[Retracted] Optimization of Substation Alarm Information Processing Based on BP Neural Network

Abstract

When a fault occurs in the power grid, a large number of alarm messages will come out. In order to reduce the damage caused by power outages, the staff should find out the location and type of faults as soon as possible based on the alarm messages uploaded from the control center. However, we are currently facing problems in handling these alarm messages such as reliability of protection actions and circuit breaker tripping, correctness of received alarm messages, and the possibility of the existence of unreceived alarm messages. The fault tolerance of the BP neural network is studied in this paper, and the fault tolerance of the network corresponds to the size of the fuzzy zone formed by the test samples. The fault tolerance of the network is improved by eliminating the fuzzy zone, and a hierarchical causal rule reasoning network with a structure of four layers is established for each element in the station. An accessible path for the candidate cause of the fault is divided into two stages as follows: from the candidate cause to the protection operation and from the protection operation to the circuit breaker trip, which are assigned different credibility contribution rates. At the same time, the artificial bee colony algorithm is combined to identify the operation mode of the substation using its reasoning and judgment capacity and make the necessary corrections to get some output results of the neural network. The results of the examples suggest that the proposed evaluation method has a small workload of calculation and can identify the faulty components with high reliability and their causes, quickly and accurately.

1. Introduction

Upon the occurrence of a fault, the alarm window on the substation automation system that monitors the man-machine interface will generate a large amount of alarm information [1], which may be mixed with erroneous alarm information. As a result, the substation operators have to spend a long time reading and analyzing the information. In order to shorten the alarm processing time, the substation alarm information processing system is needed as a necessary high-level application [2, 3]. An artificial bee colony algorithm for fault diagnosis and recovery processing of substations is put forward in this paper. The system has integrated two knowledge expression modes of rules and processes. The fault area is identified using the method of searching for passive connected areas to confirm the power outage area, and the fault point is determined accordingly [1, 4]. The fault diagnosis is carried out based on the method of taking the intersection of the protection scopes, and the protection with the fault is diagnosed and recorded in the inference process and taken as an explanation for the cause of the fault. All these diagnoses and determination are conducted based on the premise that the alarm information is entirely correct. However, in the practical conditions on site, causes such as poor contact and device failure can result in omission and false issuance of the alarm information, which can affect the accuracy of the diagnosis. Hence, the alarm information processing system is required to have certain fault tolerance, and the redundant information is used to ensure the correct diagnosis results [5, 6]. In the fault diagnosis of substations based on the artificial neural network (ANN) proposed in this paper, the inherent fault tolerance of ANN itself is used. However, no special fault tolerance study is performed. At the same time, the influence of the substation operation mode on the network structure is not taken into consideration. Therefore, it cannot be applied on site [7, 8]. The alarm processing issue is described as an unconstrained 0-1 integer programming problem, and the Tabu search algorithm is used to optimize the solution. However, when the Petri net is used for diagnosis, some changes cannot be triggered due to the operation failure of circuit breakers, which can affect the diagnosis effect to a certain extent. The genetic algorithm is used to optimize the solution. However, the calculation model is complicated, and the solution time is relatively long [9, 10]. If all the time sequence information is summarized to the dispatch center for centralized processing after the occurrence of a fault, the diagnosis model will be highly complicated and flooded with massive information, which is not conducive to timely and effective diagnosis and alarm processing. If a certain method can be adopted at the digital substation to deal with the alarms in the station in a timely manner before it is reported to the superior, it will play an essential role in ensuring the safe and stable operation of the system [11, 12].

Based on the existing studies, a substation alarm evaluation method based on the artificial bee colony algorithm optimized by the BP neural network algorithm is put forward in this paper. It is designed to identify the cause of the fault at the substation level from the alarm signal quickly. Based on the structure of the power grid, the substation, and the principles of protection, a causal rule network of four layers for each element is established. The fault source layer, the candidate cause layer, the protection operation layer, and the causality between operation layers of the circuit breaker are provided. For the possible candidate causes, the evaluation method of optimizing the BP neural network algorithm is used, which has a small workload of calculation and can obtain the faulty components and their causes with high reliability quickly and accurately upon the occurrence of a fault. The effectiveness of the proposed method is verified through the analysis of various faults in a practical substation.

