In-station master-slave differential protection method and system for photovoltaic power station

Abstract

The invention discloses an in-station master-slave differential protection method and system for a photovoltaic power station in the technical field of relay protection, and the method comprises the steps: dividing a power station line into a plurality of function partitions according to the manufacturer, the capacity and the topological structure of a component, further dividing the function subarea into a plurality of voltage subareas based on the ratio of the node real-time voltage to the rated voltage; the main differential protection module collects a current difference value at two ends of the line as an action basis; the slave differential protection module calculates the fault probability in a voltage partition based on a pre-trained random forest model in combination with the currently acquired characteristics of voltage, current, line length, equipment operation age limit and the like; and according to the relationship between the fault probability and a set threshold value, whether main differential protection action, delay action or locking is allowed is determined. According to the system, through a master-slave linkage mode, the problems of maloperation and refusal operation of traditional differential protection caused by current limiting characteristics in a photovoltaic power station are effectively solved, and the sensitivity and reliability of a protection device are remarkably improved.

Description

TECHNICAL FIELD

[0001] The application relates to an in-station master-slave differential protection method and system for a photovoltaic power station, and belongs to the technical field of relay protection. BACKGROUND

[0002] With the rapid development of photovoltaic power generation worldwide, the proportion of photovoltaic power stations in power systems continues to increase, and photovoltaic power stations have gradually become one of the important power types in power grids. Compared with traditional power generation methods such as thermal power and hydropower, photovoltaic power stations have the characteristics of controllable current output, distributed system structure, and high proportion of power electronic interfaces. However, due to their unique control characteristics and current limiting behavior of inverters, they exhibit significantly different short-circuit characteristics from traditional power generation systems in actual operation.

[0003] In terms of relay protection, the traditional differential protection method is widely used in scenarios such as power transmission lines and transformers, and its basic principle is to compare the difference between the currents at both ends of the line. When the differential current exceeds the set threshold, it is determined that there is an internal fault in the line and the protection action is triggered. However, in the photovoltaic power station scenario, due to its current limiting control characteristics, when a short-circuit fault occurs, the fault current amplitude is usually low and the rising speed is slow, which may not be enough to meet the action threshold of traditional differential protection, causing misoperation. At the same time, if the threshold value is lowered to improve sensitivity, it may also cause misoperation in non-fault situations such as system normal fluctuations and load switching.

[0004] In addition, existing differential protection schemes are mostly based on static threshold judgment and lack comprehensive recognition ability for the regional structure, electrical topology, and operating state of photovoltaic power stations. Especially in actual engineering applications, due to the large scale and complex structure of photovoltaic power stations, different component types, capacity levels, and communication topologies coexist, and the traditional differential protection scheme based on unified setting of the entire station cannot adapt to the differentiated protection needs of each region, and the protection strategy lacks pertinence and adaptive ability.

[0005] Fault current is limited, and traditional differential protection has the risk of misjudgment and refusal to operate;

[0006] Static setting threshold lacks adaptability to real-time system state;

[0007] There is a lack of intelligent auxiliary judgment mechanism based on regional division and data driving;

[0008] In a complex topology structure, the traditional protection scheme is difficult to quickly and accurately locate the fault region.

[0009] Therefore, it is urgent to propose a new master-slave differential protection method and system suitable for photovoltaic power station scenarios, with regional recognition ability and fault probability auxiliary judgment ability, to improve the accuracy, sensitivity, and reliability of protection action. SUMMARY

[0010] The present application aims at overcoming the deficiencies in the prior art, and the existing conventional differential protection has the risk of misjudgment and refusal to act, lacks self-adaptability to the real-time state of the system, and is difficult to quickly and accurately locate the fault area.

