Power distribution network operation state evaluation method

Abstract

The invention discloses a power distribution network operation state evaluation method. In order to overcome the problems that the uncertainty of a power distribution network is remarkably increased due to large-scale distributed new energy access, and the accuracy and the foresight of a traditional evaluation system are insufficient, the method comprises the steps of establishing a probability density function of energy output and load fluctuation; establishing operation risk indexes based on voltage out-of-limit and power out-of-limit for the power grid nodes; and by taking out-of-limit data as input data, clustering voltage and power out-of-limit through k-means + +, constructing a dual deep Q network to evaluate the state of the power distribution network, and monitoring and evaluating out-of-limit risks in combination with all probability density functions. According to the method, an evaluation model capable of accurately capturing and quantifying the uncertainty of the power distribution network is constructed for the uncertainty risk in the operation of the power grid so as to cope with the uncertainty fluctuation of the source side, the network side and the load side, the evaluation accuracy and the perspectiveness of the operation state of the power distribution network are remarkably improved, and the efficient and stable operation of the intelligent power grid is promoted.

Description

Technical Field

[0001] The present invention relates to the field of power grid status assessment, and in particular to a method for assessing the operating status of a distribution network. Background Art

[0002] Against the background of the rapid development of new distribution networks, their radial distribution structure and clear division of information flow and power flow make it possible for the efficient operation of the power grid. However, with the access of large-scale distributed renewable energy, the uncertainty faced by the distribution network has increased significantly, including fluctuations in renewable energy generation on the source side, the impact of important equipment and complex network structure on the grid side, and fluctuations in load and demand response on the load side. The traditional distribution network evaluation system fails to fully reflect these changes and lacks close integration with advanced technologies such as deep learning, resulting in insufficient accuracy and foresight when evaluating the operating status of the distribution network. Therefore, how to build an evaluation model that can accurately capture and quantify the uncertainty of the distribution network has become a key issue that needs to be solved urgently.

[0003] For example, a "distribution network real-time dynamic reconstruction method and system based on branched dual-depth Q network" disclosed in Chinese patent literature, with announcement number CN114282330A, includes the following process: obtaining the real-time node load and distributed power output of the distribution network; transmitting the obtained data to a dynamic DNR model constructed based on the Markov decision process MDP; the dynamic DNR model takes minimizing the network loss cost and the switch action cost as the objective function; based on the decomposition of the distribution network loop, a branched dual-depth Q network is obtained, and the dynamic DNR model is solved using a Q learning algorithm to obtain a set of switch actions that maximizes the output return of the branched dual-depth Q network, and the topology of the distribution network is updated according to the set of switch actions. As the scale of the distribution network expands, the computational complexity increases, which may affect the real-time performance and the difficulty of analyzing uncertainty problems, resulting in reduced efficiency of the solution. Summary of the invention

[0004] The present invention mainly solves the problem that the uncertainty of the distribution network is significantly increased due to the access of large-scale distributed renewable energy in the prior art, while the traditional evaluation system lacks accuracy and foresight; it provides a distribution network operation status evaluation method, and constructs an evaluation model that can accurately capture and quantify the uncertainty of the distribution network for the uncertainty risk in the operation of the power grid, so as to cope with the uncertainty fluctuations on the source, network and load sides, which significantly improves the evaluation accuracy and foresight of the distribution network operation status and promotes the efficient and stable operation of the smart grid.

[0005] The above technical problems of the present invention are mainly solved by the following technical solutions: The present invention includes: establishing probability density functions of energy output and load fluctuations; establishing operation risk indicators based on voltage over-limit and power over-limit for power grid nodes; using over-limit data as input data, clustering voltage and power over-limit through k-means++, constructing a dual-depth Q network to evaluate the state of the distribution network, and combining all probability density functions to monitor and evaluate over-limit risks. By accurately quantifying energy output and load fluctuations, combining dual-depth Q networks and K-means++ clustering, a comprehensive evaluation and intelligent optimization of the operating state of the distribution network is achieved, effectively coping with the uncertainty brought by the access of new energy, and improving the efficiency and intelligence level of power grid management.

