CN113344192B

CN113344192B - Enterprise-level motor system energy-saving optimization automatic control method and system

Info

Publication number: CN113344192B
Application number: CN202110602910.7A
Authority: CN
Inventors: 刘韧; 王根; 梁栋; 徐秉声; 周胜
Original assignee: China National Institute of Standardization
Current assignee: China National Institute of Standardization
Priority date: 2021-05-31
Filing date: 2021-05-31
Publication date: 2022-01-11
Anticipated expiration: 2041-05-31
Also published as: CN113344192A

Abstract

The invention provides an energy-saving optimization automatic control method and system for an enterprise-level motor system. The system comprises: and predicting the energy consumption data of the single discrete motor system by using the LSTM model with the improved structure. The energy-saving optimization automatic control method and the energy-saving optimization automatic control system for the enterprise-level motor system can save judgment logic judgment time, improve calculation efficiency and obtain energy efficiency and energy consumption condition prediction of each discrete motor system.

Description

Enterprise-level motor system energy-saving optimization automatic control method and system

Technical Field

The invention relates to the technical field of energy consumption management, in particular to an energy-saving optimization automatic control method and system for an enterprise-level motor system.

Background

1. Energy-saving background of motor system

The motor system is a whole composed of a driving motor, an electric control or speed regulation device, a transmission mechanism, a control (speed regulation) device, a dragged device and a system functional component. The system changes the electric energy into mechanical energy through the motor, and then realizes various functions through the dragged device doing work.

At present, the electricity consumption of motor systems in China accounts for about 60% of the total electricity consumption in China, wherein the electricity consumption of fans, pumps, compressors and air-conditioning refrigerators accounts for 10.4%, 20.9%, 9.4% and 6% of the electricity consumption in China respectively. Therefore, enterprise-level motor system optimization and energy matching are very important and are important means for improving the energy use efficiency in the industrial field. The electricity consumption of the motor system in the current industrial field accounts for 75% of the total electricity consumption of the national industry, however, the efficiency of the motor system in China is 10-30% lower than that of the motor system in the developed country, which is equivalent to the level of 70-80 years in the international 20 th century [ analysis of energy efficiency and market potential of the motor system in China, energy-saving project group of the motor system in China, mechanical industry publishing company ].

Energy conservation of motor systems has become a key to industrial energy conservation. The development trend of the motor in the future is to realize high efficiency and energy conservation, improve the efficiency, save resources and reduce environmental pollution. The energy conservation of a motor system is very important in China, the energy conservation of the motor system is listed in one of ten national energy-saving projects as early as 2008, and the application of a high-efficiency and super-efficient motor is listed in a benefit-for-the-people project in 2009.

After the paris agreement is signed, China pays more and more attention to energy conservation and emission reduction work. Actively participate in carbon emission reduction of the international society, actively conform to the global green low-carbon development trend, and the nation is deployed from the source, requires accelerating adjustment and optimization of industrial structures and energy structures, vigorously develops new energy, continues to fight against pollution prevention, attack and hardness, and the like. The method is to accelerate adjustment and optimization of an industrial structure and an energy structure, promote coal consumption to reach a peak as early as possible, vigorously develop new energy, accelerate construction of a national energy use right and carbon emission right trading market and perfect an energy consumption double-control system. Continuously drilling pollution prevention and hardness attack warfare to realize the synergistic effect of pollution reduction and carbon reduction. Large-scale homeland greening actions are carried out, and the carbon sink capacity of the ecological system is improved.

According to research, on the national level, the increase of industrial energy consumption of enterprise units with the scale of more than the current scale of China is reduced by more than 15% from 2019 of 2015, which is equivalent to saving 4.8 million tons of standard coal, and the energy cost is saved by about 4000 million yuan. The main problems are as follows:

(1) enterprises cannot mine energy-saving potential from the existing data, and have no national mandatory requirements and national subsidies, so that the enterprises are not active enough and have no power propulsion.

(2) The motor system has too many influencing factors, the energy-saving effect cannot be obtained through the direct feedback effect of a simple management means, and no means or method is developed.

(3) The existing energy-saving regulation and control means are too complex, unified management cannot be realized, and supervision on the equipment level cannot be realized.

For the reasons mentioned above, the energy consumption situation of enterprise level devices is not optimistic. The mainstream of the motor market in China is as follows: the JO2 series, the Y2 series and the YX series have efficiency far lower than that of products such as high-efficiency motors, ultra-high-efficiency motors and the like, such as permanent magnet motors, reluctance motors and the like, and the utilization rate of the high-efficiency motors is not more than 15%. Therefore, although the market potential of the existing energy-saving transformation or new high-efficiency motor system is huge, the promotion of the high-efficiency motor system and the energy-saving transformation are very important.

Through the prediction of energy-saving, not only can reduce resource consumption to the enterprise, the energy is rationally arranged, better row's production, and less discarded object introduces the carbon discharge capacity, can also the energy saving, and real novel industrialization road that lets the enterprise walk a low consumption, high benefit, low emission, high output.

2. Research background research method status

The enterprise-level energy consumption is directly related to the yield, and under the condition that the unit product uses certain energy, the energy consumption is increased when the yield is high; in addition, the requirements on equipment and energy distribution are met, otherwise, the energy consumption of unit products cannot be kept consistent; again, production equipment efficiency is optimally set. Therefore, the enterprise-level energy-saving optimization comprises three levels, firstly, the production requirements of enterprises are met, and the energy needs to meet the requirement of the yield in the month; secondly, the use of the equipment can meet the production requirement; again, the equipment needs to operate efficiently. In addition, energy consumption points in the case of changing the types of products produced by enterprises, the production in holidays, and the like are considered. Meanwhile, the enterprise-level production is necessarily a continuous time sequence process, and the method can cover the whole time sequence and has fault tolerance on special production conditions. According to the above needs, it is found that at least the following technologies are needed to be made a breakthrough in the aspect of energy-saving optimization automatic control of an enterprise-level motor system.

(1) A method for predicting and analyzing energy saving amount considering time sequence and fault tolerance;

(2) enterprise level distributed comprehensive energy consumption and analysis method;

(3) a data feedback platform under real-time monitoring and control conditions;

(4) real-time control according to multiple sets of discrete feedback signals.

