Energy storage detail data acquisition and compression control method based on batch flow fusion

Abstract

The invention discloses an energy storage detail data acquisition and compression control method based on batch stream fusion, which relates to the technical field of data compression control, and comprises the following steps: obtaining operation state parameters of an energy storage system, and carrying out state perception and hierarchical mapping on the energy storage system to obtain state levels; based on a deep reinforcement learning method, performing adaptive adjustment in combination with the state level to obtain a compression control parameter; according to the compression control parameters, compression control strategy adjustment and feedback optimization are carried out on the energy storage system based on a fuzzy reasoning method; according to the method, adaptive adjustment of the compression strategy is realized by constructing the running state potential function and combining deep reinforcement learning and fuzzy reasoning, and the problems that the compression strategy of an existing energy storage system is fixed, regulation and control response is lagged and the control precision is insufficient under the multi-state working condition are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of data compression control. More specifically, the present invention relates to a method for collecting, compressing and controlling energy storage detail data based on batch-flow fusion. Background Art

[0002] With the continuous increase in the proportion of new energy access, the energy storage system, as an important regulating device for peak shaving, frequency modulation, voltage support, energy balance and coordinated control of distributed energy in the power system, has become increasingly prominent in the automated control system. To ensure the dynamic controllability and system response ability of the energy storage device in multi-source heterogeneous operation scenarios, it is necessary to collect and continuously monitor the energy storage detail data (such as voltage, current, temperature, state of charge SOC, power change, etc.) during the operation of the battery cluster, battery management system, converter and auxiliary equipment at high frequency. However, in actual operation, the energy storage system faces the following main problems: The energy storage system usually samples the energy storage detail data at the second level or even the millisecond level. Especially under the conditions of frequent fluctuations on the grid side or rapid changes in load, the sampling frequency needs to be further increased, resulting in a sudden increase in the data traffic transmitted between the edge nodes and the central processing unit of the control system, causing network communication congestion, high data redundancy and low query efficiency.

[0003] To cope with the pressure brought by high-frequency data collection and continuous monitoring, some energy storage systems have introduced a data compression mechanism to perform dimensionality reduction, abstraction or simplification processing on the original monitoring data before data upload.

[0004] For example, a method for controlling the running speed safety based on a hybrid control strategy of reinforcement learning disclosed in the invention patent announcement with the publication number: 202411948955.X includes the following steps: S1, constructing a running process model of a high-speed train; S2, constructing a reward function of a hybrid control strategy based on reinforcement learning and fractional-order PID; S3, constructing a hybrid control strategy based on reinforcement learning and fractional-order PID, and jointly generating a final control signal through superposition processing.

[0005] For example, a method and device for controlling AGC of a thermal power unit based on a fuzzy control strategy disclosed in the invention patent announcement with the publication number: 201810142764.2 includes the following steps: obtaining a fuzzy quantity by performing fuzzy algorithm processing on the active power deviation and the change amount of the active power deviation, deriving an effective fuzzy set according to the optimized membership function and the pre-established fuzzy rules, and generating a corresponding adjustment signal by using defuzzification calculation.

[0006] Among the above disclosed technical solutions, there are at least the following technical problems: Current compression control methods are usually static and do not perform dynamic perception and adjustment according to the operating state of the energy storage system and environmental changes. Since the energy storage system has different requirements for data acquisition and compression under different operating states, static control strategies cannot meet the system's requirements for compression accuracy and storage efficiency in various states, resulting in unstable compression effects. Moreover, existing methods mostly rely on fixed rules and lack an intelligent optimization mechanism for adaptive adjustment based on the system state. Additionally, the changes in the operating state, environmental factors, etc. of the energy storage system are relatively complex, and traditional methods based on experience and rules are difficult to achieve efficient and accurate compression control. To address the above problems, the present invention proposes a solution. Summary of the Invention

[0007] To overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a method for acquiring and compressing control of energy storage detail data based on batch-flow fusion, which realizes adaptive adjustment of the compression strategy by constructing an operating state potential function and combining deep reinforcement learning and fuzzy inference, and solves the problems of fixed compression strategies, lagging regulation responses, and insufficient control accuracy of existing energy storage systems under multi-state working conditions.