[0011] To solve the above technical problems, the present application is realized by adopting the following technical solutions:

[0012] In a first aspect, a master-slave differential protection method for a photovoltaic power station is provided, comprising the following steps:

[0013] According to the component information in the photovoltaic power station, the internal power lines of the power station are functionally divided;

[0014] According to the ratio of the real-time voltage and the rated voltage of the nodes in the functional division, the voltage division is divided;

[0015] The voltage division is numbered according to the ratio of the real-time voltage and the rated voltage, and the ratio of the real-time voltage and the rated voltage of the voltage division is inversely proportional to the size of the number;

[0016] Based on the pre-trained random forest model, the fault probability in each voltage division and each functional division is calculated combined with the current collected fault feature data;

[0017] The fault probability is compared with a preset threshold value to control the permission of the master differential protection action: the size of the number and the priority order are directly proportional;

[0018] Wherein, when determining the voltage division or the functional division, a first threshold value and a second threshold value are set, and the first threshold value is greater than the second threshold value;

[0019] When the measured fault probability is higher than the first threshold value, the slave differential action protection condition is established, and the master differential protection is allowed;

[0020] When the measured fault probability is within the range of the first threshold value and the second threshold value, the slave differential action protection condition is delayed, and the master differential protection executes the action after the delay; a threshold distance is set between the first threshold value and the second threshold value, and a 1s delay is added in each distance range;

[0021] When the measured fault probability is lower than the second threshold value, the slave differential action protection condition is not established, and the master differential protection action is locked;

[0022] When determining the functional division, a limit threshold value is set, and when the measured fault probability is higher than the limit threshold value, the first threshold value and the second threshold value of the voltage division are both lowered by 10%.

[0023] Further, the function partitioning of the internal power line of the power station comprises: function partitioning according to the type of photovoltaic module, the capacity of the device, the connection relationship of the power collection line in the communication topology and the transformer convergence wiring diagram.

[0024] Further, the random forest model comprises: constructing a training data set from historical operation data of the photovoltaic power station, the training data set comprising fault samples and non-fault samples in each function partition, and the feature parameters in each sample at least including: line length, equipment operation life, voltage effective value, negative sequence voltage ratio, zero sequence voltage mutation variable, total harmonic distortion rate and peak current ratio.

[0025] Further, the pre-processing of the feature parameters comprises:

[0026] the line length is subjected to Min-Max normalization processing;

[0027] the equipment operation life is discretized into multiple levels according to a set threshold;

[0028] the voltage effective value, the negative sequence voltage ratio, the zero sequence voltage mutation variable, the total harmonic distortion rate and the peak current ratio are all directly extracted from historical data;

[0029] all the feature parameters are standardized to the interval range of 0 to 1.

[0030] Further, a plurality of training subsets are generated from the training data set using the bootstrap sampling method, each training subset being used to train an unpruned decision tree, and the node splitting of the decision tree using Gini impurity as the division standard;

[0031] the decision trees finally trained are composed into the random forest model;

[0032] the random forest model is updated by the differential protection unit every set period.

[0033] Further, when the function partition is from the differential action condition, and the voltage partition is from the differential action protection condition delay or from the differential action protection condition, the main differential protection executes action after delay.

[0034] Further, the voltage partition is numbered according to the ratio of real-time voltage to rated voltage, comprising:

[0035] when the ratio of real-time voltage to rated voltage of the node is greater than 0.95, it is classified into the first partition;

[0036] when the ratio of real-time voltage to rated voltage of the node is between 0.85 and 0.95, it is classified into the second partition;

[0037] The third partition is formed when the ratio of the real-time voltage of the node to the rated voltage is less than 0.85.

[0038] Further, the number size is proportional to the priority order, which includes:

[0039] The fault probability calculation is sequentially performed on the voltage partitions according to the descending order of the numbers, and the fault probability calculation of subsequent numbers is stopped when the condition for establishing the differential setting is met.

[0040] In a second aspect, a master-slave differential protection system for a photovoltaic power station is provided, which includes:

[0041] The area division module is configured to divide the lines in the photovoltaic power station into functional partitions according to the equipment manufacturers, batches, capacities and topological structures in the photovoltaic power station, and divide the voltage partitions of the functional partitions based on the ratio of the real-time voltage of the node to the rated voltage.

[0042] The master differential protection module is configured to collect and compare the currents at both ends of the line, and determine whether a short-circuit fault exists by judging whether the differential current exceeds the setting action value.

[0043] The fault probability calculation module is configured to calculate the real-time fault probability of each voltage partition based on a random forest model trained based on historical fault data of the photovoltaic power station, and combine the electrical and non-electrical quantity features collected in real time.

