Probability prediction method and system for safe temperature interval of echelon utilization energy storage system

Abstract

According to the echelon utilization energy storage system safety temperature interval-oriented probability prediction method and system, a temperature prediction model is established based on a graph neural network and a full connection layer, and a temperature probability distribution prediction model is established based on the graph neural network and the last Bayesian layer; establishing a temperature prediction error model; carrying out model training by utilizing the temperature and the historical value of the temperature correlation monitoring quantity; establishing a model of a temperature interval by using the temperature probability distribution in the prediction window, an absolute value of a prediction error of the temperature mean value corresponding to the temperature probability distribution in the prediction window, and a scaling factor; when the confidence coefficient mean value of the temperature interval is equal to a set threshold value, determining an optimal value of a scaling factor; and the temperature of the echelon utilization energy storage system in the operation state and the monitoring value of the temperature correlation monitoring quantity are obtained, the temperature probability distribution prediction model, the temperature prediction error model and the scaling factor optimal value are utilized, the safe temperature interval of the echelon utilization energy storage system is obtained, and a basis is provided for monitoring the operation state of the energy storage system.

Description

Technical Field

[0001] The present invention belongs to the technical field of evaluating the health state of a cascade - utilization energy storage system. Specifically, it relates to a probability prediction method and system for the safe temperature range of a cascade - utilization energy storage system. Background Art

[0002] In a new power system, the proportion of renewable energy represented by wind power and photovoltaic power is continuously increasing. Due to the intermittency and instability of natural resources such as wind energy and solar energy, the power generation of renewable energy is highly random and volatile, making it difficult to maintain the power balance of the power system. The demand for power grid peak shaving and frequency modulation surges. Traditional peak - shaving and frequency - modulation means represented by thermal power units are difficult to effectively respond. Therefore, it is urgent to introduce battery energy storage power stations into the power system to improve the stability, reliability, and efficiency of the system.

[0003] However, the high cost of batteries has become one of the main obstacles restricting the large - scale popularization and application of battery energy storage power stations in the new power system. Building a cascade - utilization battery energy storage power station based on retired batteries to participate in peak - shaving and frequency - modulation ancillary services brings new options for reducing the construction cost of power stations. However, for a cascade - utilization battery energy storage power station, ensuring safe operation is the primary issue. Retired batteries have served in various scenarios, with different degrees of degradation, uneven performance and quality, and a higher failure probability than brand - new energy storage batteries. When responding to peak - shaving and frequency - modulation ancillary services, energy storage batteries may need to charge and discharge frequently, further increasing the probability of thermal runaway. Therefore, it is urgent to establish an effective temperature monitoring method for the cascade - utilization energy storage system.

[0004] The internal electrochemical reactions of energy storage batteries are complex, and there are many factors that can affect their states. It is difficult to monitor the change state of the temperature of the energy storage system using methods driven by physical and chemical models. Existing data - driven methods often predict the maximum temperature and minimum temperature during the normal operation of the energy storage system separately to give the safe temperature range of the energy storage system, which may dilute the correlation between the upper and lower bounds of the temperature range and affect the prediction accuracy. Summary of the Invention

[0005] To solve the deficiencies in the prior art, the present invention provides a probability prediction method and system for the safe temperature range of a cascade - utilization energy storage system. By constructing the probability distribution of the temperature at the moment to be predicted, obtaining the upper and lower bounds of the temperature, and then constructing the safe temperature range of the energy storage system, it provides a basis for monitoring the operating state of the energy storage system.

[0006] The present invention adopts the following technical solutions.

[0007] The present invention proposes a probability prediction method for the safe temperature range of a cascade - utilization energy storage system, including: Obtain the historical values of the temperature of the cascade utilization energy storage system under normal operating conditions and the historical values of the temperature-related monitoring quantities; A temperature prediction model is established based on the graph neural network and the fully connected layer, and a temperature probability distribution prediction model is established based on the graph neural network and the last layer of Bayesian. A temperature prediction error model is established using a shallow machine learning method. The temperature prediction model and the temperature prediction error model are trained using the historical values of temperature and the historical values of temperature-related monitoring quantities, and the parameters of the graph neural network in the trained temperature prediction model are set as the initial parameters of the graph neural network in the temperature probability distribution prediction model; the temperature probability distribution prediction model after the initial parameters are set is trained, and the absolute value of the prediction error of the temperature mean corresponding to the temperature probability distribution in the prediction window output by the trained temperature probability distribution prediction model is output by the trained temperature prediction error model; The temperature interval model is established by using the temperature probability distribution in the prediction window, the absolute value of the prediction error of the temperature mean corresponding to the temperature probability distribution in the prediction window, and the scaling factor; when the confidence mean of the temperature interval is equal to the set threshold, the optimal value of the scaling factor is determined; The monitoring values of the temperature and temperature-related monitoring quantities of the cascade utilization energy storage system in operation are obtained, and the safe temperature range of the cascade utilization energy storage system is obtained by using the temperature probability distribution prediction model, the temperature prediction error model and the optimal value of the scaling factor.

[0008] Preferably, obtaining the historical value of the temperature of the cascade utilization energy storage system in a normal operating state and the historical value of the temperature-related monitoring quantity includes: Collect historical values of different monitoring quantities associated with temperature, including: state of charge, current, voltage, and temperature; Calculate the correlation coefficient between each associated monitoring quantity, and determine each monitoring quantity with a correlation relationship based on the correlation coefficient; Divide the historical values of temperature and temperature-related monitoring quantities into sequences of different time periods, and detect suspected abnormal data in the sequence of each time period; If other monitoring quantities that are associated with the monitoring quantity to which the suspected abnormal data belongs also have suspected abnormal data at the same time node, the suspected abnormal data is determined to be abnormal state data; otherwise, with the suspected abnormal data as the center of the sliding window, it is determined whether the absolute value of the difference between the normalized value of the abnormal data and the mean value of the normalized data in the sliding window is greater than the set abnormal threshold. If so, the suspected abnormal data is determined to be abnormal state data; after the abnormal state data is eliminated and interpolated and repaired, it is combined with the historical values of the temperature to form a multi-segment historical normal data time series.

