ELM (Extreme Learning Model) model-based online cycle prediction method for residual life of single battery

Abstract

The invention relates to the field of battery health management, in particular to an ELM model-based single battery residual life online cycle prediction method, which comprises the following steps of: extracting battery cycle data from a database, and dividing the battery cycle data into a data set; constructing an ELM (Extreme Learning Model), and training the ELM based on the data set to obtain a health degree predicted value of the single battery; performing loop iteration prediction based on the health degree prediction value of the single battery and an ELM model to obtain a residual life prediction value of the single battery; performing data updating on the database on the basis of a health degree predicted value generated by loop iteration prediction; and repeating the above process based on the database completing data updating, and updating the residual life prediction value of the single battery. According to the method, large-data-volume single batteries can be quickly predicted in batches, the prediction precision is ensured, and large-data-volume read-write storage and model closed-loop online prediction are supported.

Description

Technical Field

[0001] The present invention relates to the field of battery health management, and in particular to an online cycle prediction method for the remaining service life of a single battery based on an ELM model. Background Art

[0002] The remaining life of a battery is one of the core indicators for measuring battery health. The prediction of the remaining life is crucial to ensure reliable battery operation and prevent accidents. The definition of a battery reaching its life span generally includes the following two types: first, the battery's state of health (SOH) drops to 80%; second, the battery's internal resistance is twice the initial value. In practical applications, the first method is more commonly used, and the implementation methods of the remaining life prediction include physics-based methods and data-driven methods. With the development of battery big data, data-driven prediction methods are increasingly valued, but the actual application of data-driven prediction methods is still relatively small.

[0003] At present, the online application of data-driven prediction methods on battery big data platforms faces the following challenges: First, the scale of energy storage power plants is large, and the amount of accumulated data of single cells (such as single lithium iron phosphate batteries) is very large. It is necessary to process massive amounts of single cell data, and the prediction model needs to balance model accuracy and efficiency; Second, the existing data-driven prediction methods for the remaining life of single cells are all based on offline trained models for prediction. Model iterative updates require retraining and deployment online, causing the model to lag behind changes in actual data, and the cost of retraining, deploying and testing the model is high; Third, data-driven prediction is used to provide external prediction services. After the model is predicted, a large amount of prediction data needs to be quickly read and stored, increasing the workload of data reading and storage. Summary of the invention

[0004] In order to solve the problem that the prediction model needs to balance model accuracy and efficiency when facing a large amount of single-cell battery data, the existing prediction model cannot operate online, resulting in the model operation lagging behind the actual data changes. Each time the model is run, it is necessary to retrain and deploy the model, which increases the prediction cost, and it is necessary to complete the reading and storage of a large amount of prediction data. The present invention provides an online cycle prediction method for the remaining life of a single-cell battery based on the ELM model, and the method includes the following steps: Step S1, extracting battery cycle data from a database and dividing the battery cycle data into data sets; constructing an ELM model, training the ELM model based on the data set, and obtaining a predicted value of the health of a single battery; Step S2, based on the health prediction value of the single cell and the ELM model, perform cyclic iterative prediction to obtain the remaining life prediction value of the single cell; based on the health prediction value generated by the cyclic iterative prediction, update the data in the database; Step S3, based on the database that has completed data update, repeat the above steps S1-S2 to update the remaining life prediction value of the single battery.

[0005] Preferably, the battery cycle data is extracted from the database, specifically: A partition table is established for the CLICKHOUSE database, and the health data of the single battery and the current cycle iteration prediction number are stored based on the partition table; Extract the health data of the single cell battery from the corresponding partition table in the CLICKHOUSE database.

[0006] Preferably, the battery cycle data is divided into data sets, specifically: The first 90% of the extracted single cell battery health data is divided into a training data set, and the last 10% is divided into a test data set.

