JP6930592B2 - Warning, Irregular, Unfavorable Mode Predictors and Methods - Google Patents

Description

The present invention predicts warnings, irregularities, and unfavorable modes (eg, unit overheating, fuel consumption, malfunction due to electrical or thermal stress, failure due to aging of parts exposed to permanent burden). It is generally associated with management systems such as energy management systems and technical systems with moderate or high complexity that exhibit dynamic behavior over time and significantly deviate from normal operating mode.
In particular, the present invention relates to predictors and methods of warning, irregular, and unfavorable modes in technical systems.

Since a complex technology system is typically operated by a monitoring and control system, the system can trigger a number of warnings, such as warning signals indicating the occurrence of danger or imminent danger. Appropriate warning management is required for multiple warnings. In general, danger means operation in an inconvenient mode of operation, risk of damage due to inability to provide contract services to customers, or breach of existing rules, loss of equipment, harm to people, risk of financial loss. Mechanisms (systems and devices) that can anticipate warnings (eg, irregular or unfavorable modes) and provide suggestions and recommendations for measures to prevent alert situations are beneficial for proper warning management.

Condition monitoring has been established in the technical field. Certain parameters are monitored or estimated to determine the condition of the machine, engine, or plant. As disclosed in Non-Patent Document 1 entitled "Improved Correlation-Based Anomaly Detection Method for Condition Monitoring of Industrial Equipment Data", the detection mechanism (correlation-based detection mechanism) is capable of anomalies in industrial equipment. The sex can be detected by the worker. Also, as disclosed in Non-Patent Document 2 entitled "Introduction to Predictive Maintenance", predictive maintenance can be applied to prevent unnecessary downtime of a plant, machine or engine. Predictive maintenance incurs additional costs that trade off the monetary or other types of costs associated with a particular maintenance strategy, but it is not necessary to prove that predictive maintenance is advantageous. It relies heavily on the accuracy of forecasting, forecasting and forecasting performance of the equipment used in predictive maintenance applications.

Techniques related to the prediction of warnings, irregularities and unfavorable modes are disclosed in the related techniques. For example, Patent Document 1 discloses an irregular prediction system, that is, an information management or abnormality prediction system in a library. Patent Document 1 discloses a library system including a database, a prediction model, a warning system, a terminal, and a server, and the warning system presents a warning message when a serious condition occurs. , ICT (Information and Communication Technology) system. Patent Document 2 discloses a machine maintenance prediction method based on two or more parameter change curves, and claim 1 states that "... measures machine parameters, predicts parameter change curves, and each parameter change curve. Indicates parameters at different confidence levels ... ”, so Patent Document 2 presents ideas for prediction and uncertainty. Patent Document 3 discloses a system and a method for determining respiratory regularity such as a radar-based physiological motion sensor, which detects regularity by processing one or more frames of a respiratory waveform to detect respiratory regularity. Acquire information related to sex. Patent Document 4 discloses a system and a method for generating an alarm in a patient monitoring system, in which the system monitors parameters or tracks data values to obtain data values for a first limit or a second limit. The range is determined, and an alarm is issued when the first limit or the second limit is reached. Patent Document 5 discloses an online game fraud detection method for detecting fraud in an online game with limited information (for example, a game log).

Patent Document 6 discloses a power system monitoring and control system that detects an abnormality by state prediction that monitors the power system according to an evaluation function and predicts an appropriate state. Patent Document 7 discloses a model prediction control device that prevents failures and malfunctions in a controlled object such as a radiator cooling control system of a vehicle. Patent Document 8 discloses a feedback control device using a servomotor by simulation using a mechanical model.

In a system with a given complexity level, in a specific mode of operation, it is not possible to make predictions to determine the likelihood of error events related to warnings, irregularities, and unfavorable modes, in addition to external fluctuations. Have difficulty. In a wide range of predictions, it is difficult to accurately predict false signals by simply reproducing the behavior of a simulated system, so even a method based on a pure time domain system behavior simulation can be used accurately. It is difficult to predict. The ability to predict warnings, irregularities, and unfavorable modes allows preventive or mitigation measures to be taken to prevent error events (or unfavorable modes). Reducing the impact of error events and notifying them of the occurrence of error events with a high probability will make an important contribution to the economical, safe and efficient operation of technical systems.

Publication of Chinese patent application CN1026097989A U.S. Patent Application Publication US2009 / 0037206A1 U.S. Patent Application Publication US2010 / 0249633A1 U.S. Patent Application Publication US2011 / 0298621A1 U.S. Patent Application Publication US2005 / 0288103A1 Japanese patent application published, Japanese Patent Application Laid-Open No. 2011-24286 Japanese patent application published, Japanese Patent Application Laid-Open No. 2006-172273 Japanese patent application published, Japanese Patent Application Laid-Open No. 200-316937

Shisheng Zhong, Hui Luo, Lin Lin, Xuhun Fu, "An improved correlation-based anomaly detection approach for condition monitoring data of industrial equipment", 2016 IEEE International Conference on Prognostics and Health Management (ICPHM) R. Keith Mobley, "An Intuction to Printitive Mainence 2nd Edition", 2002, Elsevier publicer

Warning, if the "error prediction warning rate" and "non-prediction rate" are sufficiently low, the prediction range is wide enough, and the available data can be used best in consideration of the theoretical framework of recent computers and simulations. Only the prediction of irregular, unfavorable modes is valuable in different application areas.

