Converter valve waste heat digital comprehensive analysis model and method

Abstract

According to the converter valve waste heat digital comprehensive analysis model and method, working parameters of a converter valve are monitored in real time, a physical mechanism and a data driving method are combined, and a hybrid modeling system is constructed for waste heat model building and dynamic correction. A multi-scale data fusion framework is adopted to carry out data processing and mode recognition, the accuracy of model prediction is evaluated through a residual analysis mechanism, and future temperature rise prediction is carried out. According to the method, a dynamic knowledge graph is constructed based on a digital twin platform, multi-dimensional report contents including a cooling system upgrading scheme, a valve tower inspection period adjustment strategy and the like are generated, and data traceability is ensured through block chain evidence storage. According to the method, the requirements of a current converter valve waste heat analysis method for real-time monitoring and accurate prediction are improved, a feasible mode for real-time monitoring and accurate prediction of converter valve waste heat is provided, the accuracy of equipment working state feedback and fault prevention measures is ensured, and the stability of a high-voltage direct-current power transmission system is further ensured.

Description

Technical Field

[0001] The present invention belongs to the field of converter station digitization, and more specifically relates to a digital comprehensive analysis model and method for the waste heat of a converter valve. Background Art

[0002] The converter valve is a core component of a high-voltage direct current (HVDC) transmission system, and its operating state directly affects the stability and safety of the transmission system. The waste heat generated by the converter valve during operation is an important factor affecting the stable operation of the system. If the heat analysis and management of the converter valve cannot be effectively carried out, it may lead to premature wear of the converter valve and even cause serious equipment accidents.

[0003] Currently, the analysis of the waste heat of the converter valve mainly relies on physical models and experimental data. However, the limitation of this method is that it cannot monitor the equipment state in real time and make predictions. In addition, an analysis model that is more adaptable to complex and changing operating conditions is needed, so that the system can provide positive fault prevention measures while providing real-time and accurate feedback on the equipment operating state to ensure the stable operation of the system. However, existing models often only rely on theoretical calculations and do not consider the equipment state under actual operating conditions, which often cannot meet the operation requirements of high performance and high reliability.

[0004] Nowadays, with the increasing maturity of technologies such as the Internet of Things (IoT), big data analysis, and deep learning, new possibilities have been provided for the real-time monitoring and prediction of converter valves. However, how to combine these advanced technologies to establish an analysis model that can effectively monitor the waste heat of the converter valve in real time and accurately predict it, so as to provide timely feedback on the equipment state and positive fault prevention measures, remains an urgent problem to be solved. Therefore, in order to ensure the normal operation of the converter valve and the stability of the HVDC transmission system, a model and method for comprehensive analysis of the waste heat of the converter valve using digital methods are proposed. Summary of the Invention

[0005] The present invention mainly solves the problem that the existing waste heat analysis method of the converter valve lacks the ability of real-time monitoring and accurate prediction, and cannot effectively provide feedback on the equipment operating state and formulate positive fault prevention measures. In order to ensure the normal operation of the converter valve and the stability of the HVDC transmission system, a model and method for comprehensive analysis of the waste heat of the converter valve using digital methods are proposed.

[0006] To achieve the above object, the present invention is implemented by adopting the following technical solutions: including the following steps:

[0007] Data acquisition: The working parameters of the converter valve are monitored in real time through a multi-type sensor network. After the data is processed multiple times, it is transmitted to the data center through an industrial ring network architecture;

[0008] Establish a waste heat model: Integrate physical mechanisms and data-driven methods to build a hybrid modeling system, and use simulation and deep learning technologies for model parameter calibration and dynamic correction;

[0009] Data analysis: Perform data processing and pattern recognition through a multi-scale data fusion architecture, adopt a residual analysis mechanism to evaluate the accuracy of model prediction, and conduct future temperature rise prediction;

[0010] Generate a report: Build a dynamic knowledge graph based on the digital twin platform to generate multi-dimensional report content. The report version is generated differently according to user roles, and blockchain evidence preservation is used to ensure data traceability.

