A reliability-centered optimization method for the start-stop control logic of main transformer coolers

Abstract

A reliability-centric method for optimizing the start-stop control logic of main transformer coolers is proposed. First, sensor information is hardwired to the PLC, which transmits it to the power plant's MS via optical fiber. The MS and MSS are also connected via optical fiber. The MSS then uses the collected information to establish a cooling performance evaluation model for each main transformer cooler, quantifying its operational reliability. It monitors each cooler's operating condition in real time and ranks each cooler's reliability. The ranking results are sent to the MS via optical fiber, which then transmits them to the PLC. Finally, the PLC controls the start-stop relays of the corresponding coolers based on the MSS ranking results, combined with the main transformer winding and oil temperatures, and thus controls the start and stop status of the different coolers. By integrating the on-site monitoring system, the monitoring system server, and the cooler programmable logic controller, a lean and efficient cooler system is achieved, ensuring the safe and reliable operation of the hydropower plant's main transformers.

Description

Technical Field

[0001] The present invention belongs to the field of main transformer cooler start-stop control methods, and in particular relates to a main transformer cooler start-stop control logic optimization method centered on reliability. Background Art

[0002] The number of cooler starts and stops, outlet water temperature, outlet oil temperature, and operating time of a main transformer cooler all affect its reliability. Under existing cooler control logic optimization, the operating time of different coolers will inevitably vary, resulting in different cooling performance and, consequently, varying operational reliability. However, traditional main transformer cooler logic control methods focus solely on the cooler operation level, ignoring the coupling relationship between each cooler's operating conditions and its reliability, making it difficult to achieve the optimal cooler combination. Therefore, effectively managing the start-stop control logic of main transformer coolers has become a challenging task facing power industries worldwide and is also crucial for improving the operational reliability of main transformer cooler control systems. In other words, accurately estimating the cooling performance of each cooler is essential for reliability-focused main transformer cooler start-stop control logic optimization methods.

[0003] In summary, the shortcomings of the existing methods are:

[0004] 1. In the existing method, the PLC program divides the coolers into main, auxiliary, and backup coolers. The main transformer coolers use a regular rotation start-stop control logic optimization method, with each group's cycle lasting seven days to balance the operating time of each cooler. Assuming there are six coolers, in the first week, coolers 1, 2, and 3 are the main coolers, while coolers 4 and 5 are auxiliary coolers, and the sixth is the backup cooler (denoted as Rotation 1). In the second week, coolers 2, 3, and 4 are the main coolers, while coolers 5 and 6 are auxiliary coolers, and the first is the backup cooler (denoted as Rotation 2), and so on until Rotation 6. A disadvantage of this regular rotation start-stop logic optimization method is that in situations such as unit maintenance, 400V AC power failure, or main transformer system outages, the cooler rotation prior to the outage is no longer remembered. When the cooler system restarts, it will restart from Rotation 1, causing coolers 1, 2, and 3 to have higher operating times than the other coolers, increasing the probability of failure for these three coolers.

[0005] 2. Existing methods assume that each cooler operates under the same conditions and fail to consider the impact of parameters such as the main transformer cooler's outlet water temperature, outlet oil temperature, and operating time on its cooling performance (i.e., operational reliability). Each cooler's operating conditions vary, leading to varying degrees of mechanical wear and tear, and consequently, varying reliability. Existing methods fail to account for these varying operating conditions, which increases the risk of cooler failure. Summary of the Invention

[0006] The method problem to be solved by the present invention is to provide a reliability-centered main transformer cooler start-stop control logic optimization method, which can not only quantify the operating reliability of each cooler and have a high prediction accuracy in the main transformer cooler reliability index prediction, but also provide a solution to the problems in the existing technology that cooler control does not consider reliability and control time imbalance.

[0007] In order to solve the above method problems, the method scheme adopted by the present invention is:

[0008] A reliability-centered method for optimizing the start-stop control logic of a main transformer cooler includes the following steps:

[0009] Step 1: Sort the cooler operation reliability indicators:

[0010] Step 1.1. Use the monitoring system server to collect the outlet oil temperature, outlet water temperature, outlet oil pressure, outlet water pressure, operating time, main transformer oil temperature, and main transformer winding temperature information of each cooler;

[0011] Step 1.2, using the cooler reliability assessment model to calculate the reliability index of each cooler;

[0012] Step 1.3: Use the prediction model to obtain the operational reliability index of the cooler in the future time period t, and assign it to the main cooler, auxiliary cooler, and backup cooler in descending order of reliability;

[0013] Step 2: Use PLC to receive and execute instructions:

[0014] Step 2.1: Use the information collected by the digital and analog input modules of the PLC, combined with the main transformer oil temperature and winding temperature, to determine the number of coolers to be opened;

[0015] Step 2.2: The PLC sends an open signal to the relays of the coolers to be turned on in order of reliability of the coolers at intervals of T1 through the digital output module. Then the normally open nodes of the PLC are turned on in turn, and the secondary circuits of the main supply coolers are turned on in turn.

[0016] Step 3. Establish a dynamic update strategy for coolers: If only the first group of coolers is started, the monitoring system server will re-evaluate the reliability of all coolers at each rotation cycle, calculate the average reliability index, and re-sort them. The main cooler, auxiliary cooler, and backup cooler will be reselected according to the reliability level, and the coolers will be controlled with a balanced strategy.

