Bayesian network based natural disaster chain-power system-carbon emission estimation method

Abstract

The application discloses a natural disaster chain-power system-carbon emission estimation method based on a Bayesian network, uses an emergency model to perform basic modeling on a natural disaster chain-power system-carbon emission change process, uses the advantages of the Bayesian network in uncertainty event analysis to analyze the basic model, extracts key elements and the correlation between the elements from the basic model, forms a node table and an adjacency matrix, and performs reasoning calculation based on the node table and the adjacency matrix to build the Bayesian network, so that the carbon emission change analysis of the power system under the action of multiple natural disasters is realized. The method provides support for subsequent disaster resistance optimization of the power system, energy structure optimization, emission reduction and carbon fixation measures and the like, and can also be applied to other complex carbon emission process analysis.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of carbon emission estimation, and relates to a carbon emission change modeling and estimation method for multi-field complex processes, in particular to a natural disaster chain-power system-carbon emission estimation method based on a Bayesian network. BACKGROUND

[0002] Carbon emission estimation is the basis for formulating carbon emission reduction plans and constructing carbon reduction projects, and is based on basic estimation methods such as emission factor method, mass balance method and measurement method to comprehensively estimate CO2 emissions generated in production and consumption processes in society. Based on the results of carbon emission estimation, the distribution of carbon emissions in various industries and fields can be determined, making the carbon reduction plan and measures more purposeful, and the actual effect can be quantitatively evaluated.

[0003] The formation of natural disasters is an extremely complex atmospheric-geosystem movement process. Power system equipment is widely distributed, has high density and is closely coupled, and is easily affected by atmospheric-geosystem movement processes and natural disasters, especially clean energy power generation such as solar energy, wind energy and water energy and power grids connected with such power sources are more sensitive to weather, geology and environment and are more vulnerable to natural disasters. Natural disasters such as typhoons, lightning, heavy rain, forest fires and ice disasters will cause multiple equipment failures of the power system, cause power flow transfer and automatic device malfunction or refusal, and even cause large-scale power outages. The influence of natural disasters on power system equipment will eventually be reflected in the change of carbon emissions. The natural disaster chain-power system-carbon emission change is a complex continuous uncertain process involving multiple fields, and its characteristics determine that the basic estimation method of carbon emission cannot be directly applied to solve the problem, and a comprehensive method framework is needed to analyze the problem. SUMMARY

[0004] In view of the deficiencies of the prior art, the natural disaster chain-power system-carbon emission estimation method based on the Bayesian network is proposed, which uses the advantages of the Bayesian network in analyzing uncertain events, analyzes the emergency model, extracts key elements and the correlation between the elements, forms a node table and an adjacency matrix, and uses the same as a basis to build a Bayesian network for reasoning and calculation, so as to realize the carbon emission change analysis of the power system under the action of multiple natural disasters.

[0005] The natural disaster chain-power system-carbon emission estimation method based on the Bayesian network specifically includes the following steps:

[0006] Step 1: Problem abstraction and data preprocessing

[0007] s1.1, in the process of natural disaster event, the power system is considered to be the main body affected by the disaster, and also the main body of carbon emission change, based on the above relationship, the natural disaster chain, the power system and the carbon emission are integrated into a system, the system is modeled based on the system theory emergency model, the natural disaster is taken as the system input, the power system is taken as the system state, and the carbon emission change is taken as the system output, according to the historical record, the related elements and the correlation between the elements are extracted, the corresponding element set is formed, so that the natural disaster chain is separated into a single natural disaster event, and the emergency model of a single natural disaster is constituted:

[0008] E = (I, S, O) (1)

[0009] Wherein, I represents the natural disaster element set, S represents the power system element set, and O represents the carbon emission change element set.

[0010] s1.2, according to the correlation variable representing the relationship between the disasters before and after in s1.1, the emergency models of multiple natural disasters are connected together to form a "natural disaster chain-power system-carbon emission" system model, the correlation relationship between the elements in the system model is determined, the elements are taken as nodes to form a node table of the Bayesian network model, and an adjacency matrix is constructed according to the relationship between the elements.

