A Federal State Evaluation Method for Battery Energy Storage Systems
The federated learning method for battery energy storage systems addresses data privacy and security issues by using sensor fusion and encryption to achieve precise SOH and SOC estimation, improving system reliability and efficiency.
Patent Information
- Application Number
- CN202210842774.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-07-18
- Publication Date
- 2025-07-15
- Estimated Expiration
- 2042-07-18
AI Technical Summary
Existing battery energy storage systems have challenges in data privacy protection and data silos, making it difficult to achieve data sharing, affecting the accurate evaluation of health status (SOH) and state of charge (SOC), and increasing data volume leads to insufficient security and evaluation accuracy.
Multi-sensor fusion technology is used to obtain internal and external characteristic data of the battery energy storage system, combine hashing operations and homomorphic encryption technology to align data, and use tensor deep learning and FPGA to build BMS, and use federated learning to train and evaluate models through improved FedAvg algorithm and Byzantine robust algorithm to ensure data privacy and improve evaluation accuracy.
It realizes efficient joint estimation of SOH and SOC under the premise of protecting data privacy, reduces communication overhead, adapts to different application scenarios, and improves evaluation accuracy and security.
Smart Images
Figure CN115222126B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of intelligent operation and maintenance of battery energy storage systems, and particularly relates to a method for federated state assessment of battery energy storage systems. Background Art
[0002] In order to better achieve the goals of carbon peak and carbon neutrality and accelerate the positive transformation of the energy structure, vigorously developing new energy has become the consensus of all sectors at home and abroad. Energy storage, as a key point in the development and construction of energy power, plays an important supporting role in the clean, low-carbon, safe and efficient development of the system, the improvement of the grid connection and consumption capacity of renewable energy, and the reliable and economic power supply for users. Among various energy storage technologies, battery energy storage has prominent comparative advantages due to its excellent performance and rapidly declining cost. A battery energy storage system is usually composed of a large number of single cells connected in series and parallel, and is equipped with a battery management system (BMS), chassis, rack, etc. to form a battery system; the battery system is then connected to the grid or a specific power supply system through power conversion equipment and isolation transformers. However, due to the influence of the external environment during the actual operation of the battery energy storage system and being restricted by various physical and chemical reactions occurring inside, in order to ensure the safe and reliable operation of the battery energy storage system, it is very necessary to accurately evaluate its state of health (SOH) and state of charge (SOC) to overcome the shortcomings of low accuracy of current conventional algorithms. In addition, with the large-scale construction of battery energy storage systems, the amount of data has increased sharply, which puts higher requirements on data collection, transmission and processing. At the same time, the substantial increase in the amount of data makes it difficult to ensure the data security of the system, and the problem of privacy data leakage needs to be effectively solved. Moreover, due to industry competition and privacy security between different entities, there are insurmountable barriers between data sources, and the formation of "data islands" will lead to the difficulty of realizing the secure sharing of data, which greatly restricts the healthy development of battery energy storage systems. Summary of the Invention
[0003] With the further enhancement of data privacy protection and the development of artificial intelligence, a new idea of global optimization guidance under incomplete information based on federated learning has begun to attract attention, that is, through the model encapsulation and encryption of local autonomous bodies, the encapsulated models and their encrypted gradient information are integrated and trained in the cloud for joint learning, which can achieve a global optimization effect similar to that of complete data under the condition of ensuring the privacy of their respective data, and then guide the continuous evolution of the strategies of each entity.