[0007] Preferably, an ELM model is constructed and trained based on the data set to obtain the predicted value of the health of the single battery, specifically: Step S11, constructing an ELM model and initializing parameters of the ELM model; Step S12, performing function activation on the output of the ELM model after initializing the parameters; Step S13, performing least squares multiplication on the output after function activation and the label of the training data set to obtain the input weight W of the ELM model; Step S14, obtaining a predicted value of the health of the single cell based on the test data set and the input weight W; Step S15, based on the root mean square error of the health prediction value of the single cell battery, determine whether the health prediction value of the single cell battery is credible; if so, directly output the health prediction value; if not, repeat the above steps S11-S14 until the health prediction value is credible.

[0008] Preferably, the ELM model initialization parameters are specifically: Initialize the number of hidden layers of the ELM model to t, and input the length of the training data set and a random input weight into the ELM model.

[0009] Preferably, the output of the ELM model after initializing the parameters is activated by a function, specifically: Based on the training data set, the random input weights and the hidden layer bias of the ELM model, the output of the ELM model is determined as the output after the ELM model parameters are initialized; Based on the ReLU function, the output after activating the ELM model initialization parameters; Based on the test data set and the input weight W, the predicted value of the health of the single battery is obtained, specifically: Perform a dot product operation on the test data set and the input weight W to obtain the predicted health value of the single battery.

[0010] Preferably, whether the predicted value of the health of the single cell battery is credible is judged based on the root mean square error of the predicted value of the health of the single cell battery, specifically: The root mean square error of the predicted value of the health of the single battery is compared with the preset error threshold. If the root mean square error is greater than the preset error threshold, it is judged that the predicted value of the health of the single battery is unreliable; otherwise, it is judged that the predicted value of the health of the single battery is credible.

[0011] Preferably, based on the predicted value of the health of the single cell and the ELM model, a cyclic iterative prediction is performed to obtain the predicted value of the remaining life of the single cell, specifically: The predicted health value of the single battery is input into the ELM model as the initial value for cyclic iterative prediction, and a new predicted health value is obtained in each iterative prediction; Based on the new health prediction value, the remaining life prediction value of the single battery is obtained.

[0012] Preferably, based on the new health prediction value, the remaining life prediction value of the single battery is obtained, specifically: Determine whether the new health prediction value obtained by each iterative prediction is equal to or less than 80%; if so, determine the remaining life prediction value of the single battery based on the number of cycle iteration predictions currently completed by the ELM model.

[0013] Preferably, based on the health prediction value generated by the cyclic iterative prediction, the database is updated, specifically: The health prediction value generated by the cyclic iterative prediction is written into the partition table for storing the health data of the single battery in the CLICKHOUSE database, thereby realizing the data update of the CLICKHOUSE database.

[0014] Compared with the prior art, the present invention has the following beneficial effects: First, the present invention can quickly predict the single battery with large amounts of data in batches for large-scale power stations and battery big data platforms, and can ensure the accuracy of the prediction, effectively taking into account the efficiency and accuracy of big data processing; Second, the present invention only needs a small amount of historical cycle data of the single cell to predict the remaining life data of the single cell, and can continuously accumulate and correct the remaining life data obtained by each prediction, thereby improving the prediction accuracy of the remaining life of the single cell; Third, using the CLICKHOUSE database as a data storage platform can support fast reading and writing of large amounts of data, improve the efficiency of calling and extracting historical cycle data of single cells and the effect of writing back and storing prediction results; Fourth, it supports closed-loop online training and prediction of the ELM model, ensuring the synchronization of ELM model operation with actual data changes. There is no need to retrain and deploy the model every time the model is run, reducing the cost of model prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them: Figure 1 It is a flow chart of the online cycle prediction method for the remaining service life of a single cell battery based on the ELM model provided by the present invention.

[0016] Figure 2 It is a structural diagram of the ELM model in the online cycle prediction method for the remaining service life of a single cell battery based on the ELM model provided by the present invention.

