A Ship Power Grid Fault Diagnosis Method Based on DHNN Neural Network

Abstract

The present invention discloses a method for diagnosing faults in a ship's power grid based on a DHNN neural network. By classifying faults in the ship's power grid, encoding switches and protection relays, constructing a fault diagnosis system for the ship's power grid, and training the DHNN network, the DHNN network is finally used to quickly identify the fault classification to which the data input belongs when a fault occurs in the ship's power grid. This application combines the DHNN neural network, is applicable to diagnosing various faults in the ship's power grid, quickly analyzes and locates faults, ensures the efficiency and safety of fault finding, and improves the efficiency of maintenance personnel.

Description

Technical Field

[0001] The present invention relates to the field of ship power grids, and particularly to a ship power grid fault diagnosis method based on a DHNN neural network. Background Art

[0002] DHNN is the abbreviation of Discrete Hopfield Neural Networks. Compared with other neural networks, DHNN features strong dynamic performance and a simple network structure, and is a single-layer feedback network. This network abandons the hierarchical design of the original neuron network topology, boldly introduces the operating decay mechanism of the energy function, connects dynamics with the neuron network, and provides a basis for the stable operation of the neuron network. Simply put, the Hopfield neuron network is a neural network with associative memory based on the traditional neuron network. The transfer function selected by the discrete Hopfield neuron network is the threshold function, which is exactly suitable for the representation of relays and switches, that is, the two circuit states are represented by 0 and 1.

[0003] In the power system during the navigation of a ship, major power failures are not allowed. Once a dangerous situation occurs, the fault should be detected and handled in a timely manner. The prerequisite for quickly repairing the fault is to clarify the type, location, and cause of the fault. The process of observing the fault situation through the experience of the ship's crew and then analyzing the cause of the fault and repairing the fault takes a certain amount of time. Moreover, when the details of the fault are not clear, resource waste will also occur during the maintenance and replacement of equipment. To address this problem, this application proposes a solution. Summary of the Invention

[0004] Object of the Invention: The object of the present invention is to provide a ship power grid fault diagnosis method based on a DHNN neural network, which combines ship power grid fault diagnosis with artificial neurons to quickly clarify the faults in the ship power grid, thereby improving the maintenance efficiency.

[0005] Technical Solution: A ship power grid fault diagnosis method based on a DHNN neural network according to the present invention specifically includes the following steps:

[0006] Step 1: Classify the ship power grid faults into three categories: conventional faults, protection device refusal-to-operate faults, and combined faults;

[0007] Step 2: Encode the switches and relay protectors in the ship power grid, and respectively summarize the state characteristics of the respective relays and switches when the ship power grid has three types of faults.

[0008] Step 3: Build a fault diagnosis system for the ship's power grid relying on the DCS components of the ship intelligent integration platform. The fault diagnosis system mainly consists of three parts: the data acquisition unit RTU, the distributed computing unit CDU, and the human-machine graphical interaction interface HIU;

[0009] Step 4: Build a DHNN network. Specifically:

[0010] Step 4.1: Initialize the network parameters, including the number of neurons, activation function, weight matrix, and output state;

[0011] Step 4.2: Randomly select a neuron i in the network; Since the DHNN network is a single-layer neural network with feedback, the output of each neuron will be connected to non-self neurons through connection weights, that is, the output x i is multiplied by the connection weight and then acts on the j-th neuron, i≠j. After each neuron receives the input from other neurons, it is processed by the activation function and then output. Use θ i to represent its threshold function. The DHNN network selects the same activation function, and the expression is as follows:

[0012]

[0013] where f1, f2,... f n is the state activation function, sgn(x) is the sign function, and when the output is less than 0, f(x)=0;

[0014] Utilize the fact that the output of each neuron in the DHNN network is restricted by other neurons, and the nature of the mutual influence connection between neurons, so that the neural network has the ability to memorize and associate information;

[0015] Step 4.3: After summing the inputs of other neurons, process them through the activation function as the output of the neuron, while keeping the states of other neurons unchanged;

[0016] Step 4.4: Judge whether the network has reached the equilibrium state according to the stability condition. If the stability condition is reached, output the state of the network at this time; otherwise, update the state of the next neuron (i + 1) until the stable state is reached;

[0017] Step 5: Input the state characteristics of the respective relays and switches when three types of faults occur in the ship's power grid into the DHNN network for memorization;

[0018] Step 6: Input the test data of the fault to be judged into the DHNN network for diagnosis and judge the fault type.