[0008] To achieve the above object, the present invention provides the following technical solutions: A method for acquiring and compressing control of energy storage detail data based on batch-flow fusion, comprising the following steps: obtaining the operating state parameters of the energy storage system and performing state perception and hierarchical mapping on the energy storage system to obtain a state level; performing adaptive adjustment based on the deep reinforcement learning method in combination with the state level to obtain compression control parameters; and adjusting the compression control strategy of the energy storage system based on the fuzzy inference method according to the compression control parameters and performing feedback optimization.

[0009] In a preferred embodiment, obtaining the operating state parameters of the energy storage system and performing state perception and hierarchical mapping on the energy storage system to obtain a state level specifically includes: obtaining the operating state parameters of the energy storage system at the current moment, constructing a state vector, and performing differential processing on the state vectors at adjacent time points to obtain a state perception vector; constructing a sliding time window sequence based on the state perception vector at the current moment and a preset window length; for each type of state perception vector in the window sequence, respectively extracting its evolution trend in the time dimension and forming a state evolution rate vector; constructing a state potential function based on the state evolution rate vector through the principal component analysis method; and performing state perception and state hierarchical mapping according to the state potential function to obtain a state level.

[0010] In a preferred embodiment, for various state perception vectors in the window sequence, the evolution trend in the time dimension is respectively extracted to form a state evolution rate vector. Specifically: a state variable sequence is extracted according to the state perception vectors in the window sequence, and the state variable sequence includes the state of charge, the estimated value of the state of health, and the cell temperature; a time trend fitting operation is performed on each variable in the state variable sequence based on the least squares method to obtain a fitting model; the fitting slope is extracted according to the fitting model as the evolution rate of the state variable sequence to obtain the state evolution rate vector.

[0011] In a preferred embodiment, state perception and state classification mapping are performed according to the state potential function to obtain a state level. Specifically: a state potential function sequence within a preset time period is obtained and sorted in ascending order, and several candidate partition point sets are constructed, where each candidate partition point is the median of two adjacent state potential function values; for each candidate partition point, the state potential function sequence is divided into a left subset and a right subset, and the entropy values of the left subset and the right subset are respectively calculated; based on the sample quantity, the total partition entropy corresponding to each candidate partition point is calculated through the entropy value, and the information gain of the current candidate partition point is obtained based on the original entropy of the entire sequence; the candidate partition point with the largest information gain is selected from the candidate partition point set as the first-level classification threshold of the state potential function; on the basis of the first-level classification threshold division, the information gain acquisition and selection operations are repeated, and recursive division is iteratively performed until the preset classification layer number is reached to obtain the state potential function classification threshold set; based on the state potential function classification threshold set, the state potential function value is classified and mapped to obtain the state level.

[0012] In a preferred embodiment, an adaptive adjustment is performed based on the deep reinforcement learning method in combination with the state level to obtain the compression control parameter. Specifically: a Markov decision process model based on the state level is constructed; The deep deterministic policy gradient algorithm is used to iteratively train the Markov decision process model, and the interactive optimization of policy evaluation and policy improvement is realized through a dual neural network architecture until the cumulative discounted reward value of the policy function converges to a preset threshold; The trained policy function is deployed as a compression control policy mapping function, and a dynamic mapping relationship between the state level and the compression parameter group is established, where the mapping relationship is realized through an improved radial basis function neural network, and parameter boundary constraint conditions are set to ensure the feasibility of the compression operation.

[0013] In a preferred embodiment, the compression control strategy of the energy storage system is adjusted based on the compression control parameters according to the fuzzy inference method and feedback optimization is performed. Specifically: the compression control parameters are fuzzified, and the continuous input values are mapped into fuzzy linguistic values by using triangular membership functions to form a fuzzy set; a fuzzy rule base is established according to the fuzzy set, the fuzzy linguistic values and their membership degrees are used as input items, and fuzzy inference is performed according to the rule base. The Mamdani inference mechanism is used to obtain the fuzzy output variable to obtain the lossless priority value; the lossless priority value is defuzzified to obtain a clear quantization value, and a judgment is made based on a preset lossless priority threshold; according to the judgment result, a compression control strategy adjustment signal is obtained, and the corresponding lossless or lossy compression strategy selection is performed on the data collected by the energy storage system; the feedback parameters after this round of compression execution are collected to construct a compression effect scoring function, and a compression strategy performance scoring value is obtained. The feedback parameters include the actual compression ratio, the actual reconstruction error, and the actual compression delay; if the scoring value is lower than the preset threshold for several consecutive cycles, the scoring value is fed back to the fuzzy inference system to trigger the optimization of the fuzzy rule base.

