Abnormal detection method and device for steady-state operation of electric power system and medium

Abstract

The invention discloses an anomaly detection method and device for steady-state operation of an electric power system and a medium. The method comprises the steps of obtaining operation data of the electric power system; establishing a probability model containing input random impact factors according to the operation data, and obtaining probability statistical characteristics of output variables of the power system; performing state prediction and missing data replacement on the probability statistical characteristics through a long-short-term memory network to obtain complete system measurement data; and performing abnormal data detection on the complete system measurement data to obtain an abnormal detection result of the power system. According to the anomaly detection method and device for the steady-state operation of the electric power system and the medium, the probability statistical characteristics of the output variables of the electric power system are obtained by considering uncertainty factors, and the probability statistical characteristics are perfected, so that the integrity and effectiveness of data can be kept, the anomaly condition of the electric power system can be accurately identified, and the accuracy of the anomaly detection is improved. The problem that it is difficult to carry out accurate anomaly detection on a power system can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system network security, and in particular to a method, device and medium for detecting anomalies in steady-state operation of a power system. Background Art

[0002] The power system is composed of power sources, substations, transmission and distribution lines, and load centers. Building a power system based on renewable energy is a complex project. With the high proportion of renewable energy and a large number of power electronic equipment connected to the power grid, random factors have become a part that cannot be underestimated, which brings great challenges to the safe and stable operation of the new power system. The transmission network is affected by random loads and will also experience large-scale current fluctuations. In addition, the generation of a large number of random factors brings certain hidden dangers to power system planning, operation, system risk prediction and other links. At present, it is mainly based on machine learning methods, including support vector machines (SVM), decision trees, random forests, and neural networks, to establish models through training sample sets, classify and discriminate data, and thus achieve anomaly detection.

[0003] However, the output power of renewable energy sources such as wind power generation and photovoltaic power generation is greatly affected by natural factors (such as wind speed and sunshine intensity). The randomness and intermittency of these natural factors lead to strong instability in the output power of renewable energy power generation. This instability increases the complexity and uncertainty of the data, making it difficult for machine learning models based on historical data to accurately predict and identify abnormal conditions in the power system. Summary of the invention

[0004] The present invention provides a method, a device and a medium for detecting anomalies in steady-state operation of an electric power system, so as to solve the problem that it is difficult to accurately detect anomalies in the electric power system.

[0005] Obtaining operation data of power system;

[0006] Establishing a probability model including input random impact factors according to the operation data to obtain probability statistical characteristics of output variables of the power system;

[0007] The state prediction and missing data replacement of the probabilistic statistical characteristics are performed through a long short-term memory network to obtain complete system measurement data;

[0008] Abnormal data detection is performed on the complete system measurement data to obtain an abnormality detection result of the power system.

[0009] By introducing input random influencing factors, the probability model of the present invention can more comprehensively consider the uncertain factors in the operation of the power system, such as load changes, equipment aging, external environment changes, etc.; this processing method makes the model output closer to the actual situation and improves the accuracy of the evaluation; and by incorporating uncertain factors, the output of the model is no longer a single definite value, but a result containing probabilistic statistical characteristics, which helps to more comprehensively understand the operating status of the power system, thereby improving the accuracy of anomaly detection. Using long short-term memory networks to replace missing data can make full use of the contextual information in time series data, generate more reasonable and accurate data completion results, help maintain data integrity, and avoid analysis bias caused by missing data.

[0010] Compared with the prior art, the present invention obtains the probabilistic statistical characteristics of the power system output variables by considering uncertainty factors, and improves the probabilistic statistical characteristics, which helps to maintain the integrity and validity of the data, thereby accurately identifying the abnormal conditions of the power system, and thus can solve the problem of difficulty in accurately detecting abnormalities in the power system.

[0011] As a preferred solution, the state prediction and missing data replacement of the probability statistical characteristics are performed through a long short-term memory network to obtain complete system measurement data, specifically:

[0012] Performing state prediction on the probability statistical characteristics through the long short-term memory network to obtain a state prediction value;

[0013] The missing data of the state prediction value is replaced based on a preset formula set to obtain the complete system measurement data; wherein the formula set is based on the voltage phasors of the bus and the branch, and is established by calculating the active power component transmitted by the branch.

[0014] In the process of data processing and analysis, the existence of missing values ​​often leads to the accumulation and propagation of errors. By replacing missing values, the risk of error propagation can be reduced.

[0015] As a preferred solution, abnormal data detection is performed on the complete system measurement data to obtain an abnormality detection result of the power system, specifically:

[0016] Performing dimensionality reduction processing and linear transformation on the complete system measurement data to obtain an actual random impact factor data set of the power system;

[0017] Whether the power system is abnormal is judged according to the probability density value of the data in the actual random impact factor data set, and an abnormality detection result of the power system is obtained.

[0018] In the power system, due to the large number of monitoring points and the huge amount of data, direct processing of raw data may be very time-consuming and resource-intensive; through dimensionality reduction, the main information in the data can be retained while removing redundancy and noise, making subsequent analysis and judgment more efficient. Linear transformation can further process the reduced-dimensional data to make it more in line with the needs of specific analysis.

[0019] As a preferred solution, the complete system measurement data is subjected to dimensionality reduction processing and linear transformation to obtain an actual random impact factor data set of the power system, specifically:

[0020] Extract random impact factor data from the complete system measurement data to obtain a random impact factor data set;

[0021] Centralizing the random impact factor data set to obtain a first data set;

[0022] Based on the first data set, data dimension reduction is performed by calculating a covariance matrix to obtain a number of eigenvectors;

[0023] The plurality of characteristic vectors are linearly transformed to obtain the actual random impact factor data set of the power system.

[0024] This preferred solution can eliminate the bias that may be introduced in the data due to different dimensions or average levels by centralizing the random impact factor data set, making subsequent analysis more accurate. Calculating the covariance matrix and performing dimensionality reduction based on its eigenvalues ​​can retain the main features in the data while removing noise and redundant information.

