Predicting performance changes in devices subjected to intermittent loads.

Abstract

To provide a computer-implemented method, a program, and a system that generate a prediction model configured to output a prediction change in the performance index of an electrochemical device that is subject to intermittent loading, at least a portion of which is generated by one or more renewable energy sources.SOLUTION: A method for generating a predictive model that includes projecting the expected intermittent loading onto multiple basement functions to obtain the successor of each linear combination thereof and applying the successor of each linear combination to multiple fundamental changes comprises: obtaining multiple time series representing the actual intermittent loading of each, at least partially generated by one or more renewable energy sources; machine learning multiple basement functions representing the basic intermittent loading of each; and determining multiple basic changes for each basement function, including the basic changes in the performance index of the electrochemical device.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] This disclosure relates to the field of computer programs and systems, and more specifically to methods, systems, and programs for generating predictive models configured to output predicted changes in performance metrics of an electrochemical apparatus subjected to loads that are expected to be intermittent. [Background technology]

[0002] As part of the energy transition, the intermittent storage and / or release of electrical energy using electrochemical equipment is becoming increasingly important. Against this backdrop, the demand for green hydrogen is growing at a remarkable pace. In practice, a key process in green hydrogen production is supplying electricity generated from renewable energy sources to electrolytic equipment. For example, water electrolytic equipment is used to produce hydrogen in combination with renewable energy sources. Electrolytic equipment typically operates with a constant and stable input, but renewable energy sources are characterized by their fluctuating nature. Applying intermittent loads with such fluctuating profiles to electrochemical equipment can significantly alter its operation, particularly accelerating degradation of the equipment and resulting in reduced performance. The rate of degradation can have a significant impact on the operationality of a project. However, experimental testing to predict degradation is often lengthy and does not represent all renewable energy profiles.

[0003] There is a group of white-box models that integrate the physicochemical mechanisms causing the aging degradation of electrochemical equipment. On the other hand, there is a group of black-box models that attempt to predict the performance of equipment without knowledge of its internal workings. Such black-box models are sometimes based on machine learning approaches. In general, the physical phenomena within electrochemical equipment are controlled by complex multiphysics, making it difficult to provide fast, simple, real-time white-box models. On the other hand, obtaining accurate black-box models requires a huge amount of training data.

[0004] In this context, improved solutions are still needed to predict changes in performance metrics of electrochemical equipment subjected to intermittent loads. [Overview of the Initiative]

[0005] Accordingly, a computer implementation method is provided for generating a predictive model configured to output predicted changes in performance metrics of an electrochemical apparatus subjected to loads that are assumed to be intermittent and at least partially generated by one or more renewable energy sources. The method includes the step of obtaining a plurality of time series, each representing a real intermittent load that is at least partially generated by one or more renewable energy sources and configured to load an electrochemical apparatus. The method further includes machine learning a plurality of basis functions, each representing its respective underlying intermittent load. These basis functions form a projection space of the plurality of time series. The plurality of basis functions are configured to approximate each time series by successors of their respective linear combinations applied to the plurality of basis functions. The method further includes determining a plurality of underlying changes for each basis function, including the respective underlying changes in the performance metrics of the electrochemical apparatus, when the electrochemical apparatus is subjected to each underlying intermittent load represented by the basis functions. The predictive model includes projecting the expected intermittent loads onto multiple basis functions, thereby obtaining successors to each of the linear combinations applicable to the multiple basis functions to approximate the loads that are expected to be intermittent, and applying each successor to multiple underlying changes, thereby outputting the predicted changes.

[0006] This method consists of one or more of the following elements: - A process for determining each fundamental change in the performance indicators of an electrochemical apparatus, wherein the process involves: o Obtain at least one physical instance of an electrochemical apparatus; o To supply the at least one physical instance of the electrochemical apparatus to at least one actual instance of the respective fundamental intermittent loads represented by basis functions; o Perform at least one measurement of a performance indicator for at least one physical instance of an electrochemical apparatus; and o Calculate the underlying change for each based on at least one measurement. It includes; - At least one measurement of a performance indicator in at least one physical instance of an electrochemical apparatus comprises a first measurement and a second measurement, the first measurement being performed at the start of at least one actual instance of each underlying intermittent load, the second measurement being performed at the end of at least one actual instance of each underlying intermittent load, and the calculation of each underlying change comprising calculating the difference between the value obtained from the second measurement and the value obtained from the first measurement; -For at least one basis function, at least one actual instance of each underlying intermittent load represented by the basis function is composed of multiple actual instances of each underlying intermittent load; -Each time series contains its respective classification label, and machine learning of multiple basis functions involves minimizing the classification error with respect to the classification label and / or the distance metric between the basis function and the provided time series for each time series; - The distance metric is dynamic time stretching, and the classification error is cross-entropy loss; - Multiple basis sets are composed of multiple shapelets; -Each shapelet includes a length parameter, and machine learning of multiple basis functions further involves determining the length of each shapelet to minimize the error; - The machine learning of the plurality of basis functions further comprises selecting a set of subsets of a predetermined size from the determined plurality of basis functions, wherein the selected subset has a classification error in a first neighborhood of the minimized classification error and / or a distance metric in a second neighborhood of the minimized distance metric; -Performance indicators represent efficiency, temperature signal, pressure signal, ion concentration signal, or gas concentration signal in constant mode, and / or - The electrochemical apparatus consists of an electrolytic cell connected to an intermittent power supply, an electrochemical energy storage device connected to an intermittent power supply, and at least one electrochemical energy storage device connected to an intermittent load.