[0009] Preferably, each period of normal historical data time series satisfies the following relationship:

[0010] In the formula, X is the historical normal data time series, , , , ……, constitute the historical value time series of temperature, , , , ……, is the historical value time series of the th temperature-related monitored quantity, , is the total number of monitored quantities, is the length of each time series in the historical normal data time series X; In each segment of the historical normal data time series, taking the time node in the historical normal data time series as the establishment moment of the sliding window, taking as the sliding window length, taking as the prediction window length, from the moment to the moment constitutes the inside of the sliding window, and from the moment to the moment constitutes the inside of the prediction window.

[0011] Preferably, the fully connected layer is the last layer of the temperature prediction model; The Bayesian last layer is the last layer of the temperature probability distribution prediction model; Moreover, the graph neural network in the temperature prediction model has the same structure as the graph neural network in the temperature probability distribution prediction model.

[0012] Preferably, according to the historical value of the temperature and the historical value of the temperature-related monitored quantity in the sliding window, use the temperature prediction model to output the temperature of the cascade utilization energy storage system in the prediction window; The input data of the temperature prediction model satisfies the following relational formula:

[0013] In the formula, is the input data of the temperature prediction model, , , , ……, are the historical values of the temperature in the sliding window, , , , ……, are the historical values of the temperature-related monitored quantity in the sliding window, , is the total number of temperature - related monitored quantities; The output data of the temperature prediction model satisfies the following relational expression:

[0014] In the formula, is the output data of the temperature prediction model, 、 、 、……、 are the temperatures within the prediction window.

[0015] Preferably, according to the historical values of the temperature within the sliding window and the historical values of the temperature - related monitored quantities, the temperature probability distribution prediction model is used to output the temperature probability distribution of the cascade - utilization energy storage system within the prediction window; The input data of the temperature probability distribution prediction model satisfies the following relational expression:

[0016] In the formula, is the input data of the temperature probability distribution prediction model, 、 、 、……、 are the historical values of the temperature within the sliding window, 、 、 、……、 are the historical values of the temperature - related monitored quantities within the sliding window, , is the total number of temperature - related monitored quantities.

[0017] Preferably, it is set that the temperature probability distribution within the prediction window follows a Gaussian distribution The output data of the temperature probability distribution prediction model satisfies the following relational expression:

[0018] In the formula, is the output data of the temperature probability distribution prediction model, 、 、 、……、 are the temperature means within the prediction window, 、 、 、……、 are the temperature standard deviations within the prediction window.

[0019] Preferably, according to the temperature within the prediction window, the temperature prediction error model is used to output the absolute value of the prediction error of the temperature of the cascade - utilization energy storage system within the prediction window; When the temperature prediction error model learns the prediction error of temperature, the input data of the temperature prediction error model satisfies the following relational expression:

[0020] wherein, is the input data of the temperature prediction error model, , , ...... are the temperatures within the prediction window; The output data of the temperature prediction error model satisfies the following relational expression:

[0021] wherein, is the output data of the temperature prediction error model, , , ...... are the absolute values of the temperature prediction errors within the prediction window.

[0022] Preferably, using the historical values of temperature and the historical values of temperature-related monitored quantities to train the temperature prediction model and the temperature prediction error model, including: Taking the historical values of temperature and the historical values of temperature-related monitored quantities within the sliding window as the first training data set, and taking the historical values of temperature within the prediction window as the first label data; Using the first training data set and the first label data to train the temperature prediction model; during the training process, the temperature within the prediction window output by the temperature prediction model is a deterministic estimate of the historical values of temperature within the prediction window, having a corresponding relationship; Taking the temperature within the prediction window output by the temperature prediction model as the second training data set, and taking the absolute value of the difference between the temperature within the prediction window and the corresponding first label data as the second label data; Using the second training data set and the second label to train the temperature prediction error model, and the trained temperature prediction error model outputs the absolute value of the prediction error of the temperature within the prediction window.

[0023] Preferably, using the first training data set and the first label data to train the temperature probability distribution prediction model after setting the initial parameters, and according to the temperature probability distribution within the prediction window output by the trained temperature probability distribution prediction model, the absolute value of the prediction error of the temperature mean corresponding to the temperature probability distribution within the prediction window is output by the trained temperature prediction error model.

[0024] Preferably, the model of the temperature range satisfies the following relational expression:

[0025] In the formula, and are respectively the upper bound correction value and the lower bound correction value of the temperature probability distribution within the prediction window, is the temperature mean within the prediction window, is the variance of the temperature within the prediction window, is the absolute value of the prediction error of the temperature probability distribution within the prediction window, is the scaling factor; The upper bound correction value and the lower bound correction value of the temperature form a temperature interval, and the mean confidence level of the temperature interval satisfies the following relationship:

[0026] In the formula, is the mean confidence level of the temperature interval, is the total number of groups of the upper bound correction value and the lower bound correction value of the temperature, is the j-th label value, is the j-th indicator function, which takes 1 when is within the temperature interval and 0 otherwise; When the mean confidence level is equal to the set confidence level, the optimal value of the scaling factor is determined, and the set confidence level is 0.95.

[0027] Preferably, based on the current moment a retrospective window with a length of m is established, from the moment to the moment constitutes the retrospective window, and from the moment to the moment constitutes the prediction period, is the length of the prediction period; The monitoring values of the temperature and the temperature-related monitoring quantities within the retrospective window are obtained and input into the temperature probability distribution prediction model to output the temperature probability distribution of the prediction period; based on the temperature probability distribution of the prediction period, the temperature mean of the prediction period is determined; the absolute value of the prediction error of the temperature mean of the prediction period is obtained by inputting the temperature mean of the prediction period into the temperature prediction error model; The input data of the temperature prediction error model satisfies the following relationship:

[0028] In the formula, is the input data of the temperature prediction error model, , , , ……, are the temperature means within the prediction period; Using the average temperature in the prediction period, the absolute value of the prediction error of the average temperature in the prediction period, and the optimal value of the scaling factor, according to the temperature interval model, determine the upper bound correction value and the lower bound correction value of the temperature in the prediction period, and form the safe temperature interval in the prediction period.