An object of the present invention is to provide better performance than a general-purpose system having the following drawbacks.
(1) General purpose systems cannot use recent prediction, data driven, or simulation techniques for unfavorable mode prediction.
(2) The general-purpose system cannot reproduce the operation strategy for improving the prediction performance of the system for the time being.
(3) General-purpose systems cannot use physical time evolution simulation combined with system operation characteristic prediction, and best use both areas such as time operation simulation and long-term operation characteristics or meta-variable prediction. Can't.

The above drawback is that the immediate system information cannot be utilized in the best possible way, as the overall sophisticated and detailed physical model and the accurate prediction of the required signal in the overall prediction area are not practically available. It can also cause problems.

In some cases, it is best to combine prediction-based operating time evolution simulations (prediction-based discrete-time simulations) with general characteristic predictions of operation to take advantage of the special strengths of both regions. The above method can be achieved by using two types of predictions in special architectures and units with special prediction functions, and in integrated unfavorable mode / irregular predictors. The architecture can enable self-learning and improve predictive power using "predictive processing rule criteria".

An object of the present invention is to provide a device and a method for predicting warnings, irregularities, and unfavorable modes in a technical system such as an energy management system, which is superior to the general-purpose technology by tackling the problems of the general-purpose technology. It provides performance.

The present invention relates to devices and methods for predicting warning, irregular, and unfavorable modes in a technical system of complexity. That is, the present invention introduces a warning / unfavorable mode predictor having the following characteristics.
(1) Warning / unfavorable mode The predictor combines two types of predictions in a manner that provides many advantages.
(2) Warning / unfavorable mode The predictor controls the sequence and data flow in the apparatus by using a self-regulation standard (for example, a prediction processing rule standard).
(3) The warning / unfavorable mode predictor outputs warning / unfavorable mode occurrence information and related necessary actions for preventing or mitigating the warning.

The present invention provides two types of predictions (A) and (B) as follows.
(A) Type 1 prediction is called "operation time evolution simulation", and considers the simulation of the technical system featuring the calculation of future commands for the reduction control model of the system and the signals not used for the command calculation. It is based on the application of future commands in a simulation environment with advanced simulation models.
(B) Type 2 prediction is called "operation characteristic simulation" and is based on the feature quantity parameters of the prediction period and is generated by the model generator in order to predict the important parameters of the system during the considered period. The data-driven model (is) is used, or the similarity period (as defined by the similarity index) (ie, the period in which the feature amount parameter is most similar to the prediction period) is referred to.

A first aspect of the present invention is a warning / unfavorable mode predictor devised to predict a warning or unfavorable mode that deviates from normal operation in a technical system. The warning / unfavorable mode predictor is an operation command prediction module that generates an operation control command that activates a controlled object of the technical system based on a first prediction signal predetermined according to the operation policy of the technical system. The plurality of control parameters of the technical system are arranged in time series by a time evolution simulation based on a first prediction signal, an operation control command, and a second prediction signal predetermined according to an operation schedule of the technical system. An operation simulation module that generates a first prediction result showing the control pattern, a physical model relating to the physical characteristics of the controlled object of the technical system, and statistics obtained by statistically analyzing fluctuations in the control parameters of the technical system. A model database for registering a simulation model generated based on a target data-driven model, and an event that causes fluctuations in the control pattern to be controlled by the technical system by operating characteristic simulation based on the first prediction signal and the simulation model are shown. and operations characteristic simulation module for generating a second prediction result, the first prediction result and a warning / unfavorable mode detector to predict the occurrence of undesirable mode based on the second prediction result to alert the technical system And.

A second aspect of the present invention is a warning / unfavorable mode prediction method for predicting a warning or an unfavorable mode that deviates from normal operation in a technical system. In the warning / unfavorable mode prediction method, the computer of the warning / unfavorable mode predictor operates the controlled object of the technical system based on the first prediction signal predetermined according to the operation policy of the technical system. A plurality of technical systems by a process of generating control commands, a first prediction signal, an operation control command, and a time evolution simulation based on a second prediction signal predetermined according to the operation schedule of the technical system. Statistical analysis of the process of generating the first prediction result showing the control pattern in which the control parameters are arranged in time series, the physical model of the physical characteristics of the controlled object of the technical system, and the fluctuation of the control parameters of the technical system. The process of registering the simulation model generated based on the statistical data-driven model in the model database and the operation characteristic simulation based on the first prediction signal and the simulation model are used to obtain the control pattern to be controlled by the technical system. A processing process that generates a second prediction result indicating an event that causes fluctuation, and a processing process that predicts the occurrence of an unfavorable mode based on the first prediction result and the second prediction result and warns the technical system. , Is executed.

Robust and reliable predictions can be achieved for warnings, irregularities and unwanted modes in technical systems. It can also provide warning, irregular, and unfavorable mode predictions in highly complex systems. In addition, technical systems such as temperature fluctuation resistant battery management systems, fuel shortage resistant fuel tank management systems, and mechanical maintenance systems can provide predictions for good planning of countermeasures and maintenance behavior. ..