[0044] The slave differential protection module is configured to authorize and control the action of the master differential protection module according to the result of the real-time fault probability.

[0045] Compared with the prior art, the present application has the following beneficial effects:

[0046] The present application introduces a master-slave differential protection cooperative mechanism to solve the problem of misoperation or refusal of operation of the traditional differential protection in the photovoltaic power station due to the current limiting characteristics. The internal lines of the power station are divided into functional partitions based on the component manufacturer identification, capacity level and topological connection relationship, which can realize spatial positioning with clear structure. Further, the voltage partitions are divided based on the ratio of the real-time voltage of each node to the rated voltage, so that the system can dynamically identify the local voltage variation characteristics under different operating conditions. On this basis, the slave differential protection unit uses the random forest model trained based on historical data to evaluate the fault probability by combining electrical and non-electrical quantities, which enhances the recognizability of regional faults and the responsiveness of the response. The master differential protection is only allowed to execute tripping or delay action when the fault probability of the corresponding region reaches a set threshold, which effectively reduces the difficulty of setting the differential setting value, improves the sensitivity and reliability of the action, significantly suppresses the occurrence of misoperation and refusal of operation, and improves the self-adaptive ability and safety and stability level of the relay protection system of the photovoltaic power station in the complex operating environment. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 A photovoltaic power station area division diagram provided by the embodiment one of the present application is shown.

[0048] Figure 2 A random forest flowchart provided by the embodiment one of the present application is shown.

[0049] Figure 3 A decision tree example diagram provided by the embodiment one of the present application is shown.

[0050] Figure 4 A fault probability generation flowchart provided by the embodiment one of the present application is shown.

[0051] Figure 5 A master-slave differential flowchart provided by the embodiment one of the present application is shown. DETAILED DESCRIPTION

[0052] The technical solutions of the present application will be described in detail below by means of the accompanying drawings and specific embodiments. It should be understood that the specific features in the embodiments and the specific features in the embodiments are detailed descriptions of the technical solutions of the present application, and are not limitations of the technical solutions of the present application. In the case of no conflict, the technical features in the embodiments and the specific features in the embodiments can be combined with each other.

[0053] The term "and / or" in this paper is only a description of the association relationship between the associated objects, which means that there can be three kinds of relationships, for example, A and / or B, which can represent the three cases of A alone, A and B together, and B alone. In addition, the character " / " in this paper generally represents that the associated objects before and after are a kind of "or" relationship.

[0054] Embodiment one:

[0055] The embodiment provides a kind of station in for photovoltaic power station's master-slave differential protection method, such as Figure 1One of the photovoltaic power stations, according to the component information in the photovoltaic power station, the internal power line of the power station is functionally divided; first, the same line is divided into n functionally divided areas according to the topology diagram. The definition of the same line is the set of the same capacity equipment produced by the same manufacturer in the same batch, such as a photovoltaic line is 10 photovoltaic panels produced by A manufacturer and an inverter produced by B manufacturer and a cable line produced by C manufacturer with almost the same length. Another line is also the same number, capacity of photovoltaic panels, inverters and the same length of cable line produced by the above three manufacturers in the same batch. Two lines can be regarded as the same line. If one of the two lines is different, the above two lines are not the same. In addition, only the same line directly connected on the same collector line on the topology diagram can be divided into the same functionally divided area. The abstract drawing is for simple representation. The number of photovoltaic panels in the drawing is not more than one. It is assumed that the line on each collector line is the same line, so it can be divided into n functionally divided areas as shown in the figure.

[0056] The voltage partition is numbered according to the ratio of real-time voltage to rated voltage. The ratio of real-time voltage to rated voltage of the voltage partition is inversely proportional to the size of the number.