[0007] Further provided are computer implementation methods for using the predictive models generated according to the method described above for each electrochemical apparatus. This method of use includes obtaining a load that is expected to be intermittent, at least partially generated by one or more renewable energy sources, and applying the predictive model to the expected intermittent load, thereby outputting the predicted changes in performance metrics of each electrochemical apparatus when each electrochemical apparatus is subjected to the expected intermittent load. Applying the predictive model includes projecting the expected intermittent load onto a plurality of basis functions, thereby obtaining each successor of a linear combination applicable to the plurality of basis functions to approximate the expected intermittent load, and applying each successor of the linear combination to a plurality of fundamental changes, thereby outputting the predicted changes.

[0008] Furthermore, a computer program is provided that includes instructions for performing one of the methods described above.

[0009] Furthermore, a computer-readable recording medium on which this computer program is recorded is provided.

[0010] Furthermore, a system is provided which includes a processor coupled with memory and a graphical user interface, wherein computer programs are stored in the memory. [Brief explanation of the drawing]

[0011] [Figure 1] An example of the system is shown. [Figure 2-7] An example of this method is shown. [Modes for carrying out the invention]

[0012] Non-limiting embodiments are described below with reference to the attached drawings.

[0013] A computer implementation method for generating a predictive model is proposed. This predictive model is configured to output a predicted trend of performance metrics for an electrochemical apparatus, which is subjected to a load that is assumed to be intermittent. The intermittent load is generated at least partially by one or more renewable energy sources. The method includes acquiring multiple time series. Each time series represents each actual intermittent load that is at least partially generated (i.e., formed) by one or more renewable energy sources. Each actual intermittent load is configured to target an electrochemical apparatus. The method further includes a machine learning approach to identify multiple basis functions. Each basis function represents each underlying intermittent load. The multiple basis functions form a projection space of the multiple time series. Thereafter, the multiple basis functions are configured to approximate each time series by the respective successors of linear combinations applied to the multiple basis functions. The method further includes determining multiple underlying changes. The multiple fundamental changes consist of, for each basis function, the respective fundamental changes in the performance indicators of the electrochemical apparatus when the electrochemical apparatus is subjected to the respective fundamental intermittent loads represented by that basis function. The predictive model consists of projecting the assumed intermittent loads onto the multiple basis functions to obtain successors of each linear combination. Each successor of the linear combination is applied to the multiple basis functions to approximate the assumed intermittent loads. The predictive model further includes applying each successor of the linear combination to the multiple fundamental changes and outputting the changes predicted thereby.

[0014] This method constitutes an improved solution for predicting changes in performance metrics of electrochemical apparatus. This method provides a simple tool for predicting changes in performance metrics of such apparatus. Such predictions are particularly important for electrochemical apparatus subjected to intermittent loads, at least in part, generated by renewable energy sources. For example, renewable energy sources such as wind and solar resources are not always available (due to unavoidable fluctuations in the sun's position and weather conditions) and are largely unpredictable, thus resulting in intermittent loads. One or more renewable energy sources that generate at least a portion of the assumed intermittent load, and / or one or more renewable energy sources that generate at least a portion (e.g., each) of each actual intermittent load represented by multiple time series, may consist of one or any combination of solar, wind, and hydroelectric energy sources. When such loads are supplied to an apparatus, such as an electrochemical apparatus, the performance of the apparatus degrades significantly (e.g., compared to a constant load).

[0015] Changes or modifications in performance metrics are closely related to defining the initial conditions for those performance metrics and then predicting them. Such initial conditions for performance metrics, such as the efficiency of an electrochemical apparatus, are generally known for the apparatus, for example, by the manufacturer. Therefore, in this specification, the expressions "prediction of changes in performance metrics" and "prediction of performance metrics" are used interchangeably.

[0016] This method does not require complex hardware and / or measurement techniques and can rely (e.g., solely) on the processing of data simply obtained from the device and the calculations based on such data. This tool is based on the important observation that for an assumed (intermittent) load, changes in performance can be obtained as a successor to a linear combination of fundamental changes (of the performance metrics). The fundamental changes represent the influence of the fundamental load (represented by respective basis functions) on the performance changes. This method obtains a plurality of said fundamental loads by performing machine learning between a plurality of provided time series, and determines each fundamental change, for example, by performing each experiment. Each experiment consists of exposing an electrochemical device to each fundamental intermittent load and observing / measuring each change in the performance metric to determine each fundamental change.

[0017] The tool provided by this method is further beneficial in providing efficient online / offline task separation. When generating a prediction model in the offline mode, this method is ready to receive a predicted load online, project it onto a plurality of basis functions to obtain a successor to each linear combination, and apply the obtained linear combination to a plurality of already determined fundamental changes. Thereby, the online stage is simply the projection of the provided online load and the application of each linear combination to the set of determined values, which is significantly faster and has lower computational cost than the offline stage. This improves the real-time performance of this method.

[0018] Furthermore, in the context of the proposed method, it has been confirmed that the decomposition of the predicted load through the said projection becomes a successor to a linear combination of (reduced number of) basis functions, without significant loss of relevant information regarding the task at hand. Thereby, this method can predict changes in the performance metric with high accuracy.

[0019] "Electrochemical device" means a device including at least one electrochemical cell. An electrochemical cell, as is known per se, is a device that can generate electrical energy from a chemical reaction or cause a chemical reaction by using electrical energy. In the example, the electrochemical device can include at least one electrolytic cell connected to (configured / intended to be configured) an intermittent power source (and configured to consume the power received from the power source to perform an electrolysis process), an electrochemical energy storage device connected to (configured / assumed to be configured) an intermittent power source (and configured to convert the power of the power source into the chemical potential energy of the energy storage device), and an electrochemical energy storage device connected to (configured / assumed to be configured) an intermittent load (and configured to receive energy from or release energy to the load). Thereby, "intermittent load" means intermittently discharging or receiving power, in other words, positive or negative intermittent energy change. On the other hand, "intermittent power source" is an intermittent power supply, in other words, a specific form of an intermittent load. In other words, an intermittent load may relate to an electrochemical device operating in either a "receiving" mode (e.g., an electrolytic cell connected to a renewable power source) or a "sending" mode (e.g., a battery connected to an engine operating intermittently).