[0017] Figure 3 It is a prediction result diagram of the online cycle prediction method for the remaining service life of a single cell battery based on the ELM model provided by the present invention. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings. It is understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only the parts related to the present invention rather than all structures are shown in the accompanying drawings. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0019] The terms "including" and "having" and any variations thereof in the present invention are intended to cover non-exclusive inclusions. For example, a process, method, method, product or device comprising a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products or devices.

[0020] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0021] See also Figure 1 As shown, the present invention provides an online cycle prediction method for the remaining service life of a single cell battery based on an ELM model, the method comprising the following steps: Step S1, extracting battery cycle data from a database and dividing the battery cycle data into data sets; constructing an ELM model, training the ELM model based on the data set, and obtaining a predicted value of the health of a single cell battery.

[0022] Further, the battery cycle data is extracted from the database, specifically: A partition table is established for the CLICKHOUSE database, and the health data of the single battery and the current cycle iteration prediction number are stored based on the partition table; Extract the health data of the single cell battery from the corresponding partition table in the CLICKHOUSE database.

[0023] The online cycle prediction method for the remaining life of a single cell battery based on the ELM model of the present invention is implemented on the basis of a feedforward neural network and a column storage database, wherein the feedforward neural network may be, but not limited to, an ELM model (Extreme Learning Machine), and the column storage database may be, but not limited to, a CLICKHOUSE database. The CLICKHOUSE database can be used to partition and store battery cycle data, providing fast data extraction and prediction result write-back effects for the ELM model to call data online. Specifically, a partition table is established in the CLICKHOUSE database through a big data platform, and the partition table stores the health data (SOH data) of the single cell battery and the initial number of cycles corresponding to the online cycle prediction of the ELM model. Extracting the health data of the single cell battery from the corresponding partition table in the CLICKHOUSE database can improve the efficiency and accuracy of the health data call and avoid erroneous call of other data. The number of cycles is recorded according to the actual situation. If the actual situation is not recorded, the number of cycles is set to 1, and the remaining life RUL of the single cell battery can be calculated by the following formula: In the above formula, The ELM model predicts the number of cycles corresponding to the predicted health of the single battery when it is 80% of the initial health. The initial number of cycles corresponding to the online cycle prediction of the ELM model.

[0024] Furthermore, the battery cycle data is divided into data sets, specifically: The first 90% of the extracted single cell battery health data is divided into a training data set, and the last 10% is divided into a test data set.

[0025] In order to effectively and accurately train and test the ELM model, the first 90% of the health data of the single battery extracted from the CLICKHOUSE database is determined as the training data set, and the last 10% of the data is determined as the test data set, thereby ensuring the prediction accuracy of the ELM model.

[0026] Furthermore, an ELM model is constructed and trained based on the data set to obtain the predicted value of the health of the single battery, specifically: Step S11, constructing an ELM model and initializing parameters of the ELM model; Step S12, performing function activation on the output of the ELM model after initializing the parameters; Step S13, performing least squares multiplication on the output after function activation and the label of the training data set to obtain the input weight W of the ELM model; Step S14, obtaining a predicted value of the health of the single cell based on the test data set and the input weight W; Step S15, based on the root mean square error of the health prediction value of the single cell battery, determine whether the health prediction value of the single cell battery is credible; if so, directly output the health prediction value; if not, repeat the above steps S11-S14 until the health prediction value is credible.

[0027] Furthermore, the ELM model initialization parameters are as follows: Initialize the number of hidden layers of the ELM model to t, and input the length of the training data set and a random input weight into the ELM model.

[0028] Furthermore, the output of the ELM model after initializing the parameters is activated by function, specifically: Based on the training data set, the random input weights and the hidden layer bias of the ELM model, the output of the ELM model is determined as the output after the ELM model parameters are initialized; Based on the ReLU function, the output after activating the ELM model initialization parameters; Based on the test data set and the input weight W, the predicted value of the health of the single battery is obtained, specifically: Perform a dot product operation on the test data set and the input weight W to obtain the predicted health value of the single battery.