[0006] Preferably, the establishment of a probability density function based on photovoltaic output includes using shape parameters of the Beta distribution to describe the distribution change of photovoltaic output power in the range of 0 to maximum output power; establishing a probability density function based on wind turbine processing, including setting the comparison result of the cut-in wind speed and the cut-out wind speed with the rated wind speed to determine the probability density of the wind turbine output at the current wind speed; and using integral transformation to associate load fluctuation data to obtain a probability density function.

[0007] Preferably, the shape parameters of the Beta distribution are determined, matching weather and time conditions are set, the output power of the photovoltaic power station under each condition is taken, and the output power data is obtained by fitting the maximum likelihood estimation or Bayesian inference to obtain the optimal shape parameters; according to the shape parameters, the distribution function of the photovoltaic output power is constrained within the range of 0 to the maximum output power, the distribution function is normalized and constrained to the range of 0 to 1, and a probability density function based on the photovoltaic output is obtained.

[0008] Preferably, the key variable wind speed and correction factor are set; if the wind speed is between the cut-in wind speed and the cut-out wind speed, the fan is in operation; if the wind speed is less than the cut-in wind speed or greater than the cut-out wind speed, the fan is in shutdown; the wind speed in the operation state is corrected or the output power of the fan is calculated using the temperature or the historical wind speed mean, the relationship between the wind speed range and the output power is fitted, and a probability density function based on the fan output is obtained.

[0009] Preferably, the integral relationship between the inverse of the historical load standard deviation and the load fluctuation is established, wherein the process of determining the load fluctuation includes: squaring the difference between the load in different time periods and the historical load mean, and taking the inverse of the square of the historical load standard deviation; the product of the historical load standard deviation and the load fluctuation power of e is the probability density function of the load fluctuation. Preferably, the voltage over-limit risk is obtained according to the corresponding relationship between the voltage over-limit probability and degree; the voltage over-limit condition is set according to the voltage fluctuation range, and the quantized voltage offset is obtained to obtain the voltage over-limit degree; the probability and degree of the node power exceeding its safe operation threshold are calculated, and the weight factor is introduced to normalize the exceeding part, and the power over-limit risk is obtained based on the probability density function of the node power; the voltage over-limit risk value and the power over-limit risk value are used as indicators.

[0010] Preferably, the voltage over-limit risk is the integral value of the voltage at different times with the maximum over-limit voltage and the minimum over-limit voltage as endpoints; the voltage over-limit degree is the probability density of the voltage appearing in the preset value range, and the generated probability density function parameter part includes an exponential function composed of the historical voltage standard deviation and the historical voltage mean, and the variable is the voltage at different times; the closer the voltage value is to the historical voltage mean, the higher the probability density function value.

[0011] Preferably, the power over-limit probability includes a probability density function based on node power and a related function of a safe power operation threshold; the degree of power over-limit is the percentage of the difference between the power over-limit value and the standard threshold to the standard value; a weight factor is introduced to limit the correspondence between the power over-limit probability and the degree, and the weight factor takes a value in the range of 0-1.

[0012] Preferably, the setting loss function and value function are solved to construct a dual deep Q network to divide the risk labels of power grid operation; at the same time, space and reward functions are defined, combining the voltage over-limit risk with the probability density function based on load, the power over-limit risk with the probability density function based on energy processing, and combining the risk labels to evaluate the over-limit risk.

[0013] Preferably, for the power grid operation nodes, samples are selected as cluster centers, the shortest distances between all samples and the current cluster center are calculated, the maximum value is selected as the cluster center, and the risk labels are divided by the K-means++ method; hyperparameters are set for the dual-depth Q network, including but not limited to the experience pool revisit capacity and the neural network learning rate.