[0020] Thereby, the expression "electrochemical device" means a class of electrochemical devices using the same type and the same operating parameters. In particular, this method can generate a prediction model for a class of electrochemical devices, and the generated model is applicable to all devices of a class (e.g., the same model / reference from the same brand). Thereby, a universal prediction function according to this method is provided.

[0021] "Electrochemical performance indicators" means any indicator used in the art to evaluate the performance of an electrochemical apparatus. For example, a performance indicator could be the output voltage or current of the electrochemical apparatus. Alternatively, a performance indicator could be the efficiency of the apparatus. The efficiency of the apparatus could be any efficiency used in the art. For example, the efficiency of the apparatus could be the ratio of the output energy of the apparatus to the input energy of the apparatus.

[0022] Other examples include performance metrics that represent efficiency in a constant mode, temperature signals, pressure signals, ion concentration signals, or gas concentration signals. "Efficiency in a constant mode" means the ratio of the device's efficiency (i.e., when the device is subjected to an intermittent load) to its efficiency under a constant load. As mentioned above, under an intermittent load, efficiency may decrease over time compared to efficiency under a constant load. Pressure and temperature signals may be pressure / temperature signals from the device itself, for example, pressure / temperature signals from a reservoir within the device (e.g., a reservoir / pipe configured to collect / transport hydrogen (H2) and oxygen (O2) produced during electrolysis). Pressure or temperature signals can be obtained using any type of known pressure / temperature sensor. Gas concentrations may be signals representing the concentration of elements within the device, for example, signals representing the O2 concentration in an H2 reservoir / pipe and / or the H2 concentration in an O2 reservoir / pipe of an electrolytic cell. Concentration signals can be obtained using any type of known device / method, such as optical sensors, mass spectrometry, or gas chromatography. As is known, H2 and oxygen O2 obtained by electrolysis in an electrolytic cell are stored in hydrogen and oxygen containers / reservoirs (HV and OV, respectively), where they are taken up and separated from the electrolyte. The pressures between the two containers (i.e., the anode and cathode circuits) may be different. During the electrolysis process, a certain percentage (e.g., by volume) of H2 may move across a barrier placed between the containers (e.g., gas separators) to the O2 reservoir, and conversely, a certain percentage of O2 may move into the H2 reservoir. In particular, to prevent the oxygen concentration from approaching the low explosion limit of the gas mixture, a volume % of O2 in the H2 reservoir can be linked to an alarm for safety reasons. A similar alarm system can also be set up with respect to the pressure and / or temperature of each container to warn of potential overpressure / temperature in the reservoirs. The gas composition / pressure / temperature in each reservoir may vary with load. Such variations prevent the device from operating in continuous mode and force transient behavior. The device operates in transient mode rather than continuous mode. Known control methods for system dynamics are not as fast as load fluctuations.Therefore, controlling such equipment is a difficult problem. In particular, equipment subjected to intermittent loads is prone to experiencing the explosion limits of O2 in HV, overpressure, and overtemperature. Consequently, it is all the more important to have accurate and fast predictive models for electrochemical equipment exposed to this type of load.

[0023] "Changes in the performance indicators of an electrochemical apparatus" refers to the temporal behavior of those indicators while the apparatus is functioning. The changes in the performance indicators of an electrochemical apparatus may represent a decrease in efficiency and / or capacity, i.e., degradation, over time (i.e., between an initial time and a certain time). "A predictive model configured to output predicted changes in the performance indicators of an electrochemical apparatus" refers to a model that takes data about an electrochemical apparatus as input and outputs multiple values ​​representing changes in the performance indicators of that apparatus.

[0024] As is known in the field of electrical engineering, renewable energy sources often generate loads that operate for only a small portion of a 24-hour period. One or more renewable energy sources that partially generate an intermittent load may consist of (for example, a battery connected to a renewable energy source). An intermittent load may be a load that exhibits oscillatory behavior (for example, due to periodic fluctuations of the one or more renewable energy sources) and may disappear at some point in time. The concept of an intermittent load is sometimes defined inversely to a continuous load, which is usually defined as a load that operates continuously for a long period of time. In other words, a continuous load is a load that continues for a long time under the same conditions. An intermittent load, on the other hand, is a load that lasts for a short time and then stops for either a long time or a very short time. Applications of intermittent loads include electrochemical devices operating in "receiving" mode (e.g., an electrolyzer connected to renewable energy) or electrochemical devices operating in "feeding" mode (e.g., a battery connected to a motor with an intermittent load). Intermittent loads are generated at least partially by one or more renewable energy sources, such as solar energy sources, wind energy sources, geothermal energy sources, and ocean energy sources.

[0025] This method involves acquiring multiple time series representing each actual intermittent load. The actual intermittent load may be a load observed in reality. In other words, an intermittent load is a record of energy production by an electrochemical device, such as a photovoltaic or wind power system. These records are often created by so-called SCADA (Supervisory Control And Data Acquisition) systems. Furthermore, the actual intermittent load may also be a historical load, i.e., a load measured over a period in the past and stored in a database. In other words, the time series representing an actual intermittent load is the history of that intermittent load. Alternatively, the actual intermittent load may be a generated load (i.e., each time series may be artificially generated). For example, a time series may be constructed by generating wind power from wind profiles and wind turbine characteristics. Therefore, the multiple time series may consist of time series of actually observed loads, and / or time series of past loads, and / or generated time series. Alternatively, multiple time series may consist solely of time series of actually observed loads and / or past loads (i.e., for 100% of the time series) or substantially solely of them (e.g., for 90% or more of the time series).

