Model training method, power distribution network fault risk level prediction method and device

Abstract

The invention provides a model training method and a power distribution network fault risk level prediction method and device, and relates to the technical field of power grid safety. The model training method comprises the steps of obtaining historical fault data of a power distribution network, wherein the historical fault data comprises environmental characteristic factors, electrical characteristic factors and reference fault risk levels when the power distribution network breaks down; the initial Bayesian network model obtains a causal relationship between the environmental characteristic factors and between the environmental characteristic factors and a reference fault risk level by adopting a preset association rule based on historical fault data, and adjusts the direction of a directed edge of the model based on the causal relationship; based on the risk weights corresponding to the common factors and the low-frequency high-risk factors in the environmental characteristic factors, the prior probability of the nodes of the model is adjusted, and a predicted fault risk level output by the model is obtained; and iteratively training the initial Bayesian network model based on the predicted fault risk level and the reference fault risk level to obtain a correlation enhanced causal Bayesian network model with higher accuracy.

Description

Technical Field

[0001] The present application relates to the technical field of power grid security, and particularly to a model training method, a method and device for predicting the fault risk level of a distribution network. Background Art

[0002] As a high-dimensional non-linear complex system in the power system, the distribution network plays a role in distributing electric energy. At the same time, the distribution network is also an important link between users and the power system, and its safe and stable operation directly affects the reliability of user power consumption. With the continuous development of the power system, the network topology of the distribution network has become increasingly complex. Correspondingly, the types and quantities of distribution equipment are also continuously increasing. Predicting the fault risk level of the distribution network can greatly improve the reliability of the distribution network operation and the stability of user power consumption.

[0003] Currently, a Bayesian network model is usually established based on the certainty of the impact of the failure rate under a set external environment, and then the fault risk level of the distribution network is predicted through the Bayesian network model. However, when predicting the fault risk level of the distribution network in the above manner, there is a problem of inaccurate prediction. Summary of the Invention

[0004] The present application provides a model training method, a method and device for predicting the fault risk level of a distribution network to solve the problem of inaccurate prediction when predicting the fault risk level of the distribution network in the current manner.

[0005] In a first aspect, the present application provides a model training method, including: obtaining historical fault data of a distribution network, where the historical fault data includes environmental characteristic factors, electrical characteristic factors of the distribution network, and a reference fault risk level when the distribution network fails, and the environmental characteristic factors include common factors and low-frequency high-risk factors; inputting the historical fault data into an initial Bayesian network model, where the initial Bayesian network model obtains the causal relationship between the environmental characteristic factors and between the environmental characteristic factors and the reference fault risk level based on the historical fault data by using a preset association rule, and adjusts the direction of the directed edges of the initial Bayesian network model based on the causal relationship; adjusting the prior probability of the nodes of the initial Bayesian network model based on a first risk weight corresponding to the common factors and a second risk weight corresponding to the low-frequency high-risk factors to obtain a predicted fault risk level output by the initial Bayesian network model, where the nodes in the initial Bayesian network model represent environmental characteristic factors and electrical characteristic factors; and iteratively training the initial Bayesian network model based on the predicted fault risk level and the reference fault risk level to obtain an associated enhanced causal Bayesian network model.

[0006] Optionally, the antecedents of the preset association rules include common factors and low-frequency high-risk factors, and the consequents of the preset association rules include the corresponding reference fault risk levels for the antecedents. The initial Bayesian network model is based on historical fault data and uses the preset association rules to obtain the causal relationships between environmental characteristic factors and between environmental characteristic factors and the reference fault risk levels, including: obtaining the support, confidence, and lift corresponding to the low-frequency high-risk factors based on the historical fault data, antecedents, and consequents; and obtaining the causal relationships between environmental characteristic factors and between environmental characteristic factors and the reference fault risk levels based on the support, confidence, and lift.

[0007] Optionally, based on the causal relationships, adjust the directions of the directed edges of the initial Bayesian network model, including: determining strong association rules based on the support, confidence, and lift; for each directed edge of the initial Bayesian network model, determining whether there is a corresponding strong association rule; if so, adjusting the direction of the directed edge of the initial Bayesian network model according to the support, confidence, and lift of the strong association rule corresponding to the directed edge.

[0008] Optionally, before adjusting the prior probabilities of the nodes of the initial Bayesian network model based on the first risk weight corresponding to the common factors and the second risk weight corresponding to the low-frequency high-risk factors, the model training method further includes: obtaining the first risk weight according to the frequency of occurrence of the common factors; obtaining the initial risk weight corresponding to the low-frequency high-risk factors using the Laplace smoothing method; performing a weighted sum of the support, confidence, and lift corresponding to the low-frequency high-risk factors to obtain the adjusted risk weight corresponding to the low-frequency high-risk factors; and determining the second risk weight as the product of the initial risk weight and the adjusted risk weight.

[0009] Optionally, based on the causal relationships, adjust the directions of the directed edges of the initial Bayesian network model, including: obtaining the first association relationships between environmental characteristic factors and between environmental characteristic factors and the reference fault risk levels using the preset association rules; obtaining the second association relationships between environmental characteristic factors through the Spearman correlation coefficient; eliminating redundant environmental characteristic factors from the environmental characteristic factors based on the first association relationships and the second association relationships to obtain the target environmental characteristic factors; and adjusting the directions of the directed edges of the initial Bayesian network model based on the target environmental characteristic factors and the causal relationships.

[0010] Optionally, the fault risk level of the distribution network is determined based on the number of transformers and users affected after a fault occurs in the distribution network, and the fault risk levels include Risk Level One, Risk Level Two, and Risk Level Three.

[0011] In a second aspect, the present application provides a method for predicting the fault risk level of a distribution network, including: in response to a fault risk level prediction instruction for the distribution network, obtaining fault data of the distribution network, where the fault data includes environmental characteristic factors when the distribution network fails, and the environmental characteristic factors include common factors and low-frequency high-risk factors; inputting the fault data into an associated enhanced causal Bayesian network model for fault risk level prediction to obtain an objective fault risk level output by the associated enhanced causal Bayesian network model, and the associated enhanced causal Bayesian network model is trained by using the model training method described in the first aspect of the present application.

[0012] In a third aspect, the present application provides a model training device, including: an acquisition module, configured to acquire historical fault data of the distribution network, where the historical fault data includes environmental characteristic factors when the distribution network fails, electrical characteristic factors of the distribution network, and a reference fault risk level, and the environmental characteristic factors include common factors and low-frequency high-risk factors; a first processing module, configured to input the historical fault data into an initial Bayesian network model, and the initial Bayesian network model obtains the causal relationship between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level based on the historical fault data by using a preset association rule, and based on the causal relationship, adjust the direction of the directed edges of the initial Bayesian network model; based on a first risk weight corresponding to the common factor and a second risk weight corresponding to the low-frequency high-risk factor, adjust the prior probability of the nodes of the initial Bayesian network model to obtain a predicted fault risk level output by the initial Bayesian network model, and the nodes in the initial Bayesian network model represent environmental characteristic factors and electrical characteristic factors; a second processing module, configured to iteratively train the initial Bayesian network model based on the predicted fault risk level and the reference fault risk level to obtain an associated enhanced causal Bayesian network model.

[0013] Optionally, the antecedent of the preset association rule includes common factors and low-frequency high-risk factors, and the consequent of the preset association rule includes the reference fault risk level corresponding to the antecedent. When the first processing module is configured to enable the initial Bayesian network model to obtain the causal relationship between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level based on the historical fault data by using the preset association rule, it is specifically configured to: based on the historical fault data, the antecedent, and the consequent, obtain the support, confidence, and lift corresponding to the low-frequency high-risk factor; based on the support, confidence, and lift, obtain the causal relationship between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level.

