of the present application is to overcome the defects of the prior art and provide a voltage sag type identification method and device based on non-intrusive load monitoring, which is reliable and easy to identify.

[0006] The purpose of the present application can be achieved by the following technical solutions:

[0007] A voltage sag type identification method based on non-intrusive load monitoring, the method comprising: obtaining user monitoring data, extracting trajectory features from the user monitoring data, comparing the trajectory features with a pre-constructed sensitive load loss rate evaluation feature library, obtaining corresponding sensitive load loss rate, and obtaining the voltage sag type of the user based on the sensitive load loss rate, the voltage sag type reflecting the severity of the voltage sag suffered by the user;

[0008] The construction process of the sensitive load loss rate evaluation feature library comprises:

[0009] obtaining historical monitoring data, the historical monitoring data comprising voltage sequence and active power sequence when a voltage sag event occurs, constructing a voltage and active power change trajectory over time, and extracting trajectory features from the change trajectory;

[0010] calculating a corresponding sensitive load loss rate according to the user load condition when the voltage sag event occurs;

[0011] constructing training samples based on the trajectory features obtained based on the historical monitoring data and the corresponding sensitive load loss rate, and training to obtain a sensitive load loss rate evaluation model, the output of the sensitive load loss rate evaluation model being the corresponding relationship between the trajectory features and the sensitive load loss rate;

[0012] obtaining a plurality of corresponding relationships between trajectory features and sensitive load loss rates based on the sensitive load loss rate evaluation model, and generating the sensitive load loss rate evaluation feature library.

[0013] Further, the trajectory features comprise voltage sag amplitude, voltage sag duration, active power drop maximum value, active power change duration, active power recovery duration, active power difference before and after the event ends, and active power transformation time delay.

[0014] Further, the trajectory features are extracted using an MK mutation point detection algorithm.

[0015] Further, the calculation formula of the sensitive load loss rate is:

[0016]

[0017] In the formula, P1 represents the active power loss of the user due to the interruption of the production process caused by the voltage sag, P2 represents the active power recovered by the user after the voltage sag or manually recovered, and P3 represents the total rated power of the user.

[0018] Further, the sensitive load loss rate evaluation model is constructed based on a support vector machine.

[0019] Further, a particle swarm algorithm is used to optimize the parameters of the support vector machine.

[0020] Further, when the trajectory features are extracted, strong correlation trajectories of voltage and active power in the change trajectory are obtained first, and the trajectory features are extracted according to the strong correlation trajectories.

[0021] Further, the maximum mutual information coefficient is used to quantify the correlation between voltage and active power to obtain the strong correlation trajectories.

[0022] Further, the strong correlation track is obtained specifically by:

[0023] 1) For the data of voltage U and active power P in the same time period, a series of [U; P] discrete data is obtained by using a sliding window, and the maximum mutual information I is calculated * (U; P):

[0024]

[0025] In the formula, p(U), p(P) are the probability density functions of U and P respectively, and p(U, P) is the joint probability distribution function of U and P;

[0026] 2) The maximum mutual information value is normalized to obtain M(U; P):

[0027]

[0028] 3) The maximum value of the mutual information M(U; P) under different scales is selected as the MIC value:

[0029]

[0030] Wherein, T is the sample size of [U; P], MIC is [0, 1], the greater the value, the stronger the correlation of U-P.

[0031] The application also provides an electronic device, comprising one or more processors, a memory and one or more programs stored in the memory, the one or more programs comprising instructions for performing the voltage sag classification method based on non-intrusive load monitoring as described above.

[0032] Compared with the prior art, the application has the following beneficial effects:

[0033] 1) The application constructs a sensitive load loss rate evaluation feature library, which can extract features based on monitoring data, match the real load loss rate through the feature library, and further evaluate the user voltage sag severity, so that the method is simple and reliable.

[0034] 2) The application adopts the maximum mutual information coefficient to screen the U-P strong correlation time section, which can more conveniently extract the track features and improve the recognition accuracy.

[0035] 3) The application constructs a sensitive load loss rate evaluation model based on a support vector machine, and adopts a particle swarm algorithm PSO to optimize the parameters of the SVM model, so that the model has high reliability. DETAILED DESCRIPTION

[0036] Figure 1 is a schematic diagram of the overall principle of the method of the application;

[0037] Figure 2 Trajectory of U, P for voltage sag established in the embodiment;

[0038] Figure 3 The coordinate points for extracting features by applying the MK mutation point detection algorithm in the embodiment, wherein (a) is a U-P strong correlation time section, (b) is a U trajectory of the strong correlation time section, and (c) is a P trajectory of the strong correlation time section;

[0039] Figure 4 The process for establishing a sensitive load loss rate feature library based on the SVM-PSO of the application. DETAILED DESCRIPTION

[0040] The application will be described in detail below with reference to the drawings and specific embodiments. The embodiments are implemented on the premise of the technical solution of the application, and detailed implementation modes and specific operation processes are given, but the protection scope of the application is not limited to the following embodiments.