[0004] Therefore, to overcome the shortcomings of the prior art, the present invention proposes a method for federated state assessment of battery energy storage systems to effectively realize the joint estimation of SOH and SOC of battery energy storage systems. The technical solution adopted is as follows:
[0005] A method for federated state assessment of battery energy storage systems specifically includes the following steps:
[0006] Step 1: Use multi-sensor fusion technology to obtain the external characteristic data of the battery energy storage system operation; at the same time, use in-situ / non-in-situ technology to obtain the internal characteristic data, so as to obtain the internal and external characteristic data of each battery energy storage system operation; perform structured processing on the internal and external characteristic data of each battery energy storage system, and adopt a combination of hash operation and homomorphic encryption technology to align the internal and external characteristic data of different battery energy storage systems, and then use the method of tensor deep learning to extract the high-dimensional features of local data;
[0007] Step 2: Build a BMS based on FPGA, establish a method for evaluating the SOC of the battery energy storage system based on a lightweight gated recurrent unit, seal and package the locally built model, use the stochastic gradient descent method to calculate the optimal gradient information of the packaged model of each battery energy storage system, and transmit the encrypted gradient calculation information and its adjacency relationship through a distributed interaction method, and only upload the gradient calculation information of the local model to the federal cloud center to fully protect the privacy requirements of the data of each battery energy storage system;
[0008] Step 3: Use the accessed battery energy storage system data information and random scenario production strategy in the federal cloud center to establish a clustering method for the application scenarios of the battery energy storage system based on spectral clustering, and then use the improved FedAvg algorithm to optimize the global aggregation model. According to different scenario categories, establish a Byzantine-robust federated learning algorithm based on uncertainty constraints, and then form a regularized deep network with the ability of automatic knowledge acquisition and reasoning, and use a temporal convolutional neural network to predict the SOH of the battery energy storage system;
[0009] Step 4: According to the difference between the local model and the global aggregation model, apply a decision-making method for the battery energy storage system considering loss to realize the real-time update of the local model, and use the current SOH calculated by the federal cloud center to correct the local real-time SOC. The federal cloud center will also further improve the prediction accuracy of the global SOH according to the state of the local SOC, so as to achieve high-precision federal evaluation of the local SOC and the SOH of the federal cloud center.
[0010] Furthermore, the external characteristic data are voltage, current and temperature; the internal characteristic data include resistance data obtained by electrochemical impedance spectroscopy technology, crystal structure data obtained by in-situ X-ray diffraction technology, and chemical composition data obtained by in-situ Fourier transform infrared spectroscopy technology.
[0011] Further, in step 3, the improved FedAvg algorithm is used to optimize the global aggregation model. The FedAvg algorithm aggregates the client models according to weights. The more data the user uses in local training, the higher the quality of the updated model, and the greater the weight during aggregation. Then, the improved FedAvg algorithm adds a similarity evaluation index when updating the model weights to screen out data with too large distribution differences, that is, to eliminate low-quality data.
[0012] Further, the relationship expression between SOC and SOH is as follows:
[0013] At absolute capacity:
[0014]
[0015] At relative capacity:
[0016]
[0017] Among them, Q t is the current remaining power, C n is the rated capacity, and C end is the capacity at the end of battery life.
[0018] Beneficial effects:
[0019] The present invention can fully protect the privacy of local data, does not require a large amount of data transmission, reduces communication overhead, and can comprehensively consider different application scenarios and operation requirements, effectively realizing the joint estimation of SOH and SOC of the battery energy storage system. Description of the drawings
[0020] Figure 1 It is a schematic diagram of the federated state evaluation scheme of the battery energy storage system of the present invention. Specific implementation manners
[0021] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.
[0022] The present invention first describes the definitions of SOH and SOC and the relationship between the two.
[0023] 1) Definition of SOC
[0024] SOC, i.e., State of Charge, refers to the state of charge of a battery. From different perspectives such as electricity quantity and energy, there are various ways to define SOC. The commonly used calculation formula is the ratio of the current remaining electricity quantity to the current capacity.
[0025]
[0026] In the formula, Q t is the current remaining electricity quantity, and C t is the current capacity.
[0027] 2) Definition of SOH
[0028] SOH, i.e., State of Health, refers to the state of health of a battery, which is divided into two ways: absolute definition and relative definition. The absolute definition is the ratio of the current capacity of the battery to the rated capacity.
[0029]
[0030] In the formula, C n is the rated capacity.
[0031] The relative definition is divided into two ways of representation: capacity and internal resistance. When represented by capacity, it is the ratio of the difference between the current capacity of the battery and the end-of-life capacity to the difference between the rated capacity and the end-of-life capacity.