2. Artificial Bee Colony Algorithm Model

In general, the alarm information of the substation includes protection operations, circuit breaker operations, self-inspections of the protection equipment (such as sampling faults, tripping failures, and protection calculation errors), communication failures, and so on. The protection operations and the circuit breaker operations herein are mainly used to establish the substation alarm evaluation based on the artificial bee colony algorithm, supplemented by the self-inspection of the protection equipment, communication failure, etc., to explain the cause of the alarm.

2.1. Modules of the Substation Alarm Information

The alarm information of the substation generally includes protection action, circuit breaker action, self-check of protection equipment (such as sampling fault, trip failure, and protection calculation error), communication fault, etc. Here, the artificial bee colony algorithm for substation alarm evaluation is constructed based on protection action and circuit breaker action, and the alarm reason is explained by self-check of protection equipment and communication fault.

The artificial bee colony algorithm model for the evaluation of substation alarm established in this paper involves the fault source F, the fault candidate cause C, the alarm event A, the inclusive relationship between the fault source and the fault candidate cause, the causal relationship between the fault candidate cause and the alarm event, the set of accessible paths for the candidate causes of the fault, the reliability of the failure source and the candidate cause, and other concepts.

The fault source F refers to the corresponding component F_i when the power grid fails, and it can specifically be divided into fault components such as the line L_i, the transformer T_i, and the busbar B_i.

The candidate cause for a fault refers to a set of possible candidate causes C ( = F_i) = {c₁, c₂, …, c_n} when a component Fi fails. The source of the fault can contain multiple candidate causes that are independent of each other (“OR” relationship). The initiation probability of each candidate cause is P_FC (F_i, c_j), which has certain differences.

An alarm event refers to an alarm generated by the operation of the each protection and circuit breaker due to a fault, where the protection outlet operation and circuit breaker operation x are taken into consideration.

2.2. Types of Causality Relationships in the Substation

There are two types of causality relationships in the substation, that is, the causality between the fault candidate cause C and the protection operation alarm a and the causality between the protection operation alarm a and the circuit breaker operation alarm x. One fault candidate cause c_i can trigger a number of alarm signs and form the alarm signs set S_CA ( = c_i) = {a₁, a₂, …, a_m} and causality pairs (c_i, a₁), …, (c_i, a_m), which occur at the same time in general (“AND” relationship). For each causal relationship pair, the trigger probability of the alarm a_j induced by the candidate cause c_i is P_CA (c_i, a_j). Due to the difference and priority of each candidate cause and to suppress false alarms to a certain extent, they are assigned different trigger probabilities. An alarm a_j may be triggered by a number of fault candidate causes. There is a causal relationship (a, x) between the protective outlet operation and the circuit breaker operation x that triggers the circuit breaker to trip, and its trigger probability is P_AX (a, x). A protection operation a_j can have one or more corresponding trips of circuit breakers, and the circuit breaker trip set is S_AX ( = a_j) = {x₁, x₂, …, x_l}.

An accessible path of a fault candidate cause refers to a complete path composed of two subpaths due to two stages of the candidate cause c_i triggering a protection operation a_j and the protection operation triggering the trip of a specific circuit breaker x_k, [(c_i, a_j), (a_j, x_k)] or [(c_i, a_j), (a_j, x_k), …, (a_j, x_l)].

The optimized BP neural network algorithm (CF) for the occurrence of the candidate cause of the fault refers to determining the actual degree of occurrence u_i (u_i [0, 1]) based on the alarm sign set of the candidate cause c_i. In a similar way, the credibility for the occurrence of a fault source refers to that the degree of fidelity is determined based on the occurrence of the candidate cause set of the fault source. The operation probabilities of the protection a_j and the circuit breaker x_k are P (a_j) and P (x_k), respectively (if the operation is performed, it is set at 0.95; otherwise, it is set at 0.05). The cause for the lack of an alarm is its failure to operate or the failure of equipment communication.