[0041] Embodiment 1

[0042] The embodiment provides a voltage sag type identification method based on non-intrusive load monitoring, which comprises the following steps: acquiring user monitoring data, extracting trajectory features from the user monitoring data, comparing the trajectory features with a pre-constructed sensitive load loss rate evaluation feature library, obtaining corresponding sensitive load loss rates, obtaining a user voltage sag type based on the sensitive load loss rates, and reflecting the severity of voltage sag suffered by the user.

[0043] In the embodiment, the method specifically comprises the steps as shown in Figure 1 , which comprises the following contents: content 1: acquiring a strong correlation time section of voltage U and active power P; content 2: extracting features by using the MK mutation point detection; content 3: constructing a sensitive load loss rate feature library; content 4: real-time feature matching and voltage sag type identification, and realizing voltage sag degree evaluation. The specific descriptions of the contents are as follows.

[0044] (1) U-P strong correlation time section screening

[0045] According to high-precision waveform data recorded by a power quality monitoring device, the trajectory of voltage U and active power P with time during the occurrence of a voltage sag event is obtained, which is divided into five stages: a pre-event segment, a start transition segment, an event duration segment, an end transition segment, and a post-event segment, as shown in Figure 2 .

[0046] Since the change of U during the voltage sag period leads to the change of P with weak and time-delayed characteristics, the maximal information coefficient (MIC) is used to quantify the correlation between U and P, which mainly comprises three steps.

[0047] Step 1: For the U and P data of the user within the same time period, use a sliding window to obtain a series of discretized [U; P] data, and calculate the maximum mutual information as shown in Equation (1).

[0048]

[0049] In the formula, p(U) and p(P) are the probability density functions of U and P respectively, and p(U, P) is the joint probability distribution function of U and P.

[0050] Step : Normalize the maximum mutual information value to obtain M(U; P) as shown in Equation (2).

[0051]

[0052] Step 3: Select the maximum value of the mutual information M(U; P) at different scales as the MIC value as shown in Equation (3).

[0053]

[0054] Among them, T is the [U; P] sample size, MIC ∈ [0, 1], and the larger the value, the stronger the correlation between U and P.

[0055] In the specific implementation, the MIC threshold m can be determined through experiments:

[0056] When MIC(U; P) ≥ m, the corresponding section is the time section with strong U - P correlation;

[0057] When MIC(U; P) < m, the corresponding section is the time section with weak U - P correlation.

[0058] (2) Extraction of U - P trajectory features

[0059] According to (1), determine the time section with strong U - P correlation, as Figure 3 shown in (a), and obtain the U - P trajectory of the time section with strong correlation, as shown in (b) and (c). Among them, define the coordinates of the starting point, ending point of the starting transition section and ending transition section of the U trajectory as (t u1 , u1), (t u2 , u2), (t u3 , u3), (t u4 , u4), and correspondingly, the coordinates of the four mutation points of the P trajectory are (t p1 , p1), (t p2 , p2), (t p3 , p3), (t p4p4). According to the actual physical meaning of coordinates, 7 key features are extracted: C1—voltage sag amplitude, C2—voltage sag duration, C3—active power drop maximum, C4—active power change duration, C5—active power recovery duration, C6—active power difference before and after the event, C7—active power transformation delay, as shown in Table 1.

[0060] Therefore, the key point of U-P trajectory feature extraction is the detection of U-P trajectory mutation points, obtaining the accurate data of t u1 -t u4 , t p1 -t p4 , u1-u4, p1-p4. The MK mutation point detection algorithm is used to detect the mutation points of the time series of voltage U and active power P, i.e. U-T and P-T trajectories, and detect the mutation point coordinates of the transition section.

[0061] Table 1 U-P trajectory feature extraction

[0062]

[0063] For N sample data of U, a rank sequence s k is constructed as formula (4).

[0064]

[0065] In the formula, s k is the cumulative value of the voltage at the i th moment greater than the voltage at the j th moment.

[0066] Because the time series U is a random sequence, the statistical quantity UF k is defined as formula (5).

[0067]

[0068] In the formula, UF1=0, E(s k ), Var(s k ) are the mean and variance of s k .

[0069] Because U1, U1, …, U N are independent of each other and have the same continuous distribution, formula (6) can be obtained.

[0070]

[0071] UF k is the standard normal distribution, and is the statistical quantity calculated in the ascending sequence of the time series U. According to the inverse sequence of U, the process of formulas (4) ~ (6) is repeated to calculate UB k' =-UF k,k'=N,N-1,...,1, UB1=0.