[0025] As a preferred solution, judging whether the power system is abnormal is performed according to the probability density value of the data in the actual random impact factor data set, and obtaining the abnormality detection result of the power system, specifically:

[0026] According to a preset probability density formula, the probability density value of the data in the actual random influencing factor data set is calculated;

[0027] If the probability density value is less than a preset abnormality detection threshold, the abnormality detection result of the power system is defined as the power system abnormality.

[0028] This preferred solution can quantitatively evaluate the actual random impact factor data set by calculating the probability density value, avoiding the subjectivity and inconsistency of human judgment. Once the probability density value of the actual random impact factor data is lower than the preset anomaly detection threshold, the system can immediately trigger an anomaly detection alarm, which helps maintain the safety of the power system.

[0029] As a preferred solution, the probability density formula is specifically:

[0030]

[0031] Among them, α k is the coefficient of the kth Gaussian component, f(z|θ k ) is the Gaussian distribution density function, and K is the total number of Gaussian components.

[0032] As a preferred solution, a probability model including input random impact factors is established according to the operation data to obtain the probability statistical characteristics of the output variables of the power system, specifically:

[0033] Based on the operating data, a wind speed probability model is obtained by establishing a two-parameter Weibull distribution model for wind speed;

[0034] By establishing a Beta distribution model for light intensity, a light intensity probability model is obtained;

[0035] By establishing a normal distribution model of load, a load probability model is obtained;

[0036] The probability model including input random impact factors is formed by the wind speed probability model, the light intensity probability model and the load probability model, and the probability statistical characteristics about the output variables of the power system are output according to the probability model including input random impact factors.

[0037] In this preferred solution, each random influencing factor is described using the probability distribution model that best suits its characteristics; Weibull distribution is suitable for describing data with long-tail characteristics such as wind speed, Beta distribution is suitable for describing data that varies within a limited range such as light intensity, and normal distribution is often used to describe data with average trends and random fluctuations such as load. This precise modeling improves the accuracy of the description of the uncertainty of the power system; therefore, based on the constructed probability model, the probability statistical characteristics of the output variables of the power system can be effectively obtained.

[0038] The present application also provides an abnormality detection device for steady-state operation of a power system, comprising a data module, a characteristic module, a perfection module and a detection module;

[0039] Wherein, the data module is used to obtain the operation data of the power system;

[0040] The characteristic module is used to establish a probability model including input random impact factors according to the operation data to obtain the probability statistical characteristics of the output variables of the power system;

[0041] The improvement module is used to perform state prediction and missing data replacement on the probability statistical characteristics through a long short-term memory network to obtain complete system measurement data;

[0042] The detection module is used to perform abnormal data detection on the complete system measurement data to obtain an abnormality detection result of the power system.

[0043] As a preferred solution, the improvement module includes a state unit and a replacement unit;

[0044] Wherein, the state unit is used to perform state prediction on the probability statistical characteristic through the long short-term memory network to obtain a state prediction value;

[0045] The replacement unit is used to replace missing data of the state prediction value based on a preset formula set to obtain the complete system measurement data; wherein the formula set is based on the voltage phasors of the bus and the branch, and is established by calculating the active power component transmitted by the branch.

[0046] As a preferred solution, the detection module includes a data unit and an abnormality unit;

[0047] The data unit is used to perform dimensionality reduction processing and linear transformation on the complete system measurement data to obtain an actual random impact factor data set of the power system;

[0048] The abnormality unit is used to determine whether the power system is abnormal according to the probability density value of the data in the actual random impact factor data set, and obtain the abnormality detection result of the power system.

[0049] As a preferred solution, the data unit includes an extraction subunit, a processing subunit, a dimensionality reduction subunit and a transformation subunit;

[0050] Wherein, the extraction subunit is used to extract random impact factor data from the complete system measurement data to obtain a random impact factor data set;

[0051] The processing subunit is used to perform centralization processing on the random impact factor data set to obtain a first data set;

[0052] The dimension reduction subunit is used to perform data dimension reduction based on the first data set by calculating the covariance matrix to obtain a plurality of eigenvectors;

[0053] The transformation subunit is used to perform linear transformation on the plurality of characteristic vectors to obtain the actual random impact factor data set of the power system.

[0054] As a preferred solution, the abnormality unit includes a probability subunit and a judgment subunit;

[0055] Wherein, the probability subunit is used to calculate the probability density value of the data in the actual random impact factor data set according to a preset probability density formula;

[0056] The judgment subunit is used to define the abnormality detection result of the power system as the power system abnormality if the probability density value is less than a preset abnormality detection threshold.

[0057] As a preferred solution, the probability density formula is specifically:

[0058]

[0059] Among them, α k is the coefficient of the kth Gaussian component, f(z|θ k ) is the Gaussian distribution density function, and K is the total number of Gaussian components.

[0060] As a preferred solution, the characteristic module includes a wind speed unit, a light intensity unit, a load unit and an output unit;

[0061] Wherein, the wind speed unit is used to obtain a wind speed probability model by establishing a two-parameter Weibull distribution model about wind speed based on the operating data;

[0062] The light intensity unit is used to obtain a light intensity probability model by establishing a Beta distribution model about light intensity;

[0063] The load unit is used to obtain a load probability model by establishing a normal distribution model about the load;

[0064] The output unit is used to form the probability model containing input random influence factors by using the wind speed probability model, the light intensity probability model and the load probability model, and output the probability statistical characteristics about the output variables of the power system according to the probability model containing the input random influence factors.

[0065] The present application also provides a storage medium, on which a computer program is stored. The computer program is called and executed by a computer to implement the above-mentioned method for detecting anomalies in steady-state operation of a power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 It is a flow chart of an abnormality detection method for steady-state operation of a power system provided in an embodiment of the present application;

[0067] Figure 2 It is a block diagram of a novel power system provided by an embodiment of the present application;

[0068] Figure 3 It is a structural block diagram provided by an embodiment of the present application;

[0069] Figure 4 It is a data structure diagram provided by an embodiment of the present application;

[0070] Figure 5 It is a structural schematic diagram of an abnormality detection device for steady-state operation of an electric power system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0071] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0072] In the description of this application, it should be understood that the term "first" is used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" may explicitly or implicitly include one or more of the features. In the description of this application, unless otherwise specified, "several" means two or more.