[0029] Furthermore, based on the root mean square error of the predicted value of the health of the single battery, it is judged whether the predicted value of the health of the single battery is credible, specifically: The root mean square error of the predicted value of the health of the single battery is compared with the preset error threshold. If the root mean square error is greater than the preset error threshold, it is judged that the predicted value of the health of the single battery is unreliable; otherwise, it is judged that the predicted value of the health of the single battery is credible.

[0030] See also Figure 2 As shown, the present invention provides an ELM model in the online cycle prediction method of the remaining life of a single battery based on the ELM model. The ELM model consists of three parts: an input layer, a hidden layer, and an output layer. The input data is the number of cycles. , the expected output is the health of the single battery The input layer and the hidden layer are connected through the input weight W and the hidden layer bias b, and the hidden layer and the output layer are connected through the output weight β.

[0031] Set the number of hidden layers to L. The value of L can be determined through grid search. The training data set is as follows: , In the above dataset, and are the number of cycles and health, i=1, 2, ..., N; The calculation process of the ELM model can be expressed as: , j = 1, 2, ..., N In the above formula, N is the number of samples in the training data set; g is the activation function; is the input weight, ; is the output weight; is the bias of the i-th neuron in the hidden layer.

[0032] When initializing the parameters of the ELM model, the number of hidden layers of the ELM model is set to t, for example, t can be 4, and the length of the training data set and a random input weight are input to the ELM model. The random input weight can be selected according to the actual situation of the user, as long as it satisfies the operation of the ELM model, and no specific restrictions are made here. In the process of initializing the parameters, the input and The dot product of , plus the bias of the i-th neuron in the hidden layer .

[0033] In practical applications, the ReLU function (i.e., linear rectification function / corrected linear unit) can be selected as the activation function, and its implementation method is: if the input x is positive, the output of the ReLU function is x itself; if the input x is negative or zero, the output of the ReLU function is zero. Selecting the ReLU function as the activation function can simplify the computational complexity. Compared with selecting the sigmoid function or the tanh function as the activation function, the implementation of the ReLU function only involves threshold judgment, does not involve exponential operations, and has higher computational efficiency; the ReLU function can also significantly reduce the ELM model training time and computational cost, thereby accelerating the model convergence speed, improving the nonlinearity of the neural network model, and introducing necessary nonlinearity.

[0034] The output after function activation is solved by the least squares multiplication with the label of the training data set to obtain the input weight W of the ELM model, and then the dot product operation is performed on the test data set and the input weight W to obtain the health prediction value of the single cell. In actual calculations, there will be a prediction error in the health prediction value calculated each time. If the prediction error is too large, it means that the health prediction value currently calculated cannot accurately represent the true health of the single cell. Therefore, it is necessary to judge whether the health prediction value currently calculated is credible. Specifically, the root mean square error RMSE of the health prediction value is first calculated, and then the root mean square error RMSE is compared with the threshold value to judge whether the health prediction value currently calculated is credible. The calculation of the above root mean square error RMSE belongs to the conventional technical means in this field and will not be introduced in detail here.

[0035] Step S2, based on the health prediction value of the single cell battery and the ELM model, perform cyclic iterative prediction to obtain the remaining life prediction value of the single cell battery; based on the health prediction value generated by the cyclic iterative prediction, update the database data.

[0036] Furthermore, based on the predicted value of the health of the single cell and the ELM model, a cyclic iterative prediction is performed to obtain the predicted value of the remaining life of the single cell, specifically: The predicted health value of the single battery is input into the ELM model as the initial value for cyclic iterative prediction, and a new predicted health value is obtained in each iterative prediction; Based on the new health prediction value, the remaining life prediction value of the single battery is obtained.

[0037] Furthermore, based on the new health prediction value, the remaining life prediction value of the single battery is obtained, which is specifically: Determine whether the new health prediction value obtained by each iterative prediction is equal to or less than 80%; if so, determine the remaining life prediction value of the single battery based on the number of cycle iteration predictions currently completed by the ELM model.