[0014] In a preferred embodiment, triggering the optimization of the fuzzy rule base specifically includes: encoding the fuzzy rules in the form of rule vectors as genetic individuals based on the genetic algorithm. Each fuzzy rule includes the matching relationship between the fuzzy linguistic value of the input parameter and the output strategy priority; the compression effect scoring function is used as the fitness function of the genetic algorithm, and iterative evolution operations are performed on the individuals; after the iteration reaches the preset number of evolution rounds, the fuzzy rule individual set with the largest fitness value is selected as the updated fuzzy rule base and replaces the rules in the current inference system; the iterative evolution operation is repeated to form a closed-loop fuzzy rule optimization process driven by compression feedback to optimize the fuzzy rule base.

[0015] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. By introducing a state perception and hierarchical mapping mechanism, a state evolution rate vector and a state potential function are constructed based on the operating parameters of the energy storage system to realize the dynamic trend modeling and hierarchical division of the operating state. Compared with the traditional static compression strategy, it has the significant technical advantages of strong feedforward, sensitive response, and high discrimination. It can identify the system fluctuation trend and potential anomalies based on the state change rate, accurately map the current state level, so that the compression control strategy has the ability of level perception and dynamic self-adaptability, effectively improving the data compression efficiency and fidelity under multi-state working conditions, and is conducive to realizing the differential retention and intelligent control of key data.

[0016] 2. By combining deep reinforcement learning with a fuzzy inference mechanism, the state-level adaptive adjustment and feedback closed-loop optimization of the compression control parameters of the energy storage system are achieved, with the following significant technical advantages: On the one hand, the reinforcement learning strategy dynamically maps the compression parameters according to the system operation situation, realizing the strong correlation and flexible adaptability between the compression strategy and the operation state, and effectively avoiding the problem that the compression intensity is decoupled from the system state in traditional methods; on the other hand, the fuzzy inference mechanism combines with the genetic algorithm to achieve the fine determination and adaptive evolution of the compression strategy, with strong non-linear processing ability and anti-uncertainty ability, and can maintain the high robustness and efficiency of the compression strategy in a changing environment. The overall scheme constructs a closed-loop optimization system driven by the operation state and feedback by the compression effect, significantly improving the accuracy of compression strategy determination, the system resource utilization efficiency and the operation stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic flowchart of a method for collecting and compressing control of energy storage detail data based on batch-flow fusion provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0019] Embodiment 1 Figure 1 It is a schematic flowchart of a method for collecting and compressing control of energy storage detail data based on batch-flow fusion provided by an embodiment of the present application, including the following steps: S1. Obtain the operation state parameters of the energy storage system and perform state perception and hierarchical mapping on the energy storage system to obtain a state level.

[0020] In this embodiment, the operation state of the energy storage system directly affects the dynamic feature distribution of the data. Different from the traditional static sampling compression strategy, by obtaining the operation state parameters of the energy storage system, it provides the original input for the state trend modeling and the construction of perception indicators. And by obtaining the state level, it can accurately calibrate which state interval the energy storage system is currently in, enabling the compression control strategy to perform level-aware adaptive control.