[0017] Preferably, the process of establishing the reliability assessment model is as follows:

[0018] Step 1.2.1. Classify the cooler degradation characteristic state from mild to severe into four levels: I, II, III, and IV, with corresponding expert deduction values ​​of 2, 4, 8, and 10, respectively. To limit the error caused by human factors, the deduction value of the state characteristic quantity is accurately quantified using the Lagrangian interpolation method. The n+1 characteristic points of the deduction value are defined as (x0, m(x0)), …, (xn, m(xn)). The n Lagrangian interpolation results are:

[0019] (1);

[0020] Where, The first i measurement points; Indicates the i The deduction value corresponding to each measurement point; Indicates the number of feature points; y Indicates the i +1 point deduction corresponding to the measurement point; Indicates the i The Lagrangian basis function of the measurement points is calculated as follows:

[0021] (2);

[0022] Where, The first j measurement points;

[0023] Step 1.2.2: Perform entropy-based HI estimate;

[0024] Different characteristic state quantities of main transformer cooler HI Expressed as follows:

[0025] (3);

[0026] Where, Indicates the i The reliability value of each state feature; Representation characteristics i Impact factor; Representation characteristics i The deduced value of 0 indicates the worst operating condition, and 1 indicates the best operating condition. The cooler degradation characteristic state is divided into four levels from light to heavy: I, II, III, and IV. Take 1, 2, 3, and 4 respectively;

[0027] Therefore, the reliability index of the cooler HI is the weighted value of all state characteristic factors, as shown in formula (4);

[0028] (4);

[0029] Where, HI Indicates the reliability of the main transformer cooler. Indicates the i The weight of each characteristic factor; M Indicates the number of cooler state characteristics, M The cooler status characteristic quantities include the main transformer oil temperature, the main transformer winding temperature, the cooler outlet water temperature, the outlet oil temperature, the outlet water pressure, and the outlet oil pressure.

[0030] Preferably, the entropy weight method is used to calculate the weights of the characteristic quantities of each state of the cooler:

[0031] Step 1.2.2.1. Input: Number of main transformer coolers N, total number of characteristic quantities of main transformer coolers M、

[0032] Constructed by formula (5) HI Judgment matrix:

[0033] (5);

[0034] Output: The weight of the reliability index HI of the status characteristics of each main transformer cooler;

[0035] Step 1.2.2.2, Normalization: Calculate the ratio of each column matrix value to the sum of the corresponding column eigenvalues , and obtain the normalized decision matrix:

[0036] (6);

[0037] (7);

[0038] Step 1.2.2.3. Calculate the entropy of the characteristic quantities of different states of the main transformer cooler:

[0039] (8);

[0040] Step 1.2.2.4. Calculate the coefficient representing the degree of diversification:

[0041] (9);

[0042] Step 1.2.2.5: Calculate the characteristic value of each cooler HI Weight:

[0043] (10);

[0044] Step 1.2.2.6: Output the characteristic value of each cooler HI Weight.

[0045] Preferably, the prediction model construction method in step 1.3 is:

[0046] Step 1.3.1, Forget Gate: The forget gate is used to determine how much previous information will be ignored;

[0047] (11);

[0048] Where, Indicates that LSTM t The forget gate output at that moment; Represents the weight matrix of the forget gate; Indicates that LSTM t Output at time -1; Indicates that LSTM t Input at the moment; Represents the bias vector of the forget gate; Represents the sigmoid function;

[0049] Step 1.3.2, Input Gate: The input gate is used to control how much of the current input information will be stored in the memory;

[0050] (12);

[0051] (13);

[0052] (14);

[0053] Where, Indicates that LSTM t Input gate output at time; Represents the weight matrix of the input gate; represents the bias vector of the input gate; Indicates that LSTM t Candidate cell states at time instant; Indicates that LSTM t The cell state at a moment in time; The weight matrix representing the candidate cell state; A bias vector representing the candidate cell state; represents the hyperbolic tangent function; Indicates that LSTM t -1 cell state at time instant; represents point-wise multiplication;

[0054] Step 1.3.3, output gate: The output gate is used to determine the time step t What to output when

[0055] (15);

[0056] (16);

[0057] Where, Indicates that LSTM t Output gate output at time; Represents the weight matrix of the output gate; represents the bias vector of the output gate;

[0058] Based on three control gates and storage units, the LSTM neural network reads, resets, and updates the acquired long-term information.