[0014] The beneficial effects of the present invention are: 1. A distribution network operation status assessment method of the present invention realizes accurate quantification of energy output and load fluctuations by constructing a probability density function based on photovoltaic output and wind turbine processing, and using integral transformation to correlate load fluctuation data. This innovation not only improves the prediction accuracy of the power grid operation status, but also enables power grid managers to more comprehensively assess the potential risks of the power grid. At the same time, by quantifying the risks of voltage over-limit and power over-limit, the present invention provides power grid managers with more detailed risk indicators, which helps them take effective measures in a timely manner to ensure the safe and stable operation of the power grid. This method of accurate quantification and comprehensive assessment has greatly improved the efficiency and level of power grid management.

[0015] 2. A distribution network operation status evaluation method of the present invention, in the face of the uncertainty brought about by the access of large-scale distributed new energy, the present invention proposes an intelligent optimization strategy based on a dual-depth Q network and a K-means++ clustering method. Through the learning and decision-making capabilities of the dual-depth Q network, the present invention can analyze the operation status of the power grid in real time and automatically adjust the operation strategy to cope with various uncertain factors. At the same time, the introduction of the K-means++ clustering method enables the operation risks of the power grid to be more finely divided and evaluated, providing strong support for intelligent optimization. This strategy of intelligent optimization and efficient response not only improves the response capability of the power grid, but also promotes the development of the power grid towards intelligence and automation. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The present invention is a flow chart of a method for evaluating the operating status of a distribution network.

[0017] Figure 2 It is a system structure diagram of embodiment 2 of the present invention.

[0018] Figure 3 This is a diagram showing the node voltage crossing the lower limit according to the second embodiment of the present invention.

[0019] Figure 4 This is a node power lower limit diagram of embodiment 2 of the present invention.

[0020] Figure 5 This is a risk label classification result diagram of Example 2 of the present invention.

[0021] Figure 6 4 is a comparison chart of the status assessment results of Example 2 of the present invention. DETAILED DESCRIPTION

[0022] The technical solution of the present invention is further specifically described below through embodiments and in conjunction with the accompanying drawings.

[0023] Embodiment 1: A distribution network operation status evaluation method according to this embodiment is as follows: Figure 1Shown, including Step 1: Establish the probability density function of energy output and load fluctuation.

[0024] Input the basic parameters of the power system, including system dispatching operation point, network topology, line impedance parameters and node load data; the hyperparameters of the dual deep Q network, including the capacity of the experience replay pool, the expansion coefficient of the Bellman equation, the neural network learning rate, etc.

[0025] The probability density functions of photovoltaic output, wind turbine output and load fluctuation are further established including: The photovoltaic output probability density function is established as follows: Where α and β are the shape parameters of the Beta distribution, P t pv is the photovoltaic output at time t, is the maximum photovoltaic output power, Γ(.) is the gamma function.

[0026] This function uses the shape parameters α and β of the Beta distribution. In statistics, the Beta distribution is often used to describe continuous probability distributions with upper and lower limits, and is particularly suitable for describing proportions, percentages, or probability values. f(P; α, β) = some form of Beta distribution density function. Where P represents the photovoltaic output at time t, and its value varies between 0 and the maximum photovoltaic output power P_max. α and β are the shape parameters of the Beta distribution, which determine the shape of the distribution.

[0027] The probability density function of the Beta distribution is usually written as: f(P; α, β) = B(α, β)Pα-1(1-P)β-1. Where B(α, β) is the Beta function, which is related to the Gamma function and is defined as B(α, β) = Γ(α+β)Γ(α)Γ(β. Γ(.) is the Gamma function, a function widely used in mathematics, physics, and engineering. For non-negative integers n, Γ(n) = (n-1)!.

[0028] Determine shape parameters: First, the shape parameters α and β of the Beta distribution need to be determined based on historical data or expert knowledge. These parameters will determine the probability distribution of the PV output P at different values. Apply Beta distribution: Then, use the determined α and β values ​​and the probability density function formula of the Beta distribution to calculate the probability density f(P) at a given PV output P. Analysis and application: Finally, analyze the obtained probability density function to understand the statistical characteristics of the PV output, such as the possible range, mean, median, and variability. These analysis results can be used in practical applications such as power system planning, scheduling, and risk assessment.