[0032]
[0033] In the formula, C end is the capacity at the end of the battery life.
[0034] When defined by internal resistance, it is the ratio of the difference between the internal resistance at the end of the battery life and the current internal resistance to the difference between the internal resistance at the end of the battery life and the rated internal resistance.
[0035]
[0036] In the formula, R t is the current internal resistance, R n is the rated internal resistance, and R end is the internal resistance at the end of the battery life.
[0037] According to the above definitions, for SOH defined in an absolute way, generally when SOH is 0.8, that is, when the current capacity of the battery is 80% of the rated capacity, the battery life ends. When defined in a relative way, when SOH is 0, that is, when the current capacity is equal to the end-of-life capacity, or the current internal resistance is equal to the end-of-life internal resistance, the battery life ends.
[0038] 3) Relationship between SOH and SOC
[0039] Considering the inherent relationship between SOH and SOC, and the SOH definition method based on capacity, the relationship expression between the two is as follows:
[0040] When it is the absolute capacity:
[0041]
[0042] When it is the relative capacity:
[0043]
[0044] As Figure 1 shown, a method for evaluating the federated state of a battery energy storage system according to the present invention specifically includes the following steps:
[0045] Step 1: Use multi-sensor fusion technology to obtain the external characteristic data of the operation of the battery energy storage system, such as voltage, current, and temperature, etc.; at the same time, use in-situ / non-in-situ technology to obtain internal characteristic data, such as using electrochemical impedance spectroscopy technology to obtain resistance data, using in-situ X-ray diffraction technology to obtain crystal structure data, using in-situ Fourier transform infrared spectroscopy technology to obtain chemical composition data, etc., so as to obtain the internal and external characteristic data of the operation of each battery energy storage system. Furthermore, perform structured processing on the data of each battery energy storage system, and adopt a combination of hash operation and homomorphic encryption technology, that is, perform hash operation on the sample data and fuse it with the homomorphically encrypted data to complete the encryption alignment of the data of different battery energy storage systems, and then use the method of tensor deep learning to extract the high-dimensional features of the local data, that is, use tensors to model the local data, expand the vector space data to the tensor space, and construct a high-order tensor in the tensor space to capture the high-dimensional features in the data.
[0046] Step 2: Since FPGA has advantages such as high speed, repeatability, and parallel computing ability, it can make up for the deficiencies in performance when using single-chip microcomputers and DSPs as the main controllers; furthermore, in order to obtain better practical performance and effects, use deep learning for real-time online SOC evaluation, which can avoid the complex electrochemical reaction modeling of the battery and can overcome the disadvantages of poor dynamic accuracy and poor universality of the model-based method. Therefore, build a battery management system BMS based on FPGA, establish a method for evaluating the SOC of a battery energy storage system based on a lightweight gated recurrent unit, while ensuring the evaluation accuracy, reduce the complexity and calculation cost of the model. Encapsulate the locally built model, use the stochastic gradient descent method to calculate the optimal gradient information of the encapsulated model of each battery energy storage system, transmit the encrypted gradient calculation information and its adjacency relationship through a distributed interaction method, and only upload the gradient of the local model to the federated cloud center to fully protect the data privacy requirements of each battery energy storage system.
[0047] Step 3: Utilize the accessed battery energy storage system data information and random scenario production strategies in the federal cloud center to establish a clustering method for battery energy storage system application scenarios based on spectral clustering, and then optimize the global aggregation model using an improved FedAvg algorithm. The FedAvg algorithm aggregates the client models collected according to weights. The more data the user uses in local training, the higher the quality of the updated model, and the greater the weight during aggregation. The improved FedAvg algorithm adds a similarity evaluation index when updating the model weights to screen out data with overly large distribution differences, that is, eliminate low-quality data, and not all data subsets participate in federated learning. Furthermore, according to different scenario categories, establish a Byzantine-robust federated learning algorithm based on uncertainty constraints. In response to attacks from Byzantine clients, generate a sequence of numbers in the global aggregation model. When updating the model each time, calculate the changes in the sequence of numbers, and then obtain the probability distribution of the output layer of the model. According to requirements such as distance metrics, eliminate poisoning data or models outside the probability to achieve the effect of resisting attacks. Finally, form a regularized deep network with automatic knowledge acquisition and reasoning capabilities, and use a temporal convolutional neural network to predict the SOH of the battery energy storage system.