[0011] In one solution, in the data collection, the sensor network includes distributed fiber optic grating temperature sensors, ultrasonic flow meters, and platinum resistance temperature probes. Data transmission is carried out through wavelength division multiplexing technology and the Modbus RTU protocol to achieve comprehensive multi-parameter monitoring, and the data is preliminarily processed through edge computing technology to ensure the accuracy and real-time nature of the data.

[0012] In one solution, for the data collection, an intelligent acquisition terminal with FPGA acceleration function is deployed on the edge side to perform AD conversion and filtering processing on the original analog quantity, and use time-sensitive Ethernet and 5G URLLC modules to achieve efficient and low-latency data transmission, ensuring real-time monitoring and rapid response of the converter valve operating status.

[0013] In one solution, for the waste heat model, an adaptive Kalman filter is used for real-time calibration to ensure the accuracy and stability of the model. The initial model parameters are calibrated in combination with CFD simulation, and a deep state space network is used for dynamic correction to improve the prediction ability and adaptability of the model.

[0014] In one solution, for the data analysis, dynamic time warping, t-SNE dimensionality reduction, and OPTICS clustering technologies are used for data processing. The accuracy of model prediction is evaluated by designing a two-layer residual analysis mechanism, and future temperature rise prediction is carried out in combination with a hybrid predictor to ensure the reliability and accuracy of the analysis results.

[0015] In one solution, for the report generation, natural language generation technology and graph neural networks are used to mine potential association rules. The report content includes the cooling system upgrade plan, the valve tower inspection cycle adjustment strategy, and the comparison of its expected energy-saving benefits and implementation costs, supporting voice interaction query and the embedding of a simulation deduction module.

[0016] In one solution, the generated report integrates a lightweight 3D rendering engine, maps the heat dissipation bottleneck positioning result to the 3D model of the converter valve, supports viewing the heat stress distribution cloud map and the historical health degree change curve of any component by clicking, and provides visual analysis results to assist in decision-making.

[0017] Advantages of the present invention:

[0018] First of all, it realizes the real-time monitoring and accurate prediction of the converter valve. Through the Internet of Things technology and multi-type sensor networks, the working state of the converter valve is monitored in real time, and deep learning technology is used for model parameter calibration and dynamic correction, greatly improving the accuracy of the waste heat analysis of the converter valve.

[0019] Secondly, it provides a timely warning function. By using a multi-scale data fusion framework for data processing and pattern recognition, future temperature rise prediction can be carried out, early detection and prevention of possible equipment failures, providing important support for ensuring the stable operation of the power system.

[0020] Furthermore, by generating multi-dimensional report content, decision-makers can more intuitively and easily understand the cooling system upgrade plan, the valve tower inspection cycle adjustment strategy, etc., greatly improving the decision-making efficiency.

[0021] In addition, by introducing blockchain technology, the authenticity and immutability of data are guaranteed, providing a strong guarantee for the accuracy of the analysis results.

[0022] Finally, the present invention adopts a high-performance computing architecture and a big data platform for data storage and calculation, greatly improving the data processing efficiency, saving a large amount of time and human resources. Moreover, this method supports users to realize data sharing and collaborative analysis under authorization, improving the data utilization rate and the solution innovation efficiency. Description of the Drawings

[0023] Figure 1 It is the flow chart of the method of the present invention. Detailed Embodiments

[0024] To facilitate the understanding of the present invention, the present invention will be described more comprehensively below with reference to the relevant drawings. Typical embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, these embodiments are provided to make the disclosure of the present invention more thorough and comprehensive.

[0025] Unless otherwise defined, all technical and scientific terms used in this invention have the same meanings as those understood by those skilled in the technical field to which this invention belongs. The terms used in the description of this invention in this specification are only for the purpose of describing specific embodiments, and are not intended to limit this invention. For the convenience of understanding this invention, the following will describe this invention more comprehensively with reference to the relevant drawings. The drawings show typical embodiments of this invention. However, this invention can be implemented in many different forms and is not limited to the embodiments described in this invention. On the contrary, the purpose of providing these embodiments is to make the disclosure of this invention more thorough and comprehensive.