[0019] Preferably, in step 3, the data acquisition unit RTU directly transmits signals to the 32-channel digital quantity acquisition module through the on / off actions of the relay, and the 32-channel digital quantity acquisition module transmits the signals to the central console and the central diagnosis unit in the distributed computing unit through the fieldbus RS485 / CAN.

[0020] Preferably, the central diagnosis unit transmits the power grid status information sent by the data acquisition unit RTU to the data processing software for processing. The data processing software analyzes the original data under the established algorithm DHNN original program, accepts the returned results, and diagnoses and troubleshoots faults in the original power grid according to the analysis results.

[0021] Preferably, the human-computer graphic interaction interface HIU in step 3 sets parameters for the entire fault diagnosis system and initializes and starts it, and sets the scenarios in subsequent operations to ensure the user-friendliness of the entire fault diagnosis system.

[0022] Preferably, the DHNN network constructed in step 4 adopts an asynchronous working mode.

[0023] Beneficial effects: This application combines the DHNN neural network, is applicable to diagnosing various ship power grid faults, quickly analyzes and locates, ensures the efficiency and safety of finding faults, and improves the efficiency of maintenance personnel. Description of the Drawings

[0024] Figure 1 is the ship power grid operation topology diagram in this application;

[0025] Figure 2 is the DHNN network model diagram in this application;

[0026] Figure 3 is the analysis diagram of the refusal-to-operate fault in this application;

[0027] Figure 4 is the ship fault diagnosis flow chart in this application. Specific Embodiments

[0028] The following further elaborates on this application in combination with specific embodiments.

[0029] Such as Figure 1As shown in the figure, the given ship power grid has four generators G1, G2, G3, G4, four transformers T1, T2, T3, T4, six load lines: Load Line 1, Load Line 2, Load Line 3, Load Line 4, Load Line 7, Load Line 8, two main lines Line 5 and Line 6, four main switches S1, S2, S3, S4, four tie switches S12, S13, S14, S15, two disconnect switches S11, S22, and 12 protective circuit breakers S5, S6, S7, S8, S9, S10, S16, S17, S18, S19, S20, S21. For each switch, there are three protection relays to control them and cut them off when a fault occurs. These three protection relays are instantaneous protection "m", short-time delay protection "n", and long-time delay protection "f". During the actual operation of the ship, not all equipment is connected to the network at the same time. The parts listed above represent all the equipment in the network.

[0030] Number the protection relays of all switches in the given power grid in sequence, and their real-time status is represented by [0, 1], that is, the switch is off as 1 and not operating as 0; the protection relay is energized as 1 and de-energized as 0. According to the three-stage current protection, since the four main power generation switches are already the top-level protection devices, the switches S1, S2, S3, S4 only have instantaneous protection (S1m, S2m, S3m, S4m) and no short-time delay and long-time delay protection. Similarly, S5, S6, S9, S10, S16, S19, S20, S21 have no long-time delay protection, S12, S14 are two-way tie switches and have no relay protection devices, and S11, S22 are disconnect switches and have no relay protection devices, as shown in Table 1:

[0031] Table 1 Protection Relay Coding

[0032]

[0033]

[0034]

[0035] As Figure 2 shown, construct the DHNN network of the ship power grid. Since the memory sample is 54-dimensional data, the network is composed of 54 neurons. The first 32 neurons are the input relay states, and the last 22 neurons are the input switch states.

[0036] Under normal fault conditions, that is, faults occurring when the protection device operates normally. Such as short circuits of load lines 1-4, short circuits of transformers T1-T4, short circuits of load lines 7-8, and overload tripping of generators G1-G4, there are up to 14 kinds of fault situations. Here, taking line fault 1 as an example for fault diagnosis and analysis, other faults can be analyzed in the same way.