[0029] The Beta distribution is a continuous probability distribution that describes the probability distribution of a random variable with values ​​between 0 and 1. However, in the context of PV output, we can extend it to the range of 0 to P_max (maximum PV output power) by appropriate transformations (such as dividing P by P_max). Determine shape parameters α and β: These two parameters are determined by analyzing historical PV output data. The data may include the output power of PV power plants at different times and under different weather conditions. Commonly used methods include statistical methods such as maximum likelihood estimation, moment estimation, or Bayesian inference, which can be used to fit the data to obtain the optimal α and β values.

[0030] The probability density function of wind turbine output is established: It describes the probability density of wind turbine output under different wind speed conditions. Here, wind speed V is the key variable, and other parameters such as cut-in wind speed Vin, rated wind speed Vr, cut-out wind speed Vout, historical wind speed mean μ and standard deviation σ are used to define the range and distribution characteristics of wind speed.

[0031] Wind speed range judgment: First, based on the comparison of wind speed V with cut-in wind speed Vin, rated wind speed Vr and cut-out wind speed Vout, determine whether the wind turbine is in operation (V is between Vin and Vout) or shutdown (V is less than Vin or greater than Vout). Power calculation or correction: For wind speeds in operation, it may be necessary to calculate or correct the wind turbine output P based on wind speed V and other parameters (such as temperature coefficient k, historical wind speed mean μ and standard deviation σ). This part may involve complex mathematical formulas or models, such as Weibull distribution, lognormal distribution or other probability distribution functions to describe the distribution of wind speed, and convert wind speed into wind turbine output in some way (such as power curve). Probability density function construction: Based on the above calculations and corrections, a probability density function f(P|V) of wind turbine output P under a given wind speed V is constructed. This function describes the possible values ​​of wind turbine output P and their corresponding probabilities at different wind speeds.

[0032] Among them, v in is the cut-in wind speed, v r is the rated wind speed, v out is the cut-out wind speed, V is the mean of historical wind speed, σ is the standard deviation, h=(v r -v in )-1.

[0033] Determine the integral variables and parameters for the above functions: First, it is necessary to clarify the physical meaning or role of the integral variable t in the formula. In practical applications, t may represent a transformation variable of wind speed, time, or other parameters related to wind turbine output. At the same time, it is necessary to determine the meaning and source of all parameters involved in the formula (such as k, V, θ, etc.). Understand the integral expression: Analyze the function g(t; k, V, ...) inside the integral to understand how it describes the relationship between wind turbine output and parameters such as wind speed and temperature coefficient. This may involve consulting data sheets, technical specifications, or related engineering manuals provided by the wind turbine manufacturer. Set the upper and lower limits of the integral: For physical systems, the upper and lower limits of the integral should be set according to actual conditions. In the scenario of wind turbine output, the wind speed usually has a valid range (from cut-in wind speed to cut-out wind speed), so the upper and lower limits of the integral should correspond to this range.

[0034] Solve the integral: Once the integral variables, parameters, and upper and lower limits are determined, you can try to solve the integral expression. This may require the use of numerical integration methods, because the actual function form may be very complex and cannot be solved directly. Construct a probability density function: Based on the results of the integration, a probability density function of the wind turbine output can be constructed. This function will describe the possible values ​​of the wind turbine output and their corresponding probabilities under given wind speeds and other conditions. Verification and application: Finally, the accuracy and applicability of the constructed probability density function need to be verified through experimental data or historical records. Once verified, it can be applied to practical work such as planning, scheduling, and risk assessment of power systems.

[0035] The load fluctuation probability density function is established: Among them, σ load is the historical load standard deviation, μ load is the historical load average.

[0036] Step 2: Establish operation risk indicators based on voltage and power limits for grid nodes.

[0037] Establish voltage over-limit risk indicators, In the formula, Risk u,i is the voltage over-limit risk of node i, p j (u i )(j=1) is the probability that the voltage of node i exceeds the lower limit, s j (u i )(j=1) is the severity of the voltage of node i exceeding the lower limit, p j (u i )(j=2) is the probability that the voltage of node i exceeds the upper limit, s j (u i )(j=2) is the severity of the voltage at node i exceeding the upper limit.