[0072] If the significance level α = 0.05, then U α =±1.96, UF k' UB k' Positive and negative sequence curves of two statistics, U α The two critical lines are plotted in a coordinate plane. If UF k' UB k' If an intersection point occurs and lies between the critical lines, the time corresponding to the intersection point is the start time of the transition segment, and the range exceeding the critical lines is the duration of the transition segment. An intersection point position less than 0 corresponds to the start of the transition segment; an intersection point position greater than 0 corresponds to the end of the transition segment.

[0073] (3) Sensitive load loss rate feature library

[0074] a. User's actual sensitive load loss rate

[0075] To objectively describe the impact of voltage sag on user loads, the loss rate of sensitive load (LRSL) is defined as an indicator to assess the severity of voltage sag. Specifically, it refers to the proportion of active power of sensitive loads that cause production interruptions due to voltage sags to the user's total rated power. The calculation formula is shown in equation (7). Relevant data information comes from the user's actual production process.

[0076]

[0077] In the formula, P1 represents the active power lost by the user due to the interruption in production caused by the voltage dip, P2 represents the active power that the user recovers automatically or manually after the voltage dip, and P3 represents the user's total rated power.

[0078] b. Identify the correspondence between features and sensitive load loss rates

[0079] Therefore, the existing user feature samples and user LRSL calculation data are divided into training set and test set. In specific implementation, support vector machine (SVM) and particle swarm optimization (PSO) algorithm can be applied to construct a sensitive load loss rate assessment model based on SVM-PSO to determine the correspondence between feature data and load loss rate.

[0080] Let user sample (x) i ,LRSL i ), x i For the i-th set of feature data C1 to C7, LRSL i Let be the sensitive load loss rate corresponding to the i-th set of features.

[0081] When SVM solves regression problem, it finds the function relation between x and LRSL, such as equation (8), to evaluate the LRSL corresponding to any x.

[0082] LRSL = f(x) = ω T x + b (8)

[0083] In the equation, ω is normal vector, and b is displacement term.

[0084] Introducing variable ξ i , ξ * i , the solution of ω and b is converted into the objective function as equation (9), and the constraint condition as equation (10).

[0085]

[0086]

[0087] In the equation, β is weight coefficient, and δ is the absolute difference between f(x i ) and P load-i .

[0088] Using Lagrange multiplier method to equation (10), equation (11) is obtained.

[0089]

[0090] In the equation, λ is Lagrange multiplier, and K(x i , x) is kernel function.

[0091] Because Gaussian kernel function has strong interpolation ability, it is more conducive to extract local characteristics of user data, and Gaussian kernel function is selected as kernel function, such as equation (12).

[0092] K(x i , x) = exp(-γ||x i -x|| 2 ) γ > 0 (12)

[0093] In the equation, γ is Gaussian kernel function parameter.

[0094] Up to now, two optimization parameters are involved: weight coefficient β and Gaussian kernel function γ. Based on cross-validation principle, in specific implementation, particle swarm optimization (PSO) can be used to search for parameter optimization of SVM model. PSO represents particle by position, velocity and fitness, and the example updates its own position and velocity in constant iteration.

[0095] Suppose the search space of particle is D-dimensional, and n particles constitute population Y = (Y1, Y2, …, Y n ). The position of i-th particle Yi = (y i1 ,y i2 ,…,y iD ) T , the velocity of the i-th particle V i = (V i1 ,V i2 ,…,V iD ) T , the individual extreme value of the i-th particle H i = (H i1 ,H i2 ,…,H iD ) T , the global extreme value of the population composed of n particles H = (H1, H2, …, H D ) T The particle velocity and position are updated according to formula (13) and (14) in the population iteration process.

[0096]

[0097]

[0098] In the formula, d = 1, 2, …, D, i = 1, 2, …, n, k is the current iteration number, ζ is the inertia weight value, a1 and a2 are acceleration factors, and e1 and e2 are random numbers distributed between 0 and 1.

[0099] In other embodiments, other models and other model parameter optimization methods can also be used to construct the sensitive load loss rate evaluation model.

[0100] c. Establishing a sensitive load loss rate evaluation feature library

[0101] Based on a large amount of monitoring data and generation process information of known users, the above model can be applied to determine the corresponding relationship between different voltage sag U-P trajectory features and user load loss rate, thereby forming a sensitive load loss rate evaluation feature library.

[0102] (4) Voltage sag severity evaluation

[0103] For users with unknown production process information, since the true load loss rate cannot be calculated, based on the voltage sag monitoring data, the key features are extracted, the sensitive load loss rate feature library is called, and the load loss rate under the voltage sag is matched to identify the voltage sag category and evaluate the severity of the voltage sag suffered by the user.