[0026] As Figure 1 shown, this invention discloses a digital comprehensive analysis model and method for the waste heat of a converter valve. The specific steps are as follows:

[0027] Step 1: Data collection: Through devices such as sensors, various operating parameters of the converter valve are monitored in real time, and the data is transmitted to the data analysis system.

[0028] Data collection, as the basic link of converter valve waste heat analysis, requires a complete technical chain around physical parameter perception, signal transmission, and quality control in its implementation process. First, a multi-type sensor network needs to be arranged in the key heat-sensitive areas of the converter valve: Distributed fiber Bragg grating temperature sensors (accuracy ±0.5°C) are embedded at the substrate of the IGBT module, and wavelength division multiplexing technology is used to achieve 16-point temperature measurement with a single optical fiber; ultrasonic flow meters (range 0.1 - 5m 3 / s) and platinum resistance temperature probes are installed at the inlet and outlet of the cooling water, and data is output through the Modbus RTU protocol; power loss monitoring relies on the valve base electronics (VBE) to collect the thyristor conduction voltage drop and gate drive current, and combines the real-time AC side voltage transformer signal for loss integration calculation. The sensor layout needs to follow the principle of thermal-electric-current coupling to form a three-dimensional monitoring matrix at different levels of the valve tower (interlayer busbars, radiator fins, insulator surfaces). A typical 800kV converter valve needs to deploy 120 - 150 monitoring points.

[0029] During the data acquisition process, the original signal needs to go through multiple processes to ensure its usability: An intelligent acquisition terminal with FPGA acceleration function is deployed on the edge side to perform 24-bit AD conversion on the original analog quantity. A sliding window median filter is used to eliminate electromagnetic interference spikes, and a Chebyshev type II low-pass filter (cutoff frequency 500Hz) is used to suppress high-frequency noise. For digital signals, a dual-channel redundancy check mechanism is designed, and when the data difference between the two CAN bus channels exceeds 2%, an automatic retransmission is triggered. The preprocessed data is transmitted through an industrial ring network architecture. The backbone network uses TSN time-sensitive Ethernet (transmission delay < 2ms), and 5G URLLC modules are deployed at the wireless access points (air interface delay 1ms) to ensure real-time performance under a data throughput of 200Mbps level.

[0030] On the data center side, a time series database cluster (such as TDengine) is deployed to achieve millisecond-level data writing, and the time deviation of each node is controlled within ±50μs through the NTP / PTP hybrid clock synchronization protocol. For the sensor failure scenario, a robust data compensation algorithm based on the L1 norm is developed. When a certain temperature measurement point is abnormal, the temperature gradient field of adjacent nodes is used for spatial interpolation reconstruction. The finally formed standardized data stream contains multi-dimensional information such as equipment operating conditions (load rate, cooling water flow), thermal states (junction temperature, radiator temperature rise), and environmental parameters (humidity, air pressure). After being encapsulated in the Apache Avro binary format, it is pushed to the Kafka message queue to provide a high-timeliness and high-integrity input basis for subsequent modeling and analysis. The system needs to perform end-to-end verification every week during actual operation. A standard heat source generating device is used to inject known heat loads to verify that the full-link measurement error is less than 1.5%, ensuring the engineering credibility of the acquired data.

[0031] Step 2: Establish a converter valve waste heat model: By collecting the working data of the converter valve, a digital converter valve waste heat model is established.

[0032] The construction of the converter valve waste heat model needs to integrate physical mechanisms and data-driven methods to form a hybrid modeling system with dynamic correction capabilities. In the stage of establishing the basic thermodynamics framework, first, the heat conduction path is decomposed according to the converter valve structure: The thyristor-level unit is abstracted into a third-order lumped parameter model including the junction area heating body, radiator heat capacity, and coolant convection, and the thermal balance differential equations of each level are established:

[0033]

[0034] where C j , C h , C c represent the junction area, radiator, and coolant heat capacities respectively, R jh , R hc are the contact thermal resistance and convective thermal resistance, and P lossObtained by calculating the measured voltage and current through . The initial values of the model parameters are calibrated through CFD simulation. Three-dimensional fluid-structure interaction simulation is carried out using COMSOL Multiphysics, and the equivalent thermal resistance matrix under typical working conditions is extracted.