[0059] Preferably, the harmony search algorithm is used to optimize the LSTM network parameters to improve the prediction accuracy of the main transformer cooler reliability index:

[0060] Step 1.3.4: Randomly generate within the value range h Group LSTM parameters to form a harmony library, denoted as H :

[0061] (17);

[0062] Where, Represent the optimal number of LSTM neurons and learning rate respectively; For the h The absolute error between the group parameter value and the actual result; h is the size of the harmony library;

[0063] Step 1.3.5: Generate new harmony , as shown in formula (18); assuming represents the LSTM parameters in the new harmony, then have Probabilistic selection of harmony libraries H The values ​​in are 1- P H Probabilistic Selection Harmony Library H A randomly generated value outside the range of values; if H If , the pitch adjustment method of the harmony library is as shown in formula (19);

[0064] (18);

[0065] (19);

[0066] Where, Adjust bandwidth for tone; Represents a uniformly distributed random number between [0,1]; represents the pitch adjustment probability;

[0067] Step 1.3.6, harmony library update: according to the harmony in step 1.3.5, the new error is obtained. ,

[0068] like , then update the harmony library,

[0069] Will And the corresponding new harmony update to H middle;

[0070] Step 1.3.7: Determine whether the maximum number of iterations has been reached. If so, stop the iteration and obtain the optimal number of LSTM neurons and learning rate based on the current global optimal value. Otherwise, set k = k +1, go to step 1.3.5 and step 1.3.6 to continue iterating;

[0071] Step 1.3.8: Use the trained HS-LSTM model to predict the reliability of the main transformer cooler for the next seven days and calculate the average reliability over these days.

[0072] The root mean square error is used to evaluate the quality of the prediction results, as shown in formula (20);

[0073] (20);

[0074] Where, n Indicates the n A prediction moment, N Indicates the total number of prediction moments; Respectively represent the predicted n The reliability index of the main transformer cooler and the actual n 1. Reliability index of main transformer cooler; RMSE represents the root mean square error 。

[0075] The present invention can achieve the following beneficial effects:

[0076] 1. This invention proposes a reliability-centric method for optimizing the start-stop control logic of main transformer coolers. By integrating an on-site monitoring system (MS), a monitoring system server (MSS), and a cooler programmable logic controller (PLC), this method achieves a lean and efficient cooler system, ensuring the safe and reliable operation of a hydropower plant's main transformers. First, sensor information, such as the main transformer's oil and winding temperatures, and the cooler's outlet oil pressure, outlet water pressure, outlet oil temperature, and outlet water temperature, is hardwired to the PLC. The PLC then transmits this information to the power plant's MS via optical fiber, which also connects the MS and MSS. The MSS then uses this collected information to build a cooling performance evaluation model for each main transformer cooler, quantifying its operational reliability. It then monitors each cooler's operating conditions in real time and ranks each cooler's reliability. The ranking results are then sent to the MS via optical fiber, which then transmits them to the PLC. Finally, the PLC, based on the MSS ranking results and the main transformer's winding and oil temperatures, controls the corresponding cooler's start-stop relays, thereby controlling the start and stop status of each cooler. Through the actual case analysis of a power plant cooler, the results show that the reliability-centered main transformer cooler start-stop control logic optimization method invented by the present invention can take into account the operating efficiency and reliability of the cooler, analyze and decide on the operating conditions of each main transformer cooler, and determine the global optimal cooler combination.

[0077] 2. Compared with other logic optimization methods, this method can not only quantify the operating reliability of each cooler and has a higher prediction accuracy in the prediction of the reliability index of the main transformer cooler, but also provides a solution to the problems of cooler control not considering reliability and uneven control time in the existing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] The present invention will be further described below with reference to the accompanying drawings and examples:

[0079] Figure 1 This is a diagram of a characteristic variable processing method of the present invention;

[0080] Figure 2 This is the LSTM network structure diagram of the present invention;

[0081] Figure 3 The following is a graph of the prediction results collected by different prediction methods. DETAILED DESCRIPTION

[0082] The present invention provides a reliability-centered method for optimizing the start-stop control logic of a main transformer cooler. The specific scheme is as follows:

[0083] Basic introduction of main transformer cooler system:

[0084] 1) Each transformer is equipped with 6 coolers, which are named SA1, SA2, SA3, SA4, SA5 and SA6 respectively;

[0085] 2) Each cooler is equipped with an on-site monitoring system (MS), a monitoring system server (MSS), a programmable logic controller (PLC), a main transformer cooler power cabinet, a main transformer cooler control cabinet, etc.

[0086] 3) Transformer insulation oil temperature, winding temperature, cooler outlet water temperature, outlet oil temperature, outlet oil pressure, outlet water pressure and other data that can reflect the cooler's cooling performance are transmitted to the PLC through the analog input module. Digital quantities such as switch status are transmitted to the PLC through the digital input module. The PLC also records the total operating time and start and stop times of each cooler.

[0087] 4) The local monitoring system MS is connected to the PLC of the cooler control cabinet through optical fiber to collect analog signals, digital signals, total operating time and start and stop times in the PLC;

[0088] 5) The on-site monitoring system MS is connected to the monitoring system server MSS through optical fiber, and the collected analog signals, digital signals, total operating time and start and stop times are transmitted to the MSS.

[0089] 6) MSS evaluates the cooling performance of each main transformer cooler based on the collected signals and uses a hybrid learning algorithm based on the harmony search algorithm-long short-term memory network (HS-LSTM) to predict the reliability of the main transformer cooler for the next seven days. The prediction results are transmitted to the monitoring system MS via optical fiber. The MS then transmits this information to the PLC in the main transformer cooler control cabinet. The PLC combines the main transformer winding temperature and the main transformer oil temperature and controls the cooler start-stop relay through the switch output module, thereby controlling the start and stop status of the corresponding cooler. Whether the cooler needs to be restarted due to unit startup, power failure, or recovery after maintenance, the reliability-centered main transformer cooler start-stop control logic optimization method invented by the present invention first starts the cooler with higher reliability, which can more reasonably manage the operating life of the cooler.

