A Fault Diagnosis and Early Warning Method for Key Equipment of Energy Storage Power Station Based on Data Mining

Abstract

The present invention discloses a fault diagnosis and early warning method for key equipment of an energy storage power station based on data mining. This method uses the K-nearest neighbor mutual information and Apriori association rule algorithms to realize the fault diagnosis and early warning of the key equipment of the energy storage power station. Compared with the traditional alarm strategy, this method can deeply mine the implicit association between the operating parameters and fault information of the key equipment of the energy storage power station, effectively shorten the fault diagnosis time, and the maintenance suggestions given are simple and efficient, providing refined guidance for operation and maintenance personnel. Compared with other intelligent algorithm diagnosis and early warning strategies, in view of the problems of a large variety of key equipment parameters, strong coupling, and huge historical data volume in the energy storage power station, this method combines engineering experience and uses the K-nearest neighbor mutual information algorithm to screen the main influencing factors of faults, which can reduce the workload and error of manual screening and improve the execution efficiency.

Description

Technical Field

[0001] The present invention relates to the field of energy storage device fault diagnosis, and in particular to a fault diagnosis and early warning method for key equipment of an energy storage power station based on data mining. Background Art

[0002] To better achieve the "dual carbon" goal and build a new power system with new energy as the main body, in the future, renewable energy such as photovoltaic and wind power will be mainly used to replace emissions reduction. However, new energy power generation has the characteristics of intermittency, randomness, and poor schedulability. The large-scale grid connection of new energy power generation poses a severe test to the power system. The existing flexible resources are gradually unable to support the grid to accept such a high proportion of fluctuating energy. As a better flexible resource, energy storage can effectively smooth the output of new energy and provide auxiliary services such as frequency modulation and peak shaving.

[0003] Most existing energy storage systems use lithium batteries. Due to the characteristics of lithium batteries themselves, improper use may cause safety problems in the equipment. Energy storage safety is usually caused by the interaction of multiple factors, resulting in battery abuse and thermal runaway, and ultimately leading to accidents. It can be divided into several aspects such as battery body faults, operating environment defects, and lack of comprehensive energy storage system management systems. In terms of the lack of comprehensive energy storage system management systems, system management problems such as incomplete or untimely information sharing between the battery management system (BMS), power conversion system (PCS), and energy management system (EMS), improper configuration and coordination of PCS and battery protection, battery anomalies after PCS fault clearance, and conflicts between measurement devices and management systems may prevent faults from being effectively controlled in a timely manner, ultimately evolving into accidents.

[0004] The key equipment of an energy storage power station includes the battery body, battery management system BMS, power conversion system PCS, etc. There are many types and large quantities of equipment, and the energy density of energy storage batteries is high, resulting in potential safety hazards and relatively low efficiency of the original operation and maintenance mode of the energy storage power station. Existing key equipment fault diagnosis and early warning methods can only provide abnormal state alarms, with a slow response speed, and can only reflect the current abnormal state information of the equipment, unable to analyze the root cause of the equipment faults reflected by the abnormal state, making it difficult to meet the refined operation and maintenance requirements of the power station site, and even leading to energy storage safety accidents. Summary of the Invention

[0005] The purpose of the present invention is to solve the problems of slow response speed of existing fault diagnosis methods for key equipment of energy storage power stations, inability to analyze the root cause of equipment faults reflected by abnormal states, and difficulty in meeting the refined operation and maintenance requirements of the power station site, and to provide a fault diagnosis and early warning method for key equipment of an energy storage power station based on data mining.

[0006] To solve the above technical problems, the technical solution adopted by the present invention is as follows: A method for fault diagnosis and early warning of key equipment in an energy storage power station based on data mining, comprising the following steps: S01. Establish a diagnosis and early warning rule library according to historical data, and establish a corresponding operation and maintenance advice library for various rules; S02. Collect and process abnormal signals within a period of time, and perform fuzzy matching with the rule library. If the matching is successful, a fault diagnosis and early warning message is sent, otherwise only an alarm is given for this abnormal state; S03. If the system sends a diagnosis and early warning message, corresponding operation and maintenance advice is given according to different fault types deduced in step S02.