[0073] An anomaly detection method for steady-state operation of an electric power system provided in an embodiment of the present application is mainly used in situations where it is necessary to overcome the defects of low efficiency, large data size and computational difficulty of traditional electric power system analysis methods and to perform accurate anomaly detection on the electric power system.

[0074] Embodiment 1:

[0075] See also Figure 1 The embodiment of the present application provides a method for detecting anomalies in steady-state operation of a power system, including S1 to S4, and the specific implementation steps are as follows:

[0076] S1. Obtaining the operation data of the power system.

[0077] Step S1 of the embodiment of the present application is specifically as follows:

[0078] Obtain the operating data of the power system; wherein the power system is a new type of power system integrating source, grid, load and storage; and, source refers to power production, grid refers to the transmission and distribution network, load refers to power load, and storage refers to electric energy storage.

[0079] To apply the embodiments of this application, please refer to Figure 2 , Figure 2 It is a block diagram of a new power system provided in an embodiment of the present application, representing the operating framework of the new power system integrating source, grid, load and storage in this embodiment.

[0080] S2. A probability model including input random influencing factors is established based on the operating data to obtain the probability statistical characteristics of the output variables of the power system.

[0081] Step S2 of the embodiment of the present application is specifically as follows:

[0082] Based on the operating data, a two-parameter Weibull distribution model for wind speed is established to obtain a wind speed probability model.

[0083] By establishing a Beta distribution model for light intensity, a light intensity probability model is obtained;

[0084] By establishing a normal distribution model of load, a load probability model is obtained;

[0085] A probability model including input random impact factors is formed by a wind speed probability model, a light intensity probability model and a load probability model, and a probability statistical characteristic about an output variable of a power system is output according to the probability model including the input random impact factors.

[0086] Among them, the wind speed probability model is:

[0087]

[0088] The light intensity probability model is:

[0089]

[0090] The load probability model is:

[0091]

[0092] Where v represents the wind speed, k and c are the shape parameter and scale parameter of the Weibull distribution respectively;

[0093] r is the light intensity, a and b are the shape parameters of the Beta distribution, r max is the maximum light intensity at the location, Γ is the Gamma function;

[0094] P L is the power load, P L0 is the load power reference value, σ L is the load power standard deviation.

[0095] In the present embodiment S2, each random influencing factor is described by a probability distribution model that best suits its characteristics. For a wind turbine, its output power will change with the fluctuation of the local wind speed, and the Weibull distribution is suitable for describing data with long-tail characteristics such as wind speed. The output power of a photovoltaic power source is mainly affected by light intensity, so the Beta distribution is suitable for describing data that varies within a limited range such as light intensity. The normal distribution is often used to describe data with average trends and random fluctuations such as loads. This precise modeling improves the accuracy of describing the uncertainty of the power system.

[0096] Since the input random variables in the new power system are mainly composed of renewable energy and uncertain loads, the probabilistic statistical characteristics of the power system output variables can be effectively obtained based on the constructed probability model.

[0097] S3. Use long short-term memory networks to perform state prediction and replace missing data on probabilistic statistical characteristics to obtain complete system measurement data.

[0098] Step S3 of the embodiment of the present application includes S3.1 to S3.3, specifically:

[0099] S3.1. In a short time interval, it is assumed that the power system is in a quasi-steady-state operation state, and there is no large mutation in the load and generator. Therefore, for two successfully adjacent system equilibrium operating points, there exists:

[0100] f(x k+1 ,X k+1 )=0;

[0101] f(x k ,X k )=0

[0102] Among them, x k and X k represents the system variables at a certain time step (time “k”), x k+1 and X k+1 Represents the system variables at the next time step (time “k+1”).

[0103] Need to x k+1 Perform robust estimation, combine the improved PS and Huber convex fractional function GM estimator, and establish the objective function J(x), that is:

[0104]

[0105] For the parameter ρ(r S ) and r S have:

[0106]

[0107] For the parameter r i have:

[0108]

[0109] Among them, m is the total number of samples, i is the serial number of the sample, ρ(r S ) is the Huber function, its first-order derivative is related to r S Related, ψ(r S ) represents the Huber function ρ(r S) relative to the standardized residual r S The first derivative of ;

[0110] The parameter β is set to 1.5, which is efficient under Gaussian noise; r S is the standardized residual, r i represents the residual of the ith observation, y i represents the i-th observation value, is the eigenvector corresponding to the i-th observation data, is the estimated value of the target variable; By y k+1 =A k+1 x k+1 +η k+1 Derived from y k+1 represents the observation model of the system at time step k+1, x k+1 is the state vector of the system at time k+1, η k+1 is the noise term of the system at time k+1; Representation Matrix is the state matrix A of the system k+1 The transposed matrix of

[0111] s=1.4826b m median i |r i | is a robust scale estimate, when the observed value follows a normal distribution When asymptotically tends to the standard deviation σ of the normal distribution; median i |r i | is the median of all absolute residuals, where i represents an index, referring to the i-th element of a set of data; b m is a constant used to transform the median into a robust scaling estimate;

[0112] Due to η k+1 The covariance of is an identity matrix, so s asymptotically approaches 1; is through Determined on PS; Represents the weight, which is used to adjust the influence of each data point on the model estimation; z k+1 is the observed value of the system at time k+1, g(x k+1|k ) is the system state prediction at time k+1 based on time k, x k+1|k is the system state prediction value at time k+1 based on time k, x k|k-1 is the predicted value of the system state at time k based on time k-1, z k -g(x k|k-1 ) is the observation residual at time k, z kis the observed value at time k, g(x k|k-1 ) is the state prediction at time k;

[0113] And, in the proposed robust estimation, we use To constrain the influence of measurement error and structural error, and to constrain the influence of residual error through Huber function.

[0114] S3.2. Based on the objective function J(x), the state prediction is performed on the probability statistical characteristics through the trained long short-term memory network (LSTM) to obtain the state prediction value; wherein the long short-term memory network consists of an input layer, a hidden layer and an output layer, and each LSTM unit is equipped with a storage unit, and its state at period t is recorded as C t ; and in cycle t, the LSTM unit accepts three types of external information input: the current state x t , hidden state h t-1 and cell status C t-1 ;

[0115] Among them, the specific workflow of the long short-term memory network can be divided into three steps:

[0116] ① Forgetting stage. The cell state at time t-1 is selectively forgotten, that is, the unimportant information is forgotten and only the key information is remembered. The forgetting gate is used to control the degree of forgetting of cell information.