It is a block diagram which shows the overall function of the warning / unfavorable mode predictor which concerns on this invention. It is a block diagram which shows the sub-function of the model generation part included in the model database of a warning / unfavorable mode predictor. An example in which the feature amount selection is feasible is shown in relation to the two graphs showing the initial feature amount and the feature amount selected by the model generator. An example in which model generation is feasible is shown in relation to a graph showing the relationship between the predicted value and the actual value of BTin (that is, the internal temperature) by the model generation unit. FIG. 5 is a simplified block diagram showing a control / simulation model associated with a corresponding module extracted by the warning / unfavorable mode predictor shown in FIG. It is a graph which shows the SoC (charging state) pattern according to the control policy which uses a parameter on a normal day (or a low temperature day under normal operation). It is a graph which shows the SoC pattern by the outside air temperature profile on a high temperature day. It is a graph which shows the SoC pattern by the outside air temperature profile on a very hot day. It is a block diagram of a warning / unfavorable mode predictor with interpretation of two modules related to battery overheating. It is a graph which shows the improvement of the SoC pattern by the outside air temperature profile on a very hot day. It is a graph which shows the improvement of the SoC pattern about the relation signal by the outside air temperature profile on a hot day which causes the change of the battery internal temperature profile. It is a schematic diagram showing a microgrid with renewable power generation, storage, load, and control in relation to warning / unfavorable mode predictors and unreliable grid access. It is a graph which shows the tank level pattern which shows the development of a fuel tank level over time. It is a graph which shows the comparison of the tank level pattern which changes with the presence / absence of a warning / irregular / unfavorable mode predictor (AIUMP) over time. FIG. 3 is a block diagram of a warning / unfavorable mode predictor applied to a fuel tank refilling strategy using two types of predicted fuel trajectories _yocbs and _yotebs. It is a graph which shows the method of determining the similar pattern using the appropriate distance measurement unit in the feature amount space defined by the feature amount 1 and the feature amount 2. It is explanatory drawing which shows the mathematical method which maps the distance of the most similar pattern in order to determine the single weight required for calculating a prediction signal and its similarity index.

The present invention will be described in detail with reference to the accompanying drawings with examples. In a plurality of drawings, the same parts are designated by the same reference numerals, and detailed description thereof will be omitted as necessary.

1. 1. Framework of the Invention FIG. 1 is a block diagram showing the basic framework of the invention, that is, the overall function of the warning / unfavorable mode predictor 100 using the following two types of prediction signals.
(1) Type 1: Using the predicted signal, calculate robust operation control commands such as predicted load, predicted air volume / PV (photovoltaic) power generation, and predicted grid availability.
(2) Type 2: The operation control command is calculated without using the prediction signal, and the prediction signal affects the operation depending on the system and the system topology in some points. Predictive signals can be run on limited hardware for battery charging that depends on the temperature of the battery, enclosure, protection, etc. and maintain operating policies (eg, strategies and algorithms) without over-complexing. Not used for. Predictive signals relate to non-automated or semi-automated processes (eg, refueling schedules for BTS diesel vessels).
Predictive signals (state sequences) are calculated based on motion commands for physical models and / or hybrid models (incorporating physical and data-driven characteristics) as they rely on motion strategies, policies, and system modes.

Warning / unfavorable mode predictors 100 include an operation robust command prediction module 101, a time evolution operation simulation module 102 (eg, a CDF (computer-calculable fluid dynamics) simulation module), a model database module 103, an operation characteristic simulation module 104, And a warning / unfavorable mode detector 105 that executes the behavioral logic by the predictive processing rule database 106 and the output rule database 107. The model database module 103 includes a model generator 200, a physical model repository 1031, a hybrid model (physical and data driven model integration) repository 1032, and a statistical data driven model repository 1033. FIG. 2 is a block diagram showing sub-functions of the model generation unit 200 including the data preprocessing unit 201, the feature amount selection unit 202, and the core model generation unit 203.

As shown in FIG. 1, the warning / unfavorable mode predictor 100 is supplied with two types of prediction signals (ie, type 1 and type 2), type 1 for calculating behavioral strategy (or policy) parameters. Used. The operation robust command prediction module 101 predicts future commands related to the operation policy of the system. The operation simulation module 102 receives the policy parameters and outputs the physical model of the system, the operation strategy / policy (related parameters / commands from the operation robust command prediction module 101), and the prediction signal used for calculating the system command. The operation is calculated by a high-level simulation of typical time evolution considered. The reason why type 2 prediction signals are not used for command calculations causes problems that the prediction signals overly complicate command calculations and require expensive computer calculations, and under normal circumstances, predictions for system operation. This is because the effect of the signal is insignificant or negligible. However, the prediction signal can have a significant impact on system operation. Therefore, the prediction signal is considered in the operation simulation module 102.

The operation simulation module 102 outputs a sequence of predicted internal states. Contrary to the time evolution operation simulation module 102, the operation characteristic simulation module 104 considers a data driven perspective and / or system observation in the long time domain, or a key parameter in the past (in the predicted time period). Try to find periods that have characteristics that are similar to).