[0011] s1.3, reference is made to the disaster historical data to determine the possible values of each node in s1.2 and the corresponding prior probability and conditional probability. The specific steps are as follows:

[0012] s1.3.1, if the node value obtained from the data is continuous, the node value is discretized by segmentation, and the discrete value range is used to replace the continuous value as the node value. If the node value is not continuous, multiple discrete value items are selected to be combined to reduce the number of node values.

[0013] s1.3.2, if the continuous node value is discretized in s1.3.1, the continuous distribution of the node value is calculated to obtain the probability of the segmented discrete value range; if the discrete node value is combined, the probability of the combined discrete node value is calculated. Otherwise, go to s1.3.3.

[0014] s1.3.3, if the current node has no directly associated predecessor node, the prior probability of the node is obtained as the probability information of the node; if the current node has a directly associated predecessor node, the conditional probability of the node under different values of the parent node is obtained as the probability information of the node.

[0015] Step two, Bayesian network construction

[0016] A Bayesian network is constructed based on the node list and adjacency matrix obtained in step one. In the Bayesian network, the adjacency matrix records the relationships between elements such as natural disaster chains, power systems, and carbon emission changes, while the node list records the names, values, and corresponding probability information of elements at each layer. The specific steps for constructing the Bayesian network are as follows:

[0017] s2.1. Use the adjacency matrix obtained in step one to generate a Bayesian network instance. This instance has the relationships between the element nodes described by the adjacency matrix, including multiple connected and merged "natural disaster-power system-carbon emission change" sub-networks. Each sub-network corresponds to a disaster in the natural disaster chain.

[0018] s2.2 Register the names and values ​​of the elements obtained in step one to each layer of the Bayesian network instance that has formed a topological framework.

[0019] s2.3. Inject the prior probabilities and conditional probabilities of the nodes obtained in step one into the Bayesian network instance.

[0020] s2.4. Create an inference engine based on the network instance. Starting from the initial element nodes, infer layer by layer downwards based on prior probabilities to ultimately obtain the prior probabilities of all nodes in the network instance, thus obtaining a basic network. The specific process is as follows:

[0021] s2.4.1. Based on the prior probability information and conditional probability information of the input layer nodes, infer and calculate the prior probability of the input layer:

[0022]

[0023] Where i1~i k For the input layer natural disaster elements related to i k+1 The relevant k natural disaster element nodes, p(i1,i2,…,i k ) represents i1~i k The prior probability of an element in a certain state, p(i) k+1 |i1,i2,…,i k ) represents i1~i k In a certain state, i k+1 The conditional probability. s2.4.2. Based on the results of s2.4.1, continue to infer and calculate the prior probability of the state layer.

[0024]

[0025] Where s1~sl are the power system elements in the state layer and sl +1 The relevant l power system element nodes, p(i1,i2,…,i k ,s1,s2,…,s l ) represents i1~ik Element and s1~s l Prior probability of element in certain state, p(s l+1 |i1,i2,…,i k ,s1,s2,…,s l ) represents i1~i k Element and s1~s l Conditional probability of element in certain state, s l+1 .

[0026] s2.4.3, based on the results of s2.4.1 and s2.4.2, continue to infer the prior probability of the output layer:

[0027]

[0028] Where o1~o m is the m carbon emission change element nodes related to o m+1 in the output layer carbon emission change element, p(i1,i2,…,i k ,s1,s2,…,s l ,o1,o2,…,o m ) represents i1~i k Element, s1~s l Element and o1~o m Element in certain state, p(o m+1 |i1,i2,…,i k ,s1,s2,…,s l ,o1,o2,…,o m ) represents i1~i k Element, s1~s l Element and o1~o m Element in certain state, conditional probability.