[0117] ②Selective memory stage. t and h t-1 It is selective memory, that is, the more important the input information is, the more it is remembered. The input gate and tanh function are used to control the degree of memory of the input information.

[0118] ③ Output stage. In this stage, it will be determined which one is used as the output of the current state. The output gate is used to control the state of the output information. In addition, C t It will also be scaled by the tanh function to get h t .

[0119] Among them, the variable calculation formula of the long short-term memory network during specific work is:

[0120]

[0121] The prediction equation of the long short-term memory network is:

[0122] X t =L(h t-1 ,c t-1 ,[X t-1 ,X t-2 ,…X t-ρ ])

[0123] Among them, x t is the input of the current time step, h t is the hidden state of the current time step, α represents h t and x t The concatenation of , σ is the sigmoid activation function;

[0124] f t is the forget gate output of the current time step, W f is the weight matrix of the forget gate, b f is the bias term of the forget gate;

[0125] i t is the input gate output at the current time step, W i is the weight matrix of the forget gate, b i is the bias term of the forget gate;

[0126] o t is the output gate output of the current time step, W o is the weight matrix of the forget gate, b o is the bias term of the forget gate;

[0127] c t is the final cell state at the current time step, c t-1 is the final cell state at the previous time step;

[0128] h t is the hidden state of the current time step, h t-1 is the final unit state at the previous time step, and tanh is the hyperbolic tangent activation function;

[0129] L represents the preset function, [X t-1 ,X t-2 ,…X t-ρ ] is the current state x in the equation t , X t Equivalent to h t ; ρ is the time step, indicating prediction using power state.

[0130] To apply the embodiments of this application, please refer to Figure 3 , Figure 3 It is a structural block diagram provided by the embodiment of the present application, such as Figure 3 As shown in the figure, the reading and modification of the LSTM storage unit is achieved by controlling the input gate, forget gate and output gate.

[0131] S3.3. Replace missing data of the state prediction value based on a preset formula set to obtain complete system measurement data; wherein the formula set is based on the voltage phasors of the bus and the branch, and is established by calculating the active power component transmitted by the branch.

[0132] The formula set is:

[0133]

[0134] in, is the active power injection of bus n in period t; is the active power of branch nn′ in period t; is the reactive power injection of bus n in period t; is the reactive power of branch nn′ in period t; G nn′ is the conductance of branch nn′; B nn′ is the susceptance of branch nn′; is the phase difference between bus n and branch n′ in period t; and are the voltage phasors of busbar n and branch line n′, respectively. Represents the square of the voltage of bus n at time t.

[0135] To apply the embodiments of this application, please refer to Figure 4 , Figure 4 It is a data structure diagram provided in an embodiment of the present application, indicating a situation where there is missing data in the state prediction value of the power system.

[0136] In this embodiment S3.3, due to the data missing phenomenon in the measurement of the system, after obtaining the state prediction value, the measurement information of the relevant nodes can be obtained through prediction and flow calculation. Flow calculation is to calculate the distribution of active power, reactive power and voltage in the power grid when the network topology, component parameters and power generation and load parameters of the power system are known. The calculation purpose of this embodiment is to determine the steady-state operating state parameters of each part of the power system to meet the needs of system planning, operation and control; therefore, by replacing the missing values, the risk of such error propagation can be reduced, thereby obtaining complete system measurement data to accurately estimate the state of the system.

[0137] S4. Perform abnormal data detection on the complete system measurement data to obtain abnormal data detection results of the power system.

[0138] Step S4 of the embodiment of the present application includes S4.1 to S4.2, specifically:

[0139] S4.1. Extract random impact factor data from the complete system measurement data and the random impact factor monitoring system to obtain a random impact factor data set; wherein the random impact factor monitoring system is established by deploying and monitoring important nodes at various levels in the power system operating environment without affecting the normal operation of the power system, and the random impact factor monitoring system mainly includes three levels: new energy output, power load and power grid topology;

[0140] The random impact factor data set is centralized to obtain a first data set;

[0141] A covariance matrix is ​​established based on the first data set, and the eigenvalues ​​and corresponding eigenvectors of the covariance matrix are found by solving the characteristic equation to obtain several eigenvectors H; where H = h1, h2, …, h N ;

[0142] Perform linear transformation on several eigenvectors H to obtain the actual random impact factor data set X′ of the power system; where X′=(X1′,X2′,…,X′ M ).

[0143] Among them, suppose that M N-dimensional actual random impact factor vectors form a matrix X0 = (X1, X2, ..., X M ), then the formula for centralizing all random influence factor vectors in the random influence factor data set is:

[0144]

[0145] The covariance matrix is:

[0146]

[0147] The characteristic equation is:

[0148] |Cov-λI|=0

[0149] in, is the parameter mean in the actual random influence factor data set, T represents the transpose, M is a matrix related to the sample size, standardization or other preprocessing steps, λ is the eigenvalue to be solved in the covariance matrix Cov, I is the unit matrix, and the eigenvalues ​​α1, α2, …, α are decomposed. N , and the corresponding eigenvectors h1,h2,…,h N , where α1≤α2≤…≤α N .

[0150] Among them, random impact factor data is a special time series data, and its related mathematical definition is:

[0151] ① is an ordered set of n continuous data points, that is, T = [X1, X2, ..., X n ], with a length of n, for any time t, X t ∈R M×E ,m∈N + ,E∈N; where R M×E represents the M×E dimensional real number space; m is a positive integer representing the number of consecutive data points selected in the subsequence; N+ represents the set of positive integers, and N represents the set of integers.

[0152] ② A subsequence is defined as the sampling of m consecutive positions of T, that is, C = [X t-m+1 ,X t-m+2 ,…,X t ], with length m, where m≤t≤n, and t is the time index.