As an example, an EMS (energy management system) with a complex interaction between renewable storage and general purpose fuel-based power generation, with long-term fuel consumption or complex power generation consumption patterns with predicted characteristics. Can be mentioned, which will be described later together with FIG.

In FIG. 1, the warning / unfavorable mode detector 105 has internal functional operation logic associated with the prediction rule database 106 and the output rule database 107. The warning / unfavorable mode detector 105 can control the operation robust command prediction module 101, the operation simulation module 102, and the operation characteristic simulation module 104. Warning / unfavorable mode Due to the capabilities and internal knowledge of the detector 105 (ie, prediction / output rule databases 106, 107), the operation robust command prediction module 101, the operation simulation module 102, and the warning / unfavorable considering the operating characteristics. The results used for the mode detector 104 can simulate a particularly meaningful (notable) day. For example, separation of operation simulation / prediction and operation characteristic simulation / prediction because heat conduction is usually slower than electricity and deterioration of the fuel supply tank of the generator is a slow process and slower than heat conduction inside or around the EMS. Can be achieved by time scale separation.

At this time, the accumulation of the influence of the type 2 prediction signal (that is, the influence of the whole) over time should not be ignored. Therefore, the prediction signal is considered in the operation characteristic simulation / prediction. The warning / unfavorable mode detector 105 comprises two types of rule criteria, namely the predictive rule database 106 and the output rule database 107. The purpose of the prediction rule database 106 is to determine how the interaction of operational characteristic simulations / predictions with time evolution model-based simulations / predictions is performed according to operational logic guidelines. A simple example of concretely realizing this interaction will be described later in the description of the second embodiment in which the similar index is introduced. If the time interval predicted in terms of singular variables is similar to a predetermined time interval in the past, the input required for the warning / unfavorable mode detector 105 is the operation characteristic simulation module 104 rather than the time evolution operation simulation module 102. Consists of.

The output rule database 107 determines what measures to take to avoid the warning state of the system based on the prediction results related to warnings, unfavorable modes, and irregularities. The actual countermeasures will be described in the embodiment of the present invention described later.

The warning / unfavorable mode predictor 100 includes a model database module 103 that functions as a model repository. The model database module 103 stores three types of models: a physical model, a hybrid model or an integrated / mixed physical data driven model, and a pure statistical data driven model. Here, the physical model stored in the physical model repository 1031 is required for the operation robust command prediction module 101 and the time evolution operation simulation module 102. Further, the integrated / mixed physical data driven model and the pure statistical data driven model stored in the hybrid model repository 1032 and the statistical data driven model repository 1033, respectively, can be used for the operation characteristic simulation module 104.

The model database module 103 further includes a model generation unit 200, the details of which are shown in FIGS. 2, 3A and 3B. FIG. 2 shows the sub-functions of the model generation unit 200 further including the data preprocessing unit 201, the feature amount selection unit 202, and the core model generation unit 203. The data preprocessing unit 201 is used to resample and clean the data from the historical data and to organize the data related to the N time ahead data. The feature amount selection unit 202 is used to train a model in advance so that important feature amounts can be identified, and to select the top K feature amounts (K is an arbitrarily selected integer) that most influences the model. Be done. The core model generation unit 203 is used to fit the model to a small number of features and to provide the model and its model structure as the output of the model generation unit 200.

FIG. 3A shows an example in which the feature amount selection by the feature amount selection unit 202 may be realized, and FIG. 3B shows an example in which model generation by the core model generation unit 203 may be realized. FIG. 3A shows the top 15 SoCs (charged states) of the battery over time with outside / inside air temperature in relation to the two graphs showing the initial features 301 and the selected features 302 based on the importance score. An example of selecting a feature amount (K = 15) from a large number of feature amounts is shown. By data preprocessing, the high-score related features 302 are selected from all the features 301 considered and used to predict future signal values. FIG. 3B includes an example of a graph showing the relationship between the predicted value and the measured value of BTin (inside air temperature) based on the mapping feature vector. Here, reference numeral 303 indicates a true / predictive plot for prediction for a minority feature set relating to weights (parameters) resulting from a data-driven model.

FIG. 4 is a simplified block diagram showing a control / simulation model corresponding to the model extracted by the warning / unfavorable mode predictor shown in FIG. FIG. 4 shows the difference between the control model 401 (included in the robust command prediction module 101) and the advanced simulation model 402 (included in the time evolution operation simulation module 102). Here, the control model 401 (eg, lower state space model) contains the most important aspects and is a low-level complex system model for determining the optimal command, and the advanced simulation model 402 (eg, CFD). Model) is a detailed and accurate model of highly complex systems.

In particular, the control model 401 is a lean model that includes the basic and necessary aspects of a technical system for correctly calculating operation / control commands. The need for a lean model has characteristics that can be fully resolved at reasonable times and at predetermined time intervals (depending on the application, it must meet secondary optimal solutions) and feasible optimization. / There is a source in the calculation problem. The advanced simulation model 402, on the other hand, is not used for optimization, but is usually only for simulation purposes that perform straightforward calculations with few iterations (as described in the example of solving differential equations by numerical integration). To be used, it is a comprehensive and very detailed model of the technical system (considering PDEs and PDEs).