[0021] The obtaining of the operation state parameters of the energy storage system and performing state perception and hierarchical mapping on the energy storage system to obtain a state level is specifically: Obtain the operation state parameters of the energy storage system at the current moment, and the operation state parameters of the energy storage system include voltage, current, state of charge, estimated value of health state, and cell temperature; Construct a state vector based on the operating state parameters of the energy storage system, and perform a difference operation on the state vectors at adjacent time points to obtain a state perception vector; Construct a sliding time window sequence based on the state perception vector at the current moment and a preset window length; For various state perception vectors in the window sequence, extract their evolution trends in the time dimension respectively to form a state evolution rate vector; Construct a state potential function based on the state evolution rate vector through the principal component analysis method; Perform state perception and state classification mapping according to the state potential function to obtain a state level, specifically: Obtain a state potential function sequence within a preset time period, sort it in ascending order, and construct several candidate partition point sets, where each candidate partition point is the median of the state potential function values of two adjacent ones; For each candidate partition point, divide the state potential function sequence into a left subset and a right subset, and calculate the entropy values of the left subset and the right subset respectively; Calculate the total partition entropy corresponding to each candidate partition point based on the entropy value and the sample quantity, and obtain the information gain of the current candidate partition point based on the original entropy of the entire sequence; Select the candidate partition point with the largest information gain from the candidate partition point set as the first-level classification threshold of the state potential function; On the basis of the first-level classification threshold division, repeat the information gain acquisition and selection operation, and perform recursive division iteratively until the preset classification layer number is reached to obtain a state potential function classification threshold set, and the state potential function classification threshold set includes a first-level threshold, a second-level threshold, and a third-level threshold, that is 、 、 ; Based on the state potential function classification threshold set, perform classification mapping on the state potential function values to obtain a state level, and the state level includes a first level, that is, the operating state is stable, a second level, that is, the operating state has slight fluctuations, a third level, that is, the operating state deviates from normal, and a fourth level, that is, the operating state has an abnormal warning; If , it is the first level; If , it is the second level If , it is the third level If , it is the fourth level Where is the state potential function value.

[0022] The entropy values of the left subset and the right subset, the specific calculation formula is as follows:

[0023]

[0024] The specific calculation formula of the total partition entropy is as follows:

[0025] The specific calculation formula of the information gain is as follows:

[0026] Where, is the entropy of the left subset, is the entropy of the right subset, is the category of the status level defined in the subset, is the first The probability of the class state level appearing, is the first The probability of the class state level appearing, The total partition entropy at the current partition point, is the number of samples in the left subset, is the number of samples in the right subset, is the total sample size, is the information gain, is the raw entropy of the entire sequence.

[0027] It should be noted that in the state classification or control strategy selection of energy storage systems, relying solely on the current state value is not forward-looking. The state perception vector obtained through differential processing can predict whether the system is tending to be stable or deviating from the normal range, and can explicitly characterize the evolution rate, fluctuation direction and degree of each parameter in a short time. Compared with using only static values, it helps to detect system anomalies earlier, enhances the state perception's ability to respond to short-term drastic changes and non-stationary characteristics, and provides time structure information for subsequent evolution modeling.

[0028] The evolution trend of each state perception vector in the window sequence in the time dimension is extracted and a state evolution rate vector is formed, which is specifically: Extracting a state variable sequence based on the state perception vector in the window sequence, wherein the state variable sequence includes a state of charge, a health state estimation value, and a battery cell temperature; Based on the least square method, the time trend fitting operation is performed on each variable in the state variable sequence to obtain the fitting model; The fitting slope is extracted according to the fitting model as the evolution rate of the state variable sequence to obtain a state evolution rate vector.

[0029] The specific calculation formula of the fitting model is as follows:

[0030] The state evolution rate vector is specifically as follows:

[0031] The state potential function has the following specific calculation formula:

[0032] In the formula, is the fitting model, is the fitting slope, is the intercept of the fitting line at time zero, is the time position within the sliding time window, is the current time, is the sliding window width, is the set of state evolution rate vectors, is the state of charge at the time point near the evolution rate, is the estimated value of the state of health at the time point near the evolution rate, is the cell temperature at the time point near the evolution rate, is the value of the state potential function, , , are the weight coefficients respectively.

[0033] It should be noted that the state variable sequence, that is, the change trends of the state of charge, the estimated value of the state of health, and the cell temperature within the window, realizes the quantitative description of the short-term and medium- to long-term operation evolution paths of the energy storage system by extracting the state variable sequence, and constructs an operation evolution dynamic benchmark. The extracted state evolution rate vector will be used as the main input in the subsequent construction of the state potential function, determining the current data retention intensity and the key area of focus of the model. Compared with the static threshold rule, trend modeling has stronger predictability and feedforwardness, supporting more flexible and intelligent scheduling and abnormal prevention and control.

[0034] It should be noted that the state evolution rate vector is used to describe the change trend of the core state variables of the energy storage system, providing an input basis for subsequent state judgment and compression decision-making.

[0035] S2. Adaptive adjustment is performed based on the deep reinforcement learning method combined with the state level to obtain the compression control parameter.