[0153] ③ Given a T of length n and a user-defined subsequence length of m, all possible subsequences are extracted by sliding a window of size w on T with a certain step size, where the step size is generally set to 1.

[0154] It should be noted that due to the diversity, complexity, randomness and other characteristics of random influencing factors, it is difficult to summarize them comprehensively. In the embodiment of the present application, assuming that the power system is in normal operation most of the time, the anomaly of the random influencing factor is defined as a random influencing factor vector that is significantly different from the normal data at a certain time point.

[0155] In this embodiment S4.1, by centralizing the random impact factor data set, the bias that may be introduced in the data due to different dimensions or average levels can be eliminated, making subsequent analysis more accurate. Calculating the covariance matrix and performing dimensionality reduction based on its eigenvalues ​​can retain the main features in the data while removing noise and redundant information;

[0156] Moreover, in the power system, due to the large number of monitoring points and the huge amount of data, directly processing the raw data may be very time-consuming and resource-intensive; through dimensionality reduction, the main information in the data can be retained while removing redundancy and noise, making subsequent analysis and judgment more efficient. Linear transformation can further process the reduced-dimensional data to make it more in line with the needs of specific analysis.

[0157] S4.2. Calculate the probability density value of the data in the actual random impact factor data set X′ according to the preset probability density formula;

[0158] If the probability density value is less than the preset anomaly detection threshold ε, the anomaly detection result of the power system is defined as an anomaly of the power system, and an anomaly alarm is issued to the operation and maintenance personnel according to the relevant alarm strategy; wherein, the anomaly detection threshold ε is the minimum probability density of historical data, and the probability density formula is established based on the Gaussian mixture model (GMM);

[0159] Moreover, if the probability density value is greater than or equal to the anomaly detection threshold ε, it indicates that the power system is abnormal, and the actual random impact factor at this time corresponds to the system equilibrium operating point x k+1 , that is, we get x k+1 The robust estimation results of .

[0160] The probability density formula is:

[0161]

[0162] Among them, α k is the coefficient of the kth Gaussian component, f(z|θ k ) is the Gaussian distribution density function, and K is the total number of Gaussian components.

[0163] In this embodiment S4.2, by calculating the probability density value, the actual random impact factor data set can be quantitatively evaluated, avoiding the subjectivity and inconsistency of human judgment. Once the probability density value of the actual random impact factor data is lower than the preset anomaly detection threshold, the system can immediately trigger an anomaly detection alarm, which helps maintain the safety of the power system;

[0164] Moreover, the above scheme can be defined as prediction-assisted state estimation, that is: according to the GM estimator of the improved PS and Huber convex fractional function, the stable operation of the power system is analyzed, including the long short-term memory model (LSTM) state prediction step, the missing data replacement step through power flow calculation and the abnormal data detection step; the prediction-assisted state estimation uses the state space model to show the quasi-steady-state dynamic situation of the system state, which can track the state evolution, and is helpful for modern energy management centers to carry out real-time modeling and control of power systems.

[0165] Overall, this embodiment has the following beneficial effects:

[0166] By introducing input random influencing factors, the probability model of the present invention can more comprehensively consider the uncertain factors in the operation of the power system, such as load changes, equipment aging, external environment changes, etc.; this processing method makes the model output closer to the actual situation and improves the accuracy of the evaluation; and by incorporating uncertain factors, the output of the model is no longer a single definite value, but a result containing probabilistic statistical characteristics, which helps to more comprehensively understand the operating status of the power system, thereby improving the accuracy of anomaly detection. Using long short-term memory networks to replace missing data can make full use of the contextual information in time series data, generate more reasonable and accurate data completion results, help maintain data integrity, and avoid analysis bias caused by missing data.

[0167] Embodiment 2:

[0168] See also Figure 5 , an embodiment of the present application provides an abnormality detection device for steady-state operation of a power system, comprising a data module 10, a characteristic module 20, a perfecting module 30 and a detection module 40;

[0169] Wherein, the data module 10 is used to obtain the operation data of the power system;

[0170] The characteristic module 20 is used to establish a probability model including input random impact factors according to the operation data to obtain the probability statistical characteristics of the output variables of the power system;

[0171] The improvement module 30 is used to perform state prediction and missing data replacement on the probability statistical characteristics through the long short-term memory network to obtain complete system measurement data;

[0172] The detection module 40 is used to perform abnormal data detection on the complete system measurement data to obtain abnormal data detection results of the power system.

[0173] In one embodiment, the data module 10 is specifically:

[0174] Obtain the operating data of the power system; wherein the power system is a new type of power system integrating source, grid, load and storage; and, source refers to power production, grid refers to the transmission and distribution network, load refers to power load, and storage refers to electric energy storage.

[0175] To apply the embodiments of this application, please refer to Figure 2 , Figure 2 It is a block diagram of a new power system provided in an embodiment of the present application, representing the operating framework of the new power system integrating source, grid, load and storage in this embodiment.

[0176] In one embodiment, the characteristic module 20 includes a wind speed unit, a light intensity unit, a load unit, and an output unit;

[0177] The wind speed unit is used to obtain a wind speed probability model by establishing a two-parameter Weibull distribution model for wind speed based on the operating data;

[0178] A light intensity unit is used to obtain a light intensity probability model by establishing a Beta distribution model about light intensity;

[0179] A load unit is used to obtain a load probability model by establishing a normal distribution model of the load;

[0180] The output unit is used to form a probability model including input random impact factors by means of a wind speed probability model, a light intensity probability model and a load probability model, and output probability statistical characteristics of output variables of the power system according to the probability model including the input random impact factors.

[0181] Among them, the wind speed probability model is:

[0182]

[0183] The light intensity probability model is:

[0184]

[0185] The load probability model is:

[0186]

[0187] Where v represents the wind speed, k and c are the shape parameter and scale parameter of the Weibull distribution respectively;

[0188] r is the light intensity, a and b are the shape parameters of the Beta distribution, r max is the maximum light intensity at the location, Γ is the Gamma function;

[0189] P L is the power load, P L0 is the load power reference value, σ L is the load power standard deviation.