[0035] The data-driven module adopts the time-series feature enhancement method. The original working condition data (load current I d , cooling water flow , ambient temperature T amb ) are decomposed into 8 sub-bands through wavelet packet transform, and the energy entropy of each band is extracted as the state feature. A deep state space network (DSSN) is constructed, and its hidden state evolution equation is expressed as:

[0036] h t = tanh(W h [h t-1 , x t + b h )

[0037] z t = σ(W z [h t , Θ physics )

[0038] where Θ physics is the temperature prediction value output by the physical model, and the dynamic fusion of physical constraints and data features is realized through the gating mechanism. The network training adopts an improved PINN (Physics-Informed Neural Network) loss function:

[0039]

[0040] The second physical residual constraint forces the network to follow the partial differential equation of heat conduction N, enhancing the extrapolation stability. During the online operation stage of the model, an adaptive Kalman filter is deployed for real-time calibration, and the covariance matrix P k|k is updated as follows:

[0041] K k = P k|k-1 H T (HP k|k-1 H T + R) -1

[0042]

[0043] The process noise matrix Q is dynamically adjusted according to the variable operating condition frequency of the cooling system. When the flow rate fluctuation exceeds 10%, the expansion factor β of the Q matrix is triggered to be 1.5. The final model realizes multi-rate simulation on the digital twin platform. The steady-state process adopts a 1-second time step, and the transient process automatically switches to a 10-millisecond time step. It is synchronized with the real-time database through the OPC UA protocol. In the verification stage, a step response test is adopted. When the cooling flow rate drops suddenly by 30%, the maximum deviation of the predicted junction temperature of the model does not exceed ±2.3 °C, meeting the requirements of the IEC61803 standard.

[0044] Step 3, data analysis: The data analysis system obtains the waste heat situation of the converter valve under different operating states through statistical analysis of the collected data.

[0045] The core of the converter valve waste heat analysis system lies in constructing a multi-scale data fusion architecture to realize the knowledge extraction from the original time-series data to the thermal state decision-making. The data processing pipeline first performs spatio-temporal alignment on the heterogeneous data collected in Step 1: The dynamic time warping (DTW) algorithm is used to compensate for the sampling delays of different sensors. The alignment cost function is defined as:

[0046]

[0047] where \(d(x i ,y j ) is the KL divergence distance between the temperature and flow rate signals. The aligned data cube is mapped to the feature space through t-SNE dimensionality reduction, and the OPTICS clustering is used to identify typical operating condition patterns, forming a two-dimensional state matrix of load rate - cooling efficiency. For the output of the hybrid model established in Step 2, a two-layer residual analysis mechanism is designed: The basic layer calculates the absolute deviation ΔT meas between the measured temperature T phy and the predicted temperature T base of the physical model, ΔT meas =|T phy -T data |; The enhanced layer extracts the structural residual between the predicted temperature T of the data-driven model and the physical model, and captures the abnormal heat dissipation mutation through the edge detection operator with a convolution kernel of [1, -2, 1].

[0048] The thermal state evaluation adopts an improved fuzzy comprehensive evaluation method, and the membership function is defined as:

[0049]

[0050] Combined with the analytic hierarchy process (AHP) to calculate the weight vector \(W = [0.35, 0.25, 0.4]\) (junction temperature, coolant temperature rise, heat dissipation gradient), and finally the thermal risk index R thermal =W·[μ junc ,μ coolant ,μgrad T For transient process analysis, a feature correlation network based on the Maximal Information Coefficient (MIC) is constructed, and the following is calculated:

[0051]

[0052] A strong non - linear coupling relationship (MIC > 0.82) between the cooling water flow rate and the valve layer temperature difference at the 0.1 Hz frequency band is identified. A sliding - window Fourier transform is deployed in the historical data warehouse to extract the periodic performance energy spectrum features under each working condition, and a normal - condition hypersphere is constructed through Support Vector Data Description (SVDD):