[0038] Determine the influencing factors: First, determine the variables u_j that affect the voltage at node i, including the output power of the generator, the change in load demand, the change in line impedance, etc. Calculate the degree of deviation: For each random variable u_j, calculate the square of the difference between it and the expected value of the voltage μ, that is, (u_j-μ)2, which reflects the degree of voltage deviation caused by the variable.

[0039] Evaluate probability: Given a system state u, evaluate the probability P(j|u) of each random variable u_j occurring. Consider severity: Based on the system state u, determine the severity x_s(u) of the voltage over-limit (either upper or lower limit). Calculate risk comprehensively: Multiply all the above factors and sum all random variables to obtain the voltage over-limit risk Risk_i of node i. This process comprehensively considers the degree of voltage deviation, the probability of occurrence, and the severity of the over-limit.

[0040] According to the above formula, it can be seen that in the process of establishing voltage over-limit risk indicators in the power system, it is necessary to evaluate the voltage fluctuation range and over-limit conditions. This process includes the following steps: Determine the voltage range First, according to the design requirements of the power system and the operating characteristics of the equipment, determine the normal fluctuation range of the voltage, that is, U_min and U_max. Identify the voltage over-limit conditions Analyze various factors that may cause voltage over-limit, such as load changes, generator output adjustments, transmission line failures, etc.

[0041] The specific amount by which the voltage deviates from the normal range to the over-limit range is quantified. The j value is obtained through simulation calculation, statistical analysis or experimental measurement. j does not always appear directly in the form of "U_max = U_min + j". The voltage over-limit risk assessment is based on information such as the voltage fluctuation range, over-limit conditions and offset, and the potential risk of voltage over-limit to the stability of the power system and the safety of equipment is assessed. This may involve steps such as calculating risk indicators and formulating risk control strategies.

[0042] This formula is about the multiple integral of a two-variable function f(x,y), where x and y are the integral variables. Use the Fourier series (or similar trigonometric series) to approximate f(x,y) and calculate its integral over a certain area. The Fourier series can represent a function as the sum of a series of sine and cosine functions, which is particularly useful when dealing with periodic functions or systems with periodic properties. In the formula, U min and U max are the upper and lower limits of the node voltage, u t,i is the voltage amplitude of node i at time t.

[0043] f(u t,i ) is the voltage distribution probability density function, and we have: In the formula, σ U is the historical voltage standard deviation, which is a measure of the deviation between the voltage data and its mean, representing the degree of dispersion of the voltage distribution; μ U is the historical voltage mean, which is the arithmetic mean of the voltage data and represents the central trend of the voltage distribution.

[0044] The normalization constant: σ2π1 is to ensure that the probability density integral under the entire distribution is 1, that is, the sum of the probabilities of all possible voltage values ​​is 1. The exponential part: e-21(σu-μ)2 determines the shape of the probability density as the voltage value u changes. When u is close to μ, the value is large, indicating that the probability of the voltage value near the mean is high; when u is far away from μ, the value gradually decreases, indicating that the probability of the voltage value far away from the mean is low.

[0045] Establish power limit risk indicators, including: Risk P,i = p(P i )×s(P i ); In the formula, Risk P,i is the power limit risk of node i, p(P i ) is the probability of node i exceeding the power limit, s(P i ) is the severity of the power limit exceeding of node i.

[0046] Where P safe is the threshold of safe operation power of node i, f(P t,i ) is the node power probability density function.

[0047] When establishing the power over-limit risk index, if we consider the safe operation power threshold of node i (denoted as Pthreshold,i) and the probability density function of node power (denoted as fi(P)), then the risk index can be defined based on the probability of power exceeding this threshold and some measure of the excess amount. Normalization: If you want the risk index Ri to be between 0 and 1, so that you can compare different nodes or different time points, you may need to normalize the above integral. This usually involves dividing the integral result by an appropriate constant (such as the difference between the maximum value that the power may reach and the threshold multiplied by the maximum probability density in the range).