[0014] Optionally, when the first processing module is used to adjust the direction of the directed edges of the initial Bayesian network model based on causal relationships, it is specifically used to: determine strong association rules based on support, confidence, and lift; for each directed edge of the initial Bayesian network model, determine whether there is a corresponding strong association rule for the directed edge; if so, adjust the direction of the directed edge of the initial Bayesian network model according to the support, confidence, and lift of the strong association rule corresponding to the directed edge.

[0015] Optionally, the first processing module is further used to: before adjusting the prior probability of the nodes of the initial Bayesian network model based on the first risk weight corresponding to common factors and the second risk weight corresponding to low-frequency high-risk factors, obtain the first risk weight according to the frequency of occurrence of common factors; use the Laplace smoothing method to obtain the initial risk weight corresponding to low-frequency high-risk factors; perform weighted summation on the support, confidence, and lift corresponding to low-frequency high-risk factors to obtain the adjusted risk weight corresponding to low-frequency high-risk factors; determine that the second risk weight is the product of the initial risk weight and the adjusted risk weight.

[0016] Optionally, when the first processing module is used to adjust the direction of the directed edges of the initial Bayesian network model based on causal relationships, it is specifically used to: use a preset association rule to obtain a first association relationship between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level; obtain a second association relationship between environmental characteristic factors through the Spearman correlation coefficient; perform redundant environmental characteristic factor elimination on environmental characteristic factors based on the first association relationship and the second association relationship to obtain target environmental characteristic factors; adjust the direction of the directed edges of the initial Bayesian network model based on the target environmental characteristic factors and causal relationships.

[0017] Optionally, the fault risk level of the distribution network is determined based on the number of transformers and the number of users affected after the distribution network fails, and the fault risk level includes Risk Level One, Risk Level Two, and Risk Level Three.

[0018] In a fourth aspect, the present application provides a distribution network fault risk level prediction device, including: an acquisition module, configured to acquire fault data of the distribution network in response to a fault risk level prediction instruction for the distribution network, where the fault data includes environmental characteristic factors when the distribution network fails, and the environmental characteristic factors include common factors and low-frequency high-risk factors; a prediction module, configured to input the fault data into an associated enhanced causal Bayesian network model for fault risk level prediction to obtain a target fault risk level output by the associated enhanced causal Bayesian network model, and the associated enhanced causal Bayesian network model is trained by using the model training method described in the first aspect of the present application.

[0019] In a fifth aspect, the present application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0020] The memory stores computer-executable instructions;

[0021] The processor executes the computer-executable instructions stored in the memory to implement the model training method described in the first aspect of this application or the distribution network fault risk level prediction method described in the second aspect.

[0022] In a sixth aspect, this application provides a computer-readable storage medium. Computer program instructions are stored in the computer-readable storage medium. When the computer program instructions are executed, the model training method described in the first aspect of this application or the distribution network fault risk level prediction method described in the second aspect is implemented.

[0023] In a seventh aspect, this application provides a computer program product, including a computer program. When the computer program is executed, the model training method described in the first aspect of this application or the distribution network fault risk level prediction method described in the second aspect is implemented.

[0024] A model training method, a distribution network fault risk level prediction method and a device provided by this application. By obtaining historical fault data of the distribution network, the historical fault data includes environmental characteristic factors, electrical characteristic factors and reference fault risk levels when the distribution network fails. The environmental characteristic factors include common factors and low-frequency high-risk factors; input the historical fault data into an initial Bayesian network model. The initial Bayesian network model is based on the historical fault data and uses a preset association rule to obtain the causal relationship between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level. Based on the causal relationship, adjust the direction of the directed edges of the initial Bayesian network model; based on the first risk weight corresponding to the common factors and the second risk weight corresponding to the low-frequency high-risk factors, adjust the prior probability of the nodes of the initial Bayesian network model to obtain the predicted fault risk level output by the initial Bayesian network model. The nodes in the initial Bayesian network model represent environmental characteristic factors and electrical characteristic factors; based on the predicted fault risk level and the reference fault risk level, iteratively train the initial Bayesian network model to obtain an associated enhanced causal Bayesian network model. In this application, the uncertainties of common factors and low-frequency high-risk factors in environmental characteristic factors are considered. Based on the causal relationship between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level, the direction of the directed edges of the initial Bayesian network model is adjusted. Based on the first risk weight corresponding to the common factors and the second risk weight corresponding to the low-frequency high-risk factors, the prior probability of the nodes of the initial Bayesian network model is adjusted. An associated enhanced causal Bayesian network model is obtained through iterative training. The obtained associated enhanced causal Bayesian network model can more accurately predict the fault risk level of the distribution network, so as to provide effective decision-making support for the operation and management of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings here are incorporated into the specification and form a part of this specification, showing embodiments consistent with this application, and are used together with the specification to explain the principles of this application.

[0026] Figure 1 The flowchart of the model training method provided by an embodiment of this application;

[0027] Figure 2 The schematic diagram of the system structure of the distribution network provided by an embodiment of this application;

[0028] Figure 3 The schematic diagram of the structure of the traditional Bayesian network model provided by an embodiment of this application;

[0029] Figure 4 The schematic diagram of the structure of the causal Bayesian inference network model provided by an embodiment of this application;

[0030] Figure 5 The flowchart of the model training method provided by another embodiment of this application;

[0031] Figure 6 The schematic diagram of the confusion matrix provided by an embodiment of this application;

[0032] Figure 7 The flowchart of the distribution network fault risk level prediction method provided by an embodiment of this application;

[0033] Figure 8 The schematic diagram of the structure of the model training device provided by an embodiment of this application;

[0034] Figure 9 The schematic diagram of the structure of the distribution network fault risk level prediction device provided by an embodiment of this application;

[0035] Figure 10 The schematic diagram of the structure of the electronic device provided by an embodiment of this application.

[0036] Through the above accompanying drawings, specific embodiments of this application have been shown, and there will be more detailed descriptions later. These accompanying drawings and text descriptions are not intended to limit the scope of the concept of this application in any way, but to illustrate the concept of this application to those skilled in the art by referring to specific embodiments. Detailed Embodiments

[0037] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the following will clearly and completely describe the technical solutions in the embodiments of this application with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are some, but not all, of the embodiments of this application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in this application without creative efforts shall fall within the scope of protection of this application.

[0038] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use, and processing of relevant data need to comply with relevant laws, regulations, and standards, and corresponding operation entrances are provided for users to choose to authorize or refuse.

[0039] As a high-dimensional nonlinear complex system in the power system, the distribution network plays the role of distributing electric energy. At the same time, the distribution network is also an important link between users and the power system, and its safe and stable operation directly affects the reliability of user power consumption. With the continuous development of the power system, the network topology of the distribution network has become increasingly complex. Correspondingly, the types and quantities of distribution equipment are also continuously increasing, and these changes pose challenges to the stable and reliable operation of the distribution network. For example, approximately 50% of power outages are caused by distribution network faults, and the occurrence of distribution network faults is closely related to external extreme environmental factors. Predicting the fault risk level of the distribution network can greatly improve the reliability of distribution network operation and the stability of user power consumption.

[0040] The occurrence of distribution network faults is not only affected by bad weather but also related to the network structure of the distribution network, internal electrical parameters, etc. These factors often depend on each other in real situations, and there are no completely independent characteristic scenarios. There are different degrees of interaction and internal connection among the influencing factors. In practical applications, the data characteristics of these factors are not exactly equivalent to their correlation degrees with faults. For example, although bad weather belongs to a low-frequency variable, its impact on the stable operation of the distribution network is significant. Currently, a Bayesian network model is usually obtained by modeling based on the certainty of the influence of the failure rate under a set external environment, and then the fault risk level of the distribution network is predicted through the Bayesian network model. However, predicting the fault risk level of the distribution network in the above manner does not consider the occasional occurrence of extreme weather and the uncertainty of distributed power generation output. Therefore, there is a problem of inaccurate prediction.