[0007] Further, the process of establishing the diagnosis fault rule library and the operation and maintenance advice library in step S01 is as follows: S11. According to engineering experience, initially screen out the influencing parameters that may cause equipment failures from the original operation and maintenance parameters of the energy storage power station, and then select the corresponding influencing parameter data containing a complete cycle from the historical database and perform preprocessing; S12. Analyze the correlation between the initially screened influencing parameters and faults according to historical data, and determine the parameters with satisfied correlation as the main influencing factors; S13. Discretize the historical data of the main influencing factors according to the set threshold, and establish an equipment fault diagnosis and early warning rule library by using a data mining algorithm; S14. Formulate operation and maintenance advice according to the fault types corresponding to different rules in the rule library, and establish an operation and maintenance advice library.

[0008] Further, the K-nearest neighbor mutual information algorithm is used to analyze the correlation between a certain parameter and a fault. The specific method is as follows: First, set a correlation threshold θ, and then use the K-nearest neighbor mutual information algorithm to calculate the mutual information MI between a certain parameter and a fault. Compare the correlation threshold θ with the mutual information MI. If MI < θ, then discard this parameter, otherwise record this parameter.

[0009] Further, the Apriori association rule algorithm is used to determine a set of influencing parameters related to faults, so as to establish an equipment fault diagnosis and early warning rule library. The specific method is as follows: a. Set the minimum support and the minimum confidence; b. Scan the historical data of the discretized main influencing factors to find all items that meet the minimum support, which are called frequent 1-item sets; c. Based on the frequent 1-item sets, scan the historical data of the discretized main influencing factors to find frequent 2-item sets, and loop this process until there are no new frequent k + 1-item sets, where k is the number of types of main influencing factors; d. Calculate whether the confidence between item sets meets not less than the minimum confidence to determine strong associations.

[0010] Further, in the process of generating frequent k + 1-item sets from frequent k-item sets, a connection step and a pruning step are adopted to improve efficiency. The connection step is to find the set L of frequent k + 1-item sets k+1 , and connect the frequent k-item sets in L k to generate a candidate set C k+1, to ensure that the generated item sets are irrelevant, connection is only allowed when the first k - 1 items of two frequent k - item sets are the same after sorting; the pruning step is to find C k+1 the item sets that meet the minimum support. Directly scan the historical data of the main influencing factors after discretization to calculate the support of each item set. When C k+1 is greater than the set threshold, use prior knowledge to compress C k+1 .

[0011] Furthermore, step S02 is specifically as follows: S21. Collect the real - time data of the parameters screened in step S01. If an abnormal signal appears, record the signal data; S22. Start timing from the first data collected in S21, count the abnormal signals that appear within a period of time, and record the data of these abnormal signals; S23. Discretize the data recorded in steps S21 and S22 according to the set threshold, and perform fuzzy matching between the processed real - time parameter data and the diagnostic warning rule library; S24. If the matching is successful, send diagnostic warning information for different fault types and levels, otherwise only alarm the abnormal state through vibration.

[0012] Furthermore, step S03 includes two steps, specifically: S31. If the matching in step S02 is successful, while sending diagnostic warning information, accurately match the corresponding operation and maintenance suggestions in the operation and maintenance library according to different fault types; S32. Push the operation and maintenance suggestions matched in S31 through the screen.

[0013] Advantages of the present invention: The present invention uses the K - nearest neighbor mutual information and Apriori association rule algorithms to realize the fault diagnosis and warning of key equipment in energy storage power stations. Compared with traditional warning strategies, this method can deeply explore the implicit associations between the operating parameters and fault information of key equipment in energy storage power stations, effectively shorten the fault diagnosis time, and the given maintenance suggestions are simple and efficient, providing refined guidance for operation and maintenance personnel. Compared with other intelligent algorithm diagnosis and warning strategies, this method addresses the problems of a large number of parameter types, strong coupling, and huge historical data volume of key equipment in energy storage power stations. Combining engineering experience, it uses the K - nearest neighbor mutual information algorithm to screen the main influencing factors of faults, which can reduce the workload and error of manual screening and improve the execution efficiency. At the same time, the expert library established by this method supports functions such as rule expansion, modification, and deletion, and can provide support for new operation and maintenance requirements of energy storage power stations in the later stage.