[0029] The present invention also proposes a probability prediction system for the safe temperature interval of a cascade utilization energy storage system, including: An acquisition module, configured to obtain the historical values of the temperature and the historical values of the temperature-related monitoring quantities of the cascade utilization energy storage system in the normal operating state; A model establishment module, configured to establish a temperature prediction model based on a graph neural network and a fully connected layer, and establish a temperature probability distribution prediction model based on a graph neural network and a Bayesian last layer; establish a temperature prediction error model by using a shallow machine learning method; A model training module, configured to use the historical values of the temperature and the historical values of the temperature-related monitoring quantities to train the temperature prediction model and the temperature prediction error model, and set the parameters of the graph neural network in the trained temperature prediction model as the initial parameters of the graph neural network in the temperature probability distribution prediction model; train the temperature probability distribution prediction model after setting the initial parameters, and according to the temperature probability distribution within the prediction window output by the trained temperature probability distribution prediction model, obtain the absolute value of the prediction error of the temperature mean corresponding to the temperature probability distribution within the prediction window output by the trained temperature prediction error model; A scaling factor optimization module, configured to use the temperature probability distribution within the prediction window, the absolute value of the prediction error of the temperature mean corresponding to the temperature probability distribution within the prediction window, and the scaling factor to establish a model of the temperature interval; when the confidence mean of the temperature interval is equal to the set threshold, determine the optimal value of the scaling factor; A prediction module, configured to obtain the monitoring values of the temperature and the temperature-related monitoring quantities of the cascade utilization energy storage system in the operating state, and use the temperature probability distribution prediction model, the temperature prediction error model, and the optimal value of the scaling factor to obtain the safe temperature interval of the cascade utilization energy storage system.

[0030] The present invention is also a terminal, including a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of the method.

[0031] The present invention is also a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the method are implemented.

[0032] The beneficial effects of the present invention are as follows: The present invention uses a probability prediction method to obtain the probability distribution of the operating temperature of the energy storage system at the prediction moment, and based on this probability distribution, the upper and lower bounds of the temperature during the safe operation of the energy storage system are obtained, which can make full use of the correlation relationship between the upper and lower bounds. The Bayesian last layer is selected as the implementation method of probability prediction, and the uncertainty quantification mechanism is introduced only in the last layer of the neural network model, which has few modifications to the previous model and has strong compatibility and high operation efficiency in practical applications. Based on the shallow machine learning method, a mapping between the temperature prediction value and the absolute value of the temperature prediction error is established, and based on the error amount and the established interval scaling factor, the safe temperature interval of the cascade utilization energy storage system in different operating modes is optimized and adjusted to adapt to different working modes of the cascade utilization energy storage system and different error modes of the prediction model, providing more accurate reference information for decision-makers. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 is a flowchart of a probability prediction method for the safe temperature interval of a cascade utilization energy storage system proposed by the present invention; Figure 2 is a schematic structural diagram of a temperature prediction model in an embodiment of the present invention; Figure 3 is a schematic structural diagram of a temperature probability distribution prediction model in an embodiment of the present invention; Figure 4 is a schematic diagram of the safe temperature interval of the energy storage system obtained in an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0034] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the detailed implementation manners of the embodiments of the present invention. The embodiments described in this application are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the spirit of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0035] The present invention provides a probability prediction method for the safe temperature interval of a cascade utilization energy storage system. A cascade utilization energy storage system is built using retired batteries, and the monitored quantities of the cascade utilization energy storage system include temperature; as Figure 1 shown, the method includes: Step 1: Obtain the historical values of the temperature and the historical values of the temperature-related monitored quantities of the cascade utilization energy storage system in the normal operating state.

[0036] Specifically, Step 1 includes: Step 1.1: Collect historical values of different monitored quantities of the energy storage system for cascaded utilization. In the embodiment, the historical monitoring data used for example testing is from an energy storage system that has been in long-term service. It consists of 6 battery sub-modules, and there are differences in the degree of aging. The monitored quantities of the energy storage system for cascaded utilization include but are not limited to: state of charge, current, voltage, and temperature; Among them, temperature includes temperature measurement values and / or temperature extreme values. Temperature extreme values include the highest temperature value and the lowest temperature value. When temperature only includes temperature extreme values, the average value of the highest temperature value and the lowest temperature value is used as the temperature measurement value; In the embodiment, the monitored quantities of the monitoring data of the energy storage system for cascaded utilization obtained are: state of charge, current, voltage, the highest temperature value, and the lowest temperature value.

[0037] Step 1.2: Use the correlation analysis method to determine the correlation coefficients between each pair of correlated monitored quantities, and establish a correlation relationship distribution matrix using the normalized values of all correlation coefficients; sort all elements in the correlation relationship distribution matrix from large to small, and use the elements in the first 70% of the sorting as candidate values; when the candidate value is not less than 0.7, determine that the two monitored quantities are correlated monitored quantities; In the embodiment, using the grey correlation analysis method to determine the correlation relationship between historical monitoring data is a non-limiting and relatively optimal choice. The correlation relationship matrix of the present invention is not only used to distinguish which variables are strongly correlated with the variables with anomalies during data cleaning, but also used to distinguish which variables are strongly correlated with temperature.

[0038] Due to the particularity of each battery cluster in the energy storage system for cascaded utilization, a specific method needs to be adopted to first determine the temperature-correlated monitored quantities, and then use the relevant data. However, there are "dirty" data in these historical values, which need to be cleaned; Step 1.3: Divide the historical values of temperature and temperature-correlated monitored quantities into sequences of different time periods; use unsupervised machine learning methods to detect abnormal data in the sequences of different time periods to obtain abnormal sequences with abnormal data; In the embodiment, using the local outlier factor detection algorithm to detect abnormal data in the sequence is a non-limiting and relatively optimal choice; Step 1.4: If the monitored quantities that are correlated with the monitored quantity corresponding to the abnormal data also have abnormal data at the same time node, determine the abnormal data as abnormal state data; otherwise, with the abnormal data as the center of the sliding window, after normalizing the data in the sliding window, determine whether the absolute value of the difference between the normalized value of the abnormal data and the mean value of the normalized data in the sliding window is greater than the set abnormal threshold. If it is greater, the abnormal data is used as the data to be cleaned; among them, the abnormal threshold is set to 0.5; Step 1.5: Use the interpolation method to correct all the data to be cleaned.