Various examples of technical systems such as automobiles, engines, and aircraft can be mentioned. The control model 401 used for the calculation of the ideal command by the electronic control unit (ECU) is a simple model because it is considered to be a model for real-time command calculation. This calculation must be completed within a predetermined time limit that cannot be staggered. So far, in order to fully understand the idea of an invention for practical use, the following various embodiments describe the framework of the invention.

2. First Embodiment The present invention is exemplified by an embodiment relating to a microgrid system including a fuel tank and a combustion engine generator shown in FIG. FIG. 11 is a schematic diagram showing the microgrid 1101 connected to the warning / unfavorable mode predictor 1103 and the unreliable grid access 1105, wherein the microgrid 1101 is a tank 1102, a combustion engine generator (or diesel engine power generation). It comprises a machine), a battery (or energy storage) 1104, a control device, a load, and a photovoltaic (PV) unit. The first embodiment relates to the battery 1104 used for the microgrid 1101. Power from the power grid is also available, but there is no guarantee that power can be maintained at any time. In the "non-grid case", the battery 1104 is discharged to power the active load ("effective load" = load-renewable power) or a diesel engine generator and / or a renewable power source. Is charged by. Since a renewable power source or a diesel engine generator having non-linear characteristics can be used, the optimum charge / discharge pattern can be calculated. This calculation usually does not take into account the internal temperature of the battery, as it would be very complicated to consider in calculating the optimal charge / discharge command.

Next, the first embodiment of the present invention relating to the prediction of the unfavorable mode of the battery temperature will be described with reference to FIGS. 5 to 10. FIG. 5 is a graph showing a SoC (charging state) pattern according to a control policy using parameters on a normal day (or a low temperature day under normal operation). In particular, FIG. 5 shows a SoC pattern 501 for a "non-grid" period (see straight line). It is assumed that the optimal charge / discharge policy is characterized by parameters (eg, pmin1, pmin2, pmin3, pmax1, pmax2, s1, s2). Here, the internal temperature of the battery is a problem. If the internal temperature of the battery is higher than a predetermined threshold, charging will not be possible, or charging will excessively damage the battery. Also, when the battery reaches a predetermined temperature, it becomes impossible to discharge, or the discharge causes excessive damage to the battery. Therefore, in the first embodiment, events that reach the relevant threshold (for warnings, irregularities, unfavorable modes) shall be considered.

The function of the operation robust command prediction module 101 is to generate a battery charge / discharge command, which minimizes diesel fuel calculation, guarantees normal energy efficiency, and reduces the operation and maintenance cost of the microgrid 1101. It is represented by the parameters (pmin1, pmin2, pmin3, ...) Introduced as described above so as to reduce. The operational robust command prediction module 101 is implemented by the warning / unfavorable mode predictor 1103. However, it is already integrated into the system's operations / controllers. The operation robust command prediction module 101 generates commands based on a simple model such as the control model 401 shown in FIG. The time evolution operation simulation module 102 is based on a complex model, for example, the advanced simulation model 402 shown in FIG. In the case of a battery, it can be integrated into non-linear charge / discharge efficiency characteristics and the like.

FIG. 5 shows a battery charge / discharge pattern based on the assumption that the outside air temperature is low, and the battery internal temperature does not affect the battery charge / discharge process. Unlike FIG. 5, FIG. 6 shows the situation on a hot day. FIG. 6 is a graph showing SoC patterns 601 and 602 according to the outside air temperature profile 603 on a high temperature day. Here, the SoC pattern 601 relates to the low battery internal temperature, and the SoC pattern 602 relates to the battery charge power reduced by the warning when the policy parameter is not applied. In order to protect the battery from damage, the charging rate is reduced by the protection system based on the battery internal temperature threshold (which is the event of alarm 1). However, this can cause non-optimal operations, as optimal commands (eg, policy parameters) are based on the assumption that full charge rates are applied as shown in FIG. Under this circumstance, the SoC pattern 602 deviates from the "optimal" SoC pattern 601. It is possible to charge the battery further at the end of the "non-grid" period, but this is not good in terms of fuel savings.

FIG. 7 is a graph showing SoC patterns 701 and 702 according to the outside air temperature profile 703 on a very hot day. Specifically, the SoC pattern 701 is related to the low battery internal temperature, and the SoC pattern 702 is to prohibit the battery charge power reduced by the alarm 1 event or the battery discharge due to the alarm 2 event if the policy parameters are not applied. Related. On a very hot day shown in FIG. 7, the alarm 2 event occurs when the battery cannot be charged / discharged at all. The timing at which the internal temperature of the battery reaches the rechargeable temperature again becomes the event of alarm 3. In this case, the deviation of the SoC pattern 702 (that is, the actual pattern) from the SoC pattern 701 (that is, the optimum pattern) is very remarkable at the end of the power failure of the battery that maintains the high SoC value.

FIG. 8 is a block diagram of a warning / unfavorable mode predictor 100 that addresses the above issues relating to battery internal temperature warning. FIG. 8 shows the interpretation of the two modules 102 and 104 in a battery overheated state. Here, the time evolution operation simulation module 102 is used to simulate the SoC pattern of the battery, and the operation characteristic simulation module 104 is used to simulate the battery internal temperature (the evolution depends to some extent on So). used.