[0104] In other embodiments, the U-P trajectory strong correlation time section can not be screened, and the features can be directly extracted from the U-P data waveform.

[0105] The core of the method is to extract key features based on the monitoring data information of existing users through a non-intrusive monitoring method, calculate the load loss rate according to the actual voltage sag influence degree suffered by the user, and determine the corresponding relationship between the features and the load loss rate to construct a voltage sag severity evaluation feature library.

[0106] If the above method is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the present application or the parts that essentially contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0107] Embodiment 2

[0108] The embodiment provides an electronic device, including one or more processors, a memory, and one or more programs stored in the memory, the one or more programs including instructions for performing the voltage sag category identification method based on non-intrusive load monitoring as described in embodiment 1.

[0109] The above detailed the preferred embodiments of the present application. It should be understood that those skilled in the art can make many modifications and changes without creative labor based on the concept of the present application. Therefore, any technical solution obtained by logical analysis, reasoning or limited experiment based on the prior art according to the concept of the present application shall be within the protection scope determined by the claims.

Claims

1. A method for identifying voltage sag categories based on non-intrusive load monitoring, characterized in that, The method includes: acquiring user monitoring data, extracting trajectory features from the user monitoring data, comparing the trajectory features with a pre-built sensitive load loss rate assessment feature library to obtain the corresponding sensitive load loss rate, and obtaining the voltage sag type of the user based on the sensitive load loss rate, wherein the voltage sag type reflects the severity of the voltage sag suffered by the user. The construction process of the sensitive load loss rate assessment feature library includes: Historical monitoring data is acquired, including voltage and active power sequences at the time of voltage sag events. Trajectories of voltage and active power changes over time are constructed, and trajectory features are extracted from these trajectories. Calculate the corresponding sensitive load loss rate based on the user load situation when the voltage sag event occurs; Training samples are constructed using trajectory features obtained from historical monitoring data and corresponding sensitive load loss rates. A sensitive load loss rate assessment model is trained, and the output of the sensitive load loss rate assessment model is the correspondence between trajectory features and sensitive load loss rates. Based on the sensitive load loss rate assessment model, the correspondence between multiple sets of trajectory features and sensitive load loss rates is obtained, and the sensitive load loss rate assessment feature library is generated. The trajectory features include voltage sag amplitude, voltage sag duration, maximum active power drop, active power change duration, active power recovery duration, active power difference before and after the event ends, and active power conversion delay.

2. The voltage sag category identification method based on non-intrusive load monitoring according to claim 1, characterized in that, The trajectory features were extracted using the MK mutation point detection algorithm.

3. The voltage sag category identification method based on non-intrusive load monitoring according to claim 1, characterized in that, The formula for calculating the sensitive load loss rate is as follows: In the formula, P 1 represents the active power lost by the user due to a voltage sag causing a production interruption. P 2 indicates the active power that the user automatically or manually recovers after a voltage dip. P 3 indicates the total fixed power of the user.

4. The voltage sag category identification method based on non-intrusive load monitoring according to claim 1, characterized in that, The sensitive load loss rate assessment model is built based on support vector machines.

5. The voltage sag category identification method based on non-intrusive load monitoring according to claim 4, characterized in that, The parameters of the support vector machine are optimized using the particle swarm optimization algorithm.

6. The voltage sag category identification method based on non-intrusive load monitoring according to claim 1, characterized in that, When extracting the trajectory features, first obtain the strongly correlated trajectories of voltage and active power from the changing trajectories, and then extract the trajectory features based on the strongly correlated trajectories.

7. The voltage sag category identification method based on non-intrusive load monitoring according to claim 6, characterized in that, The correlation between voltage and active power is quantified using the maximum mutual information coefficient to obtain the strongly correlated trajectory.

8. The voltage sag category identification method based on non-intrusive load monitoring according to claim 7, characterized in that, Obtaining the strongly correlated trajectory specifically includes: 1) Regarding voltage within the same time period U and active power P The data is used to obtain a series of values ​​through a sliding window. U ; P Discretize the data and calculate the maximum mutual information. : In the formula, p ( U ), p ( P ) are respectively U, P The probability density function, p ( U , P )for U and P The joint probability distribution function; 2) Normalize the maximum mutual information value to obtain M ( U ; P ): 3) Select mutual information at different scales M ( U ; P The maximum value of ) is taken as the MIC value: in, T yes[ U ; P Sample size, MIC∈[0,1], the larger the value, the stronger the correlation of UP.

9. An electronic device, characterized in that, It includes one or more processors, memory, and one or more programs stored in the memory, said one or more programs including instructions for executing the voltage sag category identification method based on non-intrusive load monitoring as described in any one of claims 1-8.

Images