[0039] In the embodiment, using a linear interpolation algorithm to correct all the data to be cleaned is a non-limiting and preferred choice.

[0040] Step 1.6, after correction, remove the abnormal state data, filter out the historical values of temperature and temperature-related monitoring quantities of the cascaded energy storage system under normal operating conditions, and form multiple historical normal data time series. Each historical normal data time series satisfies the following relationship:

[0041] In the formula, X is the historical normal data time series, , , ,……, The time series of historical values of temperature, , , ,……, For the A time series of historical values of temperature-related monitoring quantities, , is the total number of monitored quantities, is the length of each time series in the historical normal data time series X; In each period of historical normal data time series, the time nodes in the historical normal data time series are Establish the moment for the sliding window, is the sliding window length, To predict the window length, from time To time Constitute a sliding window, from time To time The prediction window is constructed; the historical normal data time series is divided into two parts by using a sliding window and a prediction window. The data of the historical normal data time series in the sliding window is used as input data to obtain prediction data that corresponds one to one with the time series in the prediction window, and the data of the historical normal data time series in the prediction window is used as a label to verify and determine the error of the prediction data, thereby improving the accuracy and reliability of the model's data estimation.

[0042] Step 2: Establish a temperature prediction model based on graph neural network and fully connected layer.

[0043] Specifically, Figure 2 As shown, the temperature prediction model includes a graph neural network and a fully connected layer, where the fully connected layer is the last layer of the temperature prediction model. Specifically, based on the historical values of the temperature in the sliding window and the historical values of the temperature-related monitoring quantity, the temperature prediction model is used to output the temperature of the cascade utilization energy storage system in the prediction window; Input data for the temperature prediction model Including: historical values of temperature within a sliding window and historical values of temperature-related monitored quantities; wherein, the temperature-related monitored quantities include, but are not limited to: state of charge, current, voltage;

[0044] In the formula, is the input data of the temperature prediction model, is the establishment time of the sliding window, is the length of the sliding window, from time to time constitutes the sliding window, , , , ……, are the historical values of temperature within the sliding window, , , , ……, are the historical values of temperature-related monitored quantities within the sliding window, , is the total number of temperature-related monitored quantities; The output data of the temperature prediction model includes: temperature outside the sliding window; the output data of the temperature prediction model satisfies the following relational expression:

[0045] In the formula, is the output data of the temperature prediction model, is the establishment time of the sliding window, is the length of the prediction window, from time to time constitutes the prediction window, , , , ……, are the temperatures within the prediction window; In the embodiment, the graph neural network used is a Fourier graph neural network, and the temperature prediction model is composed of a Fourier graph neural network and a fully connected layer. The temperature-related monitored quantities in the embodiment include: state of charge, current, voltage, maximum temperature value, and minimum temperature value. The length m of the sliding window takes 6 in the embodiment, and the length n of the prediction window takes 1 in the embodiment; It should be particularly noted that the temperature prediction model can also be composed of other types of graph neural networks connected to a fully connected layer.

[0046] Step 3, establish a temperature probability distribution prediction model based on the graph neural network and the Bayesian last layer.

[0047] Specifically, as Figure 3As shown, the Bayesian last layer is the last layer of the temperature probability distribution prediction model; moreover, the graph neural network in the temperature prediction model has the same structure as the graph neural network in the temperature probability distribution prediction model.

[0048] Based on the historical values of the temperature and the historical values of the temperature-related monitored quantities within the sliding window, the temperature probability distribution of the cascade utilization energy storage system within the prediction window is output using the temperature probability distribution prediction model; based on the temperature probability distribution within the prediction window, the upper temperature bound, the lower temperature bound, and the temperature mean within the prediction period are determined.

[0049] The prediction of the safe temperature range in the prior art is usually in a deterministic prediction mode, considering the highest temperature value and the lowest temperature value in the safe temperature range as two monitored quantities, and directly predicting the values of these two monitored quantities at future moments through a time series prediction method to construct the safe temperature range; for the cascade utilization energy storage system built using retired battery clusters, the states and operating modes of each retired battery are different. This deterministic prediction method may dilute the correlation constraints existing between the highest temperature value and the lowest temperature value, and it is also difficult to obtain a safe temperature range that conforms to different operating modes, resulting in problems such as false alarms and inaccurate predictions. Therefore, the present invention proposes to use a probability prediction method to determine the safe temperature range of the cascade utilization energy storage system. By constructing a probability distribution describing the possible value situations of the temperature monitored quantities of the cascade utilization energy storage system, and based on the temperature probability distribution, the upper temperature bound and the lower temperature bound with associated constraints are determined, and then a scaling factor is constructed based on the prediction error of the temperature prediction mean to adapt to different operating modes of the cascade utilization energy storage system, facilitating a more accurate determination of the safe temperature range of the cascade utilization energy storage system.

[0050] The input data of the temperature probability distribution prediction model includes: the historical values of the temperature, the historical values of the temperature-related monitored quantities; The input data of the temperature probability distribution prediction model satisfies the following relational expression:

[0051] In the formula, is the input data of the temperature probability distribution prediction model, is the establishment moment of the sliding window, is the sliding window length, from the moment to the moment constitutes the sliding window, , , , ……, are the historical values of the temperature within the sliding window, , , , ……, is the historical value of the temperature correlation monitoring quantity within the sliding window, , is the total number of temperature correlation monitoring quantities; Since the graph neural network in the temperature prediction model has the same structure as the graph neural network in the temperature probability distribution prediction model, the input data of the temperature probability distribution prediction model is the same as the input data of the temperature prediction model; The output data of the temperature probability distribution prediction model includes: the temperature probability distribution within the prediction window; among them, the temperature probability distribution includes but is not limited to the Gaussian distribution; When the temperature probability distribution within the prediction window takes the Gaussian distribution the output data of the temperature probability distribution prediction model satisfies the following relational expression:

[0052] In the formula, is the output data of the temperature probability distribution prediction model, is the establishment time of the sliding window, is the length of the prediction window, from time to time constitutes the prediction window, , , ,……, is the temperature mean value within the prediction window, , , ,……, is the temperature standard deviation within the prediction window.