FIG. 9 is a graph showing the improvement of SoC patterns 901 and 902 by the outside air temperature profile 903 on a very high temperature day. Here, the SoC pattern 901 is related to the low battery internal temperature, and the SoC pattern 902 is related to the high outside air temperature when the optimal policy parameters are recalculated using the warning / unfavorable mode predictor 100. FIG. 9 shows the advantageous effect of the present invention on the SoC pattern 902 (ie, the battery charge / discharge pattern) which is improved as compared with the SoC pattern 702 shown in FIG. Since the warning / unfavorable mode predictor 100 outputs an optimum pattern recalculation rule considering the reduced battery charge power, the warning / unfavorable mode prediction can prevent the warning. The Warning / Unfavorable Mode Predictor 100 can calculate new policy parameters (eg, s'1, p'max1, p'min2, p'max2) as soon as it predicts an imminent warning. This keeps the energy efficiency for energy saving in an optimally considered situation where the high outside air temperature reduces the maximum battery charge power.

FIG. 10 is a graph showing improvements in the SoC patterns 1001 to 1003 in relation to the relevant signals relating to the outside air temperature profile 1007 on hot days causing variations between the battery internal temperature profiles 1008 and 1009. Here, the SoC pattern 1001 is associated with the low battery internal temperature, the SoC pattern 1002 is associated with the high battery internal temperature without warning / unfavorable mode prediction, and the SoC pattern 1003 is associated with the warning / unfavorable mode prediction and optimal policy parameters. Related to the recalculation of. Also, patterns 1004-1006 are related to diesel engine generators (DEGs), pattern 1006 shows commanded diesel engine power generation for high battery internal temperature with no warning / unfavorable mode prediction, and pattern 1005 is low. The commanded diesel engine generated power for the battery internal temperature is shown, and pattern 1004 shows the diesel engine generated power due to the high battery internal temperature due to warning / unfavorable mode prediction and recalculation of optimal policy parameters. Further, the battery internal temperature profile 1008 is created without warning / unfavorable mode prediction, and the battery internal temperature profile 1009 is created by warning / unfavorable mode prediction and recalculation of optimal policy parameters.

FIG. 10 shows microgrid signals for good insight into the reasons behind fuel savings. FIG. 10 shows a comparison of SoC patterns 1001 to 1003, SoC pattern 1001 shows a battery charge / discharge pattern on a cold day that does not cause a warning, and SoC pattern 1002 shows a battery charge on a hot day with an outside air temperature profile 1007 that causes a warning. / Discharge pattern is shown, and SoC pattern 1003 shows a battery charge / discharge pattern on a hot day due to warning / unfavorable mode prediction. In FIG. 10, the DEG power pattern 1005 corresponds to the SoC pattern 1001, the DEG power pattern 1006 corresponds to the SoC pattern 1002, and the DEG power pattern 1006 corresponds to the SoC pattern 1004.

As shown in FIG. 10, if the warning prediction is not available on a hot day, the DEG will operate for the longest time because the battery cannot be discharged due to the high battery internal temperature. Due to the non-linear nature of the generated power of the DEG, this situation is interpreted as a very inefficient use of the DEG. Also, by using warning / unfavorable mode prediction, the battery internal temperature profile 1009 is kept below the battery warning temperature (see dotted line). Failure of this boundary will trigger a warning without the use of warning / unfavorable mode prediction as shown in Battery Internal Temperature Profile 1008.

The first embodiment relating to the unfavorable mode of battery temperature can be summarized as follows.
(A) When the internal temperature of the battery is higher than a predetermined threshold value and charging is impossible, or when the battery is excessively charged and damaged, the problem of the internal temperature of the battery occurs. In addition, a problem also occurs when the internal temperature of the battery reaches a predetermined temperature and discharge is impossible, or when the discharge excessively damages the battery. Therefore, it is necessary to consider the event that the battery internal temperature reaches the relevant threshold.
(B) The role of operational robust command prediction is to generate battery charge / discharge commands. Operation robust command prediction is implemented by the warning / unfavorable mode predictor 100. However, it is already built into the system's operations / controllers. This operational robust command prediction is used to generate commands based on, for example, a simplified model such as control model 401. It is based on a time evolution operation simulation, eg, a complex model such as the advanced simulation model 402. For batteries, such as integrated non-linear charge / discharge efficiency characteristics. The role of operational characteristic simulation is to determine the battery internal temperature based on empirical values from a data-driven "no model" or "model-based" method (or regression model).
(C) The operational strategy considered for the system exemplifies reducing the charge rate based on the battery internal temperature threshold (when the threshold is reached, an alarm 1 event occurs). Typically, this event occurs on a hot day and the battery has already been charged for some time. However, the optimal command is based on the assumption that the full charge rate as shown in FIGS. 5 and 6 is applied, and the resulting pattern deviates from the optimal one in this situation. This event results in non-optimal operations.
(D) On all hot days, an alarm 2 event occurs that prevents the battery from charging / discharging at all (see FIG. 7). The timing at which the battery internal temperature reaches the rechargeable temperature again indicates the event of alarm 3. In this case, the deviation from the optimum pattern becomes very noticeable at the end of the battery failure cycle, which still maintains a high SoC.
(E) The idea is that the warning / unfavorable mode predictor 100 can predict the occurrence of an alarm and activate the corresponding rules derived from the output rule database 107. In this case, the charge rate is preferentially reduced. Due to this situation, the warning / unfavorable mode predictor 100 recalculates the expected temperature and applies the corresponding command if no unfavorable mode is detected. In this case, optimal operation can be achieved by considering the influential temperature profile.