[0041] In order to be able to predict more accurately, uncertainty needs to be considered when establishing a fault risk level prediction model. However, the structures of the Bayesian network models proposed in current research are mostly constructed based on subjective experience and intelligent algorithms, and cannot well reflect the causality and correlation between various factors. If the traditional Bayesian method is used to predict the fault risk level of a distribution network, the predicted results are somewhat subjective and deviate from the actual situation.

[0042] Based on the above problems, the present application provides a model training method, a distribution network fault risk level prediction method and device, which make full use of the operation data and external weather data corresponding to the occurrence of faults in the distribution network, deeply explore the correlation and causal relationship between external environmental characteristic factors and between environmental characteristic factors and the distribution network fault risk level, and use improved association rules to adjust and optimize the structure of the initial Bayesian network model to obtain an association-enhanced causal Bayesian network model, which can be used to more accurately predict the fault risk level of the distribution network, thereby providing effective decision-making support for the operation and management of the distribution network.

[0043] It should be noted that the model training method and the distribution network fault risk level prediction method provided in the embodiments of the present application can be applied in a server, and the server can be an independent server, or can also be a service cluster, etc.

[0044] Figure 1 It is a flowchart of the model training method provided by an embodiment of the present application. As Figure 1 shown, the method of the embodiment of the present application includes:

[0045] S101. Obtain the historical fault data of the distribution network. The historical fault data includes environmental characteristic factors, electrical characteristic factors and reference fault risk levels when the distribution network fails. The environmental characteristic factors include common factors and low-frequency high-risk factors.

[0046] In the embodiments of the present application, historical fault data of a distribution network can be obtained as training samples for training an associated enhanced causal Bayesian network model. The historical fault data of the distribution network is obtained by integrating the collected weather data and the corresponding fault data of the distribution network and performing data integration and cleaning. The historical fault data includes environmental characteristic factors, electrical characteristic factors, and reference fault risk levels when the distribution network fails. The environmental characteristic factors include common factors and low-frequency high-risk factors. Specifically, the common factors include, for example, temperature, humidity, air pressure, wind speed, precipitation, and wind direction, etc.; the low-frequency high-risk factors include, for example, typhoons, fires, and icing, etc.; the electrical characteristic factors include, for example, the voltage level of the substation. For example, the support degree index in traditional association rules can be used to distinguish common factors from low-frequency high-risk factors. The electrical characteristic factors are fixed values, while the environmental characteristic factors are uncertain, and there is no direct relationship between the two. In the embodiments of the present application, the electrical characteristic factors are used as the basic data of the initial Bayesian network model to analyze the environmental characteristic factors to optimize the initial Bayesian network model and obtain an associated enhanced causal Bayesian network model.

[0047] Optionally, the fault risk level of the distribution network is determined based on the number of transformers and users affected after the distribution network fails. The fault risk level includes Risk Level One, Risk Level Two, and Risk Level Three.

[0048] Exemplarily, the power outage losses caused by different faults in the distribution network are different. To quantify the severity of different faults, in the embodiments of the present application, based on the number of transformers and users affected after the distribution network fails, the following formula one (i.e., the weighted formula) is used to divide different fault risk levels:

[0049] R = ω g ×G + ω u ×U Formula One

[0050] Wherein, R represents the risk score; G represents the number of low-voltage users affected; U represents the number of transformers affected; ω g represents the first weight adjustment coefficient, and the value is, for example, 0.6; ω u represents the second weight adjustment coefficient, and the value is, for example, 0.4.

[0051] According to the risk score obtained from the above formula one, combined with the actual application situation and expert experience, different risk thresholds can be set, so as to divide the fault risk level into three categories, namely Risk Level One, Risk Level Two, and Risk Level Three. These three fault risk levels respectively represent that the current fault risk degree (wind direction state) of the distribution network is general, moderate, and severe. Table 1 shows the risk thresholds and the definitions of the fault risk levels.

[0052] Table 1

[0053]

[0054] It can be understood that in an actual distribution network, the external environmental characteristic factors are not completely independent, and there is a dependence relationship among the environmental characteristic factors. A fault event record contains multiple environmental characteristic factors. If the environmental characteristic factors in a record change, the fault state may also change. That is, under the action of different environmental characteristic factors, the probability that the distribution network belongs to different fault risk levels will also be different. That is, the true state of the distribution network should be a parallel structure. Figure 2 The following is a schematic diagram of the system structure of the distribution network provided by an embodiment of the present application. As Figure 2 shown, assume that j 1 represents a fault record, then J = {j 1 , j 2 , …, j i , … j n} is a database containing n fault records, where i = 1, 2, … n. A fault record consists of elements corresponding to the characteristic factors (such as temperature values, wind speed values, etc.) and the fault risk level corresponding to this fault record. The characteristic factors can be represented by the set K = {k 1 , k 2 , …, k s , … k m , L}, where s = 1, 2, … m, the number of characteristic factors contained in the set K is m, and L represents the corresponding fault risk level; among them, for each characteristic factor, it is composed of multiple elements, then J is a matrix with n rows and m + 1 columns, denoted as Jn×(m + 1). These elements can determine the state of the distribution network. In this embodiment, let X = x i1 , x i2 , …, x is , … x im be all the corresponding elements of the fault record j i , where the element x is can be any component of the characteristic factor k s . Let L = l 1 , l 2 , …, l i … l n be the set of target elements in the evaluation database, where the target element l i represents the risk level corresponding to a real fault record j i , so the components of the target element can be further assumed to be l i= F(r) ∈ {F(1), F(2), F(3)}, where 1 represents risk level one, 2 represents risk level two, and 3 represents risk level three. Therefore, an association pattern containing the risk nature of components can be described as X → F. The connection path of any set of characteristic elements is called a failure risk chain, and a failure risk chain can be expressed as F(r) = {l i |x i1 , x i2 , …, x is , …x im}.

[0055] S102. Input historical failure data into the initial Bayesian network model. The initial Bayesian network model, based on the historical failure data, uses preset association rules to obtain the causal relationships between environmental characteristic factors and between environmental characteristic factors and the reference failure risk level. Based on the causal relationships, adjust the directions of the directed edges of the initial Bayesian network model; based on the first risk weight corresponding to common factors and the second risk weight corresponding to low-frequency high-risk factors, adjust the prior probabilities of the nodes of the initial Bayesian network model to obtain the predicted failure risk level output by the initial Bayesian network model. The nodes in the initial Bayesian network model represent environmental characteristic factors and electrical characteristic factors.

[0056] It can be understood that the establishment of a traditional Bayesian network model is generally based on expert knowledge or the results of preliminary data analysis. When the structure of the Bayesian network model is determined, the initial directed edges between variables (nodes) in the network will also be determined accordingly. Figure 3 This is a schematic structural diagram of a traditional Bayesian network model provided by an embodiment of the present application. As Figure 3 shown, this Bayesian network model describes the relationships between each node (i.e., A, B, C, and F) in the network through a directed acyclic graph. The probability dependence relationships between nodes are represented by a conditional probability table between nodes. Refer to Figure 3 , a traditional Bayesian network model focuses on describing the conditional dependence between variables and does not directly express the causal relationships between variables. Due to the lack of an accurate description of the causal structure, when predicting failure risks through a traditional Bayesian network model in practical applications, its prediction results may deviate from the actual situation.