[0057] Specifically, the division of the functional partition is fixed division, which can be input into the computer after being built and saved. Then, the computer divides the voltage partition in each functional partition according to the ratio of the real-time voltage and the rated voltage of the nodes in the functional partition. Specifically, the ratio of the real-time voltage and the rated voltage of each node is calculated, and the functional partition is divided into voltage partitions through the ratio. In normal operation, the equipment is mostly operated in the rated state, so the nodes with a ratio greater than 0.95 can be divided into the same voltage partition, such as voltage partition 1.1; the nodes with a ratio greater than 0.85 but not greater than 0.95 can be divided into voltage partition 1.2; the nodes with a ratio not greater than 0.85 can be divided into voltage partition 1.3, and so on. In the figure, only regions 1.a and n.b are drawn to represent them, where a and b only represent the number of voltage partitions. The nodes included in these regions will change in real time according to the change of the node voltage. The larger the voltage partition number, the lower the node voltage, and the higher the probability of a fault region, so the priority in fault identification is also higher. In fault probability calculation, the probability is calculated in the order of voltage partition number from large to small, and when the probability requirement set by differential protection is met, the main differential protection is allowed to act. At this time, if the action region selected by the main differential protection is exactly this range, it can act immediately without the need for additional calculation of the fault probability of other voltage partitions. Although a certain time is required for the above probability calculation, due to the current limiting characteristics of the photovoltaic power station, the current will not be very large when a short circuit fault occurs, and the fault can be removed through a certain time delay. Although the calculation required by the present application requires a certain time calculation by the computer, the calculation speed can be improved through the use of algorithms and the improvement of device performance, so it does not affect the need for normal operation.

[0058] As shown in Figure 2 and Figure 3 After the division of the region, the fault features are extracted from the historical fault data set of the photovoltaic power station as the training set for the random forest algorithm to calculate the fault probability. The algorithm training and use steps are as follows:

[0059] First, the required feature parameters are obtained according to the historical operation data of the photovoltaic power station, including line length, equipment operation life, voltage effective value, negative sequence voltage ratio, zero sequence voltage mutation, total harmonic distortion rate and peak current ratio. The above feature parameters include static features and dynamic features.

[0060] For static feature preprocessing: the equipment usage time is discretized into [0-5 years), [5-10 years) and ≥10 years according to industry standards, and assigned values of 0, 0.5 and 1 respectively; the line length is processed by Min-Max normalization, with the minimum value being 0 and the maximum value being the maximum length of a single line in the station, also between 0 and 1.

[0061] For dynamic electrical quantity measurement feature extraction: voltage features: three-phase voltage effective value, negative sequence voltage ratio, zero sequence voltage mutation; current features: total harmonic distortion (THD), peak current ratio, etc. The above feature data is directly extracted from historical data and processed as a number between 0 and 1. Depending on the actual situation, the extracted fault feature data can be reduced or increased. This embodiment is described for simplicity, and assumes that the above two non-electrical quantity features and five electrical quantity features are extracted.

[0062] As shown in Figure 4 and Figure 5 , the construction of the random forest model includes: Bootstrap aggregating (Bagging): assuming there are N=10 historical data set records, 5 fault data sets, and 5 normal data sets, each data set contains the required 7 feature data described above, and M times are drawn from the historical data set with replacement to form K=50 sub-training sets; Each subset retains about 63.2% of the original data, and the remaining 36.8% is used as an OOB validation sample. It should be noted that the data used for training in this embodiment is relatively small, but for the sake of description, the value is very small, which can be modified according to the actual situation.

[0063] Set the number of candidate features when splitting each tree ; The number of features is 7, that is, 2-3 features are randomly selected from the 7 features for node splitting to generate a decision tree. To improve accuracy, feature values can be obtained from electrical quantities and non-electrical quantities. Starting from the randomly selected features, according to the pre-defined division, as shown in Figure 3 , the division criteria Y and Z are random numbers between 0 and 1. Gini impurity is used as the splitting criterion, and the leaf node sample size is allowed to be <5. If the above requirements are not met after two splits, Y and Z are randomly generated again, or they can be manually input for adjustment until the conditions are met. For each sub-training set, 50 non-pruned decision trees are independently generated. In this embodiment, there are two different categories (fault, normal) in a node, and the sample ratio of each category is, and the Gini index calculation formula is: Eventually, one category of normal and one category of fault will appear.

[0064] It should be noted that the real-time collected voltage / current data is processed as a number between 0 and 1 after the same feature extraction process. The specific processing method is reasonable according to the site requirements. Static parameters are discretized according to the binning rules of the training set.