[0037] When a short circuit fault occurs in load line 1, the branch circuit is: G1→S1→busbar→S5→load line 1. Also, since the short circuit fault of load line 1 is defined as a normal fault type, the state analysis of the relays and switches in the circuit at this time is as follows:

[0038] Protection relay S5m→1, S5n→0, S1m→0, switch S5 is disconnected;

[0039] Protection relay S5m→0, S5n→0, S1m→0, switch S1 is disconnected.

[0040] Record the states of all relays and switches in the power grid at this time, and the network memory sample data of the DHNN when the short circuit fault of load line 1 occurs can be obtained; as shown in Tables 2 and 3:

[0041] Table 2 Relay states of short circuit fault of load line 1

[0042]

[0043] Table 3 Switch states of short circuit fault of load line 1

[0044]

[0045] Sort and combine the data in Tables 2 and 3 (32 relays + 22 switches, a total of 54 states) to obtain the characteristic fault memory sample form T1:

[0046] T1 = [0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 01 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]

[0047] Fault states caused by the refusal of the protection device to operate. For the situation of the protection device refusing to operate, once a refusal-to-operate fault occurs in the ship's power grid, the states of the switches and relays in the power grid are relatively complex. Here, taking Figure 3 the refusal-to-operate fault of S7 as an example for analysis.

[0048] According to the three-stage current protection of the circuit, under normal circumstances, when S7m is energized, S7 will disconnect. However, due to a fault in S7, it fails to disconnect, that is, the state is 0, and the fault current continues to exist in the power grid and is processed by S6. Here, there are two cases: First, if the fault current reaches the instantaneous protection value of S6, the instantaneous protection of S6 will operate; Second, if the fault current does not reach its setting value, the short-time delay of S6 will be used for protection, that is, S6n controls S1 to disconnect. The analysis is as follows:

[0049] S7m → 1, SB7n → 0, S7f → 0, S1m → 0, S6 disconnects;

[0050] S6m → 1, S6n → 0, S7m → 1, S7n → 0, S7f → 0, S1m → 0, S6m → 0, S6n → 1, S1 disconnects;

[0051] Record the states of all relays and switches in the power grid at this time respectively, and the network memory sample data of DHNN during the short-circuit fault of load line 1 can be obtained; as shown in Tables 4 and 5:

[0052] Table 4 Relay states in case of S7 refusal to operate

[0053]

[0054] Table 5 Switch states in case of S7 refusal to operate

[0055]

[0056] Sort and combine the data (32 relays + 22 switches, a total of 54 states) to obtain the characteristic fault memory sample form T2;

[0057] T2 = [1 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]

[0058] The situation of combined fault states is relatively complex. If the conventional faults and the refusal-to-operate faults of protection devices are randomly combined, many kinds of fault situations will occur. Therefore, for the convenience of analysis, directly combine the short-circuit fault of load line 1 and the refusal-to-operate fault of S7. Under the combined fault state, the relay and switch states in the power grid are shown in Tables 6 and 7:

[0059] Table 6 Relay states in combined fault

[0060]

[0061] Table 7 Combined Fault Switch States

[0062]

[0063] The data is sorted and merged (relay 32 + switch 22, a total of 54 states) to obtain the characteristic fault memory sample form T3:

[0064] T3 = [1 0 0 0 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 01 0 0 0 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]

[0065] The three types of characteristic fault samples T1, T2, and T3 are sorted to obtain the memory samples of the network, as shown in Table 8:

[0066] Table 8 Data Table of Three Types of Characteristic Fault Memory Samples

[0067]

[0068] In this DHNN network, the Hebb learning rule is used to determine the weight matrix, that is, according to Δw ij = η·y i (k)·x i (k) to adjust the weights. Therefore, by adjusting the learning rate η, generally take w ij is the connection weight from neuron i to j;

[0069] So here start to adjust the weights to obtain the weight matrix table in Table 9:

[0070] Table 9 Weight Matrix Table with η = 1 / 54

[0071]

[0072] To ensure the convergence of the network when working in an asynchronous manner, the weight matrix should be a symmetric matrix. It can be seen from Table 9 that this network meets the stability conditions.