[0090] The method steps are:

[0091] Step 1: Sort the cooler operation reliability indicators:

[0092] Step 1.1. Use the monitoring system server to collect the outlet oil temperature, outlet water temperature, outlet oil pressure, outlet water pressure, operating time, main transformer oil temperature, and main transformer winding temperature information of each cooler;

[0093] Step 1.2, using the cooler reliability assessment model to calculate the reliability index of each cooler;

[0094] Step 1.3: Use the prediction model to obtain the operational reliability index of the cooler in the future time period t, and assign it to the main cooler, auxiliary cooler, and backup cooler in descending order of reliability;

[0095] Step 2: Use PLC to receive and execute instructions:

[0096] Step 2.1: Use the information collected by the digital and analog input modules of the PLC, combined with the main transformer oil temperature and winding temperature, to determine the number of coolers to be opened;

[0097] Step 2.2: The PLC sends an open signal to the relays of the coolers to be turned on in order of reliability of the coolers at intervals of T1 through the digital output module. Then the normally open nodes of the PLC are turned on in turn, and the secondary circuits of the main supply coolers are turned on in turn.

[0098] Step 3. Establish a dynamic update strategy for coolers: If only the first group of coolers is started, the monitoring system server will re-evaluate the reliability of all coolers at each rotation cycle, calculate the average reliability index, and re-sort them. The main cooler, auxiliary cooler, and backup cooler will be reselected according to the reliability level, and the coolers will be controlled with a balanced strategy.

[0099] Preferably, the process of establishing the reliability assessment model is as follows:

[0100] In general, the operating status of the main transformer cooler can be divided into normal state, low efficiency state and fault state. At present, the power industry roughly determines the current operating status of the main transformer cooler based on relevant industry standards and guidelines: the cooler degradation characteristic state is divided into four levels from light to heavy, namely I, II, III, and IV, and the corresponding expert deduction values ​​are 2, 4, 8, and 10 respectively. However, in actual applications, this method cannot quantitatively reflect the cooling performance of the main transformer cooler, and thus cannot quantitatively reflect its operating reliability, and thus cannot accurately evaluate the current operating status. In order to solve this problem, the present invention invents a new indicator to characterize the operating status of the main transformer cooler, namely the reliability HI .

[0101] In actual engineering, the cooling system mainly refers to the heat dissipation system. The selected state quantities are: cooling system motor operation, cooling device control system, cooler oil leakage, water pump and oil pump operation status, operation time, main transformer oil temperature, main transformer winding temperature, cooler outlet water temperature, outlet oil temperature, outlet water pressure, outlet oil pressure. These indicators are used to characterize the reliability of the main transformer cooler. HI . Reliability of main transformer cooler HI Evaluation can be divided into four major strategies: visual assessment, condition monitoring data, daily troubleshooting, and preventive maintenance. During daily inspections, maintenance personnel visually inspect component deformation, external damage, surface contamination, and the operating status of the water and oil pumps for intuitive judgment. Operations personnel use condition monitoring data (main transformer oil temperature, main transformer winding temperature, outlet water temperature, outlet oil temperature, outlet water pressure, and outlet oil pressure) for trend analysis, which can more directly reflect the cooling performance of the main transformer cooler and, in turn, operational reliability. Grassroots staff record the daily operation of each cooler by recording defect handling, cooler oil leakage, and operating time. They also conduct daily inspections of the cooling system motor and cooling device control system through daily maintenance to comprehensively judge the operating status of each cooler.

[0102] Therefore, the practical experience of operation and maintenance personnel promotes the reliability HI The development of a system to accurately track chiller operating efficiency and cooling performance using available multi-source monitoring data, as opposed to a crude classification of chiller operating status based on industry guidelines. HI By collecting different state characteristics that affect the operation of the cooler and reflecting its operating conditions, a refined operational reliability assessment can be performed.

[0103] In practical applications, due to the uncertainty of the evaluation of the cooler working conditions and the lack of effective data utilization, direct measurement of the cooler HI Therefore, the present invention combines several latest mathematical models with expert experience and knowledge to invent a mathematical model based on HI Estimation model. According to relevant industry standards, the operating conditions of each feature can be roughly divided into different categories. In the enterprise status assessment guidelines, the cooler degradation characteristic status is also divided into four levels from light to heavy: I, II, III, and IV, and the corresponding expert deduction values ​​are 2, 4, 8, and 10, respectively. In order to limit the errors caused by human factors and accurately quantify the deductive value of the state characteristic quantity, the present invention adopts the Lagrange interpolation technology, which is one of the representatives of a class of polynomial interpolation technologies. The specific steps are as follows:

[0104] Step 1.2.1. Classify the cooler degradation characteristic state from mild to severe into four levels: I, II, III, and IV, with corresponding expert deduction values ​​of 2, 4, 8, and 10, respectively. To limit the error caused by human factors, the deduction value of the state characteristic quantity is accurately quantified using the Lagrangian interpolation method. The n+1 characteristic points of the deduction value are defined as (x0, m(x0)), …, (xn, m(xn)). The n Lagrangian interpolation results are:

[0105] (1);

[0106] Where, The first i measurement points; Indicates the i The deduction value corresponding to each measurement point; Indicates the number of feature points; y Indicates the i +1 point deduction corresponding to the measurement point; Indicates the i The Lagrangian basis function of the measurement points is calculated as follows:

[0107] (2);

[0108] Where, The first j measurement points;

[0109] like Figure 1 As shown, in order to limit the error caused by human factors and accurately quantify the deductive value of the state characteristic quantity, the present invention adopts the Lagrange interpolation technology, which is one of the representatives of a class of polynomial interpolation technologies. Figure 1 As can be seen in the figure, expert deductions are calculated for different temperature conditions, with key characteristic points represented by gray dots. The expert judgment of field operators provides additional characteristic points (represented by light blue dots) and quantified deductions. Using these characteristic points as input to the Lagrangian interpolation technique, the remaining characteristic deductions can be accurately calculated for different temperatures.

[0110] Step 1.2.2: All the deduction values ​​for the selected state characteristics describing the reliability of the main transformer cooler can be accurately calculated in the above subsection. By subtracting all the characteristic deduction values ​​related to the above characteristic components, the reliability index proposed in the present invention can be obtained. HI . Carry out entropy weight-based HI estimate;

[0111] Different characteristic state quantities of main transformer cooler HI Expressed as follows:

[0112] (3);

[0113] Where, Indicates the i The reliability value of each state feature; Representation characteristics i Impact factor; Representation characteristics i The deduced value of 0 indicates the worst operating condition, and 1 indicates the best operating condition. The cooler degradation characteristic state is divided into four levels from light to heavy: I, II, III, and IV. Take 1, 2, 3, and 4 respectively;

[0114] Therefore, the reliability index of the cooler HI is the weighted value of all state characteristic factors, as shown in formula (4);

[0115] (4);

[0116] Where, HI Indicates the reliability of the main transformer cooler. Indicates the i The weight of each characteristic factor; M Indicates the number of cooler state characteristics, M The cooler status characteristic quantities include the main transformer oil temperature, the main transformer winding temperature, the cooler outlet water temperature, the outlet oil temperature, the outlet water pressure, and the outlet oil pressure.

[0117] Preferably, the entropy weight method is used to calculate the weights of the characteristic quantities of each state of the cooler:

[0118] Step 1.2.2.1. Input: Number of main transformer coolers N, total number of characteristic quantities of main transformer coolers M、

[0119] Constructed by formula (5) HI Judgment matrix:

[0120] (5);

[0121] Output: The weight of the reliability index HI of the status characteristics of each main transformer cooler;

[0122] Step 1.2.2.2, Normalization: Calculate the ratio of each column matrix value to the sum of the corresponding column eigenvalues , and obtain the normalized decision matrix:

[0123] (6);

[0124] (7);

[0125] Step 1.2.2.3. Calculate the entropy of the characteristic quantities of different states of the main transformer cooler:

[0126] (8);

[0127] Step 1.2.2.4. Calculate the coefficient representing the degree of diversification:

[0128] (9);

[0129] Step 1.2.2.5: Calculate the characteristic value of each cooler HI Weight:

[0130] (10);

[0131] Step 1.2.2.6: Output the characteristic value of each cooler HI Weight.

[0132] After the weights of the different state characteristics of each cooler are obtained by formula (3)-(10), the cooler's HI Value, due to the present invention HI The evaluation method combines several widely accepted models and empirical expert knowledge to estimate the reliability of the cooler more reasonably and reliably.

[0133] The above method mainly introduces the past status of the cooler based on field data and expert experience. HI Estimation is an advanced method based on multi-source data fusion and driven by both knowledge and data. In actual projects, power companies are more interested in current or future operational reliability indicators in order to optimally plan the start-stop combination of coolers.

[0134] Since the cooler state characteristic data are all time-dependent, the present invention is based on HS-LSTM HI Indicator prediction method.

[0135] Preferably, a hybrid learning algorithm based on HS-LSTM is used to predict the operating reliability index of the main transformer cooler. HI , the introduction of LSTM network avoids the serious gradient vanishing and explosion problems existing in traditional neural networks. The structure of LSTM network is as follows Figure 2 As shown in the figure, the LSTM network has three gating units. The LSTM network can update and forget information through them.

[0136] The prediction model construction method in step 1.3 is:

[0137] Step 1.3.1, Forget Gate: The forget gate is used to determine how much previous information will be ignored;

[0138] (11);

[0139] Where, Indicates that LSTM t The forget gate output at that moment; Represents the weight matrix of the forget gate; Indicates that LSTM t Output at time -1; Indicates that LSTM t Input at the moment; Represents the bias vector of the forget gate; Represents the sigmoid function;

[0140] Step 1.3.2, Input Gate: The input gate is used to control how much of the current input information will be stored in the memory;

[0141] (12);

[0142] (13);

[0143] (14);

[0144] Where, Indicates that LSTM t Input gate output at time; Represents the weight matrix of the input gate; represents the bias vector of the input gate; Indicates that LSTM t Candidate cell states at time instant; Indicates that LSTM t The cell state at a moment in time; The weight matrix representing the candidate cell state; A bias vector representing the candidate cell state; represents the hyperbolic tangent function; Indicates that LSTM t -1 cell state at time instant; represents point-wise multiplication;

[0145] Step 1.3.3, output gate: The output gate is used to determine the time step t What to output when

[0146] (15);

[0147] (16);

[0148] Where, Indicates that LSTM t Output gate output at time; Represents the weight matrix of the output gate; represents the bias vector of the output gate;

[0149] Based on three control gates and storage units, the LSTM neural network reads, resets, and updates the acquired long-term information.