[0029] Step three, Bayesian network inference update

[0030] The probability information given in the Bayesian network is an objective law under general conditions. Due to the lack of information, the carbon emission change estimation result is not accurate for specific natural disaster chains. In real cases, the values of some natural disaster elements and power system elements can often be determined through observation. Using these observation results can reduce the uncertainty of the network and form a specialized network. The observed element nodes are called evidence variables. After introducing the evidence variable information into the network, the network is updated through inference. The specific steps are as follows:

[0031] s3.1, create a new inference engine according to the Bayesian network instance.

[0032] s3.2, the observed nodes are taken as evidence variables to form an evidence vector inputting into the inference engine.

[0033] s3.3, the inference engine is used to update the probability information of the basic network node by node and layer by layer, the posterior probability of each element node and each layer is obtained by step-by-step backward inference updating, the prior probability is replaced by the posterior probability to form a specialized network, and the specific steps are as follows:

[0034] s3.3.1, the first step of inference updating is carried out according to the evidence variables in s3.2, and the posterior probability of other input layer nodes except the evidence variables is calculated by inference calculation based on the Bayes formula.

[0035] s3.3.2, the second step of inference updating is carried out on the basis of s3.3.1, and the posterior probability of the state layer nodes is calculated by inference calculation based on the Bayes formula.

[0036] s3.3.3, the third step of inference updating is carried out on the basis of s3.3.1 and s3.3.2, and the posterior probability of the output layer nodes is calculated by inference calculation based on the Bayes formula.

[0037] Step four, calculating the carbon emission change estimation result

[0038] After the evidence vector is inputted and updated by s3.2 and s3.3, the Bayes network instance has been specialized in the direction of the actual observation result, and the overall uncertainty has been reduced. The posterior probability information of the output layer in the Bayes network instance is comprehensively calculated, the carbon emission change value and the posterior probability of the output layer of the sub-network corresponding to each disaster in the natural disaster chain are calculated, and the estimation result is outputted.

[0039] The present application has the following beneficial effects:

[0040] For the complex and uncertain process of "natural disaster chain-power system-carbon emission change", it is difficult to establish the mapping relationship from natural disaster to carbon emission change for the natural disaster chain with multiple disasters in series. The present method is based on the Bayes principle, and provides a lightweight framework, which can combine the specific mechanism of each part to better quantitatively describe the correlation relationship existing in the process, quantitatively analyze the specific reflection of the change of the natural disaster element on the carbon emission change, give the specific carbon emission estimation value in the specific scene, and has good migration and expansion potential. DETAILED DESCRIPTION

[0041] Figure 1 It is a power system topology schematic diagram in the embodiment;

[0042] Figure 2 It is a sudden event model schematic diagram;

[0043] Figure 3A schematic diagram of a natural disaster chain-power system-carbon emission system model;

[0044] Figure 4 A schematic diagram of the correlation between elements in the embodiment;

[0045] Figure 5 A schematic diagram of the Bayesian network model constructed in the embodiment;

[0046] Figure 6 A natural disaster chain-power system-carbon emission estimation method based on a Bayesian network. DETAILED DESCRIPTION

[0047] The application will be further explained in connection with the accompanying drawings, in which:

[0048] The natural disaster chain in the embodiment is “heavy rain-landslide-flood”, and the corresponding power system topology is as shown in Figure 1 where the thermal power, hydropower and photovoltaic power generation systems supply power to the user load without energy exchange with the external power grid; it is considered that the system has a certain disaster resistance under the impact of natural disasters, and it is assumed that all devices in the power system are in a completely fault-free operating state before the occurrence of each disaster in the disaster chain, and the basic attributes and the influence of natural disasters are as shown in Tables 1 and 2:

[0049]

[0050] Table 1

[0051]

[0052] Table 2

[0053] The method described in the application is used to estimate the carbon emission of this example, which specifically includes the following steps:

[0054] Step 1, problem abstraction and data preprocessing

[0055] s1.1, separate the three natural disasters of heavy rain, landslide and flood from the target natural disaster chain, take the natural disasters as the system input, the power system as the system state, and the carbon emission change as the system output, extract the relevant elements and the correlation between the elements, and establish an emergency model as shown in Figure 2