[0057] Therefore, when constructing a Bayesian network model, considering the causal relationships between variables can construct a Causal Bayesian Inference Network (CBIN) model. This network model can not only capture the dependencies between variables, but more importantly, it can reveal the causal relationships between variables. Based on this,[[]] Figure 4 This is a schematic structural diagram of a causal Bayesian inference network model provided by an embodiment of the present application. As Figure 4As shown, in the causal Bayesian inference network model, for example, a directed edge A → B can be described as B = f i (A), indicating that the occurrence of A leads to the occurrence of B, where f i (·) is a causal function, representing the causal relationship between two characteristic factors. The clear causal relationship between variables enhances the interpretability of the Bayesian network and helps improve the accuracy of the prediction results. The obtained causal Bayesian inference network model can be used as the initial Bayesian network model, and the nodes in the initial Bayesian network model represent environmental characteristic factors and electrical characteristic factors.

[0058] It can be understood that traditional association rules are widely used to reveal the potential connections between different factors. In association rules, it is assumed that when the occurrence of factor X has an impact on the occurrence of factor Y, an association rule can be constructed: X → Y, where X represents the antecedent of the rule and Y represents the consequent of the rule. Since the proportion of low-frequency high-risk factors in the database is relatively low, when using traditional association rules to mine the characteristic factors affecting the operation of the distribution network, it is easy to ignore the impact of low-frequency high-risk factors on the distribution network. In order to distinguish the common factors and low-frequency high-risk factors in the database and quantify the association strength and causal relationship between them, the preset association rules in the embodiments of the present application are obtained by optimizing the support, confidence, and lift metrics in traditional association rules. Specifically, the association rule is redefined as: X c +X h → Y, and the antecedent of the rule is divided into the common factor antecedent X c and the low-frequency high-risk factor antecedent X h two parts, Y is the consequent of the rule, and this association rule is the preset association rule in the embodiments of the present application. In the embodiments of the present application, X represents the external environmental characteristic factors during the operation of the distribution network, and Y represents the fault risk level corresponding to a fault record in the database. The support, confidence, and lift of the low-frequency high-risk factors can be re-obtained through the preset association rules. Based on the obtained support, confidence, and lift of the low-frequency high-risk factors, the influence degree of different environmental characteristic factors on the distribution network risk level when they appear can be quantitatively evaluated, and at the same time, the association strength and causal relationship between various environmental characteristic factors can be analyzed. For how to use the preset association rules to obtain the causal relationship between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level, reference can be made to the subsequent embodiments.

[0059] In this step, historical fault data is input into the initial Bayesian network model. Based on the historical fault data, the initial Bayesian network model uses a preset association rule to obtain the causal relationships between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level. Based on the causal relationships, the directions of the directed edges of the initial Bayesian network model are adjusted; based on the first risk weight corresponding to common factors and the second risk weight corresponding to low-frequency high-risk factors, the prior probabilities of the nodes of the initial Bayesian network model are adjusted, so as to obtain the predicted fault risk level output by the initial Bayesian network model through the adjusted initial Bayesian network model. Among them, the adjusted initial Bayesian network model can also be understood as an Association-Enhanced Bayesian Causal Network (AEBCN) model. For how to obtain the causal relationships and based on the causal relationships, adjust the directions of the directed edges of the initial Bayesian network model, reference can be made to the subsequent embodiments.

[0060] Exemplarily, referring to the example of step S101, when a fault occurs in the distribution network, a fault risk chain can be expressed as F(r) = {l i |x i1 ,x i2 ,…,x is ,…x im}. The probability of the fault risk level P(F r ) of the distribution network at this time can be inferred through the following formula two (i.e., the Bayesian formula):

[0061] P(F r ) = P(x 1 )P(x 2 |x 1 )...P(x i |x 1 x 2 ...x i-1 ) Formula Two

[0062] Among them, P(x 1 ) represents the prior probability distribution of the first parent node x 1 ; P(x 2 |x 1 ) represents the probability of the second parent node x 2 occurring under the condition that the first parent node x 1 is given; P(x i |x 1 x 2 ...x i-1 ) represents the probability of the i-th parent node x i occurring under the condition that all leading parent nodes x 1 ,x 2,...,x i-1 The probability of occurrence under the condition of...

[0063] Characteristic factors in the fault risk chain include, for example, the substation voltage level Z, time T, temperature A, humidity B, air pressure C, wind speed D, precipitation E, and wind direction F. If the fault risk level of the distribution network is L, then under the action of the above characteristic factors, the posterior probability that the distribution network belongs to different fault risk levels can be further expressed as the following formula three:

[0064]

[0065] Among them, P(L = l) represents the probability that the distribution network belongs to the fault risk level l, and {·} represents the set of elements included in the environmental characteristic factors.

[0066] Based on the above formula two and formula three, the predicted fault risk level output by the initial Bayesian network model can be obtained.

[0067] S103. Based on the predicted fault risk level and the reference fault risk level, iteratively train the initial Bayesian network model to obtain an associated enhanced causal Bayesian network model.

[0068] In this step, after obtaining the predicted fault risk level, the predicted fault risk level and the reference fault risk level can be compared, and based on the comparison result, iteratively train the initial Bayesian network model to obtain an associated enhanced causal Bayesian network model that can more accurately output the predicted fault risk level.

[0069] The model training method provided by the embodiment of the present application obtains historical fault data of a distribution network. The historical fault data includes environmental characteristic factors, electrical characteristic factors, and reference fault risk levels when the distribution network fails. The environmental characteristic factors include common factors and low-frequency high-risk factors. The historical fault data is input into an initial Bayesian network model. Based on the historical fault data, the initial Bayesian network model uses a preset association rule to obtain the causal relationships between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level. Based on the causal relationships, the directions of the directed edges of the initial Bayesian network model are adjusted. Based on the first risk weight corresponding to the common factors and the second risk weight corresponding to the low-frequency high-risk factors, the prior probabilities of the nodes of the initial Bayesian network model are adjusted to obtain the predicted fault risk level output by the initial Bayesian network model. The nodes in the initial Bayesian network model represent environmental characteristic factors and electrical characteristic factors. Based on the predicted fault risk level and the reference fault risk level, the initial Bayesian network model is iteratively trained to obtain an associated enhanced causal Bayesian network model. In the embodiment of the present application, the uncertainties of the common factors and the low-frequency high-risk factors in the environmental characteristic factors are considered. Based on the causal relationships between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level, the directions of the directed edges of the initial Bayesian network model are adjusted. Based on the first risk weight corresponding to the common factors and the second risk weight corresponding to the low-frequency high-risk factors, the prior probabilities of the nodes of the initial Bayesian network model are adjusted. An associated enhanced causal Bayesian network model is obtained through iterative training. The obtained associated enhanced causal Bayesian network model can more accurately predict the fault risk level of the distribution network, thereby providing effective decision support for the operation and management of the distribution network.

[0070] Figure 5 It is a flowchart of the model training method provided by another embodiment of the present application. On the basis of the above embodiment, the embodiment of the present application further describes the model training method. As Figure 5 shown, the method of the embodiment of the present application may include:

[0071] S501. Obtain historical fault data of a distribution network. The historical fault data includes environmental characteristic factors and reference fault risk levels when the distribution network fails. The environmental characteristic factors include common factors and low-frequency high-risk factors.

[0072] For the specific description of this step, reference can be made to the relevant description of S101 in the embodiment shown in Figure 1 and details are not described herein again.

[0073] Considering that the antecedents of the preset association rule include common factors and low-frequency high-risk factors, and the consequent of the preset association rule includes the reference fault risk level corresponding to the antecedents, therefore, in the embodiment of the present application, Figure 1Step S102 may further include the following two steps: S502 and S503:

[0074] S502. Input the historical fault data into the initial Bayesian network model. The initial Bayesian network model obtains the support degree, confidence degree, and lift degree corresponding to the low-frequency high-risk factors based on the historical fault data, the antecedents and consequents of the preset association rules.