[0026] Each intermittent load is partially generated by one or more renewable energy sources and is configured to target an electrochemical apparatus. "Configured to target an electrochemical apparatus" means that the load is applicable to the electrochemical apparatus, i.e., it constitutes a physical configuration.

[0027] Next, the method consists of machine learning a set of basis functions that represent each of the underlying intermittent loads. These basis functions provide a finite-dimensional projection space for the time series. The method may also machine learn the basis functions using known methods to obtain the projection space for the time series. For example, the finite-dimensional projection space may be the optimal projection space. The optimal projection space is the projection space that minimizes the projection error on the projection space for a given number of dimensions (i.e., the number of basis functions). The projection error can follow any known metric. In particular, the optimal projection space may have fewer dimensions than those required to accurately define the given time series. In this embodiment, the optimal projection space provides a dimensionally reduced space for describing the time series (thus reducing complexity). Such a dimensionally reduced space is valuable enough to describe the given time series and the expected intermittent loads with sufficient accuracy using the basis functions, in order to obtain an accurate and fast predictive model. In the example, the number of basis functions may be less than 50, 25, and 15. The projection space is a linear projection space, so that multiple basis functions are configured to approximate each time series by the successor of each linear combination applied to the multiple basis functions.

[0028] In the example, multiple basis functions may consist of multiple shapelets. In other words, the method may identify multiple basis functions (using machine learning) to obtain shapelet decomposition / learning of intermittent loading. Alternatively, multiple basis functions may consist of multiple Fourier functions (i.e., providing Fourier decomposition) or multiple wavelet functions (i.e., providing wavelet decomposition). Shapelet decomposition provides an extension to Fourier decomposition and / or wavelet decomposition. Just as Fourier decomposition decomposes a signal by its (inherent) frequencies, wavelet decomposition decomposes a signal by its frequency and order. Wavelet decomposition identifies major frequencies and further provides information about the localization of major perturbations, thereby tuning the signal. Shapelet learning, on the other hand, can decompose a signal by frequency, order, and pattern. In other words, shapelet learning identifies representative patterns of a signal and directly identifies those patterns. Furthermore, shapelet learning can also simultaneously decompose a signal based on classification labels. Considering such labels when decomposing a signal is particularly important in applications to predicting performance changes of equipment targeting renewable energy sources. Such energy sources exhibit high variability with respect to weather conditions (e.g., wind, clouds, sun, etc.) and / or time of day or year (day / night, winter / summer, etc.). For example, the output profile (and pattern) of a solar panel on a sunny summer day may differ significantly from that of a solar panel on a winter night or rainy day. Therefore, it is important to consider additional parameters for decomposing patterns among similar data under equivalent operating conditions of the electrochemical apparatus. By using shapelet learning, the provided time-series metadata can be used, for example, to improve machine learning steps. At the beginning of the learning process, shapelet learning can start with a portion of the signal at hand. However, as learning is iterated (e.g., across epochs), a mixture or average of the aforementioned portions of the signal is obtained to have the best compromise.In other words, shapelet learning takes a signal, cuts it into smaller fragments, modifies the smaller fragments to reduce the Dynamic Time Warping (DTW) metric, and retains the best fragment according to the metric. In some examples, shapelet learning can use a genetic algorithm that uses DTW as the metric selection. In the example, a linear combination of shapelets is determined for different time intervals of the time series under consideration. The time intervals are determined by optimization algorithms, such as those more commonly known in the field of Generalized Additive Models (GAMs).

[0029] Next, the method determines several fundamental changes for each basis function. Each fundamental change consists of a fundamental change in the performance indicator of the electrochemical apparatus when the electrochemical apparatus is subjected to each fundamental intermittent load represented by the basis function. In other words, the method can include a test program in which the electrochemical apparatus is subjected to several (fundamental) intermittent loads. Examples of such test programs will be discussed further later.

[0030] As described above, this method is a method for generating a predictive model. The (generated) predictive model is a function that takes a hypothetical intermittent load as input and outputs a predicted change in the performance indicator of an electrochemical apparatus when subjected to the hypothetical intermittent load. To this end, this function is composed of projections of the hypothetical intermittent load onto multiple basis functions (in the first functional block). Such projections yield successors to linear combinations applicable to multiple basis functions in order to approximate the hypothetical intermittent load. The projected function is output directly by the machine learning step. The predictive model is further constructed by applying each successor to the linear combination (further output by the machine learning step) to multiple fundamental changes, thereby outputting the predicted change. Each successor to the linear combination includes a scalar coefficient for each linear combination and basis function. Each of the scalar coefficients is a coefficient applied to the respective fundamental change corresponding to the basis function.

[0031] In the example (for instance, the test program described above), determining the underlying changes for each performance metric of the electrochemical apparatus consists of obtaining at least one physical instance of the electrochemical apparatus for each basis function. A “physical instance” means an instance of a particular class (described above) of electrochemical apparatus in the real world. An instance of an electrochemical apparatus (of the aforementioned class) is defined by a set of specifications that define its behavior in terms of changes in performance under intermittent loads. The set of specifications may consist of any technical characteristics (e.g., nominal power, voltage response, materials, etc.) that control and define the behavior of the apparatus in its function. A physical instance is real-world hardware that satisfies all specifications and forms an operating version of the electrochemical apparatus.

