Method and device for evaluating running state of power distribution network

Abstract

The invention discloses a power distribution network operation state evaluation method, which relates to the technical field of power distribution network state evaluation, and comprises the following steps: inputting basic parameters of a power system, hyper-parameters in a deep Q network and initial rule parameters; establishing a deep Q network model for evaluating the running state of the power distribution network in the future, and evaluating the running state of the power distribution network with delta t as the sampling time in the time T in the future; establishing power distribution network state probability distribution in the future T time based on the well-sleeving theory; determining a reference point and a reference value of each premise input in the confidence rule base, matching to a corresponding evaluation rule through a premise input state, and outputting an evaluation result; and according to an error between an evaluation result and an actual result, introducing a cuckoo search parameter optimization method, and adjusting each rule parameter in the belief rule base. The invention further discloses a power distribution network operation state evaluation device, the balance between the operation state evaluation efficiency and the evaluation result accuracy is achieved, and a brand new calculation thought is provided for power distribution network operation state evaluation.

Description

Technical Field

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

[0002] In the power system, the distribution network is located at the end of the system and is an important link in achieving power distribution to the user end. The operating status of the distribution network will directly affect the power quality at the user end. If it cannot meet the power supply demand of the key loads on the user side, once a major power accident occurs, it may lead to serious economic losses and adverse social impacts. Through a comprehensive analysis of the distribution network operation data, the indicator model that affects the distribution network operation status is determined, and a scientific and standardized distribution network operation status evaluation mechanism is established, which not only protects the interests of users and maintains social harmony and stability, but also helps to discover the weak links in the operation of the distribution network, improve the maintenance capacity of the distribution network, and further realize the lean operation and maintenance level of the distribution network. Therefore, it is of great significance to carry out research on the evaluation of the distribution network operation status.

[0003] The evaluation of the operating status of traditional distribution networks usually focuses on economy and reliability. It is an evaluation method that selects comprehensive indicators and determines the weights according to the empirical scoring method. However, compared with the traditional distribution network, the structure of the new distribution network is more complex, the number of devices connected to the user side is greater, the volume of distribution network operation data is larger, the source is wider, and the value density is lower. The traditional evaluation method based on empirical scoring cannot accurately identify high-value data, resulting in problems such as long evaluation time and large deviations in evaluation results. The traditional evaluation method can no longer meet the current needs of distribution network operation status evaluation. At the same time, the traditional method may only provide an overall evaluation result without going deep into the operating status of each specific indicator, which makes the evaluation result lack of details and difficult to conduct targeted analysis and processing of specific problems. Since there is no separate operating status evaluation for each indicator, the traditional method may not be able to accurately identify and locate specific problem points in the distribution network.

[0004] Establishing an operating status evaluation rule base and matching the premise input after the state division to the corresponding rules is the key to solving the problem of distribution network operating status evaluation. The evaluation results of the traditional rule base model do not reflect the model's learning process of unknown knowledge. Therefore, based on this model, Professor Jianbo Yang proposed a belief rule base model (BRB). In this model, the reasoning and solving algorithms of the evaluation result confidence distribution and evidence theory (ES) are used respectively, which can not only quantitatively describe the uncertain information in the input data, but also use the existing rule base knowledge for analysis. Based on expert knowledge, the BRB model describes the lack of prior knowledge in the rule base, explaining the partial fuzziness and ignorance of the rules. When constructing a rule base, traditional methods may not be able to obtain all the necessary information or expert knowledge, resulting in gaps or incomplete knowledge in the rule base. Summary of the invention

[0005] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a method and device for evaluating the operating status of a distribution network. The present invention combines a deep Q network with a confidence rule base, which significantly improves the accuracy of the evaluation results.

[0006] The present invention adopts the following technical solutions to solve the above technical problems:

[0007] In a first aspect, the present invention provides a method for evaluating the operating status of a distribution network, comprising:

[0008] Input the basic parameters of the power system, the hyperparameters in the deep Q network, and the initial rule parameters of the confidence rule base;

[0009] A deep Q network model for evaluating the future operation status of the distribution network is established by using the basic parameters of the power system, the hyperparameters in the deep Q network, and the initial rule parameters of the confidence rule base. The operation status of the distribution network within the future T time with Δt as the sampling time is evaluated, and Δt is each sampling period.

[0010] According to the state of each sampling point, the probability distribution of the distribution network state at the future time T is established based on the well-being theory. According to the probability distribution of the distribution network, the sampling point is used as one of the premise inputs of the confidence rule base.

[0011] According to national standards and sampling points, determine the reference point and reference value of each premise input in the confidence rule base, match the reference point and reference value of the premise input with the distribution network evaluation rule, and output the evaluation result;

[0012] According to the error between the evaluation result and the actual result, the parameter optimization method of cuckoo search is introduced to adjust the parameters of each rule in the confidence rule base.

[0013] As a further optimization scheme of the distribution network operation status evaluation method described in the present invention,

[0014] The basic parameters of the input power system include system dispatching operation point, network topology, line impedance parameters and node load data;

[0015] The hyperparameters of the deep Q network include the experience replay pool capacity, the number of training times, the training step size, the neural network parameters and the corresponding number of neurons and the activation function;

[0016] The initial rule parameters of the confidence rule base include the rule weight of each rule, the weight of the premise input and the confidence distribution of each rule.