3. Energy conservation prediction current situation

3.1. Energy consumption analysis and prediction method

The energy-saving prediction coordination situation of the enterprise level and the larger scale is used in different fields and different situations, and the energy-saving algorithm is analyzed by taking building and mine safety as an example. The reason for selecting the building energy consumption analysis is that the building energy consumption period is obvious, the external influence and the artificial influence are obvious, and a periodical long-time-period data screening and cleaning method is hopefully found by summarizing the building energy consumption analysis method and is used for energy-saving prediction; the mine safety analysis is selected because the mine safety analysis is not influenced by the outside, the main energy utilization is dominated by the change of the mine safety analysis, no human factors exist, the probability is high, and a method for predicting the emergent data is expected to be found through the analysis and summary of the mine safety.

3.1.1. Building energy efficiency and energy consumption analysis method

The building energy efficiency is that under the condition of guaranteeing the existing building functions, the building is guided to reasonably use energy by the existing standard specifications, the consumption of primary energy such as electric energy and the like in the actual use process is reduced, and the energy efficiency management aims to emphasize that the energy waste is reduced and the energy utilization efficiency is improved by means of technical management and the like [ Longqi, 2.0 thinking of building energy conservation [ J ]. heating ventilation air conditioner, 2016, 46 (08): 1-12]. At present, a common energy consumption rating method is used for making a reasonable energy consumption rating on the basis of the energy consumption level of a region where a building is located and the energy-saving potential of the building, so that the building is helped to evaluate the energy consumption condition and guide the development of energy-saving work. The building energy consumption quota is a quantity index of energy sources consumed by a building to normally complete the use function of the building in a quota period, which is determined by a certain scientific method, and aims to determine the energy consumption of a sample building, evaluate the energy consumption condition of a target building by taking the energy consumption condition as a reference and judge whether energy is saved or not. The uk has evaluated hospital building Energy consumption from 1989, provided free information to users to help them reduce Energy consumption, and issued Energy Performance in the Government's Civille estimate, which divided office Buildings into 4 classes, made 2 building benchmarks based on statistical data, and determined building Energy saving rates by calculating the ratios of factors affecting building Energy consumption [ Tian W, Choudhury R.A professional Energy modules for non-domestic building segments [ J ] Energy & building, 2012,54 (NOV.: 1-11], [ Yang UK. public building situation exploration [ J ]. 2009 08): 54-55]. The VDI3807 index system published in 1994 in Germany enables benchmark value calculation and evaluation of building energy consumption. The standard value and the guide value of energy consumption are formulated by classifying the building energy consumption, the rated value of single energy consumption is respectively calculated, and then the whole energy consumption rating value is determined, so that the fairness and reasonability of the energy consumption benchmark are ensured [ Cao Yong, Wei Zheng, Liu Zi, and the like. 78-81]. The united states introduced an Energy Star Energy consumption benchmark evaluation tool [ xu wei, zhou yu. public building Energy saving modification technical guideline [ M ]. beijing: chinese architecture industry press, 2010: 219-, [ Gale Boyd, Elizabeth Dual, wall Tunnel of the ENERGY STAR energy performance indicator for marking an industrial plant manufacturing using [ J ]. Journal of Cleaner reduction, 2008,16(6):709-715] was evaluated for building performance. According to the assessment method, the influence factors such as environmental meteorological parameters and working time are used as independent variables, the energy consumption intensity is used as dependent variables, a regression equation is fitted, the purpose of predicting the building energy consumption is achieved, and the actual energy consumption condition of the building is judged according to the predicted value. Japan also evaluated their energy conservation rate by establishing the energy consumption coefficient (CEC) of the air conditioning system [ zhang scholar, liu folk. 10-12], and judging the energy utilization efficiency of the building by comparing the actual total operating energy consumption of the air conditioning system all the year around with the accumulated value of the calculation load.

The method for formulating the public building energy consumption quota is earlier researched in Beijing area, and the energy consumption quota methods of America and Germany are respectively adopted [ Caoyong, Liu Yi Min, Yudan, and the like ]. 17-24+90], and carrying out quota evaluation on the energy utilization condition of the air conditioning system of the office building in Beijing area. The method comprises the following steps of carrying out investigation and statistics on the electric energy use condition of public buildings in Beijing City, determining an energy consumption quota calculation principle, and analyzing the rationality or limitation of the energy consumption quota [ Lichao, Schlemen, Liufei, and the like. 129-133]. The Qinghua university finds that the energy consumption distribution is approximately normal distribution by counting the building energy consumption data of the sample, calculates the average value and the standard deviation of the sample building as the energy consumption quota, and corrects the quota by using the service life. In other areas, the local quota value is determined by adopting a proper method, but different energy consumption quota can be obtained only by using different energy consumption indexes to evaluate the actual energy consumption of the building due to different building types and using functions. Research finds that the current classification mode for large public buildings cannot meet the establishment requirement of the energy consumption rating, and large public buildings need to be classified more carefully [ Huangshifeng. 50-51].

The energy consumption of the building is influenced by complex factors such as the building structure, the use function, the artificial activity and the space utilization condition of the building, the climate environment of the area where the building is located, the economic development and the like, so that the high-precision energy consumption prediction is more difficult. With the continuous development of modern science and technology and the continuous deepening of intelligent power grid research, the building power consumption prediction work has made a certain progress, and the existing building energy consumption prediction method can be summarized and divided into a traditional load prediction method based on a statistical method and a modern load prediction method based on artificial intelligence. The traditional power grid load prediction model is mostly established based on a linear method, in real life, building energy consumption is influenced by various factors such as temperature and humidity of the area, and meanwhile, a statistical method has the defects of low accuracy, poor flexibility and the like, so that the traditional linear load prediction method is complex in modeling and cannot meet the precision requirement, and the practical application is greatly limited. The modern load prediction method mainly comprises a grey mathematical theory [ Zhongpinjing, Yaojiangjust, Konghui, and the like ] medium and long term power load prediction [ J ] based on a multivariable time sequence inversion self-memory model, a power system and an automatic chemical report thereof, 2017, 29 (10): 98-105], support vector machine method (SVM) [ campsis, regxin, zheng hui, etc.. short-term wind power load prediction based on improved least squares support vector machine and prediction error correction [ J ]. power system protection and control, 2015, 43 (11): 63-69] [ Song Qiang, Yang Pu. short-term power load for implementation of vector machine and particle swarm optimization [ J ]. Journal of Algorithms & computer Technology,2018,13:1-8], Genetic Algorithms (Genetic Algorithm) and neural network theory (neural networks) [ Kongyu, Zhengfeng, Huizjun, etc.. short-term load prediction method based on depth belief networks [ J ]. power system automation, 2018, 42 (05): 133-139]. In the modern load prediction method, the development of the artificial neural network is the fastest, and the literature applies the artificial neural network and the integration method to a building cold load dynamic prediction model [ Lan Wang, Eric W.M.Lee, Richard K.K. yuen.novel dynamic forecasting model for building cooling loads combining an architectural neural network and an ensemble adaptive approach [ J ]. Applied Energy,2018,228:1740-1753], so that the method not only can be used for evaluating the dynamic prediction model, but also can ensure that the proposed prediction model can be immediately used for heating, ventilation and air conditioning systems. At present, a Back Propagation (BP) neural network prediction model is most widely applied, but a conventional BP neural network has a slow convergence rate and is easy to fall into local optimization; the selection of the number of network layers and the number of nerve cells mostly depends on the experience of researchers, and a feasible theoretical method is lacked; and the prediction factor is single, and various factors influencing the load change cannot be considered comprehensively, so that the accuracy is not high in practical application, and the prediction effect can be achieved by combining with other algorithms. However, the defect that the structure of the BP neural network is not easy to determine and is easy to fall into local optimization still causes adverse effects on a prediction result, and the Radial Basis Function (RBF) neural network has the advantages of simple structure, strong global approximation capability, high learning speed and the like, and can obtain a better prediction effect. But cannot take into account the effects of real-time varying loads and sudden events.