[0032] Next, the decision may include providing at least one physical instance of the electrochemical apparatus to at least one actual instance of each underlying intermittent load represented by a basis function. The decision may further include performing at least one measurement of a performance indicator on at least one physical instance of the electrochemical apparatus and calculating each underlying change based on at least one measurement. For example, the underlying measurement may be a measurement of the maximum capacity or power of the electrochemical apparatus, and the underlying change may be the underlying degradation caused to the physical instance of the apparatus by the actual instance of each actual intermittent load.

[0033] In the embodiment, at least one measurement of the performance index in at least one physical instance of the electrochemical apparatus may include a first measurement and a second measurement. The first measurement may be performed at the start of at least one actual instance of each underlying intermittent load. The second measurement may be performed at the end of at least one actual instance of each actual intermittent load. In other words, the method can provide at least two measurements, one at the start and one at the end of each load. The calculation of each underlying change may consist of calculating the difference between the value obtained from the second measurement and the value obtained from the first measurement. Such a difference can indicate the change (i.e., variation) of the performance index corresponding to the underlying intermittent load represented by the basis function.

[0034] In the embodiment, for at least one basis function, at least one actual instance of each underlying intermittent load represented by the basis function may consist of multiple actual instances of each underlying intermittent load. In other words, such an embodiment provides the apparatus with an iterative pattern of intermittent loading. By applying iteration of intermittent loading, the generated model becomes capable of predicting changes in performance metrics over the long term. In particular, under various applications and many loading conditions, the behavior of electrochemical apparatus can be considered linear, i.e., without hysteresis. This improves the method of performing parallel predictions with multiple models instead of performing predictions at distant points in time.

[0035] In the embodiment, each time series may include its own classification label. The machine learning of multiple basis functions may, for each time series, minimize the classification error with respect to the classification label and / or the distance metric between the basis function and the provided time series. Each classification label may represent a binary classification (i.e., having two label values) or a multi-class classification. The distance (or similarity) metric represents the similarity between the time series and the basis function. The classification metric evaluates whether the fitted classification of the time series corresponds to a known classification according to the classification label.

[0036] In particular, when multiple basis functions consist of multiple shapelets as described above, the resulting shapelets best represent the classification. In such cases, the disparity metric can be a median that represents "how well" a shapelet represents a class. This metric is the lens through which each class is viewed by the shapelet. In such combinations, and with respect to classification fitting to time series, this method associates a certain number of shapelets with each class. The metric can then be used to evaluate the distance between the input time series and each shapelet of each class. The class closest to the time series is then associated with the time series. The goal of shapelet learning is to find the best shapelets that can best fit multiple time series to known classes.

[0037] In the examples, the distance metric may be a dynamic time stretching method known in the art. The classification error may be any classification metric known in the art. For example, the classification error may be a cross-entropy loss, such as an L2-penalized cross-entropy loss.

[0038] As described above, according to some examples, multiple basis functions are composed of multiple shapelets. In such examples, each shapelet may include a length parameter, and the (step) machine learning of the multiple basis functions may further include determining the length of each shapelet to minimize the error. The method may include an optimization step to obtain the optimal length of each shapelet. The optimization step may optimize the length to minimize a combination of the distance metric and / or classification error. The optimization step may follow any known optimization method in the art. In the examples, the optimization step may be configured according to a neural network or a genetic algorithm, for example, an implementation of Gendis (Vandewiele et al., "GENDIS: Genetic Discovery of Shapelets", Sensors 2021, 21(4), 1059).

[0039] Alternatively or additionally, this method may accept input defining the length of each shapelet (of a group of shapelets). Users can provide this input in any way, for example, based on domain knowledge. Such examples can utilize implementations like tslearn (see https: / / tslearn.readthedocs.io / en / stable / ) that use the provided input defining the shapelet lengths to machine learn the shapelets.

[0040] In the embodiments, machine learning of multiple basis functions may further include selecting a subset of a predetermined size from the determined basis functions. Such embodiments may be combined with the above-described examples of methods in which machine learning of multiple basis functions consists of minimizing classification errors and / or distance metrics. In such combinations, the selected subset may have a first neighborhood classification error that minimizes the classification error, and / or a second neighborhood distance metric that minimizes the distance metric.

[0041] Such examples relate to clustering algorithms that regroup the extracted shapelets (as determined basis functions) into selected shapelets that best represent the shapelet set while preserving classification power. In particular, due to computational constraints, the shapelet learning process can only test a certain number of shapelets of a certain maximum length. This length can be determined using domain knowledge or optimization, as described above. To obtain the desired number of shapelets, a larger number is set to be extracted to give the algorithm sufficient search flexibility. The shapelets can then be made equal in length by filling them with terminal values. The resulting shapelets may then be clustered into the desired number. The remaining terminal flats are then cut off. Finally, to verify that the clustering preserves the accuracy of the classifier, an inertia criterion is evaluated for the shapelets before and after clustering.

[0042] A computer implementation method is further proposed that uses the predictive models generated for each electrochemical apparatus, which are generated according to the method described above. Such an implementation method consists of obtaining assumed intermittent loads that are at least partially generated by one or more renewable energy sources. The implementation method further consists of applying the predictive model to the assumed intermittent loads. Thus, the implementation method outputs the predicted changes in performance indicators for each electrochemical apparatus when each electrochemical apparatus is subjected to the assumed intermittent loads. The application of the predictive model includes projecting the assumed intermittent loads onto a plurality of basis functions, thereby obtaining each successor of a linear combination applicable to the plurality of basis functions to approximate the assumed intermittent loads. The application of the predictive model further includes applying each successor of the linear combination to a plurality of underlying changes, thereby outputting the predicted changes.