[0036] In this embodiment, based on the classification of the operating state levels of the energy storage system, a state space is constructed, and the system operating situation is jointly characterized by using the state perception vector and the state evolution rate vector. The reinforcement learning policy network is used to output compression control parameters according to different state levels (stable, slightly fluctuating, deviating from normal, abnormal warning), so that the compression policy has good sensitivity and self - adaptability to changes in the operating state, and solves the problem that the compression intensity is decoupled from the operating state in traditional compression methods.

[0037] The compression control parameters are adaptively adjusted based on the deep reinforcement learning method combined with the state level, specifically as follows: Construct a Markov decision process model based on the state level; Adopt the deep deterministic policy gradient algorithm to iteratively train the Markov decision process model, and realize the interactive optimization of policy evaluation and policy improvement through a double - neural network architecture until the cumulative expected reward value of the policy function converges to a preset threshold; Deploy the trained policy function as the compression control policy mapping function, establish a dynamic mapping relationship between the state level and the compression parameter group, and obtain the compression control parameters corresponding to the current state level of the energy storage system, where the mapping relationship is realized through an improved radial basis function neural network, and parameter boundary constraint conditions are set to ensure the feasibility of the compression operation.

[0038] The Markov decision process model includes: A state space generated by the data state level classification rule, where the state level is dynamically defined according to the data type, data priority, and real - time requirement; An action space composed of compression control parameters, where the compression control parameters include a compression rate adjustment coefficient, an error tolerance threshold, and a sampling interval adjustment factor, and the sampling interval adjustment factor controls the sampling density in both the time dimension and the space dimension; A multi - dimensional composite reward objective function, which integrates the compressed data recovery accuracy evaluation index, the data storage consumption cost index, and the error accumulation rate index through a linear weighting method, and the recovery accuracy evaluation index adopts a fusion calculation model of structural similarity (SSIM) and peak signal - to - noise ratio (PSNR) The policy function, the specific calculation formula is as follows:

[0039] The reward objective function, the specific calculation formula is as follows:

[0040] The maximization of the cumulative expected reward value objective function, the specific calculation formula is as follows:

[0041] In the formula, is the policy function, is the state level, is the compression control action, is the reward objective function, is the data recovery accuracy, is the data storage consumption, is the error rate, , , are respectively the weight parameters of the reward function, is to maximize the cumulative expected reward value, is the discount factor, is the sampling mean of the policy function during the state transition process, is the time length boundary for policy optimization and state trajectory evaluation, is the current time.

[0042] It should be noted that the reward objective function is used to measure the impact of the data after corresponding action compression on the system availability, accuracy, and resource occupancy metrics. The policy mapping function outputs the precise numerical value of the compression control parameter defined in the action space at each state level. The sampling mean of the policy function during the state transition process is used to ensure that the compression is effective for all running trajectories and avoid overfitting to a certain state. The time length boundary for policy optimization and state trajectory evaluation is used to control the evaluation time range of the compression policy optimization and balance real-time performance and global performance. The compression rate is the ratio of the compressed data volume to the original data volume and is used to adjust the aggressiveness of the compression algorithm. The error tolerance is the maximum deviation range allowed between the compressed data and the original data. The sampling interval adjustment factor dynamically adjusts the sampling interval based on the original sampling period and is used to control the data update frequency.

[0043] S3. Adjust the compression control strategy of the energy storage system based on the fuzzy inference method according to the compression control parameters and perform feedback optimization.

[0044] In this embodiment, the compression control parameters include the compression rate, the error tolerance, and the sampling interval adjustment factor. <>

[0045] Adjust the compression control strategy of the energy storage system based on the fuzzy inference method according to the compression control parameters and perform feedback optimization, specifically as follows: Fuzzify the compression control parameters, use triangular membership functions to map continuous input values into fuzzy linguistic values, and form fuzzy sets. Each fuzzy set is divided based on the fuzzy linguistic values and preset numerical intervals. The fuzzy set of the compression rate includes low, medium, and high. The fuzzy set of the error tolerance includes extremely small, small, medium, and large. The fuzzy set of the sampling interval adjustment factor includes low, medium, and high; Establish a fuzzy rule base according to the fuzzy sets, use the fuzzy linguistic values and their membership degrees as input items, and perform fuzzy reasoning according to the rule base. Use the Mamdani reasoning mechanism to obtain the fuzzy output variable, that is, the lossless priority value; Perform defuzzification on the lossless priority value to obtain a clear quantization value, and make a judgment based on a preset lossless priority threshold. If the clear quantization value is greater than or equal to the preset lossless priority threshold, the lossless compression method is preferentially adopted; otherwise, the lossy compression method is adopted; According to the judgment result, obtain a compression control strategy adjustment signal, and perform corresponding lossless or lossy compression strategy selection on the data collected by the energy storage system; Collect the feedback parameters after this round of compression execution to construct a compression effect scoring function, and obtain the compression strategy performance scoring value. The feedback parameters include the actual compression ratio, the actual reconstruction error, and the actual compression delay; If the scoring value is lower than the preset threshold for several consecutive cycles, feedback the scoring value to the fuzzy inference system to trigger the optimization of the fuzzy rule base, so as to realize the adaptive evolution of the compression strategy determination ability.