[0056] s1.2, as shown in Figure 3 , according to the correlation variables representing the relationship between the preceding and subsequent disasters in s1.1, link the emergency models of multiple natural disasters together to form a “natural disaster chain-power system-carbon emission” system model, and clearly define the preceding and subsequent correlation between the elements in the system model, taking the elements as nodes to form a Bayesian network model node table as shown in Table 3: ​

[0057]

[0058] Table 3

[0059] The adjacency matrix is constructed in terms of the relationship between elements.

[0060] In the present embodiment, the emergent event models of three single natural disasters, i.e. rainstorm, landslide and flood, can be obtained, and then the associated variables i br , i cr are connected to form a system network as shown in Figure 4 .

[0061] s1.3, according to the disaster history data, the possible values of each node in s1.2 are determined, for the nodes with continuous values, they are divided into two discrete value segments, and for the nodes with discrete values, the options are combined to have only two possible values. After the pretreatment, all nodes have only two values, as shown in Table 4:

[0062]

[0063] Table 4

[0064] The conditional probability of all nodes is shown in Table 5, in which the value condition of the parent node is simplified as 0 and 1, 0 represents the former value in Table 4, and 1 represents the latter value in Table 4:

[0065]

[0066]

[0067]

[0068]

[0069] Table 5

[0070] Step two, construction of the Bayesian network

[0071] s2.1, using the adjacency matrix obtained in step one, a Bayesian network instance is generated by MATLAB, as shown in Figure 5 , in which each numerical label corresponds to an element in Table 5. Figure 5

[0072] s2.2, the names and value information of each layer element shown in Table 4 are registered to each layer of the Bayesian network instance which has formed a topological framework.

[0073] s2.3, the prior probability and conditional probability of the nodes shown in Table 5 are injected into the Bayesian network instance.

[0074] ​s2.4, according to the network instance to create inference engine, from the starting element node according to the prior probability to infer down layer by layer, ultimately get all the nodes in the network instance of the prior probability, as shown in table 6:

[0075]

[0076] Table 6

[0077] Step three, Bayesian network inference update

[0078] The probability information given in the Bayesian network is an objective law under general conditions. Due to the lack of information, the carbon emission change estimation result is not accurate for specific natural disaster chain. In real cases, the values of some natural disaster elements and power system elements can be determined through observation. These observation results can reduce the uncertainty of the network and form a specialized network. The observed element node is called evidence variable. After introducing the evidence variable information into the network, the network is updated by inference. The specific steps are as follows:

[0079] s3.1, according to the Bayesian network instance to create a new inference engine.

[0080] s3.2, the observed node is taken as evidence variable, and the evidence vector shown in table 7 is input into the inference engine.

[0081]

[0082] Table 7

[0083] s3.3, use the inference engine to update the probability information of the basic network node by node and layer by layer. The posterior probability of each element node and each layer is obtained by updating step by step backward inference. Replace the prior probability with the posterior probability. The posterior probability of the updated network node is shown in table 8:

[0084]

[0085]

[0086]

[0087] Table 8

[0088] Step four, calculate the carbon emission change estimation result

[0089] After the calculation of the four steps shown in table 9, the probability estimation of the carbon emission change of the power system affected by natural disasters is obtained, as shown in table 9, wherein the probability of carbon emission change is the highest in 7864-8787tCO2, and the estimated value is 7864-8787tCO2. Figure 6

