A distributed energy storage aggregation scheduling method and system considering user fatigue effect

Abstract

The application discloses a kind of distributed energy storage aggregation scheduling method and system considering user fatigue effect, method includes: based on partially observable Markov decision process, the user response behavior model considering user long-term response preference and short-term fatigue is established;Establish the user response state online inference strategy based on maximum likelihood estimation, infer the unobservable state of user;Based on user response behavior model and user response state online inference strategy, an online learning aggregation scheduling method combining distributed energy storage unit operating state constraint is established, to realize distributed energy storage aggregation scheduling.The present application comprehensively considers user response uncertainty, fatigue level and energy storage device actual response capability, provides reliable model basis for aggregation scheduling, and realizes the effective tracking of target power, so as to improve the precision and reliability of aggregation scheduling.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of smart grids, and relates to a distributed energy storage aggregation scheduling technology, in particular to a distributed energy storage aggregation scheduling method and system considering user fatigue effect. BACKGROUND

[0002] As an important part of smart grids, user-side distributed energy storage has the characteristics of dispersed access, limited capacity and uncertain output. With the popularity of energy storage products in homes and small scenarios, its potential in grid peak shaving and improving new energy consumption is increasingly prominent. However, without effective coordination mechanisms, distributed energy storage not only fails to achieve regulation, but also may exacerbate power fluctuations and affect grid safe operation and user economic benefits. In view of the uncertainty of user response in aggregation scheduling, existing research has introduced the multi-armed bandit (MAB) method to enhance the identification and adaptability to actual user response by optimizing the confidence upper bound or using power feedback signals. However, the existing methods generally ignore the dynamic evolution characteristics of user response ability with the change of scheduling frequency. Frequent scheduling may lead to short-term response attenuation and psychological resistance of users, i.e. "fatigue effect", which further increases the risk of user withdrawal. Although existing research has initially attempted to introduce fatigue factors into the regulation framework, it is generally simplified as a static parameter, which is difficult to reflect the dynamic evolution of fatigue state, limiting the long-term effectiveness of the strategy. SUMMARY

[0003] The application aims to overcome the deficiencies in the prior art and provide a distributed energy storage aggregation scheduling method and system considering user fatigue effect, which comprehensively considers user response uncertainty, fatigue level and actual response ability of energy storage devices, provides a reliable model basis for aggregation scheduling, and realizes effective tracking of target power, thereby improving the accuracy and reliability of aggregation scheduling.

[0004] Technical scheme: To achieve the above-mentioned purpose, the application