[0073] Randomly input a set of test data T4 of a conventional fault: T4 = [0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 00 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]

[0074] After passing through this network, it is attracted to T1, so it is diagnosed as a conventional fault.

[0075] The present invention has been described in detail with reference to the embodiments accompanied by drawings. Those of ordinary skill in the art can make various variations to the present invention based on the above description. Therefore, certain details in the embodiments should not constitute a limitation to the present invention, and the protection scope of the present invention will be defined by the scope defined in the appended claims.

Claims

1. A ship power grid fault diagnosis method based on DHNN neural network, characterized in that: Specifically, it includes the following steps: Step 1: Classify the ship power grid faults into three categories: conventional faults, protection device refusal-to-trip faults, and combined faults; Step 2: Encode the switches and relay protectors in the ship power grid, and respectively summarize the state characteristics of the respective relays and switches when the ship power grid has three types of faults; Step 3: Relying on the DCS component of the ship intelligent integration platform, construct a fault diagnosis system for the ship power grid. The fault diagnosis system consists of three parts: a data acquisition unit RTU, a distributed computing unit CDU, and a human-machine graphic interaction interface HIU; Step 4: Construct a DHNN network. Specifically: Step 4.1: Initialize the network parameters, including the number of neurons, activation function, weight matrix, and output state; Step 4.2: Randomly select a neuron i in the network; since the DHNN network is a single-layer neural network with feedback, the output of each neuron will be connected to non-self neurons through connection weights, that is, the output x i is multiplied by the connection weight and then acts on the j-th neuron, i≠j. After each neuron receives the input from other neurons, it is processed by the activation function and then outputs. Let θ i represent its threshold function. The DHNN network selects the same activation function, and the expression is as follows: where f1, f2,......f n is the state activation function, sgn(x) is the sign function, and f(x)=0 when the output is less than 0; Utilize the fact that the output of each neuron in the DHNN network is restricted by other neurons, and the nature of the mutual influence connection between each neuron, so that the neural network has the ability to memorize and associate information; Step 4.3: After summing the inputs of other neurons, process them through the activation function as the output of the neuron, while keeping the states of other neurons unchanged; Step 4.4: According to the stability condition, judge whether the network reaches the equilibrium state at this time. If the stability condition is reached, output the state of the network at this time, otherwise update the state of the next neuron (i + 1) until the stable state is reached; Step 5: Input the state characteristics of the respective relays and switches when the ship power grid has three types of faults into the DHNN network for memorization; Step 6: Input the test data of the fault to be judged into the DHNN network for diagnosis, and judge the fault type.

2. The ship power grid fault diagnosis method based on DHNN neural network according to claim 1, wherein: In Step 3, the data acquisition unit RTU directly transmits the signal to the 32-channel digital quantity acquisition module through the on / off action of the relay, and the 32-channel digital quantity acquisition module transmits the signal to the central console and the central diagnosis unit in the distributed computing unit through the field bus RS485 / CAN.

3. The method for diagnosing faults in a ship's power grid based on a DHNN neural network according to claim 2, characterized in that: The central diagnosis unit transmits the power grid state information sent by the data acquisition unit RTU to the data processing software for processing. The data processing software analyzes the original data under the established DHNN original program algorithm, accepts the returned result, and diagnoses and checks the faults against the original power grid according to the analysis result.

4. A ship power grid fault diagnosis method based on a DHNN neural network according to claim 1, characterized in that: In Step 3, the human-machine graphic interaction interface HIU sets parameters for the entire fault diagnosis system and initializes and starts it, and sets the scenarios in the subsequent operations to ensure the operation friendliness of the entire fault diagnosis system.

5. A ship power grid fault diagnosis method based on a DHNN neural network according to claim 1, characterized in that: The DHNN network constructed in Step 4 adopts an asynchronous working mode.

Images