[0048] In practical applications, it may also be necessary to consider introducing weight factors to adjust the contribution of different excesses to the risk. This can be achieved by modifying the integrand, such as using (P-Pthreshold,i)2·fi(P) or more complex functions to increase the penalty for larger excesses. The calculation of the above integrals may involve complex mathematical operations, especially when the probability density function fi(P) is not in a simple form. In practical applications, numerical integration methods may be required to approximate the solution. In order to calculate the risk index, it is necessary to know the probability density function fi(P) of the node power. This is usually based on a large amount of historical data or estimated through simulation models. Once the risk index is calculated, the corresponding risk management strategy and control measures can be formulated based on the index to ensure the safe and stable operation of the power system.

[0049] In order to establish the power over-limit risk index, it is necessary to comprehensively consider the power over-limit probability, the severity of the power over-limit, and other possible related factors of node i. Define a comprehensive risk index Ri, which can quantify the power over-limit risk of node i. The risk index Ri can be defined as the product of the power over-limit probability pi and the power over-limit severity si, and may include a weight factor or adjustment coefficient to adapt to different application scenarios or system characteristics.

[0050] The power over-limit probability pi indicates the possibility that the power of node i exceeds its allowed range. This value is usually derived based on historical data, simulation models or probabilistic analysis methods. The value range of pi is between 0 and 1, where 0 indicates no risk of over-limit and 1 indicates an absolute over-limit. The power over-limit severity si indicates the degree of negative impact of this event on the stability, security or economy of the power system when the power of node i exceeds the limit. The severity can be quantitative (such as the specific value or percentage of the power over-limit) or qualitative (such as a grade classification based on expert evaluation). For ease of calculation, si is usually normalized to the range of 0 to 1.

[0051] A weighting factor w can be used to adjust the sensitivity of the risk indicator to different factors. If some factors (such as the severity of the limit violation) are considered more important than others, w can be adjusted to reflect this. The value of w can also be normalized to between 0 and 1, but can also be set to other ranges as needed.

[0052] Step 3: Taking the over-limit data as input data, the voltage and power over-limits are clustered through k-means++, and a dual deep Q network is constructed to evaluate the distribution network status. All probability density functions are combined to monitor and evaluate the over-limit risk.

[0053] The k-means++ method is used to assign grid uncertainty operation risk labels with voltage over-limit and power over-limit as input, including: randomly selecting a sample as the first cluster center c1; calculating the shortest distance between each sample and the current cluster center, represented by D(x), indicating that the probability of being selected as the cluster center is large, and selecting the maximum value as the cluster center; repeatedly selecting k cluster centers; and dividing each risk label through the k-means algorithm.

[0054] Construct a double-depth Q network and consider the uncertainty of operation risk to realize the distribution network operation status evaluation, which includes: defining the state space of the double-depth Q network, there are: s n ={x n,i}; where x n,i Represents the i-th feature of the n-th group of input data. Define the action space and reward function of the dual deep Q network: A = {0, 1, 2}; In the formula, 0, 1, and 2 represent the operating state, warning state, and fault state respectively, and a t = l t Indicates that the evaluation result of the tth group of input is correct, a t ≠l t Indicates that the evaluation result is wrong.

[0055] Define the loss function and value function solution method, there are: Where, T t is the decision label at time t, T t =True means the decision is complete.

[0056] The loss function of the Predict Q network is: L(θ t )=(y t -Predict Q t ) 2 .

[0057] In practical applications, this formula may be used to monitor and control key parameters in a system to prevent them from exceeding a safe range. For example, in a power system, the power output P of each node can be monitored and used to assess the stability and over-limit risk of the system.

[0058] Embodiment 2: A distribution network operation status evaluation method according to this embodiment, such as Figure 2-6 As shown, based on the example 1, the system of this embodiment is further described by substituting an example. The system of this embodiment is composed of an IEEE33 node system and 4 photovoltaic generator sets and 4 wind turbine generator sets. Figure 2As shown, the capacity of photovoltaic power generation units is 60MW.