[0090] ​​

[0091] Table 9

Claims

1. A Bayesian network-based natural disaster chain-power system-carbon emission estimation method, characterized in that: Specifically comprising the following steps: Step one, problem abstraction and data preprocessing According to the element relationship among natural disaster chain, power system and carbon emission, the burst event model of single natural disaster is constructed: E=(I, S, O) (1) Wherein, I represents the natural disaster element set, S represents the power system element set, and O represents the carbon emission element set; The associated elements of the three are extracted, the burst event models of multiple natural disasters are connected together to form a "natural disaster chain-power system-carbon emission" system model, the correlation between the elements in the system model is clarified, the elements are taken as nodes, the values and probability information of the elements are determined, the node table of the Bayesian network model is formed; The adjacency matrix is constructed according to the relationship between the elements; Step two, construction of Bayesian network Based on the adjacency matrix obtained in step one, the Bayesian network instance is constructed, the name, value and probability information of each node are injected into the Bayesian network instance according to the information in the node table, then the prior probability of all nodes is derived to obtain a basic network; Step three, reasoning and updating of Bayesian network The observed new element value is taken as the evidence variable and injected into the Bayesian network constructed in step two, the posterior probability of each node is updated by reasoning backward again to complete the reasoning and updating of the network; Step four, calculation of carbon emission estimation result The posterior probability information of the output layer in the updated Bayesian network instance in step three is integrated, the carbon emission value and posterior probability of the output layer of the sub-network corresponding to each disaster in the natural disaster chain in the Bayesian network instance are calculated and summarized, and the estimation result is output.

2. The Bayesian network based natural disaster-chain-power system-carbon emission estimation method of claim 1, wherein: In step one, the values of the elements are derived from historical observation data, for the nodes with continuous values, the continuous values are replaced by discrete value ranges after being segmented and discretized, and the continuous distribution is calculated to obtain the probability of the segmented discrete value range; for the nodes with discrete values, the discrete values are selectively combined to reduce the number of element values, and the probability of the combined discrete node value is calculated.

3. The Bayesian network based natural disaster-chain-power system-carbon emission estimation method of claim 1, wherein: In step two, the derivation method of the prior probability of all nodes is: s2.1, according to the prior probability information and conditional probability information of the input layer nodes, the prior probability of the input layer is calculated by reasoning: Where i1~i k For the input layer natural disaster elements related to i k+1 The relevant k natural disaster element nodes, p(i1,i2,…,i k ) represents i1~i k The prior probability of an element in a certain state, p(i) k+1 |i1,i2,…,i k ) represents i1~i k In a certain state, i k+1 The conditional probability; s2.2, based on the result of s2.1, the prior probability of the state layer is calculated by reasoning; Where s1~sl are the power system elements in the state layer related to s l+1 The relevant l power system element nodes, p(i1,i2,…,i k ,s1,s2,…,s l ) represents i1~i k Elements and s1~s l The prior probability of an element in a certain state, p(s) l+1 |i1,i2,…,i k ,s1,s2,…,s l ) represents i1~i k Elements and s1~s l In a certain state, s l+1 The conditional probability; s2.3, based on the results of s2.1 and s2.2, the prior probability of the output layer is calculated by reasoning: where o1~o m are the output layer carbon emission change elements related to o m+1 , p(i1,i2,…,i k ,s1,s2,…,s l ,o1,o2,…,o m ) represents the prior probability of the i1~i k element, s1~s l element and o1~o m element in a certain state, p(o m+1 |i1,i2,…,i k ,s1,s2,…,s l ,o1,o2,…,o m ) represents the conditional probability of the i1~i k element, s1~s l element and o1~o m element in a certain state.

4. The Bayesian network based natural disaster-chain-power system-carbon emission estimation method of claim 1, wherein: In step three, the steps of network reasoning and updating are: s3.1, according to the evidence variable, the first step of reasoning and updating is carried out, and the posterior probability of other input layer nodes except the evidence variable is calculated by reasoning based on the Bayesian formula; s3.2, on the basis of s3.1, the second step of reasoning and updating is carried out, and the posterior probability of the state layer nodes is calculated by reasoning based on the Bayesian formula; s3.3, on the basis of s3.1 and s3.2, the third step of reasoning and updating is carried out, and the posterior probability of the output layer nodes is calculated by reasoning based on the Bayesian formula.

Images