[0017] As a further optimization scheme of the distribution network operation status evaluation method described in the present invention, a deep Q network model for evaluating the future distribution network operation status is established, which is as follows:

[0018] Determine the future time T of the evaluation and each sampling period Δt;

[0019] The running state of Δt is evaluated by the deep Q network, where the state space of the deep Q network is expressed as:

[0020] S={s1,s2,…,s nm}

[0021] Among them, S is the state space, n is the number of samples in the input deep Q network, m is the number of input features for each sample, and s t is the state of the agent at time t, 1≤t≤nm;

[0022] For s in the state space i Definition of i is the state of the agent at the i-th moment, and we have:

[0023] s i =(x i ,f i )

[0024] Among them, x i is the i-th input data, f i is the i-th input feature;

[0025] The action space is represented as:

[0026] A=A f ∪A y

[0027] Among them, A is the action space, A f is the action set for selecting attributes, that is, in state s t Next, sample x takes action a tAchieve selection of input feature f and subsequent actions; A y is the classification decision action set, that is, in state s t Next, sample x takes action a t Realize the prediction classification and end the decision-making process of the sample x;

[0028] In status t If the agent takes A f The action in will select the i-th input feature f from the input features i As a basis for classification; A y ={0,1,2}, where A y =0, 1, 2, respectively represent the running state, warning state and fault state in the distribution network; in state s t If the agent takes A y Centralized action, then the operating state of the distribution network in this state is divided into operating state, warning state or fault state;

[0029] The reward function is expressed as:

[0030]

[0031] Among them, r t (s t ,a t ) to take action t Post-state t The reward value obtained, r t (s t ,a t )=1 indicates that agent s t State, action taken a t Belongs to A y And the predicted result is the same as the actual label value y, r t (s t ,a t )=1 means that agent s t State, action taken a t Belongs to A y But the predicted result is different from the actual label value y, r t (s t ,a t )=0 means that agent s t In state a, the action taken t Belongs to A f , it only selects input features for subsequent actions, without classifying the input sample data;

[0032] State transfer and network update are expressed as:

[0033]

[0034] Among them, F is the stop decision identifier, (x t+1 ,f t+1 ) means continue decision-making and enter the next state, x t+1 represents t+1 input data, f t+1 represents the t+1th feature of the input data, p t represents the transfer probability at time t;

[0035]

[0036] Among them, Target Q represents the network value, r t represents the reward value at time t, γ represents the discount factor, Q(s t+1 ,a t+1 ; θ') represents the state s t+1 Take action a t+1 The Q value of a t+1 represents the action at time t+1, θ' is the Target Q network parameter value, s t+1 =F represents state s t The next state s of the agent t+1 is a terminated state;

[0037] Parameter iteration is expressed as:

[0038]

[0039] θ'=ρθ'+(1-ρ)θ

[0040] Among them, θ represents the Predict Q network parameter value, Loss(θ) represents the loss function, α is the neural network learning rate, and ρ is the Target Q network update coefficient.

[0041] As a further optimization scheme of the distribution network operation status evaluation method described in the present invention, a probability distribution of the distribution network status at the future time T based on the well-being theory is established for the status of each sampling point, including:

[0042] The well-being theory is used to obtain the probability distribution of the distribution network state in the future time T (T / Δt) period, which is:

[0043]

[0044]

[0045]

[0046] Where P(H), P(W), and P(R) represent the probability of the distribution network being in the operating state, warning state, and fault state, respectively. T is the future time of the evaluation. n(H), n(W), and n(R) represent the number of times the distribution network is in the operating state, warning state, and fault state within the time T, respectively. i (H), T i (W), T i (R) represents the duration of the i-th time in the operating state, warning state, and fault state respectively.

[0047] As a further optimization scheme of the distribution network operation status evaluation method described in the present invention, the reference point and reference value of each premise input in the confidence rule base are determined, and the premise input state is matched with the distribution network evaluation rule to output the evaluation result, including:

[0048] S41. Based on the actual operation of the distribution network, select the voltage deviation rate, load rate and future operation time as the premise input, and establish the reference state and reference point with reference to known national standards;

[0049] S42. For each premise input of each group of inputs, the premise input is divided into corresponding reference states according to the numerical value of the premise input, and the reference state of each premise input is matched with the rules in the confidence rule base, and there are:

[0050]

[0051] Then y is{(D1,β 1,k ),(D2,β 2,k ),…,(D N ,β N,k )}

[0052] With rule weightθ k and attribute weightδ1,δ2,…,δ M

[0053] Among them, x b is the feature value of the b-th premise input, M≥b≥1, M is the total number of premise features, R k is the kth rule of the confidence rule base model, is the reference value of the b-th premise input in the k-th rule of the confidence rule base model, {(D1,β 1,k ),(D2,β 2,k ),…,(D N ,β N,k )} is the output result set of the kth rule, β a,k is the confidence of the a output result of the kth rule, D ais the evaluation result corresponding to the ath rule, N≥a≥1, N is the total number of distribution network operating states, θ k is the weight of the kth rule, δ b is the weight of the bth preceding input;

[0054] S43, calculate the distribution network operation status evaluation score, which is:

[0055]

[0056] Among them, D d,k is the fuzzy score of the d-th evaluation level of the k-th rule, β d,k is the confidence of the dth evaluation level of the kth rule.