[0053] min R,c R 2 +C∑ξ i s.t.‖φ(x i )-c‖ 2 ≤R 2 +ξ i

[0054] When the new data point exceeds 1.3 times the radius of the hypersphere, a deterioration warning is triggered. The online analysis engine integrates an XGBoost regression tree and an LSTM neural network to form a hybrid predictor. The LSTM cell state update equations are as follows:

[0055] f t =σ(W f ·[h t-1 ,x t +b f )

[0056] i t =σ(W i ·[h t-1 ,x t +b i )

[0057]

[0058] o t =σ(W o ·[h t-1 ,x t +b o )

[0059] h t =o t °tanh(C t )

[0060] The tree model is responsible for capturing the discontinuous relationship between environmental parameters and heat dissipation efficiency. The prediction results of the two are fused through Bayesian Model Averaging (BMA), and the weight is calculated based on the historical window NSE coefficient: ​

[0061]

[0062] The finally generated thermal map includes real-time waste heat distribution, prediction of temperature rise in the next 5 minutes, and positioning of key heat dissipation bottlenecks, which is superimposed on the surface of the converter valve digital twin through three-dimensional topological mapping technology. The system conducts K-fold cross-validation every quarter to ensure that the waste heat assessment error rate remains within 3% under boundary conditions such as cooling system aging and sudden changes in ambient temperature, meeting the requirements of the GB / T37054-2018 standard.

[0063] Step 4: Generate a report: Generate a corresponding report based on the analysis results to provide decision-making support for formulating a reasonable energy management strategy.

[0064] The report generation system constructs a dynamic knowledge graph relying on the digital twin platform, and transforms the analysis results into multi-dimensional decision-making support information through a semantic engine. The system first extracts the thermal state assessment matrix, energy efficiency trend prediction, and abnormal event set from the data analysis module, and uses an ontology-based knowledge reasoning framework for information fusion: defines a triple relationship model of "converter valve - cooling system - environmental factors", and mines potential association rules through a graph neural network, such as identifying the implicit coupling relationship between specific load fluctuation patterns and the fouling degree of radiators. The report content generation adopts a modular template engine to automatically match the requirements of the energy management scenario - for daily operation and maintenance reports, focus on the real-time waste heat distribution thermal map, the temperature ranking of key components, and the cooling system efficiency curve; for strategic planning reports, generate a quarterly energy efficiency decay trend analysis, a radar chart of the proportion of heat dissipation energy consumption under different working conditions, and a Gantt chart for equipment life prediction.

[0065] The core chapter of the report is dynamically arranged through natural language generation (NLG) technology, and combines the LSTM+Attention model to achieve semantic mapping from data to text: extracts indicators such as the frequency of temperature exceeding the standard and the effective rate of cooling water flow adjustment from the time series database, and automatically generates the "Risk Warning Summary" chapter, marking the serial numbers of thyristors under key monitoring and their historical deterioration trajectories; an optimization suggestion generator trained based on reinforcement learning, combined with equipment parameters and industry specifications in the knowledge base, outputs executable suggestions such as cooling system upgrade plans and valve tower inspection cycle adjustment strategies, and attaches a comparison of the expected energy savings and implementation costs of different plans. The report output stage integrates a lightweight three-dimensional rendering engine, maps the heat dissipation bottleneck positioning results to the three-dimensional model of the converter valve, supports clicking to view the thermal stress distribution cloud map and historical health change curve of any component, and at the same time associates with the equipment archive to retrieve maintenance records and spare parts inventory status.

[0066] The system is built with an intelligent distribution mechanism that automatically generates differentiated report versions based on user roles: it pushes operation manuals with detailed parameters to maintenance personnel, highlighting warning thresholds and emergency response procedures; it generates a condensed strategic dashboard for management, focusing on key indicators of energy efficiency improvement and return on investment analysis. All report versions are stored on the blockchain to ensure data traceability and support voice interaction queries. When a manager asks "redundant cooling capacity under extreme high temperatures", the system automatically activates the simulation and deduction module and embeds simulation videos of the cooling system's extreme operating conditions at different temperature thresholds in the report. The report iteration mechanism is continuously optimized through online feedback learning. The annotations and evaluations of maintenance personnel on the proposed solutions are sent back to the knowledge base in real time, driving the prioritized recommendation of strategy combinations with high historical adoption rates in the next report generation.

