Applied research of artificial intelligence technology in power system Tang Huajin and Chen Hanping of Shanghai Jiaotong University claimed the following: analyzed the power system problems suitable for AI applications, and summarized four kinds of artificial intelligence technologies that are widely used, and finally pointed out that AI is in the power system. Application development trends and application prospects.
The power system is a complex, non-linear dynamic system composed of many units such as power generation equipment, transformers, transmission and distribution lines, and electrical equipment. Artificial intelligence technology is widely used to solve nonlinear problems and has irreplaceable advantages over traditional methods. At present, a variety of artificial intelligence tools have been developed at home and abroad, including expert systems (CES), artificial neural networks (ANN), fuzzy sets (FS), and heuristic search (HS), which have opened up applications in various fields of power systems. .
Application of 1AI in Power System 1.1 Operation and Control of Power System There are a large number of automatic control and manual control devices in power systems, such as relays, circuit breakers, and disconnectors. The synergy of these relatively simple local controls constitutes a complex real-time control of the entire power system.
There are two forms of protection real-time control, discrete and continuous control. Relay protection is a universal discrete control that is distributed throughout the system. The judgment of the system state (normal or accident), that is, the state assessment is the key to achieving the protection action. Because AI has logical thinking and rapid processing capabilities, it has become an important tool for online status assessment. A rule-based topology error detection algorithm is proposed to effectively use the operator's experience. Compared with Fourier transform and Kalman filtering techniques, the application of neural network for the analysis of the characteristic parameters of current and voltage waveforms has better real-time performance.
Tang Huajin, male, master student. School of Power and Energy Engineering, 200030 The correct protection settings depend on the overall analysis of the impact of equipment operation on the system, and cannot be separated from human inspiration and logical judgment. There are a lot of fuzzy knowledge and methods in the design of relay protection.
Offloading is another discrete control. Sudden loss of system components (such as an abrupt shutdown of the generator due to a fault) can cause drastic changes in system capacity. When the load exceeds the system supply capacity, the load must be reduced to avoid a wide range of power interruptions. At this time, it is necessary to control the relay in time by analyzing and heuristic knowledge of the load demand and system behavior. If the transient stability problem of the post-fault system is described by the solution of the post-fault system differential equation, there is a mathematical mapping between the failure and the transient stability. ANN has the function of approximation and parallel processing of function mapping. Therefore, ANN is suitable for the load control of power system with good adaptability and practicability. The selection of input features and the availability of samples sufficient to describe the function mapping are the key issues for load shedding control using neural networks.
Excitation control is an important part of controlling generator terminal voltage and reactive power. It is an important real-time continuous control system and plays a major role in maintaining the stability of the power system. The part that completes this function is also called the power system stabilizer (PPS). . Due to the input of large-capacity units and the application of rapid excitation system, the dynamic stability of the system becomes more and more prominent. For example, the fuzzy set theory is used in the excitation control system, which has a better control effect than the traditional PPS based linear system theory.
1.2 Management and Planning of Power Systems The role of energy management systems (EMS) in modern power systems is becoming increasingly prominent. System-wide data is transmitted to EMS through SCADA. Control signals are transmitted from EMS to each component. The whole process needs to be synchronized. This requires EMS to have real-time processing capability for a large amount of information and to be timely and correct under normal and accident conditions. Make control decisions. Monitoring and diagnosis are important functions of EMS. AI plays an important role in the field of condition monitoring and fault diagnosis. A variety of diagnostic strategies based on expert systems and neural networks have been developed at home and abroad.
Automatic Generation Control (AGC) is a centralized real-time computer control function in the operation of an interconnected power system to keep the system output and system load matching. By controlling the energy exchange between interconnected systems, the economical distribution of the load between the units (power plants) is achieved. Due to the high degree of variability of industrial loads, the use of conventional control methods has major limitations. For example, the pattern recognition of controllable signals by rice Kohonen self-organizing neural networks only responds to long-term disturbances and effectively improves AGC control quality.
Safety assessments of various types of disturbances and accidents that may occur in the power system, such as equipment damage, natural phenomena, human error and destruction, many of which are unpredictable and controllable. Therefore, the evaluation of the power system's ability to withstand interference (or accidents), that is, safety assessment is very necessary. As an important means of security assessment, neural networks have been greatly developed and applied to the static stability and dynamic stability analysis of the system.