3.1.2. Mine safety analysis method

Mine safety mainly relates to gas emission and leakage, the requirement on an algorithm is high because burst points cannot be determined, and the analysis of discrete data is very important in enterprise-level energy consumption prediction and analysis, so that the type of the burst analysis is selected for analysis. In addition, it is considered that the gas burst is a complicated, continuously varying process. For example, factors such as gas pressure, gas content, coal seam depth, ground stress, and initial velocity of gas emission are important factors in the research of gas outburst [ Liu Yan, Chen Pan, Wei Jian Ping, control of coal seam geological structure on coal and gas outburst [ J ]. coal science and technology,2010, 38(1):24-27 ] ] [ segment east. coal and gas outburst influence factor and microseismic precursor analysis [ D ]. Shenyang: the northeast university, 2009 ], can better predict gas time series data only if the influence of the factors on the gas outburst is clearly understood. In recent years, time series prediction research of gas concentration has been greatly advanced, and a number of expert scholars adopt a method of combining a grey correlation theory with a traditional neural network, so as to make certain breakthrough [ hu guang qing, ginger wave, wuhu ] a coal and gas outburst prediction model based on a grey correlation-genetic neural network [ J ]. china coal geology, 2011,23 (09): 22-26.] [ warmth, in phoenix, shore arauca ] coal and gas outburst probability neural network prediction model based on grey correlation entropy [ J/OL ]. computer application studies, 2018 (11): 1-6.]. One of the major reasons is that a conventional neural network can infinitely approximate an arbitrary complex nonlinear function as long as the conventional neural network has enough samples. The mine safety analysis mainly has two directions, one is to improve a neural network model and improve the training precision and efficiency, and the other is to perform model comparison analysis on time sequence influence and analyze the safety under the condition of a practical sequence. Kourentzes [ Barrow D K, clone S, Kourentzes N.an evaluation of Neural network intensities and model selection for time series prediction [ C ]. Dallas: International Joint Conference on Neural networks IEEE, 2013: 1-8] an integrated neural network for time series prediction was proposed in 2013 and model selection evaluation was performed. A hybrid multi-objective evolutionary algorithm is provided for training and optimizing a recurrent neural network structure for time series prediction. Then a method of selecting a single prediction model from the Pareto solution set is proposed. Jin [ Smith C, Jin Y. evolution multi-objective generation of temporal network entities for time series prediction [ J ]. neuro-compressing, 2014,143 (16): 302-311 ] in 2014, a multi-objective cyclic neural network integrated evolution algorithm for time series prediction was proposed. The method is used for classifying the financial data of the Yahoo corporation, and the LLRBFNN model is learned by using a mixed technology of back propagation and recursive least square algorithm. Das et Al [ Al-Askar H, Hussain A J, Al-Jumerily D, et Al, regulated Dynamic Self organic Neural Network implanted by the Immune Algorithm for Financial Time Series Prediction [ J ]. neuro-compressing, 2016,188: 23-30 ] an adaptive local linear optimization radial basis function neural network model for financial time series prediction was proposed in 2015.

Currently, the prediction of gas time series in China is mainly divided into three categories: 1) the support vector machine and its evolution are used for time series prediction, but the short term is mainly used. 2) The method combines a genetic algorithm with a deep neural network, and avoids the problem that the deep neural network sinks partial minimum values in time series prediction. 3) By utilizing the chaos of the gas time sequence and according to theories such as phase space reconstruction and the like, the intrinsic relation among all factors influencing the gas time sequence is excavated, and a deep neural network and other methods are used as tools for prediction. In Guanghua [ Zhao jin Xiong, in Guanghua, chaos time sequence RBF neural network model for gas concentration prediction [ J ]. proceedings of Hill-Longjiang science and technology institute, 2010,20 (02): 131-134 ] in 2010, an RBF neural network model based on a chaotic time series was proposed. After the chaos of the gas time sequence is verified, an input label of the neural network is obtained by utilizing phase space reconstruction, and a prediction model is established. An application of a Zhuyu [ Zhuyu ] constructive neural network in gas time series prediction [ D ]. Taiyuan: the tai yuan university 2010 indicates that in the prediction process of the traditional time series, the time series data has a lot of data and has non-linear and fuzzy properties, so that the efficiency is low and the precision is low. George-english and the like [ george-english, maxiaoping, lanjiayi, royal jelly ] short-term gas prediction research based on weighted LS-SVM time series [ J ]. proceedings of mining and safety engineering, 2011,28 (02): 310-314, a short-term time series prediction of gas under an LS-SVM model is provided in 2011, and the LS-SVM can be used for predicting the gas time series aiming at the problem that a traditional neural network model is heavy and easily falls into a local minimum value. Wanghe et al [ wanghe, shao liang fir, qiu yunfei ] coal mine gas content prediction model based on ant colony neural network [ J ] microcomputer information, 2011,27 (05): 197 + 198. ] in 2011, a neural network model based on the ant colony algorithm was proposed. The problem that the neural network falls into a local minimum value in the gradient descending process is effectively avoided. Liuyi jun, etc. (liuyi jun, zhao qiang, huwenli. prediction research of gas concentration based on genetic algorithm optimization BP neural network [ J ] mining safety and environmental protection, 2015,42 (02): 56-60, the prediction of the GAs concentration by combining a genetic algorithm and a BP neural network is provided in 2015, a GA-BP algorithm is established, and experiments show that the algorithm has higher precision and stability.