[0150] Preferably, the harmony search algorithm is used to optimize the LSTM network parameters to improve the prediction accuracy of the main transformer cooler reliability index:

[0151] Step 1.3.4: Randomly generate within the value range h Group LSTM parameters to form a harmony library, denoted as H :

[0152] (17);

[0153] Where, Represent the optimal number of LSTM neurons and learning rate respectively; For the h The absolute error between the group parameter value and the actual result; h is the size of the harmony library;

[0154] Step 1.3.5: Generate new harmony , as shown in formula (18); assuming represents the LSTM parameters in the new harmony, then have Probabilistic selection of harmony libraries H The values ​​in are 1- P H Probabilistic Selection Harmony Library H A randomly generated value outside the range of values; if H If , the pitch adjustment method of the harmony library is as shown in formula (19);

[0155] (18);

[0156] (19);

[0157] Where, Adjust bandwidth for tone; Represents a uniformly distributed random number between [0,1]; represents the pitch adjustment probability;

[0158] Step 1.3.6, harmony library update: according to the harmony in step 1.3.5, the new error is obtained. ,

[0159] like , then update the harmony library,

[0160] Will And the corresponding new harmony update to H middle;

[0161] Step 1.3.7: Determine whether the maximum number of iterations has been reached. If so, stop the iteration and obtain the optimal number of LSTM neurons and learning rate based on the current global optimal value. Otherwise, set k = k +1, go to step 1.3.5 and step 1.3.6 to continue iterating;

[0162] Step 1.3.8: Use the trained HS-LSTM model to predict the reliability of the main transformer cooler for the next seven days and calculate the average reliability over these days.

[0163] The root mean square error is used to evaluate the quality of the prediction results, as shown in formula (20);

[0164] (20);

[0165] Where, n Indicates the n A prediction moment, N Indicates the total number of prediction moments; Respectively represent the predicted n The reliability index of the main transformer cooler and the actual n 1. Reliability index of main transformer cooler; RMSE represents the root mean square error 。

[0166] Example 1:

[0167] A reliability-centered method for optimizing the start-stop control logic of a main transformer cooler includes the following main steps:

[0168] Step 1: Ranking of cooler operation reliability indicators

[0169] The monitoring system server (MSS) uses previously collected information, including each cooler's outlet oil temperature, outlet water temperature, outlet oil pressure, outlet water pressure, operating time, main transformer oil temperature, and main transformer winding temperature, to calculate each cooler's reliability index using the cooler reliability model described in Section 2.2.1. The predictive model described in Section 2.2.2 then calculates the cooler's reliability index for the next week and assigns it to the main cooler, auxiliary cooler, and backup cooler in descending order of reliability. For example, if the MSS calculates a reliability ranking of SA2, SA6, SA1, SA5, SA3, and SA4, then SA2, SA6, and SA1 will serve as the main cooler, SA5 and SA3 as auxiliary coolers, and SA4 as the backup cooler. Furthermore, if only the main cooler is operating at that time, the reliability indexes of SA5, SA3, and SA4 will be continuously updated based on the real-time operating status, enabling dynamic updates of the auxiliary and backup coolers. It is important that the above cooler reliability ranking needs to be confirmed by the operating personnel before it can be sent to the local monitoring system MS, which then transmits it to the PLC.

[0170] Step 2: PLC receives and executes instructions

[0171] After receiving instructions from the on-site monitoring system (MS), the programmable logic controller (PLC) first determines the number of coolers to be activated (assuming only the main supply cooler is activated) through its internal program, combining information collected by the digital and analog input modules with the main transformer oil and winding temperatures. The PLC then sends activation signals to the relays of the coolers to be activated, sequentially at 30-second intervals, through the digital output modules, in order of reliability. This causes the PLC's normally open nodes to conduct sequentially, energizing the secondary circuits of the main supply coolers.

[0172] Step 3: Cooler dynamic update strategy

[0173] If only the first group of coolers is started, the monitoring system server MSS will re-evaluate the reliability of all coolers every time a rotation cycle is reached, calculate the average reliability index, and re-sort them. It will reselect the main cooler, auxiliary cooler, and backup cooler according to the reliability level and control the coolers with a balanced strategy.