[0057] As a further optimization scheme of the distribution network operation status evaluation method described in the present invention, according to the error between the evaluation result and the actual result, the parameter optimization method of cuckoo search is introduced to adjust the parameters of each rule in the confidence rule base; the details are as follows:

[0058] According to the comparison between the evaluation result and the actual result error, the rule weight, attribute weight and confidence distribution of each rule in the confidence rule base are adjusted as follows:

[0059] min Function(θ k ,β d,k ,δ i )

[0060]

[0061] Among them, Function(θ k ,β d,k ,δ i ) represents the parameter optimization function, β d,k is the confidence of the d output results of the kth rule, θ k is the weight of the kth rule, δ i is the attribute weight of the i-th pre-feature, θ l is the rule weight of the lth rule, and L is the total number of rules;

[0062] Based on the cuckoo search algorithm, adjust the rule weight, premise input weight and confidence distribution,

[0063] have:

[0064] α=α0(x k,i -x k,best )

[0065] Among them, x k,i is the i-th solution after the k-th iteration, x k+1,iis the i-th solution after the k+1-th iteration, α is the step size coefficient, α0 refers to the step size control constant, x k,best is the optimal solution for the kth iteration, is the dot product, L(λ) is the path randomly generated by Levy flight, which obeys the Levy flight probability distribution, and we have:

[0066]

[0067]

[0068] Among them, φ is the adjustment coefficient, λ is the adjustment parameter, u and v obey the (0,1) normal distribution, and Г(.) is the standard Gamma function.

[0069] In a second aspect, an embodiment of the present invention further provides a distribution network operation status evaluation device, including a data acquisition module, an evaluation module, a probability distribution model module, a state evaluation and matching module, and a parameter optimization module; wherein,

[0070] Data acquisition module, used to collect basic parameters of the power system, hyperparameter scheduling of the deep Q network and initial rule parameters of the confidence rule base;

[0071] An evaluation module is used to establish a deep Q network model for evaluating the future operation status of the distribution network by using the basic parameters of the power system, the hyperparameters in the deep Q network, and the initial rule parameters of the confidence rule base, and to evaluate the operation status of the distribution network within the future T time with Δt as the sampling time;

[0072] The probability distribution model module is used to establish the probability distribution of the distribution network state according to the state of each sampling point, and use the sampling point as one of the prerequisite inputs of the subsequent confidence rule base according to the probability distribution of the distribution network;

[0073] The state assessment and matching module determines the reference point and reference value of each premise input in the confidence rule base according to national standards and sampling points, matches the reference point of the premise input with the distribution network assessment rule, and outputs the assessment result;

[0074] The parameter optimization module adjusts the parameters of each rule in the confidence rule base according to the error between the evaluation result and the actual result.

[0075] As a further optimization solution of the distribution network operation status evaluation device described in the present invention, the data acquisition module adopts the ATT7022E chip.

[0076] In a third aspect, an embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein when the processor executes the computer program, the steps of the distribution network operation status assessment method as described in the first aspect above or any corresponding embodiment thereof are implemented.

[0077] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the distribution network operation status assessment method as described in the first aspect above or any corresponding embodiment thereof are implemented.

[0078] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0079] The distribution network operation status evaluation method based on deep Q network and confidence rule base provided by the present invention can realize the distribution network operation status evaluation under the influence of future distribution network operation status fluctuations. Compared with the traditional evaluation method that can only perform comprehensive evaluation on the distribution network, the present invention can obtain the operation status of each evaluation indicator and provide more effective information for distribution network evaluation. In addition, the present invention can adjust the confidence rule base parameters according to the error with the actual distribution network operation status to improve the evaluation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 It is a flow chart of distribution network operation status evaluation based on deep Q network and trust rule base of the present invention.

[0081] Figure 2 It is a system structure diagram of an embodiment of the present invention.

[0082] Figure 3 It is the distribution of the operation status of the distribution network every 15 minutes in the next 6 hours of the embodiment of the present invention.

[0083] Figure 4 It is an analysis diagram of the distribution network operation status evaluation results according to an embodiment of the present invention.

[0084] Figure 5 It is a comparison chart of the evaluation results of the present invention, AHP-Critic and multi-layer perceptron MLP. DETAILED DESCRIPTION

[0085] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0086] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0087] like Figure 1 As shown, a distribution network operation status evaluation method based on a deep Q network and a trust rule base includes the following steps:

[0088] S1. Input the basic parameters of the power system and the probability distribution information of the random input variables in the system.

[0089] S11. Input basic parameters of the power system, including system dispatching operation point, network topology, line impedance parameters and node load data;

[0090] S12, hyperparameters of the deep Q network, including the capacity of the experience replay pool, the number of training times, the training step size, the neural network parameters and the corresponding number of neurons and activation functions;

[0091] S13. Initial rules of the confidence rule base, including the rule weight of each rule, the weight of the premise input and the confidence distribution of each rule.