[0043] The methods described above—namely, methods for generating a predictive model, methods for using the generated predictive model, and methods for forming a dataset—are implemented on a computer. This means that the steps (or substantially all steps) of the method are performed by at least one computer, or any system. Thus, the steps of the method are performed by the computer, possibly fully automatically or semi-automatically. In embodiments, at least some of the triggers for the steps of the method are performed by user-computer interaction. The required level of user-computer interaction depends on the balance between the expected level of automation and the need to fulfill the user's wishes. In embodiments, this level may be user-defined and / or predefined. For example, in a method for generating a predictive model, the step of acquiring multiple time series may be performed upon receiving user input. Upon receiving such input, the method can automatically perform the remaining steps and generate a predictive model. As another example, the step of acquiring the predicted intermittent loads in the method to be used may be performed upon receiving user input. Upon receiving such input, the method automatically applies the predictive model to the assumed intermittent loads and outputs the predicted change in performance metrics for each electrochemical apparatus.

[0044] A typical computer implementation of the method is to run the method on a system adapted for this purpose. This system may include a processor coupled with memory and a graphical user interface (GUI), where the memory stores a computer program containing instructions for running the method. The memory may also store a database. The memory is any hardware adapted for such storage and may consist of several physically distinct parts (e.g., one for the program and possibly one for the database).

[0045] Figure 1 shows an example of a system, which is a client computer system, such as a user's workstation.

[0046] The client computer in this embodiment includes a central processing unit (CPU) 1010 connected to an internal communication bus 1000, and random access memory (RAM) 1070 also connected to the bus. The client computer further includes a graphics processing unit (GPU) 1110 associated with video random access memory 1100 connected to the bus. The video RAM 1100 is also known in the art as a frame buffer. A mass storage device controller 1020 manages access to mass storage devices such as a hard drive 1030. Mass storage devices suitable for embodying computer program instructions and data include, for example, semiconductor memory devices such as EPROMs, EEPROMs, and flash memory devices; magnetic disks such as internal hard disks and removable disks; and all forms of non-volatile memory, including magneto-optical disks. Any of the above may be supplemented by or incorporated into a specially designed ASIC (Application-Specific Integrated Circuit). A network adapter 1050 manages access to the network 1060. The client computer may also include haptic devices 1090 such as a cursor control device and a keyboard. A cursor control device is used in a client computer to allow the user to selectively position the cursor at any desired location on the 1080-pixel display. Furthermore, the cursor control device allows the user to select various commands and input control signals. A cursor control device includes a number of signal-generating devices for inputting control signals into the system. Typically, the cursor control device is a mouse, and the mouse buttons are used to generate signals. Alternatively or additionally, the client computer system may include a sensitive pad and / or a sensitive screen.

[0047] A computer program may include instructions that can be executed by a computer, and such instructions include means for causing the system to perform the method. The program may be recordable on any data storage medium, including the system's memory. The program may be implemented, for example, in digital electronic circuits, computer hardware, firmware, software, or a combination thereof. The program may be implemented as a product embodied in a machine-readable storage device for execution by a device, for example, a programmable processor. The method steps may be performed by a programmable processor that executes a program of instructions that perform the function of the method by manipulating input data to produce an output. Thus, the processor may be programmably coupled to receive data and instructions from a data storage system, at least one input device, and at least one output device, and to transmit data and instructions. The application program may be implemented in a high-level procedural programming language or an object-oriented programming language, or, if necessary, in assembly language or machine code. In any case, the language may be a compiled language or an interpreted language. The program may be a full installation program or an update program. When the program is applied on the system, instructions for performing the method are obtained in any case. Computer programs may instead be stored and executed on servers in a cloud computing environment, with the servers communicating with one or more clients over a network. In this case, a processing unit executes instructions composed of the program, thereby executing the method on the cloud computing environment.

[0048] Next, we will explain the implementation of the method described above.

[0049] This embodiment relates to a methodology for predicting the degradation of equipment connected to intermittent upstream / downstream loads. The degradation of electrochemical systems, such as electrolytic cells, is governed by several factors, including electrode voltages that trigger degradation, the generation of thermal gradients due to starting / stopping and load fluctuations, and species diffusion (e.g., gas crossover) due to starting / stopping and load fluctuations. The implementation can capture both phenomena. The implementation imposes on the cells (of the electrochemical device) to remain at a given voltage for a given time (as defined by the amplitude of the shapelet), while reflecting the load fluctuations.

[0050] In particular, since renewable energy sources generally have periodic energy production patterns (for example, in the case of solar-driven renewable energy resources, as a response to periodic inputs from the sun), it is extremely important to be able to capture these periodic patterns as characteristics of energy production at different scales. This embodiment deals with a methodology consisting of identifying the patterns, characterizing degradation for each pattern, and associating them. Next, a predictive model forecasts the temporal shape of renewable energy production. The temporal shape is the shape projected onto the identified patterns. Projection coefficients are used to combine various degradations that characterize the patterns and give an estimate of the degradation expected in the future.

[0051] Shapelet learning generalizes Fourier decomposition and wavelet decomposition to any shape. Shapelet learning is a supervised classification machine learning approach. To implement this, the implementation uses "LearningShapelets" from the python package tslearn version 0.5.2 (see https: / / tslearn.readthedocs.io / en / stable / index.html).

[0052] In the first step, the implementation collects (i.e., retrieves) a large number of datasets with intermittent loads (e.g., 10 or more sets of data) and uses machine learning / artificial intelligence algorithms to create a base of unit patterns (s) that can be projected onto these functions. i(t)) is generated. Each unit pattern is itself a load profile with a duration ranging from a few milliseconds to several hours.