[0059] This scheme realizes a comprehensive evaluation of the distribution network operation status by combining probability density function modeling, uncertainty operation risk index construction, cluster analysis, and double deep Q-network (DDQN) and other technologies. The following are the specific steps and principles of this scheme to realize the evaluation of the power grid operation status: First, the data is input and preprocessed, including the dispatching operation point of the power system, network topology, line impedance parameters and node load data, etc. The historical output sample data of photovoltaic, wind turbine and load are obtained by using public data sets and tools such as MATLAB.

[0060] Uncertainty modeling is carried out. Beta distribution modeling is used for photovoltaic output, and shape parameters are estimated through historical data. For wind turbine output, Weibull distribution based on wind speed is used to calculate the probability density of wind turbine output in combination with the cut-in, rated and cut-out wind speeds of the wind turbine. Normal distribution modeling is used for load fluctuation, and the mean and standard deviation are calculated through historical data.

[0061] Uncertainty risk operation indicators are constructed. For voltage over-limit risk, the probability and severity of the voltage exceeding the upper and lower limits of each node are calculated, and the voltage over-limit risk is comprehensively evaluated. For power over-limit risk, the probability and severity of the power over-limit of each node are calculated, and the power over-limit risk is evaluated. Finally, the voltage and power over-limit risks of each node are summarized to obtain the cumulative risk of the entire distribution network.

[0062] k-means++ clustering: Taking voltage over-limit and power over-limit risks as input, the k-means++ algorithm is used to divide the operating status of the distribution network into different risk labels (such as low risk, medium risk, high risk, etc.).

[0063] Based on the construction and training of dual deep Q networks, the basic parameters of the power system, historical data, and risk labels are used as input features to construct the state space. Three actions, namely, operating state, alert state, and fault state, are defined, and the reward function is designed based on the correctness of the evaluation results. Through the DDQN algorithm, using hyperparameters such as the experience replay pool and the Bellman equation expansion coefficient, the network is trained to minimize the loss function and optimize the value function.

[0064] Using the trained DDQN model, the real-time input power system data is evaluated to determine its operating status (operating state, warning state or fault state). According to the evaluation results and uncertainty operation risk indicators, the operating status of the distribution network is corrected to improve the accuracy and reliability of the evaluation. The voltage and power limit conditions of each node and the overall risk level are displayed in the form of charts and other forms. Decision support is provided for grid dispatchers and operators to guide them to take appropriate measures to reduce operating risks and ensure the safe and stable operation of the grid.

[0065] After 50 solutions using random Monte Carlo simulation, the voltage lower limit and power upper limit of each node are as follows: Figure 3 and Figure 4 As shown in the figure, the risk labels divided by the k-means++ algorithm are as follows Figure 5 shown.

[0066] Whether the distribution network operation status assessment results considering the uncertainty operation risk are Figure 6 shown.

[0067] Obviously, the method proposed in the present invention can accurately perceive uncertain operation risks and correct the existing distribution network operation status assessment results, which is of great significance for the safety risk analysis of new power systems.

[0068] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0069] It should be understood that the embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope limited by the appended claims of the application.

Claims

1. A method for evaluating the operating status of a distribution network, characterized in that: include: S1: Establish the probability density function of energy output and load fluctuation; S2: Establishing operation risk indicators based on voltage and power limits for grid nodes; S3: Taking the over-limit data as input data, the voltage and power over-limits are clustered through k-means++, and a double deep Q network is constructed to evaluate the distribution network status. All probability density functions are combined to monitor and evaluate the over-limit risk.

2. A distribution network operation status evaluation method according to claim 1, characterized in that: S1.1: Establish a probability density function based on photovoltaic output, including using the shape parameters of the Beta distribution to describe the distribution changes of photovoltaic output power in the range of 0 to maximum output power; S1.2: Establish a probability density function based on wind turbine processing, including setting the comparison results of the cut-in wind speed and the cut-out wind speed with the rated wind speed to determine the probability density of the wind turbine output at the current wind speed; S1.3: Use integral transformation to correlate load fluctuation data to obtain the probability density function.