[0038] Furthermore, based on the health prediction value generated by the cyclic iterative prediction, the database is updated, specifically: The health prediction value generated by the cyclic iterative prediction is written into the partition table for storing the health data of the single battery in the CLICKHOUSE database, thereby realizing the data update of the CLICKHOUSE database.

[0039] In the process of using the ELM model for cyclic iterative prediction, after each iterative prediction obtains a new health prediction value of the single battery, it is determined whether the new health prediction value has reached (equal to or less than) 80%. If so, it indicates that the single battery has reached its life span. At this time, based on the number of cyclic iterative predictions currently completed by the ELM model and combined with the above formula , calculate the remaining life prediction value of the single cell battery. Then write the health prediction value generated by the cyclic iterative prediction into the partition table used to store the health data of the single cell battery in the CLICKHOUSE database, and use it as the training data for the next prediction of the remaining life of the single cell battery, thereby realizing the data update of the CLICKHOUSE database.

[0040] The cloud platform for large-scale power stations and battery big data collects a large amount of single battery data every day, which makes the result data of the remaining life prediction of single batteries also very large. Whether reading single battery data from the database or writing the result data back to the database, the reading and writing of such large amounts of data place high demands on the performance of the database. The existing Mysq database and Redis database are not suitable for reading and writing big data. The CLICKHOUSE database has the advantages of high concurrency, low latency, high compression ratio, and easy scalability. It can solve the reading and storage of large-scale data, so the CLICKHOUSE database is selected as the database of the ELM model.

[0041] Step S3, based on the database that has completed data update, repeat the above steps S1-S2 to update the remaining life prediction value of the single battery.

[0042] The database that has completed data update includes the health prediction values ​​generated by each iterative prediction of the ELM model. These health prediction values ​​can be used as training data for the next prediction of the ELM model, which can improve the prediction accuracy of the ELM model.

[0043] See also Figure 3 As shown, the present invention provides a prediction result diagram of a method for online cycle prediction of the remaining service life of a single cell battery based on the ELM model. Figure 3The corresponding prediction method is used to predict the single cell of a specific battery model in a laboratory environment. The corresponding CLICKHOUSE database of this prediction method contains 1705 battery cycle data, and the cycle iteration prediction of the ELM model is about the health of the single cell decreasing from 100% to 78.9%. The health of the single cell is 79.9 at the 1621st cycle iteration prediction, that is, the 1621st cycle iteration prediction is the prediction terminal. The ELM model selects the health data of the first 300 battery cycle data as the training data set, and iterates the prediction until the health of the single cell is equal to or less than 80%, which is used as the condition for the ELM model to perform cycle iteration prediction, thereby verifying the effectiveness of the ELM model in predicting the remaining life of single cells.

[0044] Continue reading Figure 3 The EML model predicts that the end of the life of the single cell battery is at the 1606th cycle, while the actual end of the life of the single cell battery is at the 1621st cycle. The difference between the two is 15 cycles, and the above difference is within the acceptable error range. In addition, the average time taken for the entire process of single cell battery life predicted by the EML model (including batch reading of data, online model training and prediction, and prediction result writing back) is 22.9 seconds (10 rounds of testing, 240 single cells predicted in each round), and the average test time for each single cell is 0.0095 seconds. It can be seen that the EML model can achieve fast and accurate prediction of the remaining life of single cells.

[0045] Through the description of the above implementation methods, technicians in this field can clearly understand that each implementation method can be implemented by adding a necessary general hardware platform, and of course can also be implemented by combining hardware and software. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a computer product, and the present invention can be in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it, and other embodiments may also be used. Although the present invention has been described in detail with reference to the aforementioned embodiments, ordinary technicians in this field should understand that they can still modify the technical solutions recorded in the aforementioned embodiments, or replace some of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An online cycle prediction method for the remaining life of a single battery based on the ELM model, characterized in that: The method comprises the following steps: Step S1, extracting battery cycle data from a database, dividing the battery cycle data into a data set; constructing an ELM model, and training the ELM model based on the data set to obtain a health prediction value of a single battery; Step S2, performing cyclic iterative prediction based on the health prediction value of the single cell and the ELM model to obtain the remaining life prediction value of the single cell; and updating the database based on the health prediction value generated by the cyclic iterative prediction; Step S3, based on the database that has completed data update, repeat the above steps S1-S2 to update the remaining life prediction value of the single battery.