[0075] Exemplarily, referring to the example of step S102, the definition of the preset association rule in the embodiments of the present application is: X c +X h →Y. The antecedents of the preset association rule include the common factor antecedent X c and the low-frequency high-risk factor antecedent X h in two parts, and Y is the consequent of the preset association rule. In the embodiments of the present application, X represents the external environment characteristic factors during the operation of the distribution network, and Y represents the fault risk level corresponding to a fault record in the database. The initial Bayesian network model can obtain the support degree, confidence degree, and lift degree corresponding to the low-frequency high-risk factors based on the historical fault data, the antecedents and consequents of the preset association rule, respectively, through the following formulas (4), (5), and (6):

[0076]

[0077] Wherein, M(·) represents the number of events containing the characteristic factor in the database; M(X h ) represents the number of fault records containing low-frequency high-risk factors; M(h ij ) represents the number of the i-th characteristic factor in the specific fault risk level j; M(h ij , Y) represents the number of the i-th characteristic factor belonging to the fault risk level j in the entire database; M(Y) represents the total number of database records.

[0078] S503. Based on the support degree, confidence degree, and lift degree, obtain the causal relationships between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level.

[0079] In this step, after obtaining the support degree, confidence degree, and lift degree corresponding to the low-frequency high-risk factors, the causal relationships between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level can be obtained based on the support degree, confidence degree, and lift degree.

[0080] S504. Use the preset association rule to obtain the first association relationship between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level; obtain the second association relationship between environmental characteristic factors through the Spearman correlation coefficient.

[0081] It can be understood that in the actual operation scenario of the distribution network, the characteristic factors of different fault risks appear at different frequencies, and there is no complete independence. There are various degrees of interaction and internal connection. When mining corresponding rules based on these similar characteristic factors, the obtained rules may be similar or repetitive. The Spearman correlation coefficient can quantify the change trend between two random variables, and can be used to reflect the correlation degree between characteristic factors, mine the potential coupling mode between characteristic factors, eliminate redundant or equivalent characteristic sets, and improve the credibility of the measurement results.

[0082] Exemplarily, a preset association rule can be used to obtain the first association relationship between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level, and the second association relationship between environmental characteristic factors can be obtained through the Spearman correlation coefficient. Among them, when obtaining the second association relationship between environmental characteristic factors through the Spearman correlation coefficient, based on the confidence of each environmental characteristic factor and the fault risk level, the Spearman correlation coefficient is used to measure the correlation between each environmental characteristic factor. For example, for a single characteristic factor X A (such as an environmental characteristic factor) and any other characteristic factor X B The Spearman correlation coefficient between them can be obtained through the following formulas (7) and (8):

[0083]

[0084] Among them, represents the difference measure between characteristic factor X A and characteristic factor X B ; N represents the number of elements in characteristic factor X A and characteristic factor X B ; and respectively represent the i-th element (1≤i≤N) in characteristic factor X A and characteristic factor X B ; and respectively represent the sorting of the i-th element in characteristic factor X A and characteristic factor X B .

[0085] S505. Eliminate redundant environmental characteristic factors from the environmental characteristic factors based on the first association relationship and the second association relationship to obtain the target environmental characteristic factors.

[0086] In this step, after obtaining the first association relationship and the second association relationship, redundant environmental characteristic factors can be eliminated from the environmental characteristic factors based on the first association relationship and the second association relationship to obtain the target environmental characteristic factors.

[0087] Considering that the direction of the directed edges of the initial Bayesian network model can be adjusted based on the target environmental characteristic factors and causal relationships, therefore, in the embodiments of the present application, Figure 1 Step S102 may further include the following two steps of S506 and S507:

[0088] S506. Determine strong association rules based on the support, confidence, and lift corresponding to the low-frequency high-risk factors included in the target environmental characteristic factors.

[0089] Exemplarily, strong association rules with a support greater than the support threshold, a confidence greater than the confidence threshold, and a lift greater than the lift threshold may be determined based on the support, confidence, and lift corresponding to the low-frequency high-risk factors included in the target environmental characteristic factors. For example, in the scenario of the fault risk level of a distribution network, it may be found that the rule from "extreme weather conditions" to "fault risk level of the distribution network" is a strong rule.

[0090] S507. For each directed edge of the initial Bayesian network model, determine whether there is a corresponding strong association rule; if so, adjust the direction of the directed edge of the initial Bayesian network model according to the support, confidence, and lift of the strong association rule corresponding to the directed edge.

[0091] Exemplarily, each directed edge of the initial Bayesian network model may be traversed to check and determine whether there is a corresponding strong association rule for the directed edge. When it is found that there is a strong association rule corresponding to a certain directed edge, the direction of the directed edge may be adjusted according to the support, confidence, and lift of the strong association rule, so as to more accurately reflect the correlation between variables.

[0092] S508. Obtain the first risk weight corresponding to the common factor according to the frequency of occurrence of the common factor included in the target environmental characteristic factors.

[0093] Exemplarily, for a common factor, the first risk weight corresponding to the common factor may be obtained according to the frequency of occurrence of the common factor. Specifically, the first risk weight corresponding to the common factor may be obtained through the following formula nine:

[0094]

[0095] where ω c.ji represents the first risk weight corresponding to the i-th common factor in the j-th characteristic factor; M(X∈X c.ji ,k j.c ) represents the number of the i-th common factor in the j-th characteristic factor containing the common factor in the total database.

[0096] S509. Obtain the initial risk weight corresponding to the low-frequency high-risk factors included in the target environmental characteristic factors by using the Laplace smoothing method; perform a weighted sum of the support, confidence, and lift corresponding to the low-frequency high-risk factors to obtain the adjusted risk weight corresponding to the low-frequency high-risk factors; determine that the second risk weight corresponding to the low-frequency high-risk factors is the product of the initial risk weight and the adjusted risk weight.

[0097] It can be understood that there are many influencing factors when the distribution network operates. In order to further quantify and reflect the influence degree of each influencing factor on the reliable operation of the distribution network, and at the same time increase the prediction accuracy of the model, it is necessary to reasonably set the weights of each influencing factor. In a real fault record, the environmental characteristic factors include common factors and low-frequency high-risk factors. Combining the influence degree and its own nature of these two factors on the reliable operation of the distribution network, different methods can be used to solve the weights of these two factors.

[0098] Since the probability of occurrence of low-frequency high-risk factors is relatively low, but the influence degree on the overall reliability of the distribution network system is relatively large, the method for obtaining the weight of common factors (Formula Nine) is not applicable to low-frequency high-risk factors. In order to be able to more accurately measure the influence degree of low-frequency high-risk factors on the distribution network system and avoid the problem that the weight calculation is zero due to data sparsity, other methods need to be used to obtain the second risk weight corresponding to the low-frequency high-risk factors. Exemplarily, the Laplace smoothing method can be used to obtain the initial risk weight corresponding to the low-frequency high-risk factors included in the target environmental characteristic factors. It can be understood that when the Laplace smoothing method is used to obtain the factor weight, by adding a constant to the factor count, it is ensured that the factor weight calculation is not zero. Combining the Laplace smoothing method, the initial risk weight corresponding to the low-frequency high-risk factors can be obtained through the following Formula Ten:

[0099]

[0100] where, ω′ h.ji represents the initial risk weight corresponding to the i-th low-frequency high-risk factor in the j-th characteristic factor; M(X ∈ X h.ji , k j.h ) represents the number of the i-th low-frequency high-risk factor in the j-th characteristic factor containing low-frequency high-risk factors in the total database; N is the smoothing parameter, and its value is, for example, 2.