The evaluation of the substation fault handling based on the artificial bee colony algorithm has four layers as follows: fault source (component), candidate cause, protection operation, and circuit breaker operation. An example of the alarm evaluation of a specific line in a substation based on the artificial bee colony algorithm is shown in Figure 1.

Figure 1 

Example of the artificial bee colony algorithm for alarm evaluation of a specific line.

The marking of the protection operation a and the circuit breaker operation x stipulates that the main protection m, local backup protection s1, and remote backup protection s2 of the line (such as L₂₁₁) are marked as , , and , respectively. The main protection m of the transformer (such as T₁) and the backup protection Section 1 s1 and Section 2 s2 are marked as , , and , respectively. The main differential protection m of the busbar (such as B₁) is marked as , and the operation of the circuit breaker (such as 211) is marked as x₂₁₁.

There are six combinations of fault candidate causes of the line L₂₁₁ on the 10 kV side: c_{211_1} (the line L₂₁₁ failure, main protection operation, and the circuit breaker 211 normal operation), c_{211_2} (the line L₂₁₁ failure, main protection failure, local backup protection operation, and the circuit breaker 211 normal operation), c_{211_3} (the line L₂₁₁ failure, remote backup protection operation, and the circuit breaker 211 normal operation), c_{211_4} (the line L₂₁₁ failure, main protection operation, and the circuit breaker 211 failure to operate), c_{211_5} (the line L₂₁₁ failure, main protection failure to operate, local backup protection operation, and the circuit breaker 211 failure to operate), and c_{211_6} (the line L₂₁₁ failure, remote backup protection operation, and the circuit breaker 211 failure to operate). There are three accessible subpaths for the candidate cause c_{211_4}: main protection operation and the circuit breaker 211 failure to operate; No. 1 main transformer backup overcurrent Section 1 protection operation and busbar tie section breaker 231 operation; No. 1 main transformer backup overcurrent stage II protection operation and the circuit breaker 201 operation. The initiation probability P_CA of each protection is 0.99, 0.96, and 0.93, respectively.

3. BP Neural Network Algorithm

3.1. The Advantages of Optimized BP Neural Network Algorithm

Based on the existing research, this paper proposes a substation alarm evaluation method based on the artificial bee colony algorithm and optimized BP neural network algorithm. The evaluation method using optimized BP neural network algorithm has small amount of calculation, and it can quickly and accurately obtain the fault components and their causes with high reliability when the fault occurs. The effectiveness of the proposed method is verified by the analysis of several practical substation fault examples.

When the power grid fails, the evaluation system at the station level receives a real-time report from each protection device and collects all the relevant alarms within a period of time. Based on the established artificial bee colony algorithm evaluation and its rules, the candidate causes for the fault are evaluated and the corresponding fault sources and causes are obtained. The optimized BP neural network algorithm evaluation method for the substation alarm is described as follows.

Step 1. Based on the collected protection operations and circuit breaker operation sets, the corresponding suspected candidate cause set C′ and its corresponding suspected fault source set F′ are identified by tracing back according to the causal relationship.

Step 2. For each suspected fault source, the credibility of all candidate causes is calculated. The calculation method is described as follows:(1)For the failure candidate cause c_i, the credibility of all accessible paths is calculated. As mentioned earlier, an accessible path of the candidate causes is composed of two accessible subpaths of (c_i, a_j) and (a_j, x_k) in general (except for some individual cases). Based on the protection principle, a component failure will inevitably lead to a protection operation and the protection operation will trigger the trip of a circuit breaker. These two links are indispensable, and the link where the fault in the relay protection leads to the protection operation is more important. Hence, the two accessible subpaths are assigned different credibility contribution rates of P_C1 and P_C2 (0.54 and 0.46, respectively), instead of multiplying the credibility of each subpath. The credibility of the accessible path = P_C1P (a_j) P_CA (c_i, a_j) + P_C2P (x_k) P_AX (a_j, x_k).(2)For the failure candidate cause c_i, the credibility of all its accessible paths is integrated. If each accessible path is in an “AND” relationship (or a conditional “AND” relationship), the averaging method is used to obtain the original credibility of the candidate cause.(3)The original credibility of each candidate cause for the fault source is multiplied by the trigger probability P_FC (F_i, c_j) and is used as the credibility of the candidate cause.