At present, no matter the process industry or the discrete manufacturing industry is oriented, research on energy efficient manufacturing mainly focuses on three aspects of energy consumption system modeling, energy efficiency analysis and evaluation and energy system optimization scheduling, and certain research results are obtained. From the research method, the early-stage research is based on a mechanism model, the global and system modeling and analysis of the whole manufacturing life process are lacked, and the research result is also restricted by the limitation of the model. The difficulty in modeling the mechanism has gained common consensus among researchers with increasing complexity of discrete manufacturing processes. However, it is worth noting that the development of industrial informatization, data acquisition and storage technologies has accumulated a huge amount of mass data for enterprises related to manufacturing processes, energy systems, equipment status, product Information, decision management, etc. [ Zhengli, Jiangyu, Qiaorhong, etc. ] mechanical engineering reports, 2010,46(21): 124-. Provides convenience for the research of energy consumption and energy efficiency of the manufacturing system.

3.2.1. Small knot

Through the analysis, the particularity of enterprise-level energy consumption analysis is considered, and the analysis of a discrete system is utilized to carry out comprehensive analysis aiming at different running conditions of different systems.

3.3. Automatic control of the present situation of research

In view of the actual situation in the field of automatic control in China at present, although the automatic control technology has been developed greatly and is widely applied in practice, the automatic control technology has a great gap with the technical level and the application degree of automatic control in developed countries abroad. The level of automatic control technology is improved in China, the investment and scientific research strength must be increased, the automatic design and the future development presetting of a novel production line are scientifically and reasonably carried out, the function of automatic information flow is particularly emphasized, the automatic control level and the application of China are improved, and the international competitiveness of enterprises in China is further improved.

From the effect of the automatic control technology in the application field in China at present, the method mainly aims to improve the operation efficiency of equipment. According to the specific condition of the development of China, the automatic control technology is developed, so that the blindness of developing the automatic control technology is avoided. However, there still exists a phenomenon that an automatic control technology lacks clear guidance on a macroscopic level in the research and development process, and economic benefits obtained in actual production are relatively low, and there also exists a phenomenon that accuracy is relatively poor, reliability is relatively low, and practicability is relatively poor on automatic equipment independently researched and developed in China. With the gradual loss of the dominance of the manual manufacturing industry in the national economic construction, the automatic production increasingly shows the characteristics of simple production operation, high product quality, high production efficiency and the like in the social production, and becomes a main mode in the enterprise production. The development of automatic control technology in China is also very unbalanced, and the automation degree of most production fields is very low, such as toys, clothes and the like.

The improvement of the automatic control level is not easy in China, so that the research and the development of a new automatic control technology are needed, the automatic modification of the production equipment of the original enterprise is needed, and the production efficiency can be improved and the cost can be reduced. Original mechanical equipment can be modified through automatic control technologies such as a numerical control technology and the like, and the automation degree of the traditional mechanical equipment is improved, so that the utilization rate and the production rate of the equipment are improved. The advantages of the computer technology are fully exerted through the improvement of the control technology on the machine tool, the automatic improvement of equipment and a production line is realized, and therefore the production efficiency is improved. The future development direction is necessarily the integration of intellectualization, networking, miniaturization and each layer.

3.3.1. Development of intelligent automatic control technology

The development of the automatic control technology level is the power and the basic force which are continuously promoted by the modern production, and in the initial stage of the automatic production, a control system is simpler and the control rule is also simpler, so that the operation can be finished by adopting a conventional control method. The intellectualization is a higher level of the development of an automatic control technology, the intellectualization mainly reflects in the function diversification and the use diversification of the control, and the intellectualization is the development direction of the future manufacturing industry. With the continuous progress of science and technology, the development direction of modern production gradually trends to the combined application of artificial intelligence and automatic control technology. The penetration of the artificial intelligence theory into the field of automatic control technology is a new development approach not only in theory but also in practice, and provides a new idea and method for the intelligent automatic control technology.

The artificial intelligence is combined with the automatic control technology, and more effective control can be adopted for the system according to the change situation in the production process. At present, the intelligent control technology is applied to production systems in many production fields, and the level and the application degree of the intelligent control technology relate to the level and the degree of automation of modern production of enterprises.

3.3.2. Development of networked and miniaturized automatic control technology

In view of the development history of automatic control technology, the automatic control technology is performed in the industrial production field for a long time. The automatic control technology provides various mechanical equipment required by industrial production with control equipment with very high reliability and performance. At the moment of rapid development of science and technology, the fields are not independently developed, but are mutually consulted to promote and even combine development to become a new development field. The development of automatic control technology certainly does not leave the way to other fields, where the impact from industrial PCs is the most severe. Networking and miniaturization are inevitable trends in future development of automatic control technology, and in the initial stage of development of automatic control technology systems, the form is very large and the price is very high. The direction of future development of automatic control technology is bound to leave with networking, and the networking technology plays an important role in modern production. Especially, the method plays a key role in the transmission and analysis of information data in the production process, and plays an effective role in taking reasonable treatment measures for the automatic control system to find safety problems, preventing faults and the like. With the continuous progress of science and technology, the development has changed much more than before, and the development is towards miniaturization and the price is gradually reduced. With the further improvement and development of the control software of the automatic control system, the market share of the control system software capable of being installed will gradually increase in the future.

3.3.3. Development of integrated automatic control technology

The development directions of control technologies such as fuzzy control, intelligent control and expert system are established in the field of modern automatic control technology, and the main characteristics of the direction automatic control technology are comprehensiveness. These control systems with special directivity are based on the theory of automatic control technology, so as to comprehensively control the whole equipment or process. The related theoretical knowledge is more, and the knowledge is not single automatic control technical knowledge, and also comprises an electronic technology, a computer technology, a mechanical technology and the like. In order to adapt to and promote social progress, the automatic control technology must be combined with related technologies to develop a new direction, so that fresh nutrients and vitality can be injected into the field of the automatic control technology, and the reliability, accuracy and efficiency of the automatic control technology can be improved. Various automatic control technologies, for example, basic technologies of automatic control technologies of various control systems, dedicated computers, and the like, are continuously developed, and new knowledge, new theory, and new technology in various fields are continuously introduced.

3.3.4. Small knot

Through the analysis, the problem of enterprise-level energy management at present needs to be solved urgently, energy consumption prediction and energy management are not simple, and further, the current system is expected to be controlled through the existing data and analysis results so as to achieve the effects of energy conservation and emission reduction. Meanwhile, with the development of industrial internet, the realization of energy conservation and emission reduction by utilizing the internet to carry out intelligent and networked management and adjustment becomes a first requirement.