[0053] In the embodiment, the graph neural network used is the Fourier graph neural network, and the temperature probability distribution prediction model consists of the Fourier graph neural network and the Bayesian last layer. The temperature correlation monitoring quantities in the embodiment include: state of charge, current, voltage, maximum temperature value and minimum temperature value. The length m of the sliding window takes 6 in the embodiment, and the length n of the prediction window takes 1 in the embodiment; in the present invention, the graph neural network in the temperature probability distribution prediction model has the same structure as the graph neural network in the temperature prediction model. Since the fully connected layer is used in the temperature prediction model, a deterministic prediction is realized, while the uncertainty quantification mechanism is introduced in the temperature probability distribution prediction model by using the Bayesian last layer to realize probability prediction. When the model is actually deployed and applied, the temperature prediction model focuses on capturing the mapping relationship between the input data and the deterministic prediction value, so as to provide a more stable and clear error calculation benchmark for the subsequent temperature prediction error model, avoid the relatively broad mean estimation caused by the uncertainty estimation of the temperature probability distribution prediction model, and reduce the performance of the temperature prediction error model.

[0054] Step 4: Establish a temperature prediction error model using a shallow machine learning method.

[0055] In the embodiment, establishing a temperature prediction error model based on a support vector regression model is a non - restrictive and relatively optimal choice.

[0056] Specifically, according to the temperature within the prediction window, the absolute value of the prediction error of the temperature of the cascade - utilization energy storage system within the prediction window is output using the temperature prediction error model.

[0057] When obtaining the upper temperature bound and the lower temperature bound corresponding to each retired battery cluster, if the same set of uncertainty evaluation criteria is used for each retired battery with different states and operating modes, it will lead to a relatively wide temperature range obtained, resulting in problems such as false alarms or inaccurate temperature ranges. Therefore, the present invention establishes a temperature prediction error model to learn the temperature prediction error pattern during the prediction time period, obtains a matching prediction error based on the temperature prediction result, and thus corrects the temperature range established based on the probability distribution at each moment to adapt to the different working modes of the cascade - utilization energy storage system; When the temperature prediction error model learns the prediction error of the temperature, the input data of the temperature prediction error model includes: the temperature during the prediction period; the input data of the temperature prediction error model satisfies the following relational expression:

[0058] In the formula, is the input data of the temperature prediction error model, is the moment when the sliding window is established, is the length of the prediction window, from the moment to the moment constitutes the prediction window, , , , ……, are the temperatures within the prediction window; The output data of the temperature prediction error model includes: the absolute value of the temperature prediction error during the prediction period; the output data of the temperature prediction error model satisfies the following relational expression:

[0059] In the formula, is the output data of the temperature prediction error model, is the moment when the sliding window is established, is the length of the prediction window, from the moment to the moment constitutes the prediction window, , , , ……, is the absolute value of the temperature prediction error within the prediction window.

[0060] Step 5: Use the historical values of temperature and the historical values of temperature-related monitored quantities to train the temperature prediction model and the temperature prediction error model. Set the parameters of the graph neural network in the trained temperature prediction model as the initial parameters of the graph neural network in the temperature probability distribution prediction model. Train the temperature probability distribution prediction model after setting the initial parameters. According to the temperature probability distribution within the prediction window output by the trained temperature probability distribution prediction model, obtain the absolute value of the prediction error of the temperature mean corresponding to the temperature probability distribution within the prediction window output by the trained temperature prediction error model.

[0061] Specifically, using the historical values of temperature and the historical values of temperature-related monitored quantities to train the temperature prediction model and the temperature prediction error model includes: Use the historical values of temperature and the historical values of temperature-related monitored quantities within the sliding window as the first training dataset, and use the historical values of temperature within the prediction window as the first label data; Use the first training dataset and the first label data to train the temperature prediction model; in the embodiment, the temperature within the prediction window output during the training process by the temperature prediction model is a deterministic estimate of the historical values of temperature within the prediction window, having a corresponding relationship; Use the temperature within the prediction window output by the temperature prediction model to form the second training dataset, and use the absolute value of the difference between the temperature within the prediction window and the corresponding first label data as the second label data; Use the second training dataset and the second label to train the temperature prediction error model. The trained temperature prediction error model outputs the absolute value of the prediction error of the temperature within the prediction window; Through the above training process, the parameters of the temperature prediction model and the parameters of the temperature prediction error model are both optimized; at this time, extract the parameters of the graph neural network in the trained temperature prediction model and set them as the initial parameters of the graph neural network in the temperature probability distribution prediction model. Make full use of the characteristic that the graph neural network structures in the temperature prediction model and the temperature probability distribution prediction model are the same. Moreover, in the embodiment, the training of the temperature prediction model and the temperature prediction error model can be carried out offline. After the transfer setting of the neural network parameters, introduce an uncertainty estimation mechanism through the Bayesian last layer without changing the topological structure of the neural network. Therefore, the training process of the temperature prediction model and the temperature prediction error model is actually the initial training process of the temperature probability distribution prediction model, and only fine-tuning training of the temperature probability distribution prediction model is required during subsequent training, which is beneficial to reducing the computational amount and improving the model training efficiency.

[0062] Specifically, the temperature probability distribution prediction model after initial parameter setting is trained using the first training dataset and the first label data. According to the temperature probability distribution within the prediction window output by the trained temperature probability distribution prediction model, the absolute value of the prediction error of the temperature mean corresponding to the temperature probability distribution within the prediction window is output by the trained temperature prediction error model. In the embodiment, the temperature probability distribution within the prediction window output is an uncertainty estimate of the temperature historical values within the prediction window and also has a corresponding relationship.