[0092] In this embodiment, opendss is used to build a circuit model to obtain input variables.

[0093] S2. Use the basic parameters of the power system, the hyperparameters in the deep Q network, and the initial rule parameters of the confidence rule base to establish a deep Q network model for evaluating the future distribution network operation status, and evaluate the distribution network operation status within the future T time with Δt as the sampling time, where Δt is each sampling period;

[0094] S21 . Determine the future time T for evaluation and each sampling period Δt.

[0095] S22. Evaluate the running state of Δt through the deep Q network, where the state space of the deep Q network is expressed as:

[0096] S = {s1, s2, ..., s nm}

[0097] Where n is the number of samples in the input model, and m is the number of input features for each sample.

[0098] For s in the state space i Definition of i is the state of the agent at the i-th moment, and we have:

[0099] si =(x i , f i )

[0100] In the formula, x i is the i-th input data, f i is the ith input feature.

[0101] The action space is represented as:

[0102] A=A f ∪A y

[0103] In the formula, A f is the action set for selecting attributes, that is, in state s t Next, sample x takes action a t Achieve selection of input feature f and subsequent actions; A y is the classification decision action set, that is, in state s t Next, sample x takes action a t Achieve prediction classification and end the decision-making process for the sample.

[0104] In status t If the agent takes A f The action of concentration will select a feature f from the input features i As the basis for classification. Action set A y = {0, 1, 2}, representing the operating state, warning state and fault state in the distribution network respectively. t If the agent takes A y The centralized action will classify the operating state of the distribution network in this state into operating state, warning state or fault state according to the existing learning experience.

[0105] The reward function is expressed as:

[0106]

[0107] In the formula, r t =1 indicates agent s t In state a, the action taken t Belongs to decision action A y And predicted

[0108] The result is the same as the actual label value y, r t = -1 means agent s t In state a, the action taken t Belongs to decision action A y But the predicted result is different from the actual label value y, r t = 0 means agent s tIn state a, the action taken t Belongs to the selection attribute action A f , only selects input features for subsequent actions, without any input sample data x t to classify.

[0109] State transfer and network update are expressed as:

[0110]

[0111] Where F is the stop decision identifier, (x t+1 , f t+1 ) indicates to continue making decisions and enter the next state.

[0112]

[0113] In the formula, s t+1 =F represents state s t The next state s of the agent t+1 The status is terminated.

[0114] Parameter iteration is expressed as:

[0115]

[0116] θ=ρθ+(1-ρ)θ

[0117] Where α is the neural network learning rate and ρ is the Target Q network update coefficient.

[0118] In this embodiment, voltage deviation, frequency deviation, load rate, controllable load ratio, controllable new energy ratio, fault rate and fault self-healing rate are selected as evaluation indicators.

[0119] S3. For the state of each sampling point, establish the probability distribution of the future distribution network state at time T based on the well-being theory, and use it as one of the prerequisite inputs of the subsequent confidence rule base;

[0120] S31. Using the well-being theory, we can find the probability distribution of states in the future T time period (T / Δt), and we have:

[0121]

[0122]

[0123]

[0124] Where P(H), P(W), and P(R) represent the probability of the distribution network being in the operating state, warning state, and fault state, respectively; T is the estimated future distribution network operation time; n(H), n(W), and n(R) represent the number of times the distribution network is in the operating state, warning state, and fault state within the time T, respectively; T i (H), T i (W), T i (R) represents the duration of the i-th time in the operating state, warning state, and fault state respectively.

[0125] In this embodiment, T is 6 h and Δt is 15 min.

[0126] S4. According to national standards and sampling points, determine the reference point and reference value of each premise input in the confidence rule base, match the reference point and reference value of the premise input with the distribution network evaluation rule, and output the evaluation result;

[0127] S41. According to the actual operation of the distribution network, select the corresponding indicators as the premise input, and establish the reference state and reference point with reference to the relevant national standards;

[0128] In this embodiment, voltage deviation, load rate and future distribution network operation status evaluation score are selected as the premise input, and their reference states and reference values ​​are shown in Table 1.

[0129] Table 1 Prerequisite input reference points and reference values

[0130]

[0131] S42. For each premise input of each group of inputs, the inputs are divided into corresponding reference states according to the numerical values, and the reference state of each premise input is matched with the rules in the confidence rule base, and the result is:

[0132]

[0133] Then y is{(D1,β 1,k ),(D2,β 2,k ),…,(D N ,β N,k )}

[0134] With rule weightθ k and attribute weightδ1,δ2,…,δ M

[0135] In the formula, R k is the kth rule of the model, is the reference value of the Mth premise input in the kth rule of the model, {(D1,β 1,k ),(D2,β2,k ),…,(D N ,β N,k )} is the output result set of the kth rule, β N,k is the confidence of the N output results of the kth rule, θ k is the weight of the kth rule, δ M is the weight of the Mth preceding input.

[0136] In this embodiment, the confidence rule base is initialized as shown in Table 2.