4. Problems and deficiencies

In summary, although an advanced artificial intelligence algorithm is used for calculation, the energy efficiency prediction and calculation method for an enterprise-level discrete motor system is still incomplete, and the application is relatively few; the existing artificial intelligence method and statistical method are too simple, because the problems of algorithm non-convergence and the like caused by excessive parameters are considered, and the statistical method has the problems of inaccuracy and the like; in the existing method, different energy utilization units in the same enterprise cannot be integrated and analyzed simply through data statistics and other methods; although the number of the platform is large, most of the platform is used for collecting data and providing no feedback control or providing no control on some system, and no suggestion is provided for production and energy distribution on the feedback of a prediction result; the control system is also complex and not completely intelligent. The foregoing is a part of the problems and deficiencies of enterprise-level energy efficiency analysis that may be involved herein.

Disclosure of Invention

The technical problem to be solved by the invention is to provide an energy-saving optimization automatic control method and system for an enterprise-level motor system, which can save judgment logic judgment time, improve calculation efficiency and obtain energy efficiency and energy consumption condition prediction of each discrete motor system.

In order to solve the technical problem, the invention provides an energy-saving optimization automatic control method for an enterprise-level motor system, which comprises the following steps: and predicting the energy consumption data of the single discrete motor system by using the LSTM model with the improved structure.

In some embodiments, in the hidden layer unit of the LSTM model with improved structure, the computing node state is optimized by:

c_t＝tanh(W_c(h_t-1,x_t)+b_c)

c_t＝c_t-1f_t+i_tc_t

wherein, c_tTo compute node states, W_cIs a matrix of weight coefficients, b_cIs an offset vector, f_tTo activate the function, t denotes the time instant.

In some embodiments, the input data for the improved structured LSTM model includes: historical energy consumption data, product factors, single motor system influence factors, motor single machine influence factors and time factors.

In some embodiments, further comprising: before the LSTM model with the improved structure is used for predicting the energy consumption data of a single discrete motor system, the training data is optimized by using a particle swarm algorithm, and the LSTM with the improved structure is trained by using the optimized training data.

In some embodiments, further comprising: after the LSTM model with the improved structure is used for predicting the energy consumption data of the single discrete motor system, the energy consumption data of the single discrete motor system is weighted and averaged, and the predicted value of the total energy consumption of the enterprise-level motor system is obtained.

In some embodiments, further comprising: and (4) calculating the influence degree of the ith motor system and a corresponding certain device on the total energy consumption by using the BP neural network.

In addition, the invention also provides an energy-saving optimization automatic control system of an enterprise-level motor system, which comprises: one or more processors; a storage device for storing one or more programs which, when executed by the one or more processors, cause the one or more processors to implement the enterprise-level motor system energy-saving optimization automation control method according to the foregoing.

After adopting such design, the invention has at least the following advantages:

and (3) considering the time sequence by using an LSTM model, modifying a hidden layer in order to improve the algorithm efficiency, setting a memory unit for each unit to update the calculation result, the input of the previous stage and the input of the previous layer, saving judgment logic judgment time, improving the calculation efficiency and obtaining the energy efficiency and energy consumption condition prediction of each discrete motor system.

Drawings

The foregoing is only an overview of the technical solutions of the present invention, and in order to make the technical solutions of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and the detailed description.

FIG. 1 is an enterprise level energy consumption hierarchy and influencing factors;

FIG. 2 is an enterprise level energy relationship diagram;

FIG. 3 is motor plant energy consumption;

FIG. 4 is a schematic diagram of discrete motor system energy consumption;

FIG. 5 illustrates the energy consumption of an enterprise-level motor system and the effect of various components on energy consumption;

FIG. 6 is a schematic illustration of a technical solution idea;

FIG. 7 is a theoretical plot of an LSTM model;

FIG. 8 is a diagram of an algorithm model for LSTM timing consideration;

FIG. 9 is a diagram of an algorithm model for LSTM timing considerations modified based on discrete system characteristics;

FIG. 10 is a schematic diagram of the ith discrete motor system prediction;

FIG. 11 is a flow chart of a predicted energy consumption of a single discrete electric machine system;

FIG. 12 is an enterprise-level motor system impact factor prediction;

FIG. 13 is a flow chart of enterprise level energy consumption reverse impact factor analysis;

FIG. 14 is an enterprise-level motor system energy consumption data acquisition network deployment;

FIG. 15 is a data acquisition and architecture diagram.

Detailed Description

The preferred embodiments of the present invention will be described in conjunction with the accompanying drawings, and it will be understood that they are described herein for the purpose of illustration and explanation and not limitation.

1. Application and implementation potential

1.1. Application scenarios

Aiming at discrete motor systems in enterprises, energy consumption data is predicted by using an improved machine learning method and fed back to the system, and the discrete motor systems are controlled by using industrial internet equipment to realize enterprise-level multi-motor cooperative work and realize efficient energy utilization.

1.2. Energy saving and emission reduction potential of enterprise-level motor system

The installed capacity of various motor systems existing nationwide is about 17 hundred million kw. The electricity consumption of the whole motor system accounts for more than 60% of the national electricity consumption. Wherein the power consumption of the fan, the pumps, the compressor and the air-conditioning refrigerator is 10.4%, 20.9%, 9.4% and 6%. Compared with the advanced foreign level, the manufacturing technology and the process of the motor have certain difference, and the difference of the transmission speed regulation and the system control technology of the motor is larger. Therefore, China has huge energy-saving potential in the aspects of improving the efficiency of the motor system and strengthening the energy-saving management of the system.

The state is accelerating to adjust and optimize industrial structures and energy structures, improving the energy utilization efficiency, reducing emission and promoting coal consumption to reach the peak as early as possible. The use of the invention inevitably provides good technical support for the energy efficiency and energy consumption dual control policy, realizes the pollution and carbon reduction synergistic effect, and helps enterprises to complete the basic foundation support of energy conservation and carbon reduction.

FIG. 1 illustrates enterprise level energy consumption hierarchy and influencing factors.

Enterprise-level energy consumption means that a plurality of devices are used in a certain process or a certain production stage after being connected in series or in parallel, or a motor system for providing power for the whole workshop and an enterprise, and the matching of the plurality of devices is also considered based on the energy consumption of the devices.

The comprehensive energy consumption of the discrete systems directly influences the energy consumption of enterprises, meanwhile, the practical matching of different discrete systems can indirectly influence the energy consumption of enterprises, and in addition, the yield, the production time, the product technology and the product type all influence the energy consumption of enterprises.

In the aspect of equipment-level energy consumption, the energy consumption conditions of equipment levels such as a motor, a frequency converter, dragging equipment and the like have many influencing factors and are relatively complex in source.