[0101] It can be understood that the acquisition of weights can reflect the importance of different factors, but it does not reflect the degree of association between factors and the fault risk level. The weights corresponding to different factors can be expressed as the prior probabilities of different nodes in the initial Bayesian network model. For the root node, its corresponding prior probability can be directly adjusted, and for the child node, the influence of prior knowledge and new data can be reflected by adjusting the parameters of its conditional probability distribution. In order to further refine the degree of improvement of different factors on the fault risk level, in this embodiment, based on support, confidence, and lift, the weighted summation method can be used to further optimize the weights corresponding to the factors. The weighted summation method is usually used to synthesize multiple numerical values into a single numerical value. Specifically, the weights corresponding to the factors can be further optimized through the following formula (11):

[0102] S = w 1 x 1 + w 2 x 2 +... + w n x n Formula (11)

[0103] Wherein, x 1 , x 2 , ···, x n represent data values; w 1 , w 2 , ···, w n represent the weights of each data value; S is the result of weighted summation.

[0104] Confidence, support, and lift measure the relevance between different factors and the fault risk level from different perspectives. Among them, confidence is used to measure the direct correlation between factors and the fault risk level; support is used to reflect the frequency of occurrence of factors in the dataset; lift is used to reveal whether there is a strong association between factors and the fault risk level. To a certain extent, the above three indicators can all measure the influence degree of factors on the fault risk level. Therefore, the weighted summation form can be used to synthesize the standardized values of the above three indicators to optimize the weights. Specifically, through the following formula (12), the support, confidence, and lift corresponding to low-frequency high-risk factors are weighted and summed to obtain the adjusted risk weight (i.e., λ) corresponding to low-frequency high-risk factors, and the second risk weight (i.e., ω h.ji ) corresponding to low-frequency high-risk factors is determined as the product of the initial risk weight and the adjusted risk weight:

[0105] λ = αSUP(h ij → y) adjust + βCON(h ij → y) adjust + γLIFT(h ij → y)adjust Formula XII

[0106] ω h.ji = λ·ω h ′ .ji Formula XIII

[0107] Among them, α, β, and γ are weight coefficients, satisfying α + β + γ = 1.

[0108] S510. Based on the first risk weight and the second risk weight, adjust the prior probability of the nodes of the initial Bayesian network model to obtain the predicted fault risk level output by the initial Bayesian network model. The nodes in the initial Bayesian network model represent environmental characteristic factors and electrical characteristic factors.

[0109] In this step, after obtaining the first risk weight corresponding to the common factors and the second risk weight corresponding to the low-frequency high-risk factors, the prior probability of the nodes of the initial Bayesian network model can be adjusted based on the first risk weight and the second risk weight, so as to obtain the predicted fault risk level output by the initial Bayesian network model.

[0110] S511. Based on the predicted fault risk level and the reference fault risk level, iteratively train the initial Bayesian network model to obtain an associated enhanced causal Bayesian network model.

[0111] The specific description of this step can be referred to Figure 1 the relevant description of S103 in the embodiment shown, which will not be repeated here.

[0112] The model training method provided by the embodiment of the present application includes: obtaining historical fault data of a distribution network, where the historical fault data includes environmental characteristic factors and reference fault risk levels when the distribution network fails, and the environmental characteristic factors include common factors and low-frequency high-risk factors; inputting the historical fault data into an initial Bayesian network model, and the initial Bayesian network model obtains the support degree, confidence degree, and lift degree corresponding to the low-frequency high-risk factors based on the historical fault data, the antecedents and consequents of preset association rules; obtaining the causal relationships between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level based on the support degree, confidence degree, and lift degree; obtaining the first association relationships between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level by using the preset association rules; obtaining the second association relationships between environmental characteristic factors by using the Spearman correlation coefficient; removing redundant environmental characteristic factors from the environmental characteristic factors based on the first association relationships and the second association relationships to obtain target environmental characteristic factors; determining strong association rules based on the support degree, confidence degree, and lift degree corresponding to the low-frequency high-risk factors included in the target environmental characteristic factors; for each directed edge of the initial Bayesian network model, determining whether there is a corresponding strong association rule; if so, adjusting the direction of the directed edge of the initial Bayesian network model according to the support degree, confidence degree, and lift degree of the strong association rule corresponding to the directed edge; obtaining the first risk weight corresponding to the common factor according to the frequency of occurrence of the common factor included in the target environmental characteristic factors, and obtaining the initial risk weight corresponding to the low-frequency high-risk factors included in the target environmental characteristic factors by using the Laplace smoothing method; performing weighted summation on the support degree, confidence degree, and lift degree corresponding to the low-frequency high-risk factors to obtain the adjusted risk weight corresponding to the low-frequency high-risk factors; determining the second risk weight corresponding to the low-frequency high-risk factors as the product of the initial risk weight and the adjusted risk weight; adjusting the prior probabilities of the nodes of the initial Bayesian network model based on the first risk weight and the second risk weight to obtain the predicted fault risk level output by the initial Bayesian network model, where the nodes in the initial Bayesian network model represent environmental characteristic factors and electrical characteristic factors; iteratively training the initial Bayesian network model based on the predicted fault risk level and the reference fault risk level to obtain an associated enhanced causal Bayesian network model.In the embodiments of the present application, the uncertainties of common factors and low-frequency high-risk factors in environmental characteristic factors are considered. Based on the support, confidence, and lift corresponding to the low-frequency high-risk factors included in the target environmental characteristic factors, strong association rules are determined. According to the support, confidence, and lift of the strong association rules corresponding to the directed edges, the directions of the directed edges of the initial Bayesian network model are adjusted; based on the first risk weight corresponding to the common factors and the second risk weight corresponding to the low-frequency high-risk factors, the prior probabilities of the nodes of the initial Bayesian network model are adjusted, and an associated enhanced causal Bayesian network model is obtained through iterative training. The obtained associated enhanced causal Bayesian network model can more accurately predict the fault risk level of the distribution network, thereby providing effective decision-making support for the operation and management of the distribution network.

[0113] On the basis of the above embodiments, the performance effect of the associated enhanced causal Bayesian network model of the embodiments of the present application can also be verified. Exemplarily, a confusion matrix is usually used to handle binary classification problems. By increasing the number of rows and columns of the confusion matrix, the confusion matrix can be applied to problems with more class values. The fault risk levels of the distribution network include risk level one, risk level two, and risk level three. Therefore, a confusion matrix can be used to measure the performance of the associated enhanced causal Bayesian network model. Figure 6 The schematic diagram of the confusion matrix provided for an embodiment of the present application is as Figure 6 shown. For example, the fault records that are actually at risk level one and are correctly predicted as risk level one are denoted as L TP1 , and the fault records that are actually at risk level one and are wrongly predicted as risk level two are denoted as L FN1,2 . In order to verify the performance effect of the associated enhanced causal Bayesian network model in this embodiment, the evaluation results can be compared with the corresponding true fault records, and the accuracy rate as shown in Formula XIV below and the recall rate as shown in Formula XV below are introduced:

[0114]

[0115] Among them, A 1 represents the accuracy rate of the associated enhanced causal Bayesian network model when predicting the fault risk level as risk level one; R 1 represents the recall rate of the associated enhanced causal Bayesian network model when predicting the fault risk level as risk level one.

[0116] Based on the above embodiments, Figure 7 The flowchart of the distribution network fault risk level prediction method provided for an embodiment of the present application is as Figure 7 shown. The method of the embodiment of the present application includes:

[0117] S701. In response to a fault risk level prediction instruction for a distribution network, obtain fault data of the distribution network. The fault data includes environmental characteristic factors when the distribution network fails, and the environmental characteristic factors include common factors and low-frequency high-risk factors.