The commonly used method in the field of safety evaluation is simulation, which simulates the static and transient response of the system in anticipation of an accident. It is a difficult point to anticipate the screening of accidents and often depends on the experience of the operating personnel. AI has a broad prospect as a predictive accident screening tool. For example, a rule-based expert system and Kohonen self-organizing neural network applied to the screening of predictive accidents can effectively combine the experience of operating personnel and the advantage of rapid screening.
The system recovery after failure recovery is an orderly coordination process, that is, the disconnected system is reconfigured in the shortest time, and power supply is restored smoothly. Improper recovery sequence may cause a new accident. The key to correct recovery action lies in the choice of recovery order. Applying heuristic search can effectively reduce the search space. Intelligent recovery technology is one of the important research directions in the power system. For example, the integrated intelligent restoration expert system combines heuristic search (genetic algorithm) and fuzzy set theory, and has made beneficial explorations.
Load forecasting is an important content and foundation of power planning. Due to the very complex nonlinear relationship between various factors including weather variables and actual loads, load forecasting is very difficult. In addition to traditional statistical analysis methods, artificial intelligence prediction techniques have emerged, mainly expert systems and neural networks. Because the neural network is suitable for solving the problem of time series forecast (especially the forecast of smooth transition process), once the power system is introduced, load forecasting becomes a main field of its application.
2 Introduction to Application Methods 2.1 Expert System (ES) Expert system is a computer program that has expert experience and knowledge in a certain field and can use such knowledge as a human expert to make decisions through reasoning. A typical expert system consists of four parts: a knowledge base, an inference engine, a knowledge acquisition mechanism, and a human-machine interface. The expert system has become the most mature artificial intelligence technology in the power system. A variety of expert systems have been developed at home and abroad for different areas of the power system: monitoring and diagnosis, power grid dispatch, predictive accident screening, and system recovery. In particular, monitoring and fault diagnosis have become ES's most important application areas in the power system.
According to different ways of storing knowledge, expert systems can be divided into different forms, namely expert systems based on shallow knowledge (experience knowledge), rules, decision trees, models, and object-oriented expert systems. Model-based knowledge representation is suitable for real-time processing, faster, simpler, and easier to maintain than other methods such as rule-based (hypothetical) or heuristic reasoning.
The bottleneck of knowledge acquisition is the main difficulty in constructing and maintaining expert systems. Somebody proposed a new way of automatic knowledge acquisition, namely machine learning, which is applied to power system switching sequence expert system. In the knowledge base construction phase, knowledge is extracted from the operating personnel's past experience without having to learn directly from the operating personnel; the knowledge base can be automatically updated and extended each time the human expert interacts with the system.
2.2 Artificial Neural Network (ANN) Artificial neural network is a simulated biological excitation system. It outputs a series of inputs through neural networks. Here, the output and input are all normalized quantities. The output is a non-linear function of the input. The value can be changed by connecting the weight of each neural network element to obtain the desired output value, which is the so-called training process.
According to different problems, neural networks with various structures and training algorithms have been applied in power systems, such as BP networks and Kohonen self-organizing neural networks.
Because neural networks have fast parallel processing capabilities and good classification capabilities, they are widely used in real-time control, monitoring and diagnosis of power systems, short-term and long-term load forecasting, and status assessment. Neural network-based load forecasting technologies have become One of the most successful applications of artificial intelligence in power systems.
The BP network structure and its algorithm are simple and easy to implement. It is a more mature method for load forecasting. A variety of improved algorithms for BP networks are proposed, such as adaptive adjustment of impulse coefficients and improvement of error functions, to accelerate convergence; the initial random weights are limited in magnitude and the local minimum problem is overcome.
2.3 Fuzzy set theory and heuristic search The human cognition world contains a large amount of uncertainty knowledge, which requires a certain amount of fuzzification of the information obtained to reduce the complexity of the problem. Fuzzy logic can be considered as an extension of multi-valued logic and can perform approximate reasoning that is difficult to achieve with traditional mathematical methods.