2. Innovation point of the invention

Innovation points 1: long-short period machine learning algorithm with timing sequence considered in improvement

Because the production schedule of an enterprise has periodicity, and production can be adjusted according to product types and yield in different time periods, the algorithm improvement is not excessively sensitive to the production energy consumption of products in different periods; secondly, the production energy consumption prediction of an enterprise needs to consider a time sequence, the time sequence needs to be efficiently matched and data converged due to long term and short term, and an LSTM time sequence model is selected for correction to obtain a model meeting the requirements finally; because the data input quantity is large, unreasonable data needs to be cleaned by an efficient method in the data screening process for efficient calculation and real-time feedback of a convergence result to a signal, so that preparation is made for subsequent calculation. The whole process also allows for integration and distribution of different motor system data. The method is not seen in a machine algorithm in the industrial field, and data verification shows that the data convergence is fast and the prediction result is accurate.

Innovation points 2: energy-saving accurate prediction of enterprise-level discrete motor system

Through the machine learning algorithm, the trained model is used for predicting the performance of a single system on the data acquired in real time, and then the whole plurality of discrete motor systems are integrated to predict and analyze the enterprise-level performance. The method is characterized in that the energy consumption of a single motor or a unit is predicted, a plurality of discrete systems are integrated, a formed enterprise-level energy consumption prediction scheme is not seen in relevant papers of the field of good prediction, and a result is accurate through small-scale data verification.

Innovation points 3: enterprise-level-based multi-motor system cooperative control

According to the production priority, the system requirements and the self attributes, the predicted data are compared with the discrete single information influence factors, the optimization scheme in the enterprise-level system is analyzed, and the multiple motor systems are cooperatively controlled in a frequency converter adjusting mode.

3. Technical difficulties of the invention

3.1. Modeling method and energy consumption prediction

Because the input parameters are more, the statistical relevance is smaller, and the relevance comprises the relevance of different levels such as an enterprise level, a system level, an equipment level and the like, a machine learning method is selected; the algorithm has high convergence difficulty due to the fact that the data are large and the relevance is poor, and meanwhile, the requirement on the calculation speed is high due to the fact that the data are required to be controlled in real time; according to different types of products produced by enterprises, the calculation period is different in length, different changes of the period are solved, different data types are analyzed and processed again, and machine learning modeling which can be adapted to both long and short periods is realized.

3.2. System cooperative control

Many theories exist in cooperative control, but the optimal energy efficiency performance can be realized after cooperative control, the predicted value of the initially set energy is met, the coordination is very difficult, parameters of a motor single system and parameters among various discrete systems need to be coordinated, the optimal output can be ensured under the condition of stable input, and the effects of energy conservation and emission reduction are realized. In addition, according to research and study, most of the existing enterprises have speed regulation and frequency modulation devices such as frequency conversion devices, the internet of things protocol is used for directly acting on the frequency converter after feedback information is obtained, and the parameter adjustment of the frequency converter is used for meeting the synergistic effect of a plurality of discrete motor systems, so that the method is one of the difficulties in the text.

4. Technical scheme of the invention

4.1. Design concept

Energy consumption data of each key part of the motor system are collected on site, and the received data are transmitted back to an energy consumption database in a wired/wireless mode; or reading corresponding data through an energy management system of an enterprise to serve as a data set of the training model.

And predicting enterprise-level data by using the energy consumption condition of the discrete motor system, comparing the existing energy consumption with the predicted energy consumption, if the energy consumption is lower than the predicted value, optimizing by using an ant colony algorithm and carrying out global optimization and system matching by using the energy consumption value of the discrete motor system and equipment layer data, judging control parameters and feeding back the control parameters to the frequency converter, and acting the frequency converter on the corresponding discrete system to realize an optimization control strategy.

From the above analysis, it can be seen that the discrete manufacturing system is a complex and large system, and extends to every workshop of the manufacturing industry, such as the building mentioned above, during which the processes of transportation, processing, assembly and the like of parts are carried out, and also the discharge of various wastes and the transmission of various information. The energy consumption of the enterprise-level motor system is characterized by more energy consumption equipment, relatively single energy consumption type and large energy consumption difference. According to the scale, the production scheduling condition and the product difference of a workshop, the used motor monomers or equipment comprising power supply and distribution, pipe network output and the like distributed according to units are very many, the energy consumption is mainly electric energy, stable data of the electric energy can be acquired by investigating and researching some enterprises, and the quality of the electric energy can be influenced by using a large number of frequency converters. Therefore, the popularization to the whole enterprise level is more complicated, and how to coordinate different workshops, different motor systems of workshop workers and even different devices in the systems are very difficult. Therefore, the discrete manufacturing system is a complex organic whole under the strong coupling effect of information such as energy and management [ infancy ] mechanical manufacturing system carbon flow dynamic model based on first-order hybrid Petri net and application [ D ] [ Master academic thesis ]. Chongqing: Chongqing university, 2012 ], and a simple schematic diagram is shown in FIG. 2.

The device layer energy consumption is as in figure 3.

The discrete motor system is formed by connecting a plurality of units in series or in parallel, partly because the normal production needs to be used for multiple purposes, and partly because the larger output needs to be provided, so that the energy consumption analysis diagram is shown in figure 4.

Referring to fig. 4, the energy consumption of the discrete motor system is composed of the energy consumption of each of the different units together. And the energy consumption of each unit specifically comprises: power transmission and distribution loss, frequency converter loss, start-stop loss, switching loss, motor body loss, transmission system loss, damping loss and friction loss.

The enterprise level energy consumption analysis diagram is shown in fig. 5.

Referring to fig. 5, the energy consumption of the enterprise-level motor system is composed of the energy consumption of different workshops together. And the energy consumption of each workshop specifically comprises production energy consumption and operation energy consumption. The operation energy consumption is influenced by various factors such as operation time, yield and product type. The production energy consumption is related to the number of devices driven by the running motor in the generation process of each workshop.

4.2. Technical scheme

The technical scheme is divided into 3 layers, namely an equipment level, a discrete motor system level and an enterprise level. Referring to fig. 6, energy consumption is brought by operation of the equipment and is reflected to the motor system in a centralized manner, and energy consumption of a plurality of discrete systems is reflected to the energy consumption of the motor system of the whole enterprise in a centralized manner. The following are respectively explained from model equipment and discrete system energy consumption prediction model establishment, enterprise-level energy consumption model establishment and control system schemes.