[0046] The specific calculation formula of the compression effect scoring function is as follows:

[0047] In the formula, is the compression strategy performance scoring value, is the target compression ratio, is the actual compression ratio, is the maximum acceptable error, is the actual reconstruction error, is the maximum acceptable compression processing delay, is the actual compression delay, 、 、 are weight coefficients.

[0048] It should be noted that, 。

[0049] Furthermore, introducing a genetic algorithm to optimize the fuzzy rule base to achieve adaptive adjustment of the compression strategy can improve the accuracy of compression strategy determination. Through the evolution of the genetic algorithm driven by the scoring function, the fuzzy rules gradually approach the high-performance interval, effectively improving the determination accuracy and matching degree of whether to prefer lossless compression. In the face of the complex and dynamically changing operating state of the energy storage system, the genetic algorithm can quickly search for the optimal rule combination, avoiding the problem that the artificially preset rules fail under some extreme working conditions. Using the real-time compression execution feedback as the fitness function, continuously adjusting the fuzzy rule base, the constructed closed-loop evolution chain has the closed-loop optimization ability driven by compression feedback.

[0050] The optimization of the trigger fuzzy rule base is specifically as follows: Based on the genetic algorithm, the fuzzy rules are encoded as genetic individuals in the form of rule vectors. Each fuzzy rule includes the matching relationship between the fuzzy linguistic values of the input parameters and the priority of the output strategy. Taking the compression effect scoring function as the fitness function of the genetic algorithm, perform iterative evolution operations on the individuals. The evolution operations include selecting the rule combination with a higher scoring value as the parent individual selection operation based on the roulette wheel method or the elitist retention strategy, using the two-point crossover or uniform crossover strategy to exchange the rule genes of the selected parents, generating the crossover operation of new rule individuals, and performing mutation operations on the fuzzy set labels or output strategy labels in the individuals to increase the population diversity. After the iteration reaches the preset number of evolution rounds, select the set of fuzzy rule individuals with the maximum fitness value as the updated fuzzy rule base, and replace the rules in the current inference system to achieve the adaptive evolution of the fuzzy inference system. Repeat the iterative evolution operation to form a closed-loop fuzzy rule optimization process driven by compression feedback, improving the environmental adaptability and determination accuracy of the compression strategy determination system.

[0051] The fuzzy rule base is specifically as follows: If the error tolerance is extremely small and the sampling interval adjustment factor is low, the compression control strategy is to preferentially use lossless compression. If the error tolerance is small and the compression rate is low, the compression control strategy is to tend to use lossless compression. If the error tolerance is large, the sampling interval adjustment factor is high, and the compression rate is high, the compression control strategy is to preferentially use lossy compression. If the error tolerance is medium, the sampling interval adjustment factor is medium, and the compression rate is high, the compression control strategy is to accept lossy compression. If the error tolerance is extremely small and the compression rate is medium or high, the compression control strategy is to must use lossless compression. If the error tolerance is medium and the sampling interval adjustment factor is low, the compression control strategy is to moderately prefer lossless compression. If the error tolerance is large, the sampling interval adjustment factor is medium or high, but the compression rate is low, the compression strategy is to use lossy compression as appropriate; If the error tolerance is small and the sampling interval adjustment factor is high, the compression control strategy is to select a lossy compression method that balances accuracy and compression rate.