3. A distribution network operation status evaluation method according to claim 2, characterized in that: In S1.1, the shape parameters of the Beta distribution are determined, matching weather and time conditions are set, the output power of the photovoltaic power station under each condition is taken, and the output power data is fitted with maximum likelihood estimation or Bayesian inference to obtain the optimal shape parameters; according to the shape parameters, the distribution function of the photovoltaic output power is constrained within the range of 0 to the maximum output power, the distribution function is normalized and constrained to the range of 0~1, and the probability density function based on the photovoltaic output is obtained.

4. A distribution network operation status evaluation method according to claim 2, characterized in that: In S1.2, set the key variable wind speed and correction factor; if the wind speed is between the cut-in wind speed and the cut-out wind speed, the fan is in operation; if the wind speed is less than the cut-in wind speed or greater than the cut-out wind speed, the fan is in shutdown; the temperature or the historical wind speed mean is used to correct or calculate the output power of the fan for the wind speed in the operating state, fit the relationship between the wind speed range and the output power, and obtain the probability density function based on the fan output.

5. A distribution network operation status evaluation method according to claim 2, characterized in that: In S1.3, an integral relationship between the inverse of the historical load standard deviation and the load fluctuation is established, where the process of determining the load fluctuation includes: squaring the difference between the load in different time periods and the historical load mean, and taking the inverse of the square of the historical load standard deviation; the product of the historical load standard deviation and the load fluctuation power of e is the probability density function of the load fluctuation.

6. A method for evaluating the operating status of a distribution network according to claim 1, characterized in that: S2.1: The voltage over-limit risk is obtained based on the corresponding relationship between the voltage over-limit probability and degree; the voltage over-limit condition is set according to the voltage fluctuation range, and the quantized voltage offset is obtained to obtain the voltage over-limit degree; S2.2: The power over-limit risk is obtained based on the probability density function of the node power by calculating the probability and degree of the node power exceeding its safe operation threshold and introducing a weight factor to normalize the exceeding part; the voltage over-limit risk value and the power over-limit risk value are used as indicators.

7. A method for evaluating the operating status of a distribution network according to claim 6, characterized in that: In S2.1, the voltage over-limit risk is the integral value of the voltage at different times with the maximum over-limit voltage and the minimum over-limit voltage as endpoints; the voltage over-limit degree is the probability density of the voltage appearing in the preset value range, and the parameter part of the generated probability density function includes an exponential function composed of the historical voltage standard deviation and the historical voltage mean, and the variable is the voltage at different times; the closer the voltage value is to the historical voltage mean, the higher the probability density function value.

8. A method for evaluating the operating status of a distribution network according to claim 6, characterized in that: In S2.2, the power over-limit probability includes a probability density function based on node power and a related function of the safe power operation threshold; the degree of power over-limit is the percentage of the difference between the power over-limit value and the standard threshold to the standard value; a weight factor is introduced to limit the corresponding relationship between the power over-limit probability and degree, and the weight factor takes a value in the range of 0-1.

9. A method for evaluating the operating status of a distribution network according to claim 1 or 2 or 3 or 4 or 5 or 6 or 7 or 8, characterized in that: In S3, the loss function and value function are set to solve and construct a dual deep Q network to divide the risk labels of power grid operation; at the same time, the space and reward functions are defined, combining the voltage over-limit risk with the probability density function based on load, the power over-limit risk with the probability density function based on energy processing, and combining the risk labels to evaluate the over-limit risk.

10. A method for evaluating the operating status of a distribution network according to claim 9, characterized in that: S3 also includes selecting samples as cluster centers for power grid operation nodes, calculating the shortest distances between all samples and the current cluster center, selecting the maximum value as the cluster center, and dividing each risk label by the K-means++ method; setting hyperparameters for the dual-depth Q network, including but not limited to the experience pool revisit capacity and the neural network learning rate.

Images