[0065] Output the classification result of a data set under each decision tree , 1 for fault, 0 for normal. And according to the result, the final fault probability value is obtained (average of 50 trees). A data set is input into the above method of 50 decision trees, each decision tree will naturally produce a result. Assuming that a data set finally produces 40 fault outputs and 10 normal outputs in the random forest decision tree, the probability of the corresponding state of the data set being in the fault state is 80%.

[0066] This embodiment provides a method of integrating 50 differentiated decision trees, which can capture the complex interaction of long-term factors such as equipment aging and instantaneous electrical characteristics, and can suppress the overfitting risk of a single tree through a probability averaging mechanism. In actual deployment, it is recommended to update 10% of the trees with new fault data every quarter to maintain the timeliness of the model.

[0067] Specifically, the above data expansion for a single node is promoted to a certain area, and only the feature data needs to be increased to obtain the fault probability of a certain voltage partition and functional partition under a certain state. When the fault probability in the voltage partition is greater than 80%, the main differential protection is allowed to directly act; when the fault probability in the voltage partition is between 60-80%, the main differential protection is allowed to act after a 1s delay, and the span setting is 20%, that is, every 20% increases the 1s delay; when the fault probability in the voltage partition is less than 20%, the main differential protection in the voltage partition is prohibited from acting. However, when the fault probability in the functional partition is higher than 90%, the fault probability requirement of the above voltage partition is reduced by 10%.

[0068] It needs to be mentioned that the differential protection detects and calculates the probability of the current fault in different areas of the entire photovoltaic power station according to various measurement data. The area division is divided into functional partitions and voltage partitions. When the current fault probability in an area is greater than a certain value, the main differential protection in this area is allowed to act. When the probability is within a certain range, the main differential protection in this area can act after a certain delay; when the probability is lower than a certain value, the main differential protection in this area is locked. The certain value compared with the probability for the functional partition can be slightly lower, and for the voltage partition can be slightly higher. If the functional partition meets the action condition of the slave differential protection, and the voltage partition does not meet the action condition, a certain delay can be allowed before the main differential protection is allowed to act. The main differential protection is still the same as the traditional differential protection, which detects the size of the differential current and the action current at all times. Through the combination of the main and slave differential protections, the main differential protection can reduce its action current setting value to some extent, thereby improving the sensitivity of the protection without increasing its misoperation.

[0069] Embodiment two:

[0070] This embodiment provides a station main-slave differential protection system for a photovoltaic power station, which comprises:

[0071] A region division module is configured to divide the lines in the photovoltaic power station into functional partitions according to the equipment manufacturers, batches, capacities and topological structures in the photovoltaic power station, and divide the voltage partitions of the functional partitions based on the ratio of the real-time voltage to the rated voltage of the nodes;

[0072] A main differential protection module is configured to collect and compare the currents at both ends of the line, determine whether the differential current exceeds the setting action value, and determine whether a short-circuit fault exists;

[0073] A fault probability calculation module is configured to calculate the real-time fault probability of each voltage partition based on a random forest model trained based on historical fault data of the photovoltaic power station and in combination with the real-time collected electrical quantity and non-electrical quantity characteristics;

[0074] A slave differential protection module is configured to authorize the action of the main differential protection module according to the result of the real-time fault probability.

[0075] The above only describes the preferred embodiments of the present application, and it should be noted that, for those skilled in the art, without departing from the technical principles of the present application, a number of improvements and modifications can be made, and these improvements and modifications should also be considered as the protection scope of the present application.

Claims

1. A method for in-station master-slave differential protection of photovoltaic power plants, characterized in that: Includes the following steps: Based on the component information in the photovoltaic power station, the internal power lines of the power station are divided into functional zones; Voltage zones are divided based on the ratio of real-time voltage to rated voltage of nodes within the functional zone. The voltage zones are numbered according to the ratio of real-time voltage to rated voltage; Based on the pre-trained random forest model, combined with the currently collected fault feature data, the fault probability in each voltage zone and each functional zone is calculated. The fault probability is compared with a preset threshold value to control the authority of the main differential protection action: specifically, the number size and priority order are directly proportional; When the measured fault probability is higher than the first threshold, the differential operation protection condition is met, and the main differential protection is allowed. When the measured fault probability is within the range of the first threshold value and the second threshold value, the differential operation protection condition is established after a delay, and the main differential protection performs the operation after the delay; the first threshold value is greater than the second threshold value, and a threshold range is set between the first threshold value and the second threshold value, with an additional 1 second delay in each range. When the measured fault probability is lower than the second threshold, the differential protection condition is not met, and the main differential protection is locked out.