[0052] It should be noted that the fuzzy rule base cannot output absolute conclusions. It provides a tendency. Therefore, a preset threshold must be used to "hard-determine" this tendency into a specific compression method. Through the adaptive evolution of the fuzzy rule base, especially by introducing a genetic algorithm for optimization, the system can adjust the compression control strategy in real time according to the feedback compression effect scoring function. This enables the system to automatically optimize the compression strategy and maintain high performance under the dynamically changing operating state of the energy storage system without manual intervention. By optimizing the fuzzy rule base with a genetic algorithm, the accuracy of compression strategy determination can be effectively improved, avoiding the limitations of rule setting, especially when facing extreme working conditions. The genetic algorithm can quickly search for and find the optimal rule combination, thereby achieving fine adjustment of the compression strategy and improving the stability and performance of the system. Using a fuzzy inference mechanism and Mamdani inference rules can handle the non-linear relationships of parameters such as compression rate, error tolerance, and sampling interval, avoiding the deficiencies of traditional control methods when dealing with complex inputs. The determination of the lossless priority level can more accurately reflect the actual system requirements and improve the accuracy of decision-making. Since the fuzzy control method itself has the ability to tolerate uncertainty and noise, the system can effectively cope with various uncertain factors during the operation of the energy storage system, such as real-time compression error, environmental changes, etc., and provide a more robust control strategy. Through the precise selection of lossless and lossy compression strategies, the system can select the most suitable compression method according to the actual operating state, optimize the use of computing resources and storage space, avoid over-compression or over-retaining data, and improve the overall performance of the energy storage system.

[0053] The above formulas are all dimensionless and take their numerical calculations. The formula is obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0054] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product.

[0055] Those of ordinary skill in the art can realize that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0056] In addition, in each embodiment of this application, the functional modules can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0057] As mentioned above, the above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0058] Finally: The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should all be included in the protection scope of the present invention.

Claims

1. A method for collecting, compressing, and controlling energy storage detail data based on batch and stream fusion, characterized in that It includes the following steps: Obtain the operating state parameters of the energy storage system, perform state perception and hierarchical mapping on the energy storage system to obtain the state level; Construct a Markov decision process model based on the state level, and use the deep reinforcement learning method for training and optimization to generate a policy function for compression control. According to the policy function, adaptively generate compression control parameters according to different state levels; Perform fuzzification processing on the compression control parameters, dynamically select and adjust the compression strategy based on the fuzzy inference rule base, and generate a compression control strategy adjustment signal; Execute the corresponding data compression operation based on the compression control strategy adjustment signal, and collect the feedback parameters after compression execution for optimization.

2. The method for collecting, compressing and controlling energy storage detail data based on batch-stream fusion according to claim 1, wherein The step of obtaining the operating state parameters of the energy storage system, performing state perception and hierarchical mapping on the energy storage system to obtain the state level is specifically as follows: Obtain the state perception vector through differential processing, and extract its evolution trend based on the sliding time window to generate the state evolution rate vector; Construct a state potential function based on the state evolution rate vector through the principal component analysis method; Perform state perception and state hierarchical mapping according to the state potential function to obtain the state level.

3. The method for collecting, compressing and controlling energy storage detail data based on batch-stream fusion according to claim 2, wherein Obtain the state perception vector through differential processing, and extract its evolution trend based on the sliding time window to generate the state evolution rate vector, specifically as follows: Obtain the operating state parameters of the energy storage system at the current moment, construct a state vector, and perform differential processing on the state vectors at adjacent time points to obtain the state perception vector; According to the state perception vector, construct a sliding time window sequence based on the current moment and the preset window length, and extract the evolution trend of each type of state perception vector in the window sequence in the time dimension to form the state evolution rate vector.

4. The method for collecting, compressing and controlling energy storage detail data based on batch-flow fusion according to claim 3, wherein The step of extracting the evolution trend of each type of state perception vector in the window sequence in the time dimension to form the state evolution rate vector is specifically as follows: Obtain the state perception vector sequence within the sliding time window, extract the state variable sequence including at least the state of charge, the estimated value of the health state, and the cell temperature from it; perform time trend fitting on each variable in the state variable sequence based on the least squares method to generate the corresponding fitting model; extract the fitting slope from the fitting model to construct the state evolution rate vector representing the dynamic evolution trend of the state variable.