4.2.1. Time series prediction model considering long and short periods and emergency

Early artificial neural networks integrated abstract data to build a model with neurons as units, and repeated revisions of weights between nodes to obtain a final prediction result, although accurate prediction can be performed for multi-hidden-layer complex problems, time series cannot be considered. In order to solve the defects that the traditional neural network model is inflexible in structure, insensitive to time series data, easy to fall into local minimum and the like, the LSTM model is used as a basic model for model reconstruction. The LSTM model is shown in fig. 7.

As can be seen in FIG. 7, the LSTM cell itself contains one storage cell to store long-term information, while utilizing three logical gating cells: the input gate, the output gate and the forgetting gate are used for controlling the data flow. The logic gate units are independent, and do not transmit the self-behaviors to other neurons, but are responsible for the memory module part of the whole network to modify the magnitude of the weight at the edge. In the calculation process, the function of a neuron is recorded based on the Cell state, meanwhile, gate updating information is input, a forgetting gate selectively forgets information irrelevant to updating content, an output gate stores the information of the last time point into a hidden layer, and result prediction is carried out when the information is output in the calculation. As can be seen by the description, this method takes time into account, but is computationally complex and slow. To ensure that the computation of a single cell can be accelerated, we consider simplifying the interior of the hidden layer.

For a given time sequence x ═ x (x)₁,x₂,…,x_n) After applying standard LSTM iteration, the hidden layer sequence h ═ (h) can be calculated₁,h₂,…,h_n) And the output sequence y ═ y (y)₁,y₂,…,y_n) Respectively as follows:

h_t＝f_n(W_xhx_t+W_hhx_t-1+b_h)

y_t＝W_hyh_t+b_y

wherein: w is a weight coefficient matrix, b is an offset vector, f_nFor the activation function, the index t is the time instant.

Entry of the door status: i.e. i_t＝σ(W_xix_t+W_hih_t-1+W_cic_i-1+b_i)

Forget the door state: f. of_t＝σ(W_xfx_t+W_hfh_t-1+W_cfc_t-1+b_f)

Cell state: c. C_t＝f_tc_t-1+i_t-1tanh(W_xcx_t+W_hch_t-1+b_c)

h_t＝o_ttanh(c_t)

The corresponding state is as in fig. 8.

In order to ensure the calculation speed, the states of the individual gates are adjusted, the data are transmitted to the maximum extent while taking into account the remaining portions, the cell state is optimized in the following manner,

c_t＝tanh(W_c(h_t-1,x_t)+b_c)

c_t＝c_t-1f_t+i_tc_t

according to the formula, the current cell state is updated by utilizing the previous neuron to the maximum extent according to the characteristics of the discrete motor system, and the operation amount is reduced. The operation stability of the discrete motor at a short moment changes and updates the state mainly depending on the influence of the previous moment, so that the information is amplified in a single operation and transmitted as far as possible. This form of cell state corresponds to the state diagram shown in FIG. 9, which is clearly much simpler to see.

Along with the change in cell state, the expression of the other gates evolved as:

i_t＝σ(W_xi(h_t-1,x_t)+b_i)

f_t＝σ(W_xf(h_t-1,x_f)+b_f)

o_t＝σ(W_xo(h_t-1,x_o)+b_o)

the transfer process also uses the tanh function, so h_tAnd is not changed. It can also be seen from the expression that the number of times of accumulation and multiplication is reduced after improvement, and the outputThe output part selects and reserves the state of the input gate and the state of the cell, and accumulates the part forgotten gate through a sigma function to meet the requirement of transmitting the related information downwards, thereby improving the efficiency of each calculation.

The above is a modification to the algorithm portion, and the modified discrete system computing structure is shown in fig. 10.

Referring to fig. 10, the discrete data of the ith discrete motor system at the time before t is input to the modified discrete system LSTM model. These discrete data may be: historical energy consumption data, production requirements/product requirements, individual motor system impact factors (power, frequency converter, load capacity, load rate, efficiency), motor stand-alone impact factors (current, electricity, three-phase imbalance, efficiency, etc.), time factors (production duration, rework duration, etc.).

And inputting the discrete data into the corrected LSTM model of the discrete system, and predicting the energy consumption and the yield of the ith discrete motor system of the model at the time t.

Meanwhile, the input data sources in the level comprise yield data, energy consumption data, time data, motor body data, motor system data and the like of enterprises, the data are not directly related, the input quantity is large, and in order to ensure that the best result can be found in LSTM calculation at the fastest speed, a particle swarm algorithm is used for classifying and cleaning the data to obtain the optimal data in the same class, and then the optimized data are transmitted into a model to be solved in space. The optimizing clustering mode aims at realizing the lowest energy consumption value by using each mode to a target function of clustering analysis, and the mathematical model is expressed as follows:

in the formula, x_ipP-th attribute representing that i-th variable affects energy efficiency, c_jpThe p attribute which is the j domain center;

performing local and global search in the domain according to the expression, calculating a clustering center and an objective function value for each variable, transforming a result set, and using a solution set for updating the pheromone matrix, wherein the solution set comprises the following steps:

τ_ij(t+n)＝(1-ρ)τ_ij(t)+Δτ_ij

the algorithm flow diagram for the discrete motor system is shown in fig. 11.

Referring to fig. 11, the algorithm flow of the discrete motor system includes the following operational steps: acquiring data and executing data preprocessing; searching class center points by K-Mean clustering; initializing data information; performing local optimization of the ant colony algorithm and optimizing parameter information; global optimization is carried out, and data accuracy is improved; training an LSTM network; judging whether the precision requirement/the maximum iteration number is met, if not, carrying out global optimization again; obtaining a default optimized network model based on the existing data; referring to the test data set, and checking the existing data network model; performing data de-normalization; and taking the discrete motor system data as an enterprise-level energy consumption input.

4.2.2 integration and Enterprise-level energy consumption of discrete systems

With the discrete data of each motor system, the motor systems of the whole enterprise are integrated. The partial input becomes single, and the enterprise-level energy consumption is the energy consumption prediction result of each discrete system:

wherein n is the number of discrete systems, w_iIs the energy consumption of the i-th motor system, t_iThe service life of the ith motor system is long, and W is the total energy consumption of the enterprise motor system.

According to the training result of the previous model, a plurality of influence factors can act on the energy consumption, the existing energy consumption is used as the model input in the part, the LTSM model is reversely used for operation, and the influence degree of the ith motor system and a corresponding certain device on the energy consumption can be known. The integrated process diagram is shown in fig. 12 corresponding to the LSTM process.

Referring to FIG. 12, first, different discrete data are input to the LSTM model for energy consumption prediction. And then performing prediction of total energy consumption of the enterprise-level motor system. And estimating the influence rate of each discrete system by using the BP reverse network. The influence rate includes: the influence rate of the motor unit parameters, the influence rate of the motor body parameters and the influence rate of other influence factors.