3. 3. Second Embodiment Next, a second embodiment of the present invention relating to prediction of an unfavorable mode (or alarm) in a fuel tank that consumes fuel will be described with reference to FIGS. 11 to 16. FIG. 12 is a graph showing a tank level pattern 1201 showing the progress of the fuel tank over time. Here, the warning or unfavorable mode is an event in tank 1102 of a diesel engine that consumes fuel. FIG. 12 shows a typical tank level pattern 1201 according to a periodic refill schedule during the m [i-5] to m [i] period. Warning or unfavorable modes (ie, "no fuel" events) can be prevented by regular refill schedules. However, in the period after the m [i] period (indicated by the star symbol), a warning or unfavorable mode occurs for a short period of time due to fuel availability.

FIG. 13 is a graph showing a comparison of tank level patterns 1301 and 1302 that change with the presence / absence of warning / irregular / unfavorable mode prediction (AIUMP) over time. FIG. 13 shows the effect of warning / irregular / unfavorable mode prediction (AIUMP) introducing tank level pattern 1302 instead of “early” tank level pattern 1301. Since AIUMP can predict the unfavorable mode 1304 (indicating a tank running out of fuel) that occurs during the period (indicated by the star symbol) in the tank level pattern 1301, the output rule database 107 is described as " Output a rule for "early refilling at least 2 days early" to perform early refilling action 1303 and prevent the tank from running out of fuel.

ここで、μは類似指標（即ち、０と１の間の値）を示し、過去のタンク枯渇期間が予測される期間に類似しているかを示す。類似指標が０の場合、ＴＦＬＴ予測は、純粋な時間発展オペレーションシミュレーションのみに基づく。類似指標が１の場合、最大類似度を有する過去の状況の重み付け平均が予測に用いられる。 FIG. 14 shows Warning / Irregular / Unfavorable Mode Predictor (AIUMP), that is, Warning / Unfavorable Mode Predictor 100 applied to a fuel tank refilling strategy using _{two types of predicted fuel trajectories yocbs} and _yotebs. In the block diagram of, AIUMP combines two types of simulations / predictions to predict the tank fuel level locus (TFLT) and detect the possibility of unfavorable mode occurrence. Specifically, the operation characteristic simulation module 104 _{calculates the predicted fuel locus y ocbs} , and the time evolution operation simulation module 101 calculates the _{predicted fuel locus y otebs.} The warning / unfavorable mode detector 105 controls the sequence and control / data flow inside AIUMP _{to calculate a single predicted tank fuel level trajectory y pred} by the following formula.

Here, μ indicates a similarity index (ie, a value between 0 and 1), indicating whether the past tank depletion period is similar to the expected period. If the similarity index is 0, the TFLT prediction is based solely on pure time evolution operation simulations. When the similarity index is 1, the weighted average of the past situations with the highest similarity is used for the prediction.

FIG. 15 is a graph showing a method of determining a similar pattern using _{appropriate distance measurements (for example, d 1 to d 5} ) in the feature space defined by the features 1 and 2. Here, reference numeral 1501 indicates a related region of the feature quantity space in the past period, reference numeral 1502 indicates an undetermined point in the period of the feature quantity space to be predicted for the related characteristic, and reference numeral 1503 has a similarity characteristic. Indicates the proximity point in the feature space indicating the past period. FIG. 16 is an explanatory diagram showing a mathematical method of mapping the distance of the maximum similarity pattern to determine the similarity index required to calculate a single weight and prediction signal. Here, reference numeral 1601 indicates a section to be mapped in the distance space, and reference numeral 1602 indicates a section in the weight space.

15 and 16 show details of possible calculation executions. Since the TFLT is predicted by the operation characteristic simulation module 104, the feature signal for the predicted period is determined and compared with the past situation. Feature signals are shown as points in two-dimensional space to facilitate comprehension. The more similar to the past situation (the degree of approximation is evaluated by proper distance measurement), the more the relevant parameters of the situation contribute more to the prediction of TFLT. Due to the limitations of consideration, the distance to the feature of the predicted situation (defined by the interval and period) is at least d _limit for consideration. FIG. 16 shows how this calculation is mathematically achieved by appropriately mapping the _{distance interval [0; limit} ] (1601) to the weight interval [0; 1] (1602). To calculate the similarity index, the mean weights are determined and the nonlinear function s () is applied to achieve the first similarity even if perfect similarity is not achieved. Normalized weights are used to calculate the _{predicted fuel locus yocbs} by combining different situations in the past due to the high priority given to the maximum similarity situation. _{Also, the predicted fuel locus yotebs} are calculated by simulating the operation of the microgrid (used to solve hybrid differential algebraic expressions and the like). If past simulations show many differences (defined by distance measurements in feature space), predictions are based solely on _yotebs.