[0063] Step 6: Establish a model for the temperature interval using the temperature probability distribution within the prediction window, the absolute value of the prediction error of the temperature mean corresponding to the temperature probability distribution within the prediction window, and the scaling factor. When the mean confidence level of the temperature interval is equal to the set threshold, determine the optimal value of the scaling factor.

[0064] Specifically, the model for the temperature interval satisfies the following relationship:

[0065] In the formula, and are the upper temperature correction value and the lower temperature correction value of the temperature probability distribution within the prediction window respectively, is the temperature mean within the prediction window, is the variance of the temperature within the prediction window, is the absolute value of the prediction error of the temperature probability distribution within the prediction window, is the scaling factor.

[0066] The upper temperature correction value and the lower temperature correction value form the temperature interval, and the mean confidence level of the temperature interval satisfies the following relationship:

[0067] In the formula, is the mean confidence level of the temperature interval, is the total number of groups of the upper temperature correction value and the lower temperature correction value, is the j-th label value, is the j-th indicator function, which takes 1 when is within the temperature interval and 0 otherwise.

[0068] When the mean confidence level is equal to the set confidence level, determine the optimal value of the scaling factor ; in the embodiment, the set confidence level is 0.95.

[0069] Step 7: Obtain the monitoring values of the temperature and the temperature-related monitoring quantities of the cascade utilization energy storage system in the operating state, and use the temperature probability distribution prediction model, the temperature prediction error model, and the optimal value of the scaling factor to obtain the safe temperature interval of the cascade utilization energy storage system.

[0070] The monitored values of the temperature and temperature-related monitored quantities of the energy storage system for cascade utilization in the operating state are real-time operating data; if the actually monitored temperature is outside the predicted safe temperature range, it indicates that the energy storage system for cascade utilization may be in an abnormal operating state and an alarm needs to be issued.

[0071] Based on the current moment A retrospective window with a length of m is established, starting from the moment to the moment to form a retrospective window, starting from the moment to the moment to form a prediction period, is the length of the prediction period; Obtain the monitored values of the temperature and temperature-related monitored quantities within the retrospective window and input them into the temperature probability distribution prediction model to output the temperature probability distribution of the prediction period; when the temperature probability distribution of the prediction period follows a Gaussian distribution, determine the temperature mean of the prediction period; input the temperature mean of the prediction period into the temperature prediction error model to obtain the absolute value of the prediction error of the temperature mean of the prediction period; use the temperature mean of the prediction period, the absolute value of the prediction error of the temperature mean of the prediction period, and the optimal value of the scaling factor to determine the upper temperature correction value and the lower temperature correction value of the prediction period according to the temperature interval model, and form the safe temperature range of the prediction period.

[0072] When predicting the safe temperature range based on the real-time operating data of the energy storage system for cascade utilization, the input data of the temperature prediction error model includes: the temperature mean within the prediction period; the input data of the temperature prediction error model satisfies the following relationship:

[0073] In the formula, is the input data of the temperature prediction error model, is the current moment, is the length of the prediction period, starting from the moment to the moment to form a prediction period, , , , ……, are the temperature means within the prediction period.

[0074] In the embodiment, the safe temperature range obtained through the test data is as Figure 4 shown. This figure is composed of the results predicted based on continuous multiple groups of test data. It can be seen from Figure 4 that the finally obtained safe temperature range has good accuracy in most cases, can basically cover the highest temperature value and the lowest temperature value, and effectively avoids the phenomenon of frequent false alarms.

[0075] The present invention also proposes a probability prediction system for the safe temperature range of a second-life energy storage system, including: An acquisition module, configured to obtain historical values of the temperature and historical values of temperature-related monitoring quantities of the second-life energy storage system in a normal operating state; A model establishment module, configured to establish a temperature prediction model based on a graph neural network and a fully connected layer, and establish a temperature probability distribution prediction model based on a graph neural network and a Bayesian last layer; establish a temperature prediction error model by using a shallow machine learning method; A model training module, configured to use the historical values of the temperature and the historical values of the temperature-related monitoring quantities to train the temperature prediction model and the temperature prediction error model, set the parameters of the graph neural network in the trained temperature prediction model as the initial parameters of the graph neural network in the temperature probability distribution prediction model; train the temperature probability distribution prediction model after setting the initial parameters, and according to the temperature probability distribution within the prediction window output by the trained temperature probability distribution prediction model, obtain the absolute value of the prediction error of the temperature mean corresponding to the temperature probability distribution within the prediction window output by the trained temperature prediction error model; A scaling factor optimization module, configured to establish a model of the temperature range by using the temperature probability distribution within the prediction window, the absolute value of the prediction error of the temperature mean corresponding to the temperature probability distribution within the prediction window, and the scaling factor; determine the optimal value of the scaling factor when the mean confidence level of the temperature range is equal to a set threshold; A prediction module, configured to obtain the monitored values of the temperature and the temperature-related monitoring quantities of the second-life energy storage system in an operating state, and obtain the safe temperature range of the second-life energy storage system by using the temperature probability distribution prediction model, the temperature prediction error model, and the optimal value of the scaling factor.

[0076] The present disclosure may be a system, a method, and / or a computer program product. The computer program product may include a computer-readable storage medium having thereon computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0077] A computer-readable storage medium can be a tangible device that can hold and store instructions for use by an instruction execution device. A computer-readable storage medium may be, for example—but not limited to—an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disc (DVD), a memory stick, a floppy disk, a mechanically encoded device, such as a punched card or raised structures in grooves having instructions stored thereon, and any suitable combination of the foregoing. The computer-readable storage medium as used herein is not construed as being a transitory signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagated through a waveguide or other transmission medium (e.g., an optical pulse through an optical fiber cable), or an electrical signal transmitted through a wire.