[0190] In the characteristic module 20 of this embodiment, each random influencing factor is described by a probability distribution model that best suits its characteristics. For a wind turbine, its output power will change with the fluctuation of the local wind speed, and the Weibull distribution is suitable for describing data with long-tail characteristics such as wind speed. The output power of a photovoltaic power source is mainly affected by light intensity, so the Beta distribution is suitable for describing data that varies within a limited range such as light intensity. The normal distribution is often used to describe data with average trends and random fluctuations such as loads. This precise modeling improves the accuracy of describing the uncertainty of the power system.

[0191] Since the input random variables in the new power system are mainly composed of renewable energy and uncertain loads, the probabilistic statistical characteristics of the power system output variables can be effectively obtained based on the constructed probability model.

[0192] In one embodiment, the perfection module 30 includes a target unit, a state unit, and a replacement unit;

[0193] Among them, the target unit is used to assume that the power system is in a quasi-steady-state operation state within a short time interval, and there is no large mutation in the load and generator. Therefore, for two successfully adjacent system balanced operating points, there exists:

[0194] f(x k+1 ,X k+1 )=0;

[0195] f(x k ,X k )=0

[0196] Among them, x k and X k represents the system variables at a certain time step (time “k”), x k+1and X k+1 Represents the system variables at the next time step (time “k+1”).

[0197] Need to x k+1 Perform robust estimation, combine the improved PS and Huber convex fractional function GM estimator, and establish the objective function J(x), that is:

[0198]

[0199] For the parameter ρ(r S ) and r S have:

[0200]

[0201] For the parameter r i have:

[0202]

[0203] Among them, m is the total number of samples, i is the serial number of the sample, ρ(r S ) is the Huber function, its first-order derivative is related to r S Related, ψ(r S ) represents the Huber function ρ(r S ) relative to the standardized residual r S The first derivative of ;

[0204] The parameter β is set to 1.5, which is efficient under Gaussian noise; r S is the standardized residual, r i represents the residual of the ith observation, y i represents the i-th observation value, is the eigenvector corresponding to the i-th observation data, is the estimated value of the target variable; By y k+1 =A k+1 x k+1 +η k+1 Derived from y k+1 represents the observation model of the system at time step k+1, x k+1 is the state vector of the system at time k+1, η k+1 is the noise term of the system at time k+1; Representation Matrix is the state matrix A of the system k+1 The transposed matrix of

[0205] s=1.4826b m median i |ri | is a robust scale estimate, when the observed value follows a normal distribution When asymptotically tends to the standard deviation σ of the normal distribution; median i |r i | is the median of all absolute residuals, where i represents an index, referring to the i-th element of a set of data; b m is a constant used to transform the median into a robust scaling estimate;

[0206] Due to η k+1 The covariance of is an identity matrix, so s asymptotically approaches 1; is through Determined on PS; Represents the weight, which is used to adjust the influence of each data point on the model estimation; z k+1 is the observed value of the system at time k+1, g(x k+1|k ) is the system state prediction at time k+1 based on time k, x k+1|k is the system state prediction value at time k+1 based on time k, x k|k-1 is the predicted value of the system state at time k based on time k-1, z k -g(x k|k-1 ) is the observation residual at time k, z k is the observed value at time k, g(x k|k-1 ) is the state prediction at time k;

[0207] And, in the proposed robust estimation, we use To constrain the influence of measurement error and structural error, and to constrain the influence of residual error through Huber function.

[0208] The state unit is used to predict the state of the probability statistical characteristics based on the objective function J(x) through the trained long short-term memory network (LSTM) to obtain the state prediction value; wherein the long short-term memory network consists of an input layer, a hidden layer and an output layer, and each LSTM unit is equipped with a storage unit, and its state at period t is recorded as C t ; and in cycle t, the LSTM unit accepts three types of external information input: the current state x t , hidden state h t-1 and cell status C t-1 ;

[0209] Among them, the specific workflow of the long short-term memory network can be divided into three steps:

[0210] ① Forgetting stage. The cell state at time t-1 is selectively forgotten, that is, the unimportant information is forgotten and only the key information is remembered. The forgetting gate is used to control the degree of forgetting of cell information.

[0211] ②Selective memory stage. t and h t-1 It is selective memory, that is, the more important the input information is, the more it is remembered. The input gate and tanh function are used to control the degree of memory of the input information.

[0212] ③ Output stage. In this stage, it will be determined which one is used as the output of the current state. The output gate is used to control the state of the output information. In addition, C t It will also be scaled by the tanh function to get h t .

[0213] Among them, the variable calculation formula of the long short-term memory network during specific work is:

[0214]

[0215] The prediction equation of the long short-term memory network is:

[0216] X t =L(h t-1 ,c t-1 ,[X t-1 ,X t-2 ,…X t-ρ ])

[0217] Among them, x t is the input of the current time step, h t is the hidden state of the current time step, α represents h t and x t The concatenation of , σ is the sigmoid activation function;

[0218] f t is the forget gate output of the current time step, W f is the weight matrix of the forget gate, b f is the bias term of the forget gate;

[0219] i t is the input gate output at the current time step, W i is the weight matrix of the forget gate, b i is the bias term of the forget gate;

[0220] o t is the output gate output of the current time step, W o is the weight matrix of the forget gate, b o is the bias term of the forget gate;

[0221] ct is the final cell state at the current time step, c t-1 is the final cell state at the previous time step;

[0222] h t is the hidden state of the current time step, h t-1 is the final unit state at the previous time step, and tanh is the hyperbolic tangent activation function;

[0223] L represents the preset function, [X t-1 ,X t-2 ,…X t-ρ ] is the current state x in the equation t , X t Equivalent to h t ; ρ is the time step, indicating prediction using power state.

[0224] To apply the embodiments of this application, please refer to Figure 3 , Figure 3 It is a structural block diagram provided by the embodiment of the present application, such as Figure 3 As shown in the figure, the reading and modification of the LSTM storage unit is achieved by controlling the input gate, forget gate and output gate.

[0225] The replacement unit is used to replace missing data of the state prediction value based on a preset formula set to obtain complete system measurement data; wherein the formula set is based on the voltage phasors of the busbar and the branch line and is established by calculating the active power component transmitted by the branch line.