[0137] Table 2 Initial parameters of the confidence rule base

[0138]

[0139]

[0140] S43. After matching the corresponding rules, the corresponding distribution network operation status evaluation score is obtained, which is:

[0141]

[0142] Where D n,k is the fuzzy score of the nth evaluation level of the kth rule, β n,k is the confidence of the nth evaluation level of the kth rule.

[0143] S5. Based on the error between the evaluation result and the actual result, the parameter optimization method of cuckoo search is introduced to adjust the parameters of each rule in the confidence rule base.

[0144] S51. According to the error comparison between the evaluation result and the actual result in S4, the rule weight, attribute weight and confidence distribution of each rule in the confidence rule base are adjusted, as follows:

[0145] min Function(θ k ,β n,k ,δ i )

[0146]

[0147] In the formula, β n,k is the confidence of the n output results of the kth rule, θ k is the weight of the kth rule, δ i , is the attribute weight of the i-th preceding feature.

[0148] In this embodiment, the minimum MSE of the evaluation result score and the actual distribution network operation status score is selected as the objective function.

[0149] S52. Based on the cuckoo search algorithm, the rule weight, premise input weight and confidence distribution are adjusted, and:

[0150]

[0151] α=α0(x k,i -x k,best )

[0152] In the formula, x k,i is the i-th solution after the k-th iteration, α0 refers to the step size control constant, usually 0.01, x k,best is the optimal solution for the kth iteration, is the dot product, L(λ) is the path randomly generated by Levy flight, which obeys the Levy flight probability distribution, and we have:

[0153]

[0154]

[0155] Where u and v follow the (0,1) normal distribution, λ is usually taken as 1.5, and Г(.) is the standard Gamma function.

[0156] Embodiment 1:

[0157] The system of this embodiment is based on the actual distribution network structure of a certain place, such as Figure 2 As shown in the figure, the distribution network has 149 nodes, a load capacity of 115MVA, and a photovoltaic capacity of 14.825MVA. The main lines include 10kV and 20kV.

[0158] The output results of the deep Q network evaluation model based on the confidence rule base evaluation model of the distribution network operation status distribution with 15-minute sampling points in the next 6 hours and the voltage deviation, load rate and future distribution network operation status score are as follows: Figure 3 and Figure 4 shown.

[0159] The evaluation results of the present invention are compared with those of AHP-Critic and multi-layer perceptron MLP. Figure 5 , as shown in Table 3.

[0160] Table 3 Comparison of evaluation methods

[0161]

[0162] Obviously, the method proposed in the present invention has higher accuracy and better computing efficiency than traditional classification evaluation methods, which is of great significance for the evaluation of the operating status of new distribution networks.

[0163] The present invention also discloses a distribution network operation status evaluation device, including a data acquisition module, an evaluation module, a probability distribution model module, a status evaluation and matching module, and a parameter optimization module; wherein,

[0164] Data acquisition module, used to collect basic parameters of the power system, hyperparameters of the deep Q network and initial rule parameters of the confidence rule base;

[0165] An evaluation module is used to establish a deep Q network model for evaluating the future operation status of the distribution network by using the basic parameters of the power system, the hyperparameters in the deep Q network, and the initial rule parameters of the confidence rule base, and to evaluate the operation status of the distribution network within the future T time with Δt as the sampling time;

[0166] The probability distribution model module is used to establish the probability distribution of the distribution network state according to the state of each sampling point, and use the sampling point as one of the prerequisite inputs of the subsequent confidence rule base according to the probability distribution of the distribution network;

[0167] The state assessment and matching module determines the reference point and reference value of each premise input in the confidence rule base according to national standards and sampling points, matches the reference point of the premise input with the distribution network assessment rule, and outputs the assessment result;

[0168] The parameter optimization module adjusts the parameters of each rule in the confidence rule base according to the error between the evaluation result and the actual result.

[0169] The data acquisition module uses ATT7022E chip.

[0170] The data acquisition module includes a unit for collecting basic parameters of the power system, a unit for collecting hyperparameters of the deep Q network, and a unit for collecting initial rule parameters of the confidence rule base, wherein:

[0171] The unit for collecting basic parameters of the power system collects basic parameters of the power system, and the basic parameters of the power system include system dispatching operation point, network topology, line impedance parameters and node load data;

[0172] The unit for collecting hyperparameters of the deep Q network collects hyperparameters in the deep Q network, and the hyperparameters in the deep Q network include the capacity of the experience replay pool, the number of training times, the training step size, the neural network parameters and the corresponding number of neurons and the activation function;

[0173] The unit for collecting initial rule parameters of the confidence rule base collects initial rule parameters of the confidence rule base, and the initial rule parameters of the confidence rule base include a rule weight of each rule, a weight of a premise input, and a confidence distribution of each rule.