In recent years, the application of fuzzy set theory in power system has made rapid development, including power flow calculation, system planning, fuzzy control and other fields. For load variation and power production uncertainty, a fuzzy value is used to represent the membership function of an uncertain load in the actual set, and a fuzzy model of the optimal power flow in the power system is established, namely the fuzzy optimal power flow (FOPF).
2.4 Heuristic Search Genetic Algorithm (GA) and Simulated Annealing (SA) Algorithms are two heuristic searches that have gradually emerged in recent years. They generate new solutions randomly and retain better results, and avoid falling into local minimums. Get the global optimal solution or approximate optimal solution. GA is a solution to the optimization problem represented by a set of digit strings, and the string is optimized by genetic operators, ie selection, hybridization, and mutation operations. The SA generates new solutions in the neighborhood of the known solution, and gradually reduces the size of the neighboring region until it approaches the global optimal solution. Both methods can be used to solve optimization problems of arbitrary objective functions and constraints, and have achieved satisfactory results in energy engineering, economics, and power.
Genetic algorithm is a search algorithm based on natural selection and genetic mechanism. It has less requirements for optimal design. It does not require differentiable, nor continuous, for the objective function. It only requires the problem to be computable, and its search always covers the entire solution. Space can effectively avoid the combination of "explosive" problems and local minimum solutions of conventional mathematical methods, and has a strong practical value.
At present, there are still many problems to be solved in the application of heuristic search, such as the selection of the search termination criterion, the termination of the quick deviation from the optimal solution, and failure to stop it in time will lead to over calculation and will not improve the quality of the solution. The selection of genetic factors in GA and the cooling rate in SA is a key factor affecting the performance of the algorithm and must be properly adjusted, otherwise a local optimal solution may be obtained.
3AI development trends in power systems Hybrid intelligence Currently, the four main tools in artificial intelligence (expert systems, ANN, fuzzy set theory, and heuristic search) have their own advantages and limitations, and they lack a common and effective method for power. The various areas of the system. Hybrid intelligence, that is, the integration of multiple intelligent technologies, has become one of the important development directions of AI.
Distributed artificial intelligence (DAI) technology is a branch of artificial intelligence research developed in the 1980s and is accompanied by the development of parallel distributed computing, including distributed problem solving (DPS), parallel artificial intelligence (PAI), Multi-agent and other content. The application of DAI in power systems is currently focused on the use of multi-agent technology.
The improvement of the structure and algorithm of the neural network itself is also an important task in the development of AI. In recent years, the proposed ellipsoidal unit neural network has opened up new directions for fault diagnosis. Compared with the classical BP network, the ellipsoidal element network has the advantages of generalization boundedness, good rejection performance, high fault classification accuracy, and especially in the diagnosis of multiple fault simultaneity, it has better pattern recognition ability than the BP network.
4 Conclusions AI has achieved a healthy development in the application of power systems. While more mature technologies such as expert systems have been put into practical use, various intelligent technologies have been researched and explored. With the introduction of competition mechanism in China's power construction and power market, the increase of uncertainties and complexity of operation, the application prospect of AI in power systems will be even broader. â–¡
We have 2 types of Large Diameter Hose: Frac Hose and Layflat Hose.
Frac hoses were produced through-the-weave, easy cleaning and no drying required.
Designed for long service life. Tough and durable with exceptional resistance to abrasion
And cutting for use on wide variety of environment conditions.
Protected against mechanical damages.
Very low elongation (compared with vulcanized process) under pressure, prevents hose from [snaking".
High resistance to heat, fuels, chemicals, UV, ozone, weathering,hydrolysis,and microbiological attack.
Tube & cover: thermoplastic, polyurethane (TPU)
Reinforcement: High tenacity polyester filament, circular weaving.
The special extrusion through-the-weave process gives a very strong bonding between cover and lining as well as firmly encapsulation the woven polyester.
Temperature range from -50℃ to +70℃.
We can supply frac hoses with size: 8``,10``,12`` and 16``.
Large Diameter Hose
Large Diameter Hose,25Mm Reel Hose,32Mm Reel Hose,Large Diameter Rubber Hose
NANTONG SENTIAN FIRE-FIGHTING EQUIPMENT CO.,LTD. , http://www.firehosefactory.com