[0226] The formula set is:

[0227]

[0228] in, is the active power injection of bus n in period t; is the active power of branch nn′ in period t; is the reactive power injection of bus n in period t; is the reactive power of branch nn′ in period t; G nn′ is the conductance of branch nn′; B nn′ is the susceptance of branch nn′; is the phase difference between bus n and branch n′ in period t; and are the voltage phasors of busbar n and branch line n′, respectively. Represents the square of the voltage of bus n at time t.

[0229] To apply the embodiments of this application, please refer to Figure 4 , Figure 4It is a data structure diagram provided in an embodiment of the present application, indicating a situation where there is missing data in the state prediction value of the power system.

[0230] In the replacement unit of this embodiment, due to the data missing phenomenon in the measurement of the system, after obtaining the state prediction value, the measurement information of the relevant nodes can be obtained through prediction and flow calculation. Flow calculation is to calculate the distribution of active power, reactive power and voltage in the power grid when the network topology, component parameters and power generation and load parameters of the power system are known. The calculation purpose of this embodiment is to determine the steady-state operating state parameters of each part of the power system to meet the needs of system planning, operation and control; therefore, by replacing the missing values, the risk of error propagation can be reduced, thereby obtaining complete system measurement data to accurately estimate the state of the system.

[0231] In one embodiment, the detection module 40 includes an extraction subunit, a processing subunit, a dimension reduction subunit, a transformation subunit, a probability subunit, and a judgment subunit;

[0232] Among them, the extraction subunit is used to extract random impact factor data from the complete system measurement data and the random impact factor monitoring system to obtain a random impact factor data set; wherein, the random impact factor monitoring system is established by deploying and monitoring important nodes at various levels in the power system operation environment without affecting the normal operation of the power system, and the random impact factor monitoring system mainly includes three levels: new energy output, power load and power grid topology;

[0233] A processing subunit, used for performing centralization processing on the random impact factor data set to obtain a first data set;

[0234] The dimension reduction subunit is used to establish a covariance matrix based on the first data set, and find the eigenvalues ​​and corresponding eigenvectors of the covariance matrix by solving the characteristic equation to obtain a number of eigenvectors H; where H = h1, h2, ..., h N ;

[0235] The transformation subunit is used to perform linear transformation on several eigenvectors H to obtain the actual random impact factor data set X′ of the power system; where X′=(X1′,X2′,…,X′ M ).

[0236] Among them, suppose that M N-dimensional actual random impact factor vectors form a matrix X0 = (X1, X2, ..., X M ), then the formula for centralizing all random influence factor vectors in the random influence factor data set is:

[0237]

[0238] The covariance matrix is:

[0239]

[0240] The characteristic equation is:

[0241] |Cov-λI|=0

[0242] in, is the parameter mean in the actual random influence factor data set, T represents the transpose, M is a matrix related to the sample size, standardization or other preprocessing steps, λ is the eigenvalue to be solved in the covariance matrix Cov, I is the unit matrix, and the eigenvalues ​​α1, α2, …, α are decomposed. N , and the corresponding eigenvectors h1,h2,…,h N , where α1≤α2≤…≤α N .

[0243] Among them, random impact factor data is a special time series data, and its related mathematical definition is:

[0244] ① is an ordered set of n continuous data points, that is, T = [X1, X2, ..., X n ], with a length of n, for any time t, X t ∈R M×E ,m∈N + ,E∈N; where R M×E represents the M×E dimensional real number space; m is a positive integer representing the number of consecutive data points selected in the subsequence; N + represents the set of positive integers, and N represents the set of integers.

[0245] ② A subsequence is defined as the sampling of m consecutive positions of T, that is, C = [X t-m+1 ,X t-m+2 ,…,X t ], with length m, where m≤t≤n, and t is the time index.

[0246] ③ Given a T of length n and a user-defined subsequence length of m, all possible subsequences are extracted by sliding a window of size w on T with a certain step size, where the step size is generally set to 1.

[0247] It should be noted that due to the diversity, complexity, randomness and other characteristics of random influencing factors, it is difficult to summarize them comprehensively. In the embodiment of the present application, assuming that the power system is in normal operation most of the time, the anomaly of the random influencing factor is defined as a random influencing factor vector that is significantly different from the normal data at a certain time point.

[0248] The extraction subunit, processing subunit, dimensionality reduction subunit and transformation subunit of this embodiment can eliminate the deviation that may be introduced in the data due to different dimensions or average levels by centralizing the random influencing factor data set, making subsequent analysis more accurate. Calculating the covariance matrix and performing dimensionality reduction based on its eigenvalues ​​can retain the main features in the data and remove noise and redundant information;

[0249] Moreover, in the power system, due to the large number of monitoring points and the huge amount of data, directly processing the raw data may be very time-consuming and resource-intensive; through dimensionality reduction, the main information in the data can be retained while removing redundancy and noise, making subsequent analysis and judgment more efficient. Linear transformation can further process the reduced-dimensional data to make it more in line with the needs of specific analysis.

[0250] The probability subunit is used to calculate the probability density value of the data in the actual random impact factor data set X′ according to a preset probability density formula;

[0251] The judgment subunit is used to define the abnormal detection result of the power system as an abnormality of the power system if the probability density value is less than a preset abnormal detection threshold ε, and issue an abnormal alarm to the operation and maintenance personnel according to the relevant alarm strategy; wherein the abnormal detection threshold ε is the minimum probability density of historical data, and the probability density formula is established based on the Gaussian mixture model (GMM);

[0252] Moreover, if the probability density value is greater than or equal to the anomaly detection threshold ε, it indicates that the power system is abnormal, and the actual random impact factor at this time corresponds to the system equilibrium operating point x k+1 , that is, we get x k+1 The robust estimation results of .

[0253] The probability density formula is:

[0254]

[0255] Among them, α k is the coefficient of the kth Gaussian component, f(z|θ k ) is the Gaussian distribution density function, and K is the total number of Gaussian components.