[0048] Step 4: According to the differences between the local model and the global model, apply a decision-making method for the battery energy storage system considering losses to achieve real-time updates of the local model, and use the current SOH calculated by the federal cloud center to correct the local real-time SOC. The cloud center will also further improve the prediction accuracy of the global SOH based on the status of the local SOC, thereby achieving high-precision federated evaluation of the local SOC and the SOH of the federal cloud center.
[0049] Those skilled in the art can easily understand that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.
Claims
1. A method for evaluating the federal state of a battery energy storage system, characterized in that, Specifically, it includes the following steps: Step 1: Use multi-sensor fusion technology to obtain the external characteristic data of the battery energy storage system operation; at the same time, use in-situ / non-in-situ technology to obtain the internal characteristic data, so as to obtain the internal and external characteristic data of each battery energy storage system operation; perform structured processing on the internal and external characteristic data of each battery energy storage system, and use a combination of hash operation and homomorphic encryption technology to align the internal and external characteristic data of different battery energy storage systems, and then use the method of tensor deep learning to extract the high-dimensional features of local data; Step 2: Build a battery management system BMS based on FPGA, establish a method for evaluating the SOC of the battery energy storage system based on a lightweight gated recurrent unit, seal and package the locally built model, use the stochastic gradient descent method to calculate the optimal gradient information of the packaged model of each battery energy storage system, and transmit the encrypted gradient calculation information and its adjacency relationship through a distributed interaction method, and only upload the gradient calculation information of the local model to the federal cloud center to fully protect the privacy needs of the data of each battery energy storage system; Step 3: Use the accessed battery energy storage system data information and random scenario production strategy in the federal cloud center to establish a clustering method for the application scenarios of the battery energy storage system based on spectral clustering, and then optimize the global aggregation model. According to different scenario categories, establish a Byzantine-robust federated learning algorithm based on uncertainty constraints, and then form a regularized deep network with the ability of automatic knowledge acquisition and reasoning, and use a temporal convolutional neural network to predict the SOH of the battery energy storage system; Step 4: According to the difference between the local model and the global aggregation model, apply a decision-making method for the battery energy storage system considering loss to realize the real-time update of the local model, and use the current SOH calculated by the federal cloud center to correct the local real-time SOC. The federal cloud center will also further improve the prediction accuracy of the global SOH according to the state of the local SOC, so as to achieve a high-precision federal evaluation of the local SOC and the SOH of the federal cloud center.
2. The method for evaluating the federal state of a battery energy storage system according to claim 1, characterized in that: The external characteristic data are voltage, current and temperature; the internal characteristic data include resistance data obtained by electrochemical impedance spectroscopy technology, crystal structure data obtained by in-situ X-ray diffraction technology, and chemical composition data obtained by in-situ Fourier transform infrared spectroscopy technology.
3. A method for evaluating the federal state of a battery energy storage system according to claim 1, characterized in that: In Step 3, the improved FedAvg algorithm is used to optimize the global aggregation model. The FedAvg algorithm aggregates the client models collected according to the weights. The more data the user uses in local training, the higher the quality of the updated model, and the greater the weight during aggregation. Then the improved FedAvg algorithm adds a similarity evaluation index when updating the model weights to screen out the data with too large distribution differences, that is, eliminate the inferior data.
4. A method for federated state evaluation of a battery energy storage system according to claim 3, characterized in that: The relationship expression between SOC and SOH is as follows: At absolute capacity: At relative capacity: Among them, Q t is the current remaining power, C n is the rated capacity, C end is the capacity at the end of battery life.