[0078] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to respective computing / processing devices, or can be downloaded to an external computer or an external storage device through a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network may include a copper transmission cable, an optical fiber transmission, a wireless transmission, a router, a firewall, a switch, a gateway computer, and / or an edge server. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in each computing / processing device.

[0079] Computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state-setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, etc., and conventional procedural programming languages such as the "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider). In some embodiments, by using the state information of the computer-readable program instructions to customize an electronic circuit, such as a programmable logic circuit, a field-programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit can execute the computer-readable program instructions to implement various aspects of the present disclosure.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications or equivalent replacements can still be made to the specific embodiments of the present invention, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered by the protection scope of the claims of the present invention.

Claims

1. A probability prediction method for the safe temperature range of a second-life energy storage system, characterized in that Including: Obtaining historical values of the temperature and historical values of temperature-related monitoring quantities of the second-life energy storage system in a normal operating state; Based on a graph neural network and a fully connected layer, establishing a temperature prediction model, and based on a graph neural network and a Bayesian last layer, establishing a temperature probability distribution prediction model; using a shallow machine learning method to establish a temperature prediction error model; Using the historical values of the temperature and the historical values of the temperature-related monitoring quantities to train the temperature prediction model and the temperature prediction error model, setting the parameters of the graph neural network in the trained temperature prediction model as the initial parameters of the graph neural network in the temperature probability distribution prediction model; training the temperature probability distribution prediction model after setting the initial parameters, and according to the temperature probability distribution within the prediction window output by the trained temperature probability distribution prediction model, obtaining the absolute value of the prediction error of the temperature mean corresponding to the temperature probability distribution within the prediction window output by the trained temperature prediction error model; Using the temperature probability distribution within the prediction window, the absolute value of the prediction error of the temperature mean corresponding to the temperature probability distribution within the prediction window, and a scaling factor to establish a model of the temperature range; when the mean confidence level of the temperature range is equal to a set threshold, determining the optimal value of the scaling factor; Obtaining the monitored values of the temperature and the temperature-related monitoring quantities of the second-life energy storage system in an operating state, and using the temperature probability distribution prediction model, the temperature prediction error model, and the optimal value of the scaling factor to obtain the safe temperature range of the second-life energy storage system.

2. The probability prediction method for the safe temperature range of the second-life energy storage system according to claim 1, wherein Obtaining the historical values of the temperature and the historical values of the temperature-related monitoring quantities of the second-life energy storage system in a normal operating state includes: Collecting historical values of different monitoring quantities related to the temperature, and the monitoring quantities include: state of charge, current, voltage, temperature; Calculating the correlation coefficients between the respective related monitoring quantities, and based on the correlation coefficients, determining the monitoring quantities having a correlation relationship; Dividing the historical values of the temperature and the temperature-related monitoring quantities into sequences of different time periods, and detecting suspected abnormal data existing within each time period sequence; If other monitoring quantities having a correlation relationship with the monitoring quantity to which the suspected abnormal data belongs also have suspected abnormal data at the same time node, then determining the suspected abnormal data as abnormal state data; otherwise, taking the suspected abnormal data as the center of a sliding window, determining whether the absolute value of the difference between the normalized value of the abnormal data and the mean value of the normalized data within the sliding window is greater than a set abnormal threshold, and if it is greater, determining the suspected abnormal data as abnormal state data; removing the abnormal state data and performing interpolation repair, and together with the historical value of the temperature, constituting a multi-segment historical normal data time series.

3. The probability prediction method for the safe temperature range of the second-life energy storage system according to claim 2, wherein Each segment of the historical normal data time series satisfies the following relational expression: Wherein, X is a historical normal data time series, , , , ……, constitute the historical value time series of temperature, , , , ……, is the historical value time series of the th temperature-related monitored quantity, , is the total number of monitored quantities, is the length of each time series in the historical normal data time series X; In each historical normal data time series, using the time nodes in the historical normal data time series as the establishment moment of the sliding window, using as the sliding window length, using as the prediction window length, from the moment to the moment to form within the sliding window, from the moment to the moment to form the prediction window.

4. The probability prediction method for the safe temperature range of the second-life energy storage system according to claim 1, wherein The fully connected layer is the last layer of the temperature prediction model; The Bayesian last layer is the last layer of the temperature probability distribution prediction model; Moreover, the graph neural network in the temperature prediction model has the same structure as the graph neural network in the temperature probability distribution prediction model.

5. The probability prediction method for the safe temperature range of the energy storage system for cascaded utilization according to claim 3, wherein Based on the historical values of the temperature and the historical values of the temperature-related monitoring quantities within the sliding window, the temperature of the energy storage system for cascaded utilization within the prediction window is output by using the temperature prediction model; The input data of the temperature prediction model satisfies the following relational expression: Wherein, is the input data of the temperature prediction model, , , , ……, are the historical values of the temperature within the sliding window, , , , ……, are the historical values of the temperature correlation monitoring quantities within the sliding window, , is the total number of the temperature correlation monitoring quantities; The output data of the temperature prediction model satisfies the following relational expression: In the formula, is the output data of the temperature prediction model, , , , ……, are the temperatures within the prediction window.

6. The probability prediction method for the safe temperature range of the energy storage system for cascaded utilization according to claim 3, wherein Based on the historical values of the temperature and the historical values of the temperature-related monitoring quantities within the sliding window, the temperature probability distribution of the energy storage system for cascaded utilization within the prediction window is output by using the temperature probability distribution prediction model; The input data of the temperature probability distribution prediction model satisfies the following relational expression: Wherein, is the input data of the temperature probability distribution prediction model, , , , ……, are the historical values of the temperature within the sliding window, , , , ……, are the historical values of the temperature correlation monitoring quantities within the sliding window, , is the total number of the temperature correlation monitoring quantities.

7. The probability prediction method for the safe temperature range of the energy storage system for cascaded utilization according to claim 6, wherein Assume that the temperature probability distribution within the prediction window follows a Gaussian distribution , and the output data of the temperature probability distribution prediction model satisfies the following relationship: In the formula, is the output data of the temperature probability distribution prediction model, , , , ……, are the average temperatures within the prediction window, , , , ……, are the standard deviations of the temperatures within the prediction window.