[0256] The probability subunit and judgment subunit of this embodiment can quantitatively evaluate the actual random impact factor data set by calculating the probability density value, avoiding the subjectivity and inconsistency of human judgment. Once the probability density value of the actual random impact factor data is lower than the preset anomaly detection threshold, the system can immediately trigger an anomaly detection alarm, which helps maintain the safety of the power system;

[0257] Moreover, the above scheme can be defined as prediction-assisted state estimation, that is: according to the GM estimator of the improved PS and Huber convex fractional function, the stable operation of the power system is analyzed, including the long short-term memory model (LSTM) state prediction step, the missing data replacement step through power flow calculation and the abnormal data detection step; the prediction-assisted state estimation uses the state space model to show the quasi-steady-state dynamic situation of the system state, which can track the state evolution, and is helpful for modern energy management centers to carry out real-time modeling and control of power systems.

[0258] Overall, this embodiment has the following beneficial effects:

[0259] By introducing input random influencing factors, the probability model of the present invention can more comprehensively consider the uncertain factors in the operation of the power system, such as load changes, equipment aging, external environment changes, etc.; this processing method makes the model output closer to the actual situation and improves the accuracy of the evaluation; and by incorporating uncertain factors, the output of the model is no longer a single definite value, but a result containing probabilistic statistical characteristics, which helps to more comprehensively understand the operating status of the power system, thereby improving the accuracy of anomaly detection. Using long short-term memory networks to replace missing data can make full use of the contextual information in time series data, generate more reasonable and accurate data completion results, help maintain data integrity, and avoid analysis bias caused by missing data.

[0260] Embodiment three:

[0261] An embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the method for detecting abnormalities in steady-state operation of a power system;

[0262] Wherein, the abnormality detection method for steady-state operation of a power system, if implemented in the form of a software functional unit and used as an independent product, can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0263] The above are preferred embodiments of the present invention. It should be noted that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for detecting anomalies in steady-state operation of a power system, characterized in that: include: Obtaining operation data of power system; Establishing a probability model including input random impact factors according to the operation data to obtain probability statistical characteristics of output variables of the power system; The state prediction and missing data replacement of the probabilistic statistical characteristics are performed through a long short-term memory network to obtain complete system measurement data; Abnormal data detection is performed on the complete system measurement data to obtain an abnormality detection result of the power system.

2. The method for detecting anomalies in steady-state operation of a power system according to claim 1, characterized in that: The state prediction and missing data replacement of the probability statistical characteristics are performed through the long short-term memory network to obtain the complete system measurement data, specifically: Performing state prediction on the probability statistical characteristics through the long short-term memory network to obtain a state prediction value; The missing data of the state prediction value is replaced based on a preset formula set to obtain the complete system measurement data; wherein the formula set is based on the voltage phasors of the bus and the branch, and is established by calculating the active power component transmitted by the branch.

3. The method for detecting anomalies in steady-state operation of a power system according to claim 1, characterized in that: Perform abnormal data detection on the complete system measurement data to obtain an abnormality detection result of the power system, specifically: Performing dimensionality reduction processing and linear transformation on the complete system measurement data to obtain an actual random impact factor data set of the power system; Whether the power system is abnormal is judged according to the probability density value of the data in the actual random impact factor data set, and an abnormality detection result of the power system is obtained.

4. The method for detecting anomalies in steady-state operation of a power system according to claim 3, characterized in that: The complete system measurement data is subjected to dimensionality reduction processing and linear transformation to obtain the actual random impact factor data set of the power system, specifically: Extract random impact factor data from the complete system measurement data to obtain a random impact factor data set; Centralizing the random impact factor data set to obtain a first data set; Based on the first data set, data dimension reduction is performed by calculating a covariance matrix to obtain a number of eigenvectors; The plurality of characteristic vectors are linearly transformed to obtain the actual random impact factor data set of the power system.

5. The method for detecting anomalies in steady-state operation of a power system according to claim 3, characterized in that: Whether the power system is abnormal is judged according to the probability density value of the data in the actual random impact factor data set, and an abnormality detection result of the power system is obtained, specifically: According to a preset probability density formula, the probability density value of the data in the actual random influencing factor data set is calculated; If the probability density value is less than a preset abnormality detection threshold, the abnormality detection result of the power system is defined as the power system abnormality.

6. A method for detecting anomalies in steady-state operation of a power system according to claim 5, characterized in that: The probability density formula is specifically: Among them, α k is the coefficient of the kth Gaussian component, f(z|θ k ) is the Gaussian distribution density function, and K is the total number of Gaussian components.

7. The method for detecting anomalies in steady-state operation of a power system according to claim 1, characterized in that: A probability model including input random impact factors is established according to the operation data to obtain the probability statistical characteristics of the output variables of the power system, which are specifically: Based on the operating data, a wind speed probability model is obtained by establishing a two-parameter Weibull distribution model for wind speed; By establishing a Beta distribution model for light intensity, a light intensity probability model is obtained; By establishing a normal distribution model of load, a load probability model is obtained; The probability model including input random impact factors is formed by the wind speed probability model, the light intensity probability model and the load probability model, and the probability statistical characteristics about the output variables of the power system are output according to the probability model including input random impact factors.

8. An abnormality detection device for steady-state operation of a power system, characterized in that: Includes data module, feature module, improvement module and detection module; Wherein, the data module is used to obtain the operation data of the power system; The characteristic module is used to establish a probability model including input random impact factors according to the operation data to obtain the probability statistical characteristics of the output variables of the power system; The improvement module is used to perform state prediction and missing data replacement on the probability statistical characteristics through a long short-term memory network to obtain complete system measurement data; The detection module is used to perform abnormal data detection on the complete system measurement data to obtain an abnormality detection result of the power system.

9. The abnormality detection device for steady-state operation of a power system according to claim 8, characterized in that: The improvement module includes a state unit and a replacement unit; Wherein, the state unit is used to perform state prediction on the probability statistical characteristic through the long short-term memory network to obtain a state prediction value; The replacement unit is used to replace missing data of the state prediction value based on a preset formula set to obtain the complete system measurement data; wherein the formula set is based on the voltage phasors of the bus and the branch, and is established by calculating the active power component transmitted by the branch.

10. A storage medium, characterized in that: The storage medium stores a computer program, which is called and executed by a computer to implement an abnormality detection method for steady-state operation of a power system as described in any one of claims 1 to 7 above.

Images