[0053] In the second step, the implementation runs a test program. The test program has the same number of test phases as the number of unit patterns obtained in the first step. Each phase of the test program consists of applying a load corresponding to one unit pattern several times. The typical duration of each test ranges from 1 hour to 10,000 hours. The test period can extend to one year. At the end of each phase, the degradation of the device is monitored by measuring the efficiency compared to a constant mode. In this embodiment, the observed degradation is divided by the number of unit patterns in that phase to obtain the unit pattern (s i Deterioration around (d i Calculate ).

[0054] To speed up testing, the implementation involves performing multiple parallel tests on separate devices. This parallel strategy is based on the assumption that system degradation is not affected by previous lifespan (i.e., there is no hysteresis). This is a valid assumption, especially for electrolytic cells, where linear degradation is observed over time. Failures cause changes in the slope of degradation, so cells / stacks are replaced well before failure occurs.

[0055] Figure 2A outlines the first and second steps. In step 1, the implementation collects a large dataset by accessing the intermittent loading library 201. Next, the implementation uses machine learning and artificial intelligence algorithms to identify the best family of eigenvectors that describe the intermittent profile obtained by the method. The implementation can identify such eigenvectors using different techniques such as Fourier transforms, wavelets, or shapelets. As a result of this step, the embodiment obtains a set of identified eigenvectors (or patterns) that can be used to describe the intermittent loading. iA set consisting of can be obtained. This set is composed of 10 eigenvectors. Next, the embodiment applies each specific pattern of the set to test a device (e.g., an electrolytic cell), and the eigen degradation d caused by each pattern i is determined.

[0056] In the third step, the degradation of the device connected to the new intermittent load profile S(t) can be predicted. The new load is approximated as follows by a linear combination of the unitary patterns generated in step 1. P(t)=S i=1.n a i s i (t)

[0057] And by summing the individual degradations of the unitary patterns used and weighting them with the coefficients used for the conversion of the unitary patterns of the new load into a linear combination, the degradation (d) can be predicted as follows. d =S i=1..n a i d i

[0058] Figure 2B shows an overview of the third step. In this example, a new profile P(t) is provided to the method. The profile may be related to daily, monthly, or annual power / load. The load is normalized. The implementation calculates the spectral transform of the provided profile using, for example, Fourier, wavelet, or shapelet, and then predicts the degradation as described above. The embodiment can estimate the confidence level by calculating the distance between the original (i.e., the provided profile) profile and the reconfigured profile (i.e., using the calculated spectral transform).

[0059] This embodiment can be used in a similar manner to predict the degradation of other types of devices, such as electrolytic cell families connected to intermittent power sources (e.g., renewable energy, hybrid systems, etc.), electrochemical energy storage devices connected to intermittent power sources, and / or electrochemical energy storage devices connected to intermittent loads (e.g., electric motors). The embodiment can also predict other output signals such as temperature, pressure, ion concentration, and / or gas in a similar manner.

[0060] Next, we will discuss the quantitative aspects of the implementation.

[0061] As is well known, shapelet extraction is a method for extracting characteristic patterns from a set of signals. In other words, it identifies the basic shape of a signal by learning shapelets.

[0062] Figure 3 shows the base signals used to generate the training and test data. Figures 4A and 4B show the time series of the training and test sets, respectively. Two-thirds of the dataset is used as the training set, and the remaining one-third is used as the test set.

[0063] The shapes identified by the implementation at two points in the process are evaluated to explain the identification process. The implementation is performed by tslearn, a Python package that provides machine learning tools for time series analysis. The tslearn package implements a neural network that modifies its weights each time it passes through the entire dataset, called an epoch, to optimize its performance. The two points in evaluation are at different epoch counts, one before convergence (e.g., 3000 epochs) and the other when convergence is considered to have been achieved (e.g., 10000 epochs).

[0064] Figure 5 shows the case at the first moment, i.e., 3000 epochs. Looking at the fitness curve 510, we can see that it has not yet reached convergence. Furthermore, the extracted (i.e., identified) shapelet 520 is a mixture of the base signals used to generate the training and test days, as shown in Figure 3.

[0065] The second moment, i.e., 10,000 epochs, is shown in Figure 6. The fitness curve 610 is lower and flatter compared to 510 in Figure 5. This indicates that convergence has been achieved. Furthermore, the extracted shapelet 620 shows almost the pure base signal used to generate the training and test days, as shown in Figure 3.

[0066] Comparing the accuracy of both training sessions (3000 epochs and 10000 epochs), there is a 0.6% increase from the first case (Figure 5) to the second case (Figure 6). Here, accuracy measures the classification performance of the shapelets, not their proximity to the base signal. The accuracy of the shapelets' shapes is not necessary for the shapelet-learning classifier; rather, it depends on the differences between them. However, the accuracy of the identified shapes improves classification performance.

[0067] These observations indicate that the implementation converges to the correct shape. In other words, the learned shapelet algorithm certainly identifies and extracts the correct basic pattern from the set of signals.

[0068] Figures 7A-C show examples of extracted shapelets: Figure 7A is the identified sinusoidal signal, Figure 7B is the identified triangular signal, and Figure 7C is the identified rectangular signal.

[0069] In this implementation, datasets representing renewable energy production in different locations and under different conditions are collected and labeled by macroscopic conditions (day / night, winter / summer, solar / wind, climate, etc.). For these collected datasets or examples, a machine learning method can use a metric called "dynamic time stretching" to calculate the distance or similarity between subsections or shapelets of this example. The length (or duration) of these shapelets is determined either by an optimization method or by domain knowledge. The group of shapelets that is closest to all other possible subsections and best represents the user-given classification is retained or extracted. Classification is performed by logistic regression constructed as a neural network, and the weights are optimized by stochastic gradient descent with respect to L2-penalized cross-entropy loss.