[0174] It should be noted that in order to prevent all coolers from stopping, the present invention proposes an iterative start-stop strategy: assuming that initially SA2, SA6, and SA1 are the main coolers, SA5 and SA3 are auxiliary coolers, and SA4 is the backup cooler, and only the first group of coolers is started. After reordering, SA3, SA5, and SA2 are the main coolers, SA4 and SA6 are auxiliary coolers, and SA1 is the backup cooler. In this case, we stop SA1 first, and start SA3 after SA1 stops. After SA3 starts, we stop SA6, and after SA6 stops, we start SA5. In other words, the previous round of coolers is stopped in order of cooler operation reliability from low to high, and the new round of coolers is started in order of cooler operation reliability from high to low.

[0175] In actual operation, steps 1 to 3 are continuously iterated to dynamically determine the global optimal cooler combination.

[0176] Use the model to verify:

[0177] Taking the outlet oil temperature, water temperature, oil pressure, water pressure and other data collected by the sensor of a main transformer cooler as an example, the present invention illustrates the effectiveness of a reliability-centered main transformer cooler start-stop control logic optimization method and a main transformer cooler reliability index prediction method based on a hybrid learning method through case analysis.

[0178] According to the main transformer cooler operation reliability evaluation model of the present invention, the cooler's past operation reliability index is calculated, and the cooler reliability evaluation index for the next 7 days is predicted by the hybrid learning method based on HS-LSTM. Figure 3 In addition, to facilitate quantitative analysis and intuitively demonstrate the accuracy of various methods, Table 1 shows the root mean square error values ​​corresponding to the HS-LSTM method proposed in this invention and the ARIMA and RNN methods.

[0179] Table 1 Comparison of RMSE of different prediction methods

[0180]

[0181] Table 1 shows that the proposed HS-LSTM method achieves RMSE values ​​that are 80.69% and 72.65% lower than those of the ARIMA and RNN prediction methods, respectively. This indicates that the proposed method achieves superior performance in predicting the reliability indicators of the main transformer cooler. Therefore, it can be concluded that the proposed hybrid ensemble learning method can accurately predict the reliability indicators of the main transformer cooler.

[0182] like Figure 3As shown in the figure, in the first rotation cycle, the reliability evaluation results are ranked from high to low as follows: SA5, SA2, SA6, SA3, SA1, SA4. Then SA5, SA2, SA6 will serve as the main coolers, SA3 and SA1 will serve as auxiliary coolers, and SA4 will serve as the backup cooler. On the last day of the first cycle, only the main cooler will be turned on. Figure 2 As shown in the figure, after seven days of reordering all the coolers, the reliability evaluation results are ranked from highest to lowest: SA5, SA2, SA3, SA6, SA4, SA1. SA5, SA2, and SA3 will serve as the primary coolers, SA6 and SA4 will serve as auxiliary coolers, and SA1 will serve as the backup cooler. To prevent a complete shutdown of the coolers, SA6 will be shut down first, and then SA3 will be turned on after SA6 stops.

[0183] The above embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the present invention. The scope of protection of the present invention shall be the methods and solutions set forth in the claims, including equivalent alternatives to the method features set forth in the claims. Equivalent alternatives and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A reliability-centered main transformer cooler start-stop control logic optimization method, characterized in that The following steps are involved: Step 1: Sort the cooler operation reliability indicators: Step 1.

1. Use the monitoring system server to collect the outlet oil temperature, outlet water temperature, outlet oil pressure, outlet water pressure, operating time, main transformer oil temperature, and main transformer winding temperature information of each cooler; Step 1.2, using the cooler reliability assessment model to calculate the reliability index of each cooler; Step 1.3: Use the prediction model to obtain the operational reliability index of the cooler in the future time period t, and assign it to the main cooler, auxiliary cooler, and backup cooler in descending order of reliability; Step 2: Use PLC to receive and execute instructions: Step 2.1: Use the information collected by the digital and analog input modules of the PLC, combined with the main transformer oil temperature and winding temperature, to determine the number of coolers to be opened; Step 2.2: The PLC sends an open signal to the relays of the coolers to be turned on in order of reliability of the coolers at intervals of T1 through the digital output module. Then the normally open nodes of the PLC are turned on in turn, and the secondary circuits of the main supply coolers are turned on in turn. Step 3. Establish a dynamic update strategy for coolers: If only the first group of coolers is started, the monitoring system server will re-evaluate the reliability of all coolers at each rotation cycle, calculate the average reliability index, and re-sort them. The main cooler, auxiliary cooler, and backup cooler will be reselected according to the reliability level, and the coolers will be controlled with a balanced strategy.

2. The reliability-centered main transformer cooler start-stop control logic optimization method according to claim 1 is characterized in that: The process of establishing the reliability assessment model is as follows: Step 1.2.