5. The method for collecting, compressing and controlling energy storage detail data based on batch and stream fusion according to claim 2, wherein The step of performing state perception and state hierarchical mapping according to the state potential function to obtain the state level is specifically as follows: Obtain the state potential function sequence within the preset time period and sort it; Generate a candidate partition point set based on the sorted state potential function sequence, where each candidate partition point is obtained from the median of adjacent potential function values; For each candidate partition point, divide the sequence into two subsets and calculate the entropy value of each subset; Calculate the information gain of each candidate partition point based on the entropy value and the sample quantity, and determine the optimal partition point according to the principle of maximizing the information gain; Use the optimal partition point to perform level division on the state potential function sequence, and output the state level classification result of the energy storage system.

6. The method for collecting, compressing and controlling energy storage detail data based on batch-stream fusion according to claim 5, wherein The specific steps of using the optimal partition point to perform level division on the state potential function sequence and output the state level classification result of the energy storage system are as follows: (1) Obtain the sequence of state potential functions, and iteratively determine the hierarchical threshold set based on the principle of maximizing information gain; (2) By means of recursive partitioning, calculate the information gain of candidate partitioning points level by level, and select the partitioning point with the maximum information gain as the hierarchical threshold of the current level; (3) Repeat step (2) until the preset hierarchical number of levels is reached, and generate a complete hierarchical threshold set of state potential functions; (4) Perform level mapping on the real-time state potential function values according to the hierarchical threshold set, and output the state level of the energy storage system.

7. The method for collecting, compressing and controlling energy storage detail data based on batch-stream fusion according to claim 6, wherein The deep reinforcement learning method is used for training and optimization to generate a policy function for compression control. According to different state levels, compression control parameters are adaptively generated through the policy function, specifically: The deep deterministic policy gradient algorithm is used to iteratively train the Markov decision process model, and the interaction optimization of policy evaluation and policy improvement is realized through a dual neural network architecture until the cumulative discounted reward value of the policy function converges to a preset threshold; Deploy the trained policy function as a compression control policy mapping function, establish a dynamic mapping relationship between the state level and the compression parameter group, and obtain the compression control parameters corresponding to the state level of the current energy storage system, where the mapping relationship is realized through an improved radial basis function neural network, and parameter boundary constraint conditions are set to ensure the feasibility of the compression operation.

8. The method for collecting, compressing and controlling energy storage detail data based on batch-flow integration according to claim 1, wherein The compression control parameters are fuzzified, and the compression policy is dynamically selected and adjusted based on the fuzzy inference rule base to generate a compression control policy adjustment signal, specifically: Fuzzify the compression control parameters, and use the triangular membership function to map the continuous input values into fuzzy linguistic values to form a fuzzy set; Establish a fuzzy rule base according to the fuzzy set, use the fuzzy linguistic values and their membership degrees as input items, and perform fuzzy inference according to the rule base. The Mamdani inference mechanism is used to obtain the fuzzy output variable to obtain the lossless priority value; Defuzzify the lossless priority value to obtain a clear quantization value, and make a judgment based on a preset lossless priority threshold; According to the judgment result, obtain a compression control policy adjustment signal, and perform corresponding lossless or lossy compression policy selection on the data collected by the energy storage system.

9. The method for collecting, compressing and controlling energy storage detail data based on batch-stream fusion according to claim 7, wherein Perform corresponding data compression operations based on the compression control policy adjustment signal, and collect the feedback parameters after compression execution for optimization, specifically: Collect the feedback parameters after compression execution to construct a compression effect scoring function, and obtain the compression policy performance scoring value. The feedback parameters include the actual compression ratio, the actual reconstruction error, and the actual compression delay; If the scoring value is lower than the preset threshold for several consecutive cycles, feedback the scoring value to the fuzzy inference system to trigger the optimization of the fuzzy rule base.

10. The method for collecting, compressing, and controlling energy storage detail data based on batch-stream fusion according to claim 9, wherein The triggering of the optimization of the fuzzy rule base is specifically: Based on the genetic algorithm, encode the fuzzy rules in the form of rule vectors as genetic individuals. Each fuzzy rule includes the matching relationship between the fuzzy linguistic values of the input parameters and the output policy priority; Use the compression effect scoring function as the fitness function of the genetic algorithm, and perform iterative evolution operations on the individuals; After the iteration reaches the preset number of evolution rounds, select the set of fuzzy rule individuals with the largest fitness value as the updated fuzzy rule base, and replace the rules in the current inference system; Repeat the iterative optimization operation to form a closed-loop fuzzy rule optimization process based on compressed feedback drive to optimize the fuzzy rule base.

Images