[0118] Exemplarily, the fault risk level prediction instruction can be input by a user to an electronic device implementing the embodiments of this method, or automatically triggered by the electronic device implementing the embodiments of this method, or sent by other devices to the electronic device implementing the embodiments of this method. The electronic device implementing the embodiments of this method obtains the fault data of the distribution network in response to the fault risk level prediction instruction for the distribution network. The fault data includes environmental characteristic factors when the distribution network fails, and the environmental characteristic factors include common factors and low-frequency high-risk factors. Among them, the environmental characteristic factors include common factors and low-frequency high-risk factors. Specifically, common factors include, for example, temperature, humidity, air pressure, wind speed, precipitation, and wind direction, etc.; low-frequency high-risk factors include, for example, typhoons, fires, and icing, etc.; electrical characteristic factors include, for example, the voltage level of the substation, etc.

[0119] S702. Input the fault data into an associated enhanced causal Bayesian network model for fault risk level prediction to obtain the target fault risk level output by the associated enhanced causal Bayesian network model. The associated enhanced causal Bayesian network model is trained by using the model training method in any of the above method embodiments.

[0120] In this step, after obtaining the fault data of the distribution network, the fault data can be input into the associated enhanced causal Bayesian network model for fault risk level prediction to obtain the target fault risk level output by the associated enhanced causal Bayesian network model.

[0121] The distribution network fault risk level prediction method provided by the embodiments of the present application obtains the fault data of the distribution network in response to the fault risk level prediction instruction for the distribution network. The fault data includes environmental characteristic factors when the distribution network fails, and the environmental characteristic factors include common factors and low-frequency high-risk factors; input the fault data into the associated enhanced causal Bayesian network model for fault risk level prediction to obtain the target fault risk level output by the associated enhanced causal Bayesian network model. Since the prediction accuracy of the associated enhanced causal Bayesian network model in the embodiments of the present application is higher, therefore, the fault risk level of the distribution network can be predicted more accurately through the associated enhanced causal Bayesian network model, so as to provide effective decision support for the operation and management of the distribution network.

[0122] The following is an embodiment of the device of the present application, which can be used to execute the embodiment of the method of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the embodiment of the method of the present application.

[0123] Figure 8 The structural schematic diagram of the model training device provided by an embodiment of the present application. As Figure 8 shown, the model training device 800 of the embodiment of the present application includes: an acquisition module 801, a first processing module 802, and a second processing module 803. Among them:

[0124] The acquisition module 801 is used to acquire historical fault data of the distribution network. The historical fault data includes environmental characteristic factors, electrical characteristic factors of the distribution network, and reference fault risk levels when the distribution network fails. The environmental characteristic factors include common factors and low-frequency high-risk factors.

[0125] The first processing module 802 is used to input the historical fault data into the initial Bayesian network model. The initial Bayesian network model is based on the historical fault data and uses a preset association rule to obtain the causal relationships between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level. Based on the causal relationships, adjust the direction of the directed edges of the initial Bayesian network model; based on the first risk weight corresponding to the common factors and the second risk weight corresponding to the low-frequency high-risk factors, adjust the prior probabilities of the nodes of the initial Bayesian network model to obtain the predicted fault risk level output by the initial Bayesian network model. The nodes in the initial Bayesian network model represent environmental characteristic factors and electrical characteristic factors;

[0126] The second processing module 803 is used to iteratively train the initial Bayesian network model based on the predicted fault risk level and the reference fault risk level to obtain an associated enhanced causal Bayesian network model.

[0127] Optionally, the antecedent of the preset association rule includes common factors and low-frequency high-risk factors, and the consequent of the preset association rule includes the reference fault risk level corresponding to the antecedent. When the first processing module 802 is used for the initial Bayesian network model to obtain the causal relationships between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level based on the historical fault data by using the preset association rule, it can specifically be used for: based on the historical fault data, the antecedent, and the consequent, obtain the support degree, confidence degree, and lift degree corresponding to the low-frequency high-risk factors; based on the support degree, confidence degree, and lift degree, obtain the causal relationships between environmental characteristic factors and between environmental characteristic factors and the reference fault risk level.

[0128] Optionally, when the first processing module 802 is used to adjust the direction of the directed edges of the initial Bayesian network model based on the causal relationships, it can specifically be used for: based on the support degree, confidence degree, and lift degree, determine strong association rules; for each directed edge of the initial Bayesian network model, determine whether there is a corresponding strong association rule; if so, adjust the direction of the directed edge of the initial Bayesian network model according to the support degree, confidence degree, and lift degree of the strong association rule corresponding to the directed edge.

[0129] Optionally, the first processing module 802 may also be configured to: before adjusting the prior probability of the nodes of the initial Bayesian network model based on the first risk weight corresponding to the common factors and the second risk weight corresponding to the low-frequency high-risk factors, obtain the first risk weight according to the frequency of occurrence of the common factors; adopt the Laplace smoothing method to obtain the initial risk weight corresponding to the low-frequency high-risk factors; perform weighted summation on the support, confidence, and lift corresponding to the low-frequency high-risk factors to obtain the adjusted risk weight corresponding to the low-frequency high-risk factors; determine that the second risk weight is the product of the initial risk weight and the adjusted risk weight.

[0130] Optionally, when the first processing module 802 is used to adjust the direction of the directed edges of the initial Bayesian network model based on the causal relationship, it may specifically be configured to: adopt a preset association rule to obtain the first association relationship between the environmental characteristic factors and between the environmental characteristic factors and the reference fault risk level; obtain the second association relationship between the environmental characteristic factors through the Spearman correlation coefficient; perform redundant environmental characteristic factor elimination on the environmental characteristic factors based on the first association relationship and the second association relationship to obtain the target environmental characteristic factors; adjust the direction of the directed edges of the initial Bayesian network model based on the target environmental characteristic factors and the causal relationship.

[0131] Optionally, the fault risk level of the distribution network is determined based on the number of transformers and the number of users affected after the distribution network fails, and the fault risk level includes Risk Level One, Risk Level Two, and Risk Level Three.

[0132] The device according to the embodiment of the present application can be used to execute the solution of the model training method in any of the above method embodiments, and its implementation principle and technical effects are similar, which will not be elaborated here.

[0133] Figure 9 It is a schematic structural diagram of a distribution network fault risk level prediction device provided by an embodiment of the present application. As Figure 9 shown, the distribution network fault risk level prediction device 900 according to the embodiment of the present application includes: an acquisition module 901 and a prediction module 902. Among them:

[0134] The acquisition module 901 is configured to obtain the fault data of the distribution network in response to a fault risk level prediction instruction for the distribution network, where the fault data includes environmental characteristic factors when the distribution network fails, and the environmental characteristic factors include common factors and low-frequency high-risk factors.

[0135] The prediction module 902 is configured to input the fault data into the associated enhanced causal Bayesian network model for fault risk level prediction, and obtain the target fault risk level output by the associated enhanced causal Bayesian network model, where the associated enhanced causal Bayesian network model is trained by using the model training method in any of the above method embodiments.

[0136] The device according to the embodiment of the present application can be used to execute the solution of the distribution network fault risk level prediction method in any of the above method embodiments, and its implementation principle and technical effect are similar, which will not be elaborated here.

[0137] Figure 10 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application. As Figure 10 shown, the electronic device 1000 may include: at least one processor 1001 and a memory 1002.

[0138] The memory 1002 is used to store a program. Specifically, the program may include program code, and the program code includes computer execution instructions.

[0139] The memory 1002 may include a high-speed random access memory (Random Access Memory, RAM), and may also include a non-volatile memory, such as at least one disk memory.

[0140] The processor 1001 is used to execute the computer execution instructions stored in the memory 1002 to implement the model training method or the distribution network fault risk level prediction method described in the foregoing method embodiments. Among them, the processor 1001 may be a central processing unit (Central Processing Unit, CPU), or an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits configured to implement the embodiments of the present application. Specifically, when implementing the model training method or the distribution network fault risk level prediction method described in the foregoing method embodiments, the electronic device may be an electronic device with processing functions such as a server.