8. The probability prediction method for the safe temperature range of the energy storage system for cascaded utilization according to claim 5, wherein Based on the temperature within the prediction window, the absolute value of the prediction error of the temperature of the energy storage system for cascaded utilization within the prediction window is output by using the temperature prediction error model; When the temperature prediction error model learns the prediction error of the temperature, the input data of the temperature prediction error model satisfies the following relational expression: In the formula, is the input data of the temperature prediction error model, , , , ……, are the temperatures within the prediction window; The output data of the temperature prediction error model satisfies the following relational expression: In the formula, is the output data of the temperature prediction error model, , , , ……, are the absolute values of the temperature prediction errors within the prediction window.

9. The probability prediction method for the safe temperature range of the energy storage system for cascaded utilization according to claim 8, wherein Training the temperature prediction model and the temperature prediction error model by using the historical values of the temperature and the historical values of the temperature-related monitoring quantities, including: Taking the historical values of the temperature and the historical values of the temperature-related monitoring quantities within the sliding window as the first training data set, and taking the historical values of the temperature within the prediction window as the first label data; Training the temperature prediction model by using the first training data set and the first label data; during the training process, the temperature within the prediction window output by the temperature prediction model is a deterministic estimate of the historical values of the temperature within the prediction window, having a corresponding relationship; Taking the temperature within the prediction window output by the temperature prediction model as the second training data set, and taking the absolute value of the difference between the temperature within the prediction window and the corresponding first label data as the second label data; Training the temperature prediction error model by using the second training data set and the second label, and the trained temperature prediction error model outputs the absolute value of the prediction error of the temperature within the prediction window.

10. The probability prediction method for the safe temperature range of the energy storage system for cascaded utilization according to claim 9, wherein The temperature probability distribution prediction model after the initial parameters are set is trained using the first training data set and the first label data. According to the temperature probability distribution within the prediction window output by the trained temperature probability distribution prediction model, the absolute value of the prediction error of the temperature mean corresponding to the temperature probability distribution within the prediction window is output by the trained temperature prediction error model.

11. The probability prediction method for the safe temperature range of the energy storage system for cascade utilization according to claim 7, wherein the model of the temperature range satisfies the following relational expression: Wherein, and are the upper bound correction value and the lower bound correction value of the temperature probability distribution within the prediction window respectively, is the temperature mean within the prediction window, is the variance of the temperature within the prediction window, is the absolute value of the prediction error of the temperature probability distribution within the prediction window, is the scaling factor; The temperature upper bound correction value and the temperature lower bound correction value form a temperature range, and the mean confidence level of the temperature range satisfies the following relational expression: In the formula, is the mean confidence level of the temperature range, is the total number of groups of the upper temperature bound correction value and the lower temperature bound correction value, is the j-th label value, is the j-th indicator function, and this indicator function takes 1 when is within the temperature range, and takes 0 otherwise; When the mean confidence level is equal to the set confidence level, determine the optimal value of the scaling factor , and the set confidence level is 0.

95.

12. The probability prediction method for the safe temperature range of the energy storage system for cascade utilization according to claim 1, wherein Based on the current moment Establish a backtracking window of length m, starting from the moment to the moment to form a backtracking window, starting from the moment to the moment to form a prediction period, where is the length of the prediction period; The monitored values of the temperature and the temperature-related monitored quantity within the retrospective window are obtained and input into the temperature probability distribution prediction model to output the temperature probability distribution of the prediction period; the temperature mean of the prediction period is determined based on the temperature probability distribution of the prediction period; the temperature mean of the prediction period is input into the temperature prediction error model to obtain the absolute value of the prediction error of the temperature mean of the prediction period. The input data of the temperature prediction error model satisfies the following relational expression: Wherein, is the input data of the temperature prediction error model, , , , ……, are the average temperatures during the prediction period; Using the temperature mean of the prediction period, the absolute value of the prediction error of the temperature mean of the prediction period, and the optimal value of the scaling factor, according to the temperature range model, determine the temperature upper bound correction value and the temperature lower bound correction value of the prediction period, and form the safe temperature range of the prediction period.

13. A probability prediction system for the safe temperature range of a second-life energy storage system, which is used to implement the probability prediction method for the safe temperature range of the second-life energy storage system according to any one of claims 1 to 12, characterized in that, including: An acquisition module for obtaining the historical values of the temperature and the temperature-related monitored quantity of the energy storage system for cascade utilization in the normal operating state; A model establishment module for establishing a temperature prediction model based on a graph neural network and a fully connected layer, and establishing a temperature probability distribution prediction model based on a graph neural network and a Bayesian last layer; using a shallow machine learning method to establish a temperature prediction error model; A model training module for training the temperature prediction model and the temperature prediction error model using the historical values of the temperature and the temperature-related monitored quantity, setting the parameters of the graph neural network in the trained temperature prediction model as the initial parameters of the graph neural network in the temperature probability distribution prediction model; training the temperature probability distribution prediction model after the initial parameters are set, and according to the temperature probability distribution within the prediction window output by the trained temperature probability distribution prediction model, the absolute value of the prediction error of the temperature mean corresponding to the temperature probability distribution within the prediction window is output by the trained temperature prediction error model; A scaling factor optimization module for using the temperature probability distribution within the prediction window, the absolute value of the prediction error of the temperature mean corresponding to the temperature probability distribution within the prediction window, and the scaling factor to establish a model of the temperature range; when the mean confidence level of the temperature range is equal to the set threshold, determine the optimal value of the scaling factor; A prediction module for obtaining the monitored values of the temperature and the temperature-related monitored quantity of the energy storage system for cascade utilization in the operating state, and using the temperature probability distribution prediction model, the temperature prediction error model, and the optimal value of the scaling factor to obtain the safe temperature range of the energy storage system for cascade utilization.

14. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used for storing instructions; The processor is used for operating according to the instructions to execute the steps of the method described in any one of claims 1-12.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, the steps of the method described in any one of claims 1-12 are implemented.

Images