[0174] The evaluation module includes a deep Q network model unit, which is used to evaluate the deep Q network model of the future distribution network operation status, as follows:

[0175] Determine the future time T of the evaluation and each sampling period Δt;

[0176] The running state of Δt is evaluated by the deep Q network, where the state space of the deep Q network is expressed as:

[0177] S={s1,s2,…,s nm}

[0178] Among them, S is the state space, n is the number of samples in the input deep Q network, m is the number of input features for each sample, and s t is the state of the agent at time t, 1≤t≤nm;

[0179] For s in the state space i Definition of i is the state of the agent at the i-th moment, and we have:

[0180] s i =(x i ,f i )

[0181] Among them, x i is the i-th input data, f i is the i-th input feature;

[0182] The action space is represented as:

[0183] A=A f ∪A y

[0184] Among them, A is the action space, A f is the action set for selecting attributes, that is, in state s t Next, sample x takes action a t Achieve selection of input feature f and subsequent actions; A y is the classification decision action set, that is, in state s t Next, sample x takes action a t Realize the prediction classification and end the decision-making process of the sample x;

[0185] In status t If the agent takes A f The action in will select the i-th input feature f from the input features i As a basis for classification; A y ={0,1,2}, where A y=0, 1, 2, respectively represent the running state, warning state and fault state in the distribution network; in state s t If the agent takes A y Centralized action, then the operating state of the distribution network in this state is divided into operating state, warning state or fault state;

[0186] The reward function is expressed as:

[0187]

[0188] Among them, r t (s t ,a t ) to take action t Post-state t The reward value obtained, r t (s t ,a t )=1 indicates that agent s t State, action taken a t Belongs to A y And the predicted result is the same as the actual label value y, r t (s t ,a t )=-1 means that agent s t State, action taken a t Belongs to A y But the predicted result is different from the actual label value y, r t (s t ,a t )=0 means that agent s t In state a, the action taken t Belongs to A f , it only selects input features for subsequent actions, without classifying the input sample data;

[0189] State transfer and network update are expressed as:

[0190]

[0191] Among them, F is the stop decision identifier, (x t+1 , f t+1 ) means continue decision-making and enter the next state, x t+1 represents t+1 input data, f t+1 represents the t+1th feature of the input data, p t represents the transfer probability at time t;

[0192]

[0193] Among them, Target Q represents the network value, rt represents the reward value at time t, γ represents the discount factor, Q(s t+1 , a t+1 ; θ′) represents the state s t+1 Take action a t+1 The Q value of a t+1 represents the action at time t+1, θ′ is the Target Q network parameter value, s t+1 =F represents state s t The next state s of the agent t+1 is a terminated state;

[0194] Parameter iteration is expressed as:

[0195]

[0196] θ′=ρθ′+(1-ρ)θ

[0197] Among them, θ represents the Predict Q network parameter value, Loss(θ) represents the loss function, α is the neural network learning rate, and ρ is the Target Q network update coefficient.

[0198] The probability distribution model module includes a distribution network state probability distribution unit, which establishes the probability distribution of the distribution network state at the future time T based on the well-being theory, including:

[0199] The well-being theory is used to obtain the probability distribution of the distribution network state in the future time T (T / Δt) period, which is:

[0200]

[0201]

[0202]

[0203] Where P(H), P(W), and P(R) represent the probability of the distribution network being in the operating state, warning state, and fault state, respectively. T is the future time of the evaluation. n(H), n(W), and n(R) represent the number of times the distribution network is in the operating state, warning state, and fault state within the time T, respectively. i (H), T i (W), T i (R) represents the duration of the i-th time in the operating state, warning state, and fault state respectively.

[0204] The state evaluation and matching module includes an advance input unit, a matching unit, and a distribution network operation state evaluation unit, among which:

[0205] The advance input unit is used to select the voltage deviation rate, load rate and future operation time as the premise input according to the actual distribution network operation situation, and establish the reference state and reference point with reference to the known national standards;

[0206] The matching unit is used to divide each premise input of each group of inputs into corresponding reference states according to the numerical value of the premise input, and match the reference state of each premise input with the rules in the confidence rule base, including:

[0207]

[0208] Then y is{(D1,β 1,k ),(D2,β 2,k ),…,(D N ,β N,k )}

[0209] With rule weightθ k and attribute weightδ1,δ2,…,δ M

[0210] Among them, x b is the feature value of the b-th premise input, M≥b≥1, M is the total number of premise features, R k is the kth rule of the confidence rule base model, is the reference value of the b-th premise input in the k-th rule of the confidence rule base model, {(D1,β 1,k ),(D2,β 2,k ),…,(D N ,β N,k )} is the output result set of the kth rule, β a,k is the confidence of the a output result of the kth rule, D a is the evaluation result corresponding to the ath rule, N≥a≥1, N is the total number of distribution network operating states, θ k is the weight of the kth rule, δ b is the weight of the bth preceding input;

[0211] The distribution network operation status evaluation unit is used to calculate the distribution network operation status evaluation score, which is:

[0212]

[0213] Among them, D d,k is the fuzzy score of the d-th evaluation level of the k-th rule, β d,k is the confidence of the dth evaluation level of the kth rule.