[0141] Optionally, the electronic device 1000 may further include a communication interface 1003. In a specific implementation, if the communication interface 1003, the memory 1002, and the processor 1001 are independently implemented, the communication interface 1003, the memory 1002, and the processor 1001 may be interconnected through a bus and communicate with each other. The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc., but it does not mean that there is only one bus or one type of bus.

[0142] Optionally, in a specific implementation, if the communication interface 1003, the memory 1002, and the processor 1001 are integrated on a single chip, the communication interface 1003, the memory 1002, and the processor 1001 can complete communication through an internal interface.

[0143] This application also provides a computer-readable storage medium storing computer program instructions. When the processor executes the computer program instructions, the solutions of the above model training method or the solutions of the distribution network fault risk level prediction method are implemented.

[0144] This application also provides a computer program product including a computer program, which when executed implements the solutions of the above model training method or the solutions of the distribution network fault risk level prediction method.

[0145] The above computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), magnetic memory, flash memory, a magnetic disk, or an optical disc. The readable storage medium can be any available medium accessible by a general-purpose or special-purpose computer.

[0146] An exemplary readable storage medium is coupled to the processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application-specific integrated circuit. Of course, the processor and the readable storage medium can also exist as discrete components in the model training device or the distribution network fault risk level prediction device.

[0147] Those of ordinary skill in the art will understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments; and the foregoing storage medium includes various media such as ROM, RAM, magnetic disk, or optical disc that can store program code.

[0148] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A model training method, characterized in that: include: Acquire historical fault data of the distribution network, the historical fault data including environmental characteristic factors when the distribution network fails, electrical characteristic factors of the distribution network and a reference fault risk level, the environmental characteristic factors including common factors and low-frequency high-risk factors; The historical fault data is input into an initial Bayesian network model. The initial Bayesian network model, based on the historical fault data, uses preset association rules to obtain the causal relationship between the environmental characteristic factors and between the environmental characteristic factors and the reference fault risk level. Based on the causal relationship, the direction of the directed edge of the initial Bayesian network model is adjusted; based on the first risk weight corresponding to the common factor and the second risk weight corresponding to the low-frequency high-risk factor, the prior probability of the node of the initial Bayesian network model is adjusted to obtain the predicted fault risk level output by the initial Bayesian network model, and the nodes in the initial Bayesian network model represent the environmental characteristic factors and the electrical characteristic factors; Based on the predicted fault risk level and the reference fault risk level, the initial Bayesian network model is iteratively trained to obtain an associated enhanced causal Bayesian network model.

2. The model training method according to claim 1, characterized in that: The antecedent of the preset association rule includes the common factors and the low-frequency high-risk factors, the posterior term of the preset association rule includes the reference fault risk level corresponding to the antecedent, and the initial Bayesian network model uses the preset association rule based on the historical fault data to obtain the causal relationship between the environmental characteristic factors and between the environmental characteristic factors and the reference fault risk level, including: Based on the historical fault data, the antecedent item and the consequent item, obtaining the support, confidence and improvement corresponding to the low-frequency high-risk factor; Based on the support, the confidence and the improvement, the causal relationship between the environmental characteristic factors and between the environmental characteristic factors and the reference fault risk level are obtained.

3. The model training method according to claim 2, characterized in that: The adjusting the direction of the directed edge of the initial Bayesian network model based on the causal relationship includes: Determining a strong association rule based on the support, the confidence and the lift; For each directed edge of the initial Bayesian network model, determining whether there is a corresponding strong association rule for the directed edge; If so, the direction of the directed edge of the initial Bayesian network model is adjusted according to the support, confidence and lift of the strong association rule corresponding to the directed edge.

4. The model training method according to claim 2, characterized in that: Before adjusting the prior probability of the nodes of the initial Bayesian network model based on the first risk weight corresponding to the common factor and the second risk weight corresponding to the low-frequency high-risk factor, the method further includes: Obtaining the first risk weight according to the frequency of occurrence of the common factors; The Laplace smoothing method is used to obtain the initial risk weight corresponding to the low-frequency high-risk factor; the support, confidence and lift corresponding to the low-frequency high-risk factor are weighted and summed to obtain the adjusted risk weight corresponding to the low-frequency high-risk factor; the second risk weight is determined to be the product of the initial risk weight and the adjusted risk weight.

5. The model training method according to any one of claims 1 to 4, characterized in that: The adjusting the direction of the directed edge of the initial Bayesian network model based on the causal relationship includes: Using the preset association rule to obtain a first association relationship between the environmental characteristic factors and between the environmental characteristic factors and the reference fault risk level; Obtaining the second correlation relationship between the environmental characteristic factors through the Spearman correlation coefficient; Eliminate redundant environmental characteristic factors from the environmental characteristic factors based on the first association relationship and the second association relationship to obtain target environmental characteristic factors; Based on the target environment characteristic factors and the causal relationship, the directions of the directed edges of the initial Bayesian network model are adjusted.

6. The model training method according to any one of claims 1 to 4, characterized in that: The fault risk level of the distribution network is determined based on the number of transformers and users affected by a fault in the distribution network, and the fault risk level includes risk level one, risk level two and risk level three.

7. A method for predicting the risk level of distribution network faults, characterized in that: include: In response to a fault risk level prediction instruction for the distribution network, acquiring fault data of the distribution network, the fault data including environmental characteristic factors when a fault occurs in the distribution network, the environmental characteristic factors including common factors and low-frequency high-risk factors; The fault data is input into the association enhanced causal Bayesian network model to predict the fault risk level, and the target fault risk level output by the association enhanced causal Bayesian network model is obtained. The association enhanced causal Bayesian network model is trained using the model training method described in any one of claims 1 to 6.

8. A model training device, characterized in that: include: An acquisition module is used to acquire historical fault data of the distribution network, wherein the historical fault data includes environmental characteristic factors when the distribution network fails, electrical characteristic factors of the distribution network, and a reference fault risk level, wherein the environmental characteristic factors include common factors and low-frequency high-risk factors; A first processing module is used to input the historical fault data into an initial Bayesian network model, wherein the initial Bayesian network model uses preset association rules to obtain the causal relationship between the environmental characteristic factors and between the environmental characteristic factors and the reference fault risk level based on the historical fault data, and adjusts the direction of the directed edge of the initial Bayesian network model based on the causal relationship; adjusts the prior probability of the nodes of the initial Bayesian network model based on the first risk weight corresponding to the common factors and the second risk weight corresponding to the low-frequency high-risk factors, and obtains the predicted fault risk level output by the initial Bayesian network model, wherein the nodes in the initial Bayesian network model represent the environmental characteristic factors and the electrical characteristic factors; The second processing module is used to iteratively train the initial Bayesian network model based on the predicted fault risk level and the reference fault risk level to obtain an associated enhanced causal Bayesian network model.

9. A distribution network fault risk level prediction device, characterized in that: include: An acquisition module, configured to acquire, in response to a fault risk level prediction instruction for the distribution network, fault data of the distribution network, wherein the fault data includes environmental characteristic factors when a fault occurs in the distribution network, and the environmental characteristic factors include common factors and low-frequency high-risk factors; A prediction module is used to input the fault data into an association enhanced causal Bayesian network model to predict the fault risk level, and obtain a target fault risk level output by the association enhanced causal Bayesian network model, wherein the association enhanced causal Bayesian network model is trained using the model training method described in any one of claims 1 to 6.

10. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 7.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed, the method according to any one of claims 1 to 7 is implemented.

12. A computer program product, comprising a computer program, characterized in that When the computer program is executed, the method according to any one of claims 1 to 7 is implemented.

Images