2. The in-station master-slave differential protection method for photovoltaic power plants according to claim 1, characterized in that, The functional zoning of the power lines inside the power station includes dividing the functional zones according to the type of photovoltaic modules, the capacity of the devices, the connection relationship of the collector wires in the communication topology, and the substation busbar wiring diagram.

3. The in-station master-slave differential protection method for photovoltaic power plants according to claim 1, characterized in that, The random forest model includes: constructing a training dataset from historical operating data of photovoltaic power plants. The training dataset includes fault samples and non-fault samples in each functional partition, and the feature parameters in each sample include at least: line length, equipment operating years, effective voltage value, negative sequence voltage ratio, zero sequence voltage mutation, total harmonic distortion rate, and peak current ratio.

4. The in-station master-slave differential protection method for photovoltaic power plants according to claim 3, characterized in that, Preprocessing the feature parameters includes: The line length is normalized using Min-Max. The service life of the equipment is discretized into multiple levels according to a set threshold. The effective voltage value, negative sequence voltage ratio, zero sequence voltage abrupt change, total harmonic distortion rate, and peak current ratio were all directly extracted from historical data. Standardize all feature parameters to the range of 0 to 1.

5. The in-station master-slave differential protection method for photovoltaic power plants according to claim 3, characterized in that, Multiple training subsets are generated from the training dataset using a bootstrap sampling method. Each training subset is used to train an unpruned decision tree, and the node splitting of the decision tree uses Gini impurity as the splitting criterion. The final trained decision trees are used to form the random forest model. The differential protection unit updates a portion of the decision trees in the random forest model at set intervals.

6. The in-station master-slave differential protection method for photovoltaic power plants according to claim 1, characterized in that, When the differential protection condition for the functional zone is met, but the differential protection condition for the voltage zone is met after a delay or is not met, the main differential protection will perform the action after the delay.

7. The in-station master-slave differential protection method for photovoltaic power plants according to claim 1, characterized in that, The voltage zones are numbered according to the ratio of real-time voltage to rated voltage, including: the ratio of real-time voltage to rated voltage of the voltage zone is inversely proportional to the number. A node is classified as the first partition when the ratio of its real-time voltage to its rated voltage is greater than 0.

95. A node is classified as belonging to the second partition when the ratio of its real-time voltage to its rated voltage is between 0.85 and 0.

95. A node is classified as the third partition when the ratio of its real-time voltage to its rated voltage is less than 0.

85.

8. The in-station master-slave differential protection method for photovoltaic power plants according to claim 1, characterized in that, The proportionality between the number size and priority order includes: The fault probability calculation is performed on the voltage zones in descending order of their numbers, and the calculation stops when the condition for differential setting is met.

9. The in-station master-slave differential protection method for photovoltaic power plants according to claim 1, characterized in that, When measuring the functional zones, a limit threshold is set. When the measured fault probability is higher than the limit threshold, the first and second threshold values ​​of the voltage zone are both reduced by 10%.

10. A master-slave differential protection system for a photovoltaic power station, characterized in that, The method for in-station master-slave differential protection of photovoltaic power plants as described in any one of claims 1-9 includes: The area division module is used to divide the lines within the photovoltaic power station into functional zones based on the equipment manufacturers, batches, capacity, and topology, and to divide the functional zones into voltage zones based on the ratio of the real-time voltage to the rated voltage of the nodes. The main differential protection module is used to collect and compare the current at both ends of the line to determine whether the differential current exceeds the set action value, so as to determine whether there is a short circuit fault. The fault probability calculation module is used to calculate the real-time fault probability of each voltage zone based on the random forest model trained from the historical fault data of the photovoltaic power station, combined with the real-time collected electrical and non-electrical quantity characteristics. The differential protection module is used to authorize and control the operation of the main differential protection module based on the real-time fault probability results.

Images