[0014] Effect: By real - time monitoring the operating status of key equipment in energy storage power stations, when a fault is about to occur or has occurred, multiple warning or alarm signals collected within a certain time range are fuzzy - matched with the rule library, potential hazards that may exist in the system operation process are analyzed, the abnormal or fault types of the energy storage system are judged timely and accurately, abnormal working condition restrictions, fault protection, and audible and visual alarms are automatically implemented, and corresponding operation and maintenance suggestions are given. Description of the Drawings

[0015] Figure 1 This is the overall flowchart of the present invention;

[0016] Figure 2 This is the flowchart for screening the main influencing factors of faults using K-nearest neighbor mutual information;

[0017] Figure 3 This is the flowchart for data mining using the Apriori algorithm;

[0018] Figure 4 This is the diagram of partial faults and corresponding operation and maintenance suggestions;

[0019] Figure 5 This is the trigger mechanism of the rule base of the present invention and the specific trigger path diagram. Detailed Embodiment

[0020] The present invention will be further described below in conjunction with the drawings and specific embodiments.

[0021] Embodiment 1

[0022] When a power grid fault occurs, the EMS can receive the sequence records of the actions of power grid switches and relay protection devices sent by different RTUs, substation automation systems, and other systems, and arrange these events in chronological order and save them as historical information in the historical database. According to the different key equipment of the energy storage power station, this method can be mainly divided into two parts: PCS diagnosis and early warning and battery system diagnosis and early warning.

[0023] (1) PCS Diagnosis and Early Warning

[0024] Diagnosis and early warning are carried out according to the operating status and real-time data of the PCS, and operations such as shutting down the PCS for maintenance, tightening connectors, replacing faulty modules, and even withdrawing for overhaul are performed. The main fault types include the following:

[0025] a) AC voltage too high or too low

[0026] b) AC frequency too high or too low

[0027] c) DC voltage too high or too low

[0028] d) Inverter overload, overheat, short circuit

[0029] e) Radiator overheat

[0030] f) Inverter islanding

[0031] g) DSP fault

[0032] (2) Battery System Diagnosis and Early Warning

[0033] When the battery system is operating, if analog quantities such as the voltage, current, and temperature of the battery exceed the safety protection limit values, local fault isolation can be implemented to remove the problematic battery pack from operation and report the protection information simultaneously. The main fault types are as follows:

[0034] a) Battery overvoltage

[0035] b) Battery undervoltage

[0036] c) Battery overtemperature

[0037] d) Battery low temperature

[0038] e) Battery overcurrent

[0039] f) SOC out of limit

[0040] As Figure 1 shown, a method for fault diagnosis and early warning of key equipment in an energy storage power station based on data mining mainly includes the following specific implementation steps:

[0041] 1. Establishment and maintenance of "two libraries": Establish a diagnostic and early warning rule library based on historical data, and establish a corresponding operation and maintenance suggestion library for various rules. The rule library and operation and maintenance library can be supplemented, deleted, modified, etc. through background maintenance according to different application scenario requirements;

[0042] 2. Equipment fault diagnosis and early warning: Process abnormal signals within a period of time and perform fuzzy matching with the rule library. If the matching is successful, a fault diagnosis and early warning message is sent; otherwise, only an alarm is issued for this abnormal state;

[0043] 3. Push of operation and maintenance suggestions: If the system sends a diagnostic and early warning message, corresponding operation and maintenance suggestions are given according to the fault type deduced in step 2, as Figure 4 shown.

[0044] According to the specific implementation manner of the present invention, the process of establishing and maintaining the "two libraries" in step 1 includes four steps, which are specifically as follows:

[0045] (1) Based on engineering experience, initially screen out the influencing parameters that cause equipment faults in the energy storage power station from a large number of operation and maintenance parameters. Select data containing a complete cycle from the historical database and perform preprocessing, including the processing of abnormal values and missing values;

[0046] (2) Determine the correlation between the parameters initially screened out according to historical data analysis and the faults, and determine the parameters with satisfied correlation as the main influencing factors;

[0047] (3) Discretize the historical data of the above parameters according to the set threshold, and establish a rule library using a data mining algorithm. This rule library can be expanded as needed later;

[0048] (4) Develop operation and maintenance suggestions based on the fault types corresponding to different rules in the rule library, and establish an operation and maintenance suggestion library. This operation and maintenance library can be expanded as needed in the future.

[0049] The correlation analysis method in Step 2 is the K-nearest neighbor mutual information algorithm, and the data mining algorithm for establishing the rule library is the Apriori association rule algorithm.

[0050] Mutual information (MI) is a method used to measure the degree of mutual dependence between two random variables, which is rooted in the concept of entropy in information theory. The greater the mutual information between two random variables, the stronger the correlation. Selecting the main influencing factors of faults based on mutual information does not consider the impact of multi-variable coupling on equipment faults, and has disadvantages such as large high-dimensional computation and low accuracy. The K-nearest neighbor mutual information method can avoid directly calculating the probability density and has a good evaluation effect on complex non-linear relationships.

[0051] From the definition of mutual information, it can be seen that the smaller the mutual information MI( x , y ), the less the common information content of the sequences X and Y , and the smaller the sequence correlation; conversely, the greater the mutual information MI( x , y ), the more the common information content of the sequences X and Y , and the greater the sequence correlation. Therefore, the main influencing factors of the faults of the key equipment in the energy storage power station can be determined according to the size of MI.

[0052] As Figure 2 shown, the process of using the K-nearest neighbor mutual information method to screen the main influencing factors of faults is as follows: First, set the correlation threshold θ, then use the K-nearest neighbor mutual information algorithm to calculate the mutual information MI between a certain parameter and the fault, compare the correlation threshold θ with the mutual information MI. If MI < θ, then discard this parameter; otherwise, record this parameter. If calculating multiple main influencing factors, repeat the above process to calculate the correlation between other factors and the fault.

[0053] Based on the above description of the K-nearest neighbor mutual information algorithm, make further explanations for (1) and (2) in Step 1:

[0054] The key equipment of the energy storage power station includes the battery body, BMS, PCS, etc. The equipment parameters are of various types, have strong coupling, and the historical data volume is huge, making it difficult to determine the main influencing factors leading to failures. Therefore, in combination with engineering experience, this invention preliminarily screens out the influencing parameters that cause equipment failures from many operation and maintenance parameters such as current, voltage, frequency, temperature, and pressure. And data containing a complete cycle is selected from the historical database for preprocessing, including the processing of outliers and missing values. Then a correlation threshold is set, that is, if a parameter exceeds this threshold, it is considered that the parameter has a strong enough correlation with the failure and can be used as the main cause of the failure. Finally, according to the historical data, the K-nearest neighbor mutual information algorithm is used to analyze the correlation between the preliminarily screened parameters and equipment failures, and 3-5 parameters that exceed the correlation threshold are determined as the final parameters for establishing the rule base.

[0055] Association rules can find the internal association mechanism between two or more variables and predict events through the established association mechanism. When key equipment such as PCS fails, the EMS system log will record the fault-related information and store it in the historical database. Association rules can be applied to extract the dependence and correlation between fault types and fault phenomena by analyzing historical data, obtain the pattern characteristics of various faults, and guide the diagnosis and early warning process.

[0056] Generally, two key indicators, support and confidence, are used to measure the results of association analysis. Specifically, taking item set X and item set Y as an example, support S refers to the probability that item set X and Y occur simultaneously. Confidence C is refers to the probability that item set X occurs when item set Y occurs. Among them, an item set is a set of items. An item set containing k items is called an k item set. For example, the set {A, B, C} is a 3-item set.

[0057] Generally speaking, the minimum support s_min and the minimum confidence c_min are used to define the thresholds of support and confidence. An item set that satisfies the support not less than the minimum support is called a frequent item set, and a rule that satisfies both the support not less than the minimum support and the confidence not less than the minimum confidence is called a strong rule.

[0058] The purpose of mining association rules in this embodiment is to obtain the pattern characteristics of various faults, that is, a set of signals corresponding to the faults. As shown in Figure 3 , the specific process is as follows:

[0059] a. Set the minimum support and minimum confidence; b. Scan the historical data of the main influencing factors after discretization, and find all items that meet the minimum support, which are called frequent 1-itemsets; c. Based on the frequent 1-itemsets, scan the historical data of the main influencing factors after discretization to find frequent 2-itemsets, and loop this process until there are no new frequent k+1-itemsets, where k is the number of types of main influencing factors; d. Calculate whether the confidence between item sets meets the requirement of not less than the minimum confidence to determine strong associations.

[0060] Since the entire database needs to be scanned completely each time a new item set is searched, the execution efficiency is low. Therefore, in the process of generating frequent k+1-itemsets from frequent k-itemsets, a join step and a pruning step are adopted to improve efficiency. The join step is to find the set L of frequent k+1-itemsets k+1 , and join the frequent k-itemsets in L k to generate a candidate set C k+1 . To ensure that the generated item sets are not irrelevant, two frequent k-itemsets can be joined if and only if the first k-1 items after sorting are the same; the pruning step is to find the item sets in C k+1 that meet the minimum support. Directly scan the historical data of the main influencing factors after discretization to calculate the support of each item set. When C k+1 is greater than the set threshold, use prior knowledge to compress C k+1 .

[0061] Since the association rule algorithm can only process boolean data, while the values of parameters such as current, voltage, and temperature of the key equipment in the energy storage power station are all continuous, the historical data of the parameters selected in step (2) needs to be discretized according to the set threshold. Taking current as an example, when the current exceeds the set threshold, it is recorded as "1", otherwise it is recorded as "0". Then, using the discretized data, the Apriori association rule algorithm combines the join step and the pruning step with the confidence formula to establish a diagnostic and early warning rule base.

[0062] According to the specific implementation manner of the present invention, the above-mentioned equipment fault diagnosis and early warning process in step 2 includes 4 steps, which are specifically as follows:

[0063] (1) Collect the real-time data of the parameters selected in step 1. If an abnormal signal appears, record the signal data;

[0064] (2) Start timing from the first data collected in (1), count the abnormal signals that appear within a period of time, and record the data of these abnormal signals;

[0065] (3) Discretize the data recorded in (1) and (2) according to the set threshold, and perform fuzzy matching between the processed parameter real-time data and the diagnostic and early warning rule base;

[0066] (4)If the match is successful, different diagnostic warning messages such as sound, light, vibration, and pushed images are sent for different fault types and levels. Otherwise, only vibration is used to alarm the abnormal state. Figure 5 According to the specific embodiments of the present invention, the above-mentioned step 3 operation and maintenance suggestion pushing process includes 2 steps, which are specifically as follows:

[0067] According to the specific embodiments of the present invention, the above-mentioned step 3 operation and maintenance suggestion pushing process includes 2 steps, which are specifically as follows:

[0068] (1)If the match in step 2 is successful, while sending diagnostic warning messages, corresponding operation and maintenance suggestions are accurately matched in the operation and maintenance library according to different fault types;

[0069] (2)Push the operation and maintenance suggestions matched in (1) through the screen.

[0070] According to the specific embodiments of the present invention, taking the battery module failure as an example, the entire process of this method is further explained:

[0071] Combined with engineering experience, analyze the historical operation data of signals such as battery current and voltage, battery compartment temperature and pressure, and battery system fire overlimit collected by the K-nearest neighbor mutual information algorithm to obtain parameters with a relatively large MI between battery cell voltage, battery compartment temperature, battery system fire, etc. and battery module failure. Discretize the above parameters according to the set threshold, and the processed results are: battery cell overvoltage is recorded as "1", battery compartment overheating is recorded as "1", battery system fire overlimit alarm is recorded as "1", and battery module failure is recorded as "1". Determine the strong association rule of {battery cell overvoltage, high battery temperature, battery system fire alarm} → battery module failure according to the Apriori association rule algorithm.

[0072] When the energy storage power station is operating, when any one of {battery cell overvoltage, high battery temperature, battery system fire alarm} appears in the real-time collected parameters, the system starts timing and automatically performs fuzzy matching with the rule library. If two other abnormal signals appear within 30 seconds, the rule library is triggered under the combined alarm signal, and warning is given by means of sound, light, vibration, pushing images, etc., and information such as maintenance suggestions is pushed, which can be used as a reference for operation and maintenance and repair personnel.

[0073] It should be noted that the above content has elaborated in detail the principle and implementation method of the present invention. This content is only used to help understand the method and its core idea of the present invention. The protection scope of the present invention is not limited to the above embodiments. Those of ordinary skill in the art can make various other improvements and deformations that do not deviate from the essence of the present invention according to the disclosed content of the present invention, and these changes are still within the protection scope of the present invention.

Claims

1. A fault diagnosis and early warning method for key equipment of an energy storage power station based on data mining, characterized in that: It includes the following steps: S01. Establish a diagnostic warning rule base according to historical data, and establish a corresponding operation and maintenance suggestion base for various rules; S02. Collect and process abnormal signals within a period of time, and perform fuzzy matching with the rule base. If the matching is successful, a fault diagnosis warning message is sent; otherwise, only an alarm is issued for this abnormal state; S03. If the system sends a diagnostic warning message, corresponding operation and maintenance suggestions are given according to the different fault types deduced in step S02. The process of establishing the diagnostic fault rule base and the operation and maintenance suggestion base in step S01 is as follows: S11. According to engineering experience, initially screen out the influencing parameters that may cause equipment failures from the original operation and maintenance parameters of the energy storage power station, and then select the corresponding influencing parameter data containing a complete cycle from the historical database and perform preprocessing; S12. Determine the correlation between the initially screened influencing parameters and the faults according to historical data analysis, and determine the parameters with satisfied correlation requirements as the main influencing factors; S13. Discretize the historical data of the main influencing factors according to the set threshold, and use data mining algorithms to establish an equipment fault diagnosis warning rule base; S14. Formulate operation and maintenance suggestions according to the fault types corresponding to different rules in the rule base, and establish an operation and maintenance suggestion base. The K-nearest neighbor mutual information algorithm is used to analyze the correlation between a certain parameter and the fault. The specific method is as follows: First, set the correlation threshold θ, and then use the K-nearest neighbor mutual information algorithm to calculate the mutual information MI between a certain parameter and the fault. Compare the correlation threshold θ with the mutual information MI. If MI < θ, then discard this parameter; otherwise, record this parameter. The Apriori association rule algorithm is used to determine a set of influencing parameters related to the fault, so as to establish an equipment fault diagnosis warning rule base. The specific method is as follows: a. Set the minimum support and the minimum confidence; b. Scan the historical data of the discretized main influencing factors to find all items that meet the minimum support, which are called frequent 1-item sets; c. Based on the frequent 1-item sets, scan the historical data of the discretized main influencing factors to find frequent 2-item sets, and loop this process until there are no new frequent k+1-item sets, where k is the number of types of main influencing factors; d. Calculate whether the confidence between each frequent item set meets the requirement of not less than the minimum confidence to determine strong associations.

2. The method for fault diagnosis and early warning of key equipment in an energy storage power station based on data mining according to claim 1, wherein: In the process of generating frequent k+1-item sets from frequent k-item sets, a connection step and a pruning step are adopted to improve efficiency. The connection step is to find the set of frequent k+1-item sets Lk+1, and connect the frequent k-item sets in Lk to generate a candidate set Ck+1. To ensure that the generated item sets are not related, connection is only allowed when the first k-1 items of two frequent k-item sets are the same after sorting; the pruning step is to find the item sets in Ck+1 that meet the minimum support. Directly scan the historical data of the discretized main influencing factors to calculate the support of each item set. When Ck+1 is greater than the set threshold, use prior knowledge to compress Ck+1.

3. The method for fault diagnosis and early warning of key equipment in an energy storage power station based on data mining according to claim 1, wherein: Step S02 is specifically as follows: S21. Collect the real-time data of the parameters screened in step S01. If an abnormal signal appears, record the signal data; S22. Start timing from the first data collected in S21, count the abnormal signals that appear within a period of time, and record the data of these abnormal signals; S23. Discretize the data recorded in steps S21 and S22 according to the set threshold, and perform fuzzy matching between the processed parameter real-time data and the diagnostic warning rule library; S24. If the matching is successful, send diagnostic warning information for different fault types and levels, otherwise only alarm the abnormal state through vibration.

4. The method for fault diagnosis and early warning of key equipment in an energy storage power station based on data mining according to claim 1, wherein: Step S03 includes two steps, specifically: S31. If the matching in step S02 is successful, while sending the diagnostic warning information, accurately match the corresponding operation and maintenance suggestions in the operation and maintenance library according to different fault types; S32. Push the operation and maintenance suggestions matched in S31 through the screen.

Images