In this process we use the BP back-propagation algorithm after passing LSTM, which is generated after the inputs are activated by weighted summation and activation functions. Since the output value and the actual value will generate an error, the error is propagated reversely by means of partial derivative (gradient), and the partial derivative is called a residual error. And correcting the weight of each layer connection through residual errors, and continuously iterating until the result is satisfied. The BP back propagation algorithm is based on LSTM data input, linking energy consumption results and influencing factors.

This partial flow chart is shown in fig. 13.

Referring to fig. 13, the process of enterprise level energy consumption reverse impact factor analysis includes the following steps: inputting a predicted value of energy consumption of a discrete motor system, and performing data preprocessing; initializing data information; the LTSM model integrates and calculates the total energy consumption of a plurality of discrete systems; judging whether the energy consumption requirements of enterprises are met; taking enterprise energy consumption requirements, the existing energy consumption indexes, the number of discrete systems and important parameters as input; calculating the existing LSTM model by adopting a BP reverse network; global optimization is carried out, and data accuracy is improved; judging whether the precision requirement and the maximum iteration number are met; obtaining the influence rate of a discrete system; selecting the main parameters of the system with the highest influence rate as input; BP reverse network calculation; and obtaining the influence rate of important equipment.

4.2.3 implementation of real-time control System

From the above algorithm, the energy consumption influence of each discrete system is obtained after passing through the LSTM, and the BP inverse algorithm is used to obtain the magnitude of the energy consumption influence of each component inside each discrete system, that is, the influence of the energy consumption influence from the minimum unit to the discrete system to the integrated system is obtained. At this time, according to the production requirements of enterprises, for example: and (4) controlling the integrated system, the discrete system and the components according to the influence factors. In this section, the product type of the enterprise and the working system of the enterprise and the system need to be considered, so that the actual situation can be better simulated, and the real-time control can be carried out. The back propagation is also time-based, so that the memory of the sequence needs to be preserved, and the matrix gradient change of each layer is consistent.

For such situations, the following requirements are satisfied when taking values:

(1) the data is discrete but continuous. When the product type is switched or different work systems such as rest are considered, the data are included in the data input process, otherwise, the data are discontinuous and the prediction fails when the situation is met for the first time in the machine learning process. I.e. the loss function C is to be the partial derivative of the error and weight matrix at each time point and the gradient is to be decreased after the residual transfer is satisfied.

(2) The data is input by a particle swarm algorithm. The particle swarm algorithm can distinguish different classes and search the optimal values of the different classes, but if local and global optimization cannot be successfully found in the particle swarm algorithm, the data needs to be cleaned again. The method is used for processing a singular point at the time t in a time series data through a moving average line method. Since these data must be included from real production data, too much difference in magnitude or index will result in misconvergence and outliers in order to get them togetherTaking into account that x is used_averThe singular values are replaced, so that the singular values not only contain individual information, but also contain information of local and global data, and the influence of the singular values can be gradually reflected in the calculation process.

Where b represents the time of the anomalous sampling point.

If the enterprise is shut down or holidays, time still exists as input, but the energy consumption value of the discrete motor system is not input, the data cannot be simply skipped, the result prediction is not considered to be inaccurate, and the data is processed through exponential smoothing. At x_tThe data set has missing data and is smoothed as follows.

Where step is the step value, x_t+stepIndicates the predicted value of the missing part, a_tb_tc_tRespectively, the smoothing parameters, the derivatives of different orders in the formula represent different smoothing values, and alpha is a weight function.

Generally, the data volume of an enterprise level is very large, the method generally requires iteration after a data set is scanned and repeated after calculation of different levels and then comprehensive calculation, and usually, thousands of iterations are caused under the condition that data is complete, and the efficiency is greatly influenced by the huge data volume. By the method, the ant colony algorithm clustering analysis is used, the initial value is well screened, and the efficiency is greatly improved. The flow chart is as above.

4.2.4 Enterprise-level motor system energy consumption acquisition and deployment

The energy consumption analysis of an equipment layer, a discrete system layer and an enterprise layer is used for determining the acquisition of information in various aspects such as equipment, tasks, processes, auxiliary production and the like, so that the whole enterprise can complete the conditions of digital fine adjustment.

Referring to fig. 14, the overall scheme herein includes three levels of acquisition, communication and data processing, which respectively correspond to hardware data acquisition, discrete system data communication and network transmission, background database data processing, and adopt a distributed processing and centralized monitoring mode. The enterprise is converted from decentralized management into networked and intelligent control, and meanwhile, the acquired data can be directly used for intelligent control and can also provide a data basis for professionals and managers.

Referring to fig. 15, the feeding and device acquisition for the terminal or for the discrete motor system is performed by a hand-held or smart device.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the present invention in any way, and it will be apparent to those skilled in the art that the above description of the present invention can be applied to various modifications, equivalent variations or modifications without departing from the spirit and scope of the present invention.

Claims

1. An energy-saving optimization automatic control method for an enterprise-level motor system is characterized by comprising the following steps:

predicting the energy consumption data of a single discrete motor system by using an LSTM model with an improved structure;

in a hidden layer unit of the LSTM model with an improved structure, the state of a computing node is optimized in the following mode:

c_t＝tanh(W_c(h_t-1,x_t)+b_c)

c_t＝c_t-1f_t+i_tc_t

wherein, c_tTo compute node states, W_cIs a matrix of weight coefficients, b_cIs an offset vector, f_tT represents the time instant for activating the function;

further comprising:

after the LSTM model with the improved structure is used for predicting the energy consumption data of a single discrete motor system, the energy consumption data of the single discrete motor system is weighted and averaged to obtain the predicted value of the total energy consumption of the enterprise-level motor system;

further comprising:

and (4) calculating the influence degree of the ith motor system and a corresponding certain device on the total energy consumption by using the BP neural network.

2. The method of claim 1, wherein the input data of the LSTM model of the improved structure comprises: historical energy consumption data, product factors, single motor system influence factors, motor single machine influence factors and time factors.

3. The enterprise-level motor system energy-saving optimization automatic control method according to claim 1, further comprising:

before the LSTM model with the improved structure is used for predicting the energy consumption data of a single discrete motor system, the training data is optimized by using a particle swarm algorithm, and the LSTM with the improved structure is trained by using the optimized training data.

4. An energy-saving optimization automatic control system for an enterprise-level motor system is characterized by comprising:

one or more processors;

a storage device for storing one or more programs,

when executed by the one or more processors, cause the one or more processors to implement the enterprise-level motor system energy conservation optimization automation control method of any one of claims 1 to 3.