2. The method according to claim 1, characterized in that The extracting of battery cycle data from the database is specifically as follows: A partition table is established for the CLICKHOUSE database, and health data of the single battery and the current predicted number of cycle iterations are stored based on the partition table; The health data of the single cell battery is extracted from the corresponding partition table in the CLICKHOUSE database.

3. The method according to claim 2, characterized in that The dividing of the battery cycle data into data sets is specifically as follows: The first 90% of the extracted single cell battery health data is divided into a training data set, and the last 10% is divided into a test data set.

4. The method according to claim 3, characterized in that: The ELM model is constructed, and based on the data set, the ELM model is trained to obtain the health prediction value of the single battery, specifically: Step S11, constructing an ELM model and initializing parameters of the ELM model; Step S12, performing function activation on the output of the ELM model after initializing the parameters; Step S13, performing least squares multiplication on the output activated by the function and the label of the training data set to obtain the input weight W of the ELM model; Step S14, obtaining a predicted value of the health of a single cell based on the test data set and the input weight W; Step S15, judging whether the predicted value of the health of the single cell battery is credible based on the root mean square error of the predicted value of the health of the single cell battery; If yes, the health prediction value is directly output; if no, the above steps S11-S14 are repeated until the health prediction value is credible.

5. The method according to claim 4, characterized in that The ELM model initialization parameters are specifically: Initialize and set the number of hidden layers of the ELM model to t, and input the length of the training data set and a random input weight into the ELM model.

6. The method according to claim 5, characterized in that The function activation is performed on the output after the ELM model parameters are initialized, specifically: Determine the output of the ELM model based on the training data set, the random input weights and the hidden layer bias of the ELM model, and use this as the output of the ELM model after initializing the parameters; Based on the ReLU function, the output after activating the initialization parameters of the ELM model; The health prediction value of the single cell battery is obtained based on the test data set and the input weight W, specifically: A dot product operation is performed on the test data set and the input weight W to obtain a predicted value of the health of the single cell.

7. The method according to claim 6, characterized in that The step of judging whether the predicted value of the health of the single cell battery is credible based on the root mean square error of the predicted value of the health of the single cell battery is specifically as follows: Comparing the root mean square error of the predicted value of the health of the single battery with a preset error threshold, and if the root mean square error is greater than the preset error threshold, determining that the predicted value of the health of the single battery is unreliable; Otherwise, it is determined that the predicted value of the health status of the single cell is credible.

8. The method according to claim 1, characterized in that The method of performing cyclic iterative prediction based on the predicted health value of the single cell battery and the ELM model to obtain the predicted remaining life of the single cell battery is specifically as follows: Inputting the predicted health value of the single battery as an initial value into the ELM model for cyclic iterative prediction to obtain a new predicted health value obtained by each iterative prediction; Based on the new health prediction value, the remaining life prediction value of the single battery is obtained.

9. The method according to claim 8, characterized in that The remaining life prediction value of the single battery is obtained based on the new health prediction value, specifically: Determine whether the new health prediction value obtained by each iterative prediction is equal to or less than 80%; if so, determine the remaining life prediction value of the single cell based on the number of cycle iterative predictions currently completed by the ELM model.

10. The method according to claim 9, characterized in that The updating of the database based on the predicted health value generated by the cyclic iterative prediction is specifically as follows: The health prediction value generated by the cyclic iterative prediction is written into a partition table in the CLICKHOUSE database for storing health data of single cells, thereby realizing data update of the CLICKHOUSE database.

Images