1. Classify the cooler degradation characteristic state from mild to severe into four levels: I, II, III, and IV, with corresponding expert deduction values ​​of 2, 4, 8, and 10, respectively. To limit the error caused by human factors, the deduction value of the state characteristic quantity is accurately quantified using the Lagrangian interpolation method. The n+1 characteristic points of the deduction value are defined as (x0, m(x0)), …, (xn, m(xn)). The n Lagrangian interpolation results are: （1）； Where, The first i measurement points; Indicates the i The deduction value corresponding to each measurement point; Indicates the number of feature points; y Indicates the first i +1) The deduction value corresponding to the measurement point; Indicates the i The Lagrangian basis function of the measurement points is calculated as follows: （2）； Where, The first j measurement points; Step 1.2.2: Perform entropy-based HI estimate; Different characteristic state quantities of main transformer cooler HI Expressed as follows: （3）； Where, Indicates the i The reliability value of each state feature; Representation characteristics i Impact factor; Representation characteristics i The deduced value of 0 indicates the worst operating condition, and 1 indicates the best operating condition. The cooler degradation characteristic state is divided into four levels from light to heavy: I, II, III, and IV. Take 1, 2, 3, and 4 respectively; Therefore, the reliability index of the cooler HI is the weighted value of all state characteristic factors, as shown in formula (4); （4）； Where, HI Indicates the reliability of the main transformer cooler. Indicates the i The weight of each characteristic factor; M Indicates the number of cooler state characteristics, M The cooler status characteristic quantities include the main transformer oil temperature, the main transformer winding temperature, the cooler outlet water temperature, the outlet oil temperature, the outlet water pressure, and the outlet oil pressure.

3. The reliability-centered main transformer cooler start-stop control logic optimization method according to claim 2 is characterized in that: The entropy weight method is used to calculate the weights of each state characteristic of the cooler: Step 1.2.2.

1. Input: Number of main transformer coolers N, total number of characteristic quantities of main transformer coolers M、 Constructed by formula (5) HI Judgment matrix: （5）； Output: The weight of the reliability index HI of the status characteristics of each main transformer cooler; Step 1.2.2.2, Normalization: Calculate the ratio of each column matrix value to the sum of the corresponding column eigenvalues , and obtain the normalized decision matrix: （6）； （7）； Step 1.2.2.

3. Calculate the entropy of the characteristic quantities of different states of the main transformer cooler: （8）； Step 1.2.2.

4. Calculate the coefficient representing the degree of diversification: （9）； Step 1.2.2.5: Calculate the characteristic value of each cooler HI Weight: （10）； Step 1.2.2.6: Output the characteristic value of each cooler HI Weight.

4. The reliability-centered main transformer cooler start-stop control logic optimization method according to claim 1 is characterized in that: The prediction model construction method in step 1.3 is: Step 1.3.1, Forget Gate: The forget gate is used to determine how much previous information will be ignored; （11）； Where, Indicates that LSTM t The forget gate output at that moment; Represents the weight matrix of the forget gate; Indicates that LSTM t Output at time -1; Indicates that LSTM t Input at the moment; Represents the bias vector of the forget gate; Represents the sigmoid function; Step 1.3.2, Input Gate: The input gate is used to control how much of the current input information will be stored in the memory; （12）； （13）； （14）； Where, Indicates that LSTM t Input gate output at time; Represents the weight matrix of the input gate; represents the bias vector of the input gate; Indicates that LSTM t Candidate cell states at time instant; Indicates that LSTM t The cell state at a moment in time; The weight matrix representing the candidate cell state; A bias vector representing the candidate cell state; represents the hyperbolic tangent function; Indicates that LSTM t -1 cell state at time instant; represents point-wise multiplication; Step 1.3.3, output gate: The output gate is used to determine the time step t What to output when （15）； （16）； Where, Indicates that LSTM t Output gate output at time; Represents the weight matrix of the output gate; represents the bias vector of the output gate; Based on three control gates and storage units, the LSTM neural network reads, resets, and updates the acquired long-term information.

5. The reliability-centered main transformer cooler start-stop control logic optimization method according to claim 4 is characterized in that: The harmony search algorithm is used to optimize the LSTM network parameters to improve the prediction accuracy of the main transformer cooler reliability index: Step 1.3.4: Randomly generate within the value range h Group LSTM parameters to form a harmony library, denoted as H : （17）； Where, Represent the optimal number of LSTM neurons and learning rate respectively; For the h The absolute error between the group parameter value and the actual result; h is the size of the harmony library; Step 1.3.5: Generate new harmony , as shown in formula (18); assuming represents the LSTM parameters in the new harmony, then have Probabilistic selection of harmony libraries H The values ​​in are 1- P H Probabilistic Selection Harmony Library H A randomly generated value outside the range of values; if H If , the pitch adjustment method of the harmony library is as shown in formula (19); （18）； （19）； Where, Adjust bandwidth for tone; Represents a uniformly distributed random number between [0,1]; represents the pitch adjustment probability; Step 1.3.6, harmony library update: according to the harmony in step 1.3.5, the new error is obtained. , like , then update the harmony library, Will And the corresponding new harmony update to H middle; Step 1.3.7: Determine whether the maximum number of iterations has been reached. If so, stop the iteration and obtain the optimal number of LSTM neurons and learning rate based on the current global optimal value. Otherwise, set k = k +1, go to step 1.3.5 and step 1.3.6 to continue iterating; Step 1.3.8: Use the trained HS-LSTM model to predict the reliability of the main transformer cooler for the next seven days and calculate the average reliability over these days. The root mean square error is used to evaluate the quality of the prediction results, as shown in formula (20); （20）； Where, n Indicates the n A prediction moment, N Indicates the total number of prediction moments; Respectively represent the predicted n The reliability index of the main transformer cooler and the actual n 1. Reliability index of main transformer cooler; RMSE represents the root mean square error 。

Images