[0214] The parameter optimization module includes an adjustment unit, which is used to adjust the parameters of each rule in the confidence rule base according to the error between the evaluation result and the actual result, including:

[0215] According to the comparison between the evaluation result and the actual result error, the rule weight, attribute weight and confidence distribution of each rule in the confidence rule base are adjusted as follows:

[0216] min Function(θ k , β d，k， δ i )

[0217]

[0218] Among them, Function(θ k , β d，k , δ i ) represents the parameter optimization function, β d，k is the confidence of the d output results of the kth rule, θ k is the weight of the kth rule, δ i is the attribute weight of the i-th pre-feature, θ l is the rule weight of the lth rule, and L is the total number of rules;

[0219] Based on the cuckoo search algorithm, adjust the rule weight, premise input weight and confidence distribution,

[0220] have:

[0221] α=α0(x k，i -x k，best )

[0222] Among them, x k，i is the i-th solution after the k-th iteration, x k+1，i is the i-th solution after the k+1-th iteration, α is the step size coefficient, α0 refers to the step size control constant, x k，best is the optimal solution for the kth iteration, is the dot product, L(λ) is the path randomly generated by Levy flight, which obeys the Levy flight probability distribution, and we have:

[0223]

[0224]

[0225] Among them, φ is the adjustment coefficient, λ is the adjustment parameter, u and v obey the (0, 1) normal distribution, and Γ(.) is the standard Gamma function.

[0226] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0227] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0228] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0229] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0230] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0231] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for evaluating the operating status of a distribution network, characterized in that: include: Input the basic parameters of the power system, the hyperparameters in the deep Q network, and the initial rule parameters of the confidence rule base; A deep Q network model for evaluating the future operation status of the distribution network is established by using the basic parameters of the power system, the hyperparameters in the deep Q network, and the initial rule parameters of the confidence rule base. The operation status of the distribution network within the future T time with Δt as the sampling time is evaluated, and Δt is each sampling period. According to the state of each sampling point, the probability distribution of the distribution network state at the future time T is established based on the well-being theory. According to the probability distribution of the distribution network, the sampling point is used as one of the premise inputs of the confidence rule base. According to national standards and sampling points, determine the reference point and reference value of each premise input in the confidence rule base, match the reference point and reference value of the premise input with the distribution network evaluation rule, and output the evaluation result; According to the error between the evaluation result and the actual result, the parameter optimization method of cuckoo search is introduced to adjust the parameters of each rule in the confidence rule base.

2. A distribution network operation status evaluation method according to claim 1, characterized in that: The basic parameters of the input power system include system dispatching operation point, network topology, line impedance parameters and node load data; The hyperparameters of the deep Q network include the experience replay pool capacity, the number of training times, the training step size, the neural network parameters and the corresponding number of neurons and the activation function; The initial rule parameters of the confidence rule base include the rule weight of each rule, the weight of the premise input and the confidence distribution of each rule.

3. A distribution network operation status evaluation method according to claim 1, characterized in that: A deep Q network model is established to evaluate the future operation status of the distribution network, as follows: Determine the future time T of the evaluation and each sampling period Δt; The running state of Δt is evaluated by the deep Q network, where the state space of the deep Q network is expressed as: S={s1,s2,...,s nm } Among them, S is the state space, n is the number of samples in the input deep Q network, m is the number of input features for each sample, and s t is the state of the agent at time t, 1≤t≤nm; For s in the state space i Definition of i is the state of the agent at the i-th moment, and we have: s i =(x i ,f i ) Among them, x i is the i-th input data, f i is the i-th input feature; The action space is represented as: A=A f ∪A y Among them, A is the action space, A f is the action set for selecting attributes, that is, in state s t Next, sample x takes action a t Achieve selection of input feature f and subsequent actions; A y is the classification decision action set, that is, in state s t Next, sample x takes action a t Realize the prediction classification and end the decision-making process of the sample x; In status t If the agent takes A f The action in will select the i-th input feature f from the input features i As a basis for classification; A y ={0,1,2}, where A y =0, 1, 2, respectively represent the running state, warning state and fault state in the distribution network; in state s t If the agent takes A y Centralized action, then the operating state of the distribution network in this state is divided into operating state, warning state or fault state; The reward function is expressed as: Among them, r t (s t ,a t ) to take action t Post-state t The reward value obtained, r t (s t ,a t )=1 indicates that agent s t State, action taken a t Belongs to A y And the predicted result is the same as the actual label value y, r t (s t ,a t )=-1 means that agent s t State, action taken a t Belongs to A y But the predicted result is different from the actual label value y, r t (s t ,a t )=0 means that agent s t In state a, the action taken t Belongs to A f , it only selects input features for subsequent actions, without classifying the input sample data; State transfer and network update are expressed as: Among them, F is the stop decision identifier, (x t+1 ,f t+1 ) means continue decision-making and enter the next state, x t+1 represents t+1 input data, f t+1 represents the t+1th feature of the input data, p t represents the transfer probability at time t; Among them, Target Q represents the network value, r t represents the reward value at time t, γ represents the discount factor, Q(s t+1 ,a t+1 ; θ') represents the state s t+1 Take action a t+1 The Q value of a t+1 represents the action at time t+1, θ' is the Target Q network parameter value, s t+1 =F represents state s t The next state s of the agent t+1 is a terminated state; Parameter iteration is expressed as: θ′=ρθ′+(1-ρ)θ Among them, θ represents the Predict Q network parameter value, Loss(θ) represents the loss function, α is the neural network learning rate, and ρ is the Target Q network update coefficient.