[0067] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, the storage medium can be a magnetic disk, an optical disc, a read-only memory (ROM), or a random access memory (RAM), etc.

[0068] It should be understood that the detailed description of the technical solutions of the present invention with the help of the preferred embodiments above is illustrative rather than restrictive. Those of ordinary skill in the art can modify the technical solutions recorded in each embodiment or perform equivalent replacements on some of the technical features based on reading the specification of the present invention; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present invention.

Claims

1. A digital comprehensive analysis model and method for the waste heat of a converter valve, characterized in that: It includes the following steps: Data acquisition: The working parameters of the converter valve are monitored in real time through a multi-type sensor network. After the data undergoes multiple processes, it is transmitted to the data center through an industrial ring network architecture; Establishing a waste heat model: Integrating physical mechanisms and data-driven methods, constructing a hybrid modeling system, and using simulation and deep learning technologies to calibrate and dynamically correct model parameters; Data analysis: Conducting data processing and pattern recognition through a multi-scale data fusion architecture, using a residual analysis mechanism to evaluate the accuracy of model predictions, and making future temperature rise predictions; Generating a report: Building a dynamic knowledge graph based on a digital twin platform, generating multi-dimensional report content. The report version is generated differently according to user roles, and blockchain evidence preservation is used to ensure data traceability.

2. The digital comprehensive analysis model and method for the waste heat of a converter valve according to claim 1, wherein: In the data acquisition described above, the sensor network includes distributed fiber optic grating temperature sensors, ultrasonic flow meters, and platinum resistance temperature probes. Data transmission is carried out through wavelength division multiplexing technology and the Modbus RTU protocol to achieve comprehensive multi-parameter monitoring, and the data is preliminarily processed through edge computing technology to ensure the accuracy and real-time nature of the data.

3. The digital comprehensive analysis model and method for the waste heat of a converter valve according to claim 2, characterized in that: In the data acquisition described above, an intelligent acquisition terminal with FPGA acceleration function is deployed on the edge side to perform AD conversion and filtering processing on the original analog quantity, and use time-sensitive Ethernet and 5G URLLC modules to achieve efficient and low-latency data transmission, ensuring real-time monitoring and rapid response to the operating status of the converter valve.

4. The digital comprehensive analysis model and method for the waste heat of a commutation valve according to claim 1, wherein: For the waste heat model described above, an adaptive Kalman filter is used for real-time calibration to ensure the accuracy and stability of the model. The initial model parameters are calibrated in combination with CFD simulation, and a deep state space network is used for dynamic correction to improve the prediction ability and adaptability of the model.

5. The digital comprehensive analysis model and method for the waste heat of a commutation valve according to claim 1, characterized in that: In the data analysis described above, dynamic time warping, t-SNE dimensionality reduction, and OPTICS clustering technologies are used for data processing. The accuracy of model predictions is evaluated by designing a two-layer residual analysis mechanism, and future temperature rise predictions are made in combination with a hybrid predictor to ensure the reliability and accuracy of the analysis results.

6. The digital comprehensive analysis model and method for the waste heat of a commutation valve according to claim 1, characterized in that: In the report generation described above, natural language generation technology and graph neural networks are used to mine potential association rules. The report content includes the cooling system upgrade plan, the valve tower inspection cycle adjustment strategy, and the comparison of its expected energy-saving benefits and implementation costs, supporting voice interaction queries and the embedding of simulation deduction modules.

7. A digital comprehensive analysis model and method for the waste heat of a converter valve according to claim 6, characterized in that: In the report generation described above, a lightweight 3D rendering engine is integrated to map the heat dissipation bottleneck location result to the 3D model of the converter valve, supporting the click to view the thermal stress distribution cloud map and the historical health degree change curve of any component, providing visual analysis results to assist decision-making.

Images