Step 3. For a suspected fault source, the maximum value of the credibility of all candidate causes is selected. If it exceeds the preset evaluation threshold value (such as 0.55), it is considered that the component is faulty.

Step 4. Repeat Step 2 and Step 3 to process all suspected fault sources.

3.2. The Reason of Missing Alarm

The reliability of fault source occurrence is to determine the true degree of fault source occurrence according to the occurrence of candidate cause set of the fault source. The action probabilities of protection AJ and circuit breaker XK are p (AJ) and P (XK), respectively (0.95 for protection AJ and 0.05 for circuit breaker XK). The reason of missing alarm is that it refuses to operate or its equipment communication fails.

4. Experimental Analysis and Results

4.1. Construction of the Practical Substation Alarm Evaluation Model

In this paper, a practical substation is taken as an example to establish an alarm evaluation model. The main wiring structure of the substation is shown in Figure 2.

Figure 2 

Main wiring structure of a substation.

The artificial bee colony algorithms for the evaluation of the lines on the left of 10 kV, the busbar B21, and the transformer T1 are shown in Figure 3 to 5. The initiation probabilities of each candidate cause for the busbar B21 are 0.99, 0.97, and 0.95, respectively; the initiation probability of each candidate cause for the transformer T1 is 0.99, 0.97, 0.95, and 0.93, respectively.

Figure 3 

Alarm evaluation of each line on the left side of 10 kV based on the artificial bee colony algorithm (including fault scenario 1).

Figure 4 

Alarm evaluation of the busbar B21 based on the artificial bee colony algorithm (including fault scenario 2).

Figure 5 

Alarm evaluation of the transformer T₁ based on the artificial bee colony algorithm (including fault scenario 2).

In Figure 4, the candidate cause c_{B21_1} of the busbar B₂₁ suggests that the busbar B₂₁ is faulty, the busbar differential protection is active, and the circuit breakers are normal. The c_{B21_2} indicates that the busbar B₂₁ is faulty, the busbar differential protection fails, and the circuit breakers are normal, which are dependent on the 10 kV side backup protection operation of the transformer T₁ and the backup protection operation on the opposite side of lines 211∼214 (receiving end R). Hence, it is necessary to obtain the protection operation and circuit breaker operation information on the opposite side of each line through wide-area communication. The c_{B21_3} indicates that the busbar B₂₁ is faulty, the busbar differential protection operation is performed, the circuit breaker 201 fails to operate, and the circuit breaker 211 fails to operate (or the other circuit breakers with one or multiple feeders fail to operate), and its protection includes the busbar differential main protection, the 110 kV side backup protection of the transformer T₁, and the backup protection on the opposite side of the line 211. The 3rd protection corresponding to c_{B21_3} (the backup protection on the opposite side of the line 211 or the backup protection on the opposite side of the other lines 212 to 214) has certain conditions and restrictions to be involved in the evaluation: the circuit breaker 211 is detected to fail to operate; the backup protection operation and the circuit breaker operation information on the opposite side of the line can be obtained through the wide-area communication. Hence, the three protections corresponding to c_{B21_3} have a conditional “AND” relationship. Based on this conditional rule definition, the number of combinations of the circuit breaker 201 and circuit breaker failure to operate on each line can be reduced effectively, thereby reducing the number of candidate causes for the busbar B₂₁.