4. Third Embodiment The third embodiment follows the second embodiment. Warning / Irregular / Unfavorable Mode Prediction (AIUMP) is used to plan and predict the required maintenance actions. Therefore, by combining the two methods (that is, operation characteristic calculation and time evolution operation simulation), wear of machine parts is predicted.

Finally, the present invention is not limited to the above-described embodiment, and modifications and design changes can be made within the scope of the invention defined by the accompanying claims.

The present invention is applied to predict warnings, irregularities, and unfavorable modes that may occur in energy management systems and transportation systems by using two types of simulations together with a model database. The present invention is also applicable to other management / maintenance such as power management and communication management.

100 Warning / Unfavorable Mode Predictor 101 Operation Robust Command Prediction Module 102 Time Evolution Operation Simulation Module 103 Model Database 1031 Physical Model Repository 1032 Hybrid Model (Composite Physical Data Driven Model) Repository 1033 Statistical Data Driven Model Repository 104 Operation Characteristics Simulation Module 105 Warning / Unfavorable mode detector 106 Prediction rule database 107 Output rule database 108 Internal logic control and data signal 200 Model generation unit 201 Data preprocessing unit 202 Feature selection unit 203 Core model generation unit 401 Control model 402 Advanced simulation model 1101 Microgrid 1102 Tank 1103 Warning / Unfavorable Mode Detector 1104 Battery 1105 Unreliable Grid Access

Claims (7)

A warning / unfavorable mode predictor that predicts a warning that deviates from normal operation or an unfavorable mode in a technical system.
An operation command prediction module that generates an operation control command for operating a controlled object of the technical system based on a first prediction signal predetermined according to an operation policy of the technical system.
A time series of a plurality of control parameters of the technical system is performed by a time development simulation based on the first prediction signal, the operation control command, and a second prediction signal predetermined according to the operation schedule of the technical system. An operation simulation module that generates a first prediction result showing the control patterns arranged in
A model database that registers a physical model relating to the physical characteristics of the controlled object of the technical system and a simulation model generated based on a statistical data-driven model obtained by statistically analyzing fluctuations of the control parameters of the technical system. When,
An operation characteristic simulation module that generates a second prediction result indicating an event that causes a fluctuation in the control pattern of the controlled object of the technical system by the operation characteristic simulation based on the first prediction signal and the simulation model.
Warning / unfavorable mode prediction comprising a warning / unfavorable mode detector that predicts the occurrence of the unfavorable mode based on the first prediction result and the second prediction result and warns the technical system. vessel. The model database presets a plurality of importance scores for a plurality of features involved in the technical system, and generates the simulation model based on a predetermined number of features selected according to the plurality of importance scores. The warning / unfavorable mode predictor according to claim 1, further comprising a model generator. The warning / unfavorable mode detector describes the first prediction result and the second prediction result in accordance with the prediction rules relating to the interaction between the simulation and the prediction and the output rules of the unfavorable mode in the technical system. The warning / unfavorable mode predictor according to claim 1, which outputs an unfavorable mode. The operation simulation module generates the first prediction result showing the first prediction locus, and the operation characteristic simulation module generates the second prediction result showing the second prediction locus for the operation schedule of the technical system. Then, the operation of the technical system is performed according to the prediction formula using the first prediction locus and the second prediction locus, and a similar index indicating whether the past period is similar to the prediction period in the operation schedule of the technical system. The warning / unfavorable mode predictor according to claim 1, which calculates a single prediction trajectory that controls the schedule. A warning / unfavorable mode prediction method that predicts a warning or an unfavorable mode that deviates from normal operation in a technical system, and by a computer of the warning / unfavorable mode predictor.
A processing process for generating an operation control command for operating a controlled object of the technical system based on a first prediction signal predetermined according to an operation policy of the technical system.
A time series of a plurality of control parameters of the technical system is performed by a time evolution simulation based on the first prediction signal, the operation control command, and a second prediction signal predetermined according to the operation schedule of the technical system. The process of generating the first prediction result showing the control patterns arranged in
A simulation model generated based on a physical model relating to the physical characteristics of the controlled object of the technical system and a statistical data-driven model obtained by statistically analyzing fluctuations of the control parameters of the technical system is registered in the model database. A processing process to generate a second prediction result indicating an event that causes a fluctuation in the control pattern of the controlled object of the technical system by the operation characteristic simulation based on the first prediction signal and the simulation model.
A warning / unfavorable mode prediction method for executing a processing process of predicting the occurrence of the unfavorable mode based on the first prediction result and the second prediction result and warning the technical system. A claim that a plurality of importance scores are preset for a plurality of feature quantities involved in the technical system, and the simulation model is generated based on a predetermined number of feature quantities selected according to the plurality of importance scores. 5. The warning / unfavorable mode prediction method according to 5. Whether the prediction formula using the first prediction result showing the first prediction trajectory and the second prediction result showing the second prediction trajectory and the past period are similar to the prediction period in the operation schedule of the technical system. The warning / unfavorable mode prediction method according to claim 5, wherein a single prediction trajectory that controls the operation schedule of the technical system is calculated according to a similar index indicating.

Images