[0070] Experiments in electrolytic cells require significant time and resources and are subject to operational constraints; therefore, implementations may use approaches that do not explore all possible shapelet lengths. In such an approach, embodiments may use a clustering algorithm to regroup the "extracted shapelets" into k "selected shapelets" that best represent the shapelets. The clustering algorithm can select k selected shapelets so as to retain either x% of the classification power by logistic regression or x% of the total distance between shapelets. The value x may be determined by constraints of the experimental design that characterize the degradation associated with each "selected shapelet".

[0071] The experimental design can be simplified to any two accompanying experiences. The first is a baseline scenario that serves as a reference point for comparison with the second. The second consists of N iterations of each "selected shapelet." N can be determined by the central limit theorem (usually 30 or greater, assuming a Gaussian distribution) so that the characterized degradation is statically representative.

Claims

1. A computer implementation method for generating a predictive model configured to output predicted changes in performance indicators of an electrochemical apparatus subjected to assumed intermittent loads, which is generated at least partially by one or more renewable energy sources, wherein the method comprises: - A process in which a computer acquires multiple time series, each of which represents an actual intermittent load, each generated at least in part by one or more renewable energy sources and configured to load an electrochemical device; - A process of machine learning a plurality of basis functions, each representing a fundamental intermittent load, wherein the plurality of basis functions form a projection space of a plurality of time series, and the plurality of basis functions are configured to approximate each of the time series by successors of the respective linear combinations applied to the plurality of basis functions, and - When the electrochemical apparatus is subjected to each of the fundamental intermittent loads represented by the basis function, the computer determines a plurality of fundamental changes for each basis function, including each fundamental change in the performance indicator of the electrochemical apparatus. The prediction model includes: - A step of projecting the assumed intermittent load onto the plurality of basis functions, thereby obtaining a continuation of each linear combination applicable to the plurality of basis functions in order to approximate the assumed intermittent load. A method comprising the step of applying the successors of each of the linear combinations to the plurality of basic changes, thereby outputting a predicted change.

2. Determining the fundamental changes in each of the performance indicators of the electrochemical apparatus is possible for each basis function. - Obtain at least one physical instance of the electrochemical apparatus, - Provide at least one physical instance of the electrochemical apparatus to at least one actual instance of each fundamental intermittent load represented by a basis function, - Perform at least one measurement of a performance indicator on at least one physical instance of the electrochemical apparatus, and - The method according to claim 1, comprising calculating each of the underlying changes based on at least one measurement.

3. The method according to claim 2, wherein at least one measurement of a performance indicator in at least one physical instance of the electrochemical apparatus includes a first measurement and a second measurement, the first measurement being performed at the start of the at least one actual instance of the respective basic intermittent load, the second measurement being performed at the end of the at least one actual instance of the respective basic intermittent load, and the calculation of the respective basic change includes calculating the difference between a value obtained from the second measurement and a value obtained from the first measurement.

4. The method according to claim 2 or 3, wherein for at least one basis function, at least one actual instance of each of the underlying intermittent loads represented by the basis function comprises a plurality of actual instances of each of the underlying intermittent loads.

5. Each time series contains its own classification label, and machine learning with multiple basis functions is performed for each time series. - Classification error with respect to the classification label, and / or - Distance metric between the basis function and the provided time series The method according to claim 1, comprising minimizing.

6. - The distance metric is a dynamic time stretching method, and / or - The classification error is the cross-entropy loss. The method according to claim 5.

7. The method according to claim 1, wherein the plurality of basis functions consist of a plurality of shapelets.

8. Each shapelet includes a length parameter, and machine learning of the aforementioned multiple basis functions further - Including determining the length of each shapelet to minimize error, The method according to claim 7.

9. The machine learning of the aforementioned multiple basis functions is further - This includes selecting a subset of a predetermined size from the plurality of basis functions determined above, The selected subset is - The classification error in the first neighborhood of the minimized classification error, and / or - Having a distance metric for the second neighbor of the minimized distance metric, The method according to claim 7.

10. The aforementioned performance indicators - Efficiency in constant mode, - temperature signal, - Pressure signal, - Ion concentration signal, or - Represents the gas concentration signal, The method according to claim 1.

11. The aforementioned electrochemical apparatus, - Electrolytic cell connected to intermittent power supply, - An electrochemical energy storage device connected to an intermittent power supply, and - Electrochemical energy storage device connected to an intermittent load The method according to claim 1, comprising at least one of the following.

12. A computer implementation method for using a predictive model generated by the method described in claim 1 in each electrochemical apparatus, This method is - The computer acquires a hypothetical intermittent load, which is generated at least partially by one or more renewable energy sources. - The computer applies the prediction model to the assumed intermittent load, thereby outputting the predicted changes in the performance indicators of each electrochemical device when each electrochemical device is subjected to the assumed intermittent load. The application of the prediction model includes o Project the assumed intermittent load onto multiple basis functions, thereby obtaining successors to each of the linear combinations applicable to the multiple basis functions to approximate the assumed intermittent load, and Applying the successors of each of the above linear combinations to multiple fundamental changes, thereby outputting a predicted change. Methods that include...

13. A computer program comprising instructions for performing the method described in claim 1.

14. A computer-readable recording medium having the computer program described in claim 13 recorded on it.

15. A system including a processor coupled to memory, wherein the memory stores the computer program described in claim 13.

Images