In Figure 5, the candidate cause c_{T1_3} of the transformer T₁ suggests that the transformer T1 fails, the main transformer differential protection operation is performed, the circuit breaker 201 fails to operate, and the circuit breaker 211 fails to operate (or the other circuit breakers with one or more feeders fail to operate). There is also a conditional “AND” relationship between the protection alarms. The case of c_{T1_4} is similar, which is not shown in the figure.

In the following section, the calculation process for the reliability of each of the six candidate causes when the line L_{21_1} fails is described. The accessible path of c_{211_1} is , and the accessible paths corresponding to the other candidate causes are shown in Figure 3.

u_{211_4} has 3 accessible subpaths, which are abbreviated as p₄₁, p₄₂, and p₄₃; then, the following can be obtained:

4.2. Single Fault Accompanied by Protection and Circuit Breaker Failing to Operate and Lack of Information

In Figure 2, the 10 kV side line L₂₁₁ has a failure (f₁) and the alarm is issued as follows: the communication failure of the backup Section 2 protection equipment on the 10 kV side of the No. 1 main transformer is detected; the line L211 main protection operation is received; No. 1 main transformer back-up over-current protection is operated; the breaker 231 trips; the breaker 201 trips. The alarm assignment values based on the artificial bee colony algorithm are shown in Figure 3 above. The credibility of each candidate cause for the line L₂₁₁ is calculated as follows:

The six candidate causes for the line L₂₁₁ are ranked according to their degrees of credibility. The credibility of the candidate cause c_{211_4} is the highest, which has exceeded the evaluation threshold value of 0.55.

Similarly, the credibility of each candidate cause for the line L₂₁₂ is calculated as follows: u_{212_1} = 0.0490, u_{212_2} =  0.0480, u_{212_3} = 0.0470, u_{212_4} = 0.4440, u_{212_5} = 0.4345, and u_{212_6} = 0.4249, none of which exceeds the evaluation threshold value. The other lines are handled based on the similar method.

Finally, it is determined that the line L₂₁₁ is faulty, the cause is c_{211_4}, and the result is correct. It can be observed that the line L₂₁₁ backup protection fails to operate, the circuit breaker x₂₁₁ fails to operate, the 10 kV side backup protection of the No. 1 main transformer operates correctly, and the circuit breaker 201 is tripped. However, it is missing due to communication failure, and a total of 3 bits of information error are found. Nevertheless, the correct judgment is still made.

The main protection operation of the line in the candidate cause c_{211_1} is performed. Although circuit breaker 211 fails to operate, its credibility is 0.5253. It still has a certain degree of credibility, which is consistent with the protection principle. If the method of multiplying the credibility of each section in the subpath is used, the credibility may be extremely low, which suggests that it is relatively appropriate to assign different credibility contribution rates to the two stages of the path.

If the line L₂₁₂ backup protection malfunctions and the circuit breaker x₂₁₂ does not operate at this point, then u_{212_5} = 0.7047 and the line L₂₁₂ fault will be misjudged. At this point, it is necessary to check whether the circuit breaker x₂₁₂ has a tripping failure. If it does not fail, a false alarm of is suspected. There are some problems with this evaluation when the protection is malfunctioning. When the reliability of the component is between 0.55 and 0.80, the evaluation situation needs to be analyzed more carefully and prudently.

4.3. Double Faults Accompanied by Two Circuit Breakers Failing to Operate

Transformer T₁ and busbar B₂₁ fail at the same time (refer to f₂ and f₃ in Figure 2), and circuit breakers 201 and 211 fail to operate. The alarm sequence is as follows: T₁ differential main protection operation is performed; the circuit breaker 111 trips; the circuit breaker 131 trips; the circuit breaker 301 trips; B₂₁ main protection operation is performed; the circuit breakers 212, 213, 214, and 231 trip in turn.

Firstly, the credibility of each candidate cause for the busbar B21 is calculated as follows:

For c_{B21_2}, if only the protection of is taken into consideration, then u_{B21_2} = 0.5253. If the backup protection and circuit breaker operation on the opposite side of the line L₂₁₁ can be obtained, that is, is taken into consideration, then u_{B21_2} = 0.7188.