4. A distribution network operation status evaluation method according to claim 3, characterized in that: According to the state of each sampling point, the probability distribution of the distribution network state at the future time T is established based on the well-being theory, including: The well-being theory is used to obtain the probability distribution of the distribution network state in the future time T (T / Δt) period, which is: Where P(H), P(W), and P(R) represent the probability of the distribution network being in the operating state, warning state, and fault state, respectively. T is the future time of the evaluation. n(H), n(W), and n(R) represent the number of times the distribution network is in the operating state, warning state, and fault state within the time T, respectively. i (H), T i (W), T i (R) represents the duration of the i-th time in the operating state, warning state, and fault state respectively.

5. A distribution network operation status evaluation method according to claim 3, characterized in that: Determine the reference point and reference value of each premise input in the confidence rule base, match the premise input state with the distribution network evaluation rule, and output the evaluation results, including: S41. Based on the actual operation of the distribution network, select the voltage deviation rate, load rate and future operation time as the premise input, and establish the reference state and reference point with reference to known national standards; S42. For each premise input of each group of inputs, the premise input is divided into corresponding reference states according to the numerical value of the premise input, and the reference state of each premise input is matched with the rules in the confidence rule base, and there are: Then y is{(D1,β 1，k ),(D2,β 2，k ),…,(D N ,β N，k )} With rule weight θ k and attribute weight δ1,δ2,…δ M Among them, x b is the feature value of the b-th premise input, M≥b≥1, M is the total number of premise features, R k is the kth rule of the confidence rule base model, is the reference value of the b-th premise input in the k-th rule of the confidence rule base model, {(D1, β 1，k ), (D2,β 2，k ),…,(D N , β N，k )} is the output result set of the kth rule, β a，k is the confidence of the a output result of the kth rule, D a is the evaluation result corresponding to the ath rule, N≥a≥1, N is the total number of distribution network operating states, θ k is the weight of the kth rule, δ b is the weight of the bth preceding input; S43, calculate the distribution network operation status evaluation score, which is: Among them, D d，k is the fuzzy score of the d-th evaluation level of the k-th rule, β d，k is the confidence of the dth evaluation level of the kth rule.

6. A method for evaluating the operating status of a distribution network according to claim 3, characterized in that: According to the error between the evaluation result and the actual result, the parameter optimization method of cuckoo search is introduced to adjust the parameters of each rule in the confidence rule base; the details are as follows: According to the comparison between the evaluation result and the actual result error, the rule weight, attribute weight and confidence distribution of each rule in the confidence rule base are adjusted as follows: min Function(θ k ,b d，k ,d i ) Among them, Function(θ k , β d，k , δ i ) represents the parameter optimization function, β d，k is the confidence of the d output results of the kth rule, θ k is the weight of the kth rule, δ i is the attribute weight of the i-th pre-feature, θ l is the rule weight of the lth rule, and L is the total number of rules; Based on the cuckoo search algorithm, adjust the rule weight, premise input weight and confidence distribution, have: α=α0(x k，i -x k，best ) Among them, x k,i is the i-th solution after the k-th iteration, x k+1,i is the i-th solution after the k+1-th iteration, α is the step size coefficient, α0 refers to the step size control constant, x k,best is the optimal solution for the kth iteration, is the dot product, L(λ) is the path randomly generated by Levy flight, which obeys the Levy flight probability distribution, and we have: Among them, φ is the adjustment coefficient, λ is the adjustment parameter, u and v obey the (0,1) normal distribution, and Г(.) is the standard Gamma function.

7. A distribution network operation status evaluation device, characterized in that: It includes data acquisition module, evaluation module, probability distribution model module, state evaluation and matching module, and parameter optimization module; among them, Data acquisition module, used to collect basic parameters of the power system, hyperparameter scheduling of the deep Q network and initial rule parameters of the confidence rule base; An evaluation module is used to establish a deep Q network model for evaluating the future operation status of the distribution network by using the basic parameters of the power system, the hyperparameters in the deep Q network, and the initial rule parameters of the confidence rule base, and to evaluate the operation status of the distribution network within the future T time with Δt as the sampling time; The probability distribution model module is used to establish the probability distribution of the distribution network state according to the state of each sampling point, and use the sampling point as one of the prerequisite inputs of the subsequent confidence rule base according to the probability distribution of the distribution network; The state assessment and matching module determines the reference point and reference value of each premise input in the confidence rule base according to national standards and sampling points, matches the reference point and reference value of the premise input with the distribution network assessment rules, and outputs the assessment results; The parameter optimization module adjusts the parameters of each rule in the confidence rule base according to the error between the evaluation result and the actual result.

8. A distribution network operation status evaluation device according to claim 7, characterized in that: The data acquisition module uses ATT7022E chip.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: When the processor executes the computer program, the steps of the method for evaluating the operating status of a distribution network as described in any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for evaluating the operating status of a distribution network as described in any one of claims 1 to 6 are implemented.

Images