If only the subpaths p_B31 and p_B32 are taken into consideration, then u_{B21_3} = (0.7937 + 0.9405)/2 × 0.95 = 0.8237.

If the backup protection and circuit breaker operation on the opposite side of the line L₂₁₁ can be obtained, that is, the subpath p_B33 is taken into consideration, then u_{B3_3} =  0.54 × 0.95 × 0.99 + 0.46 × 0.95 × 0.99 = 0.9405 and u_{B21_3} = (0.7937 + 0.9405 + 0.9405)/3 × 0.95 = 0.8470.

The candidate cause cB213 has the highest credibility. Hence, it is determined that the busbar B₂₁ is faulty, the cause is c_{B21_3}, and the result is correct.

The reliability of each candidate cause for the transformer T₁ is calculated as follows:c_{T1_2} has 3 subpaths, and their credibility is as follows: 0.54 ×  0.95 × 0.99 + 0.46 × (0.95 × 0.99 + 0.95 × 0.99)/2 = 0.9405; 0.54 × 0.95 × 0.99 + 0.46 × 0.05 × 0.99 = 0.5306; 0.54 × 0.95 ×  0.99 + 0.46 × 0.95 × 0.99 = 0.9405.

For c_{T1_3} and u_{T1_3} = 0.7962, if the wide-area information such as the backup protection operation on the opposite side of line 211 after the circuit breaker 201 fails to operate and the circuit breaker tripping operation is taken into consideration, then u_{T1_3} = 0.8710 and u_{T1_4} = 0.4389.

If there is no wide-area communication available, the credibility of the candidate cause c_{T1_1} is the highest; otherwise, the credibility of c_{T1_3} is the highest. Based on both the candidate causes, it can be determined that the transformer T₁ fails and that the circuit breaker 201 fails to operate. Hence, the result is correct.

From the evaluation result of example 2, it can be seen that if the conditions for wide-area communication are available, the wide-area information on the opposite side of the line can be used to enhance the credibility in determination of the transformer and busbar faults.

5. Discussion

The alarm information processing system of the substation has relatively good fault tolerance with one input error, which can totally meet the online requirements. The established diagnosis model is not subjected to be affected by the influence of changes in the operation mode. In addition, before the information enters the diagnosis model, the artificial bee colony algorithm is used to identify the operation mode of the substation. In this way, when there is any change in the operating mode, it is not required to carry out restructuring and training, which have enhanced the versatility of the model. The diagnosis network of the whole station is established based on the diagnosis model for the lines, busbars, and transformers. Therefore, on the one hand, the system scale control is less than 20 inputs and 20 outputs, which makes it easy to converge, and the training time is short; on the other hand, the corresponding models for various types of protection devices can be established and trained so that when the diagnosis network of various substations is constructed, these diagnosis models can be used directly without retraining, with relatively good versatility.

6. Conclusions

In this paper, a method of substation alarm evaluation based on the artificial bee colony algorithm is put forward. Based on the power grid and substation structure and protection principles, an alarm evaluation causal rules network with a structure of four layers is established with a clear physical meaning. The degrees of the occurrence of different candidate causes for the same fault are different in terms of fuzziness, and the two stages of an accessible path of the candidate cause are assigned various contribution degrees. The averaging method is used to obtain the fusion result of the credibility of each candidate cause for the same fault, which has a relatively small workload of calculation. For transformers and busbars, the addition of the wide-area communication is more conducive to alarm evaluation. The results of the numerical examples suggest that the proposed evaluation method has relatively high accuracy and certain fault tolerance. However, in the case of false alarms, the fault tolerance of the proposed method still needs to be further improved.

Data Availability

The data used to support the findings of this study are available from the author upon request.

Conflicts of Interest

The author declares no conflicts of interest.

Acknowledgments

This research study was sponsored by the National Natural Science Foundation of China under contract (61772398 and 61972239) and the Key Research and Development Program Projects of Shaanxi Province (2019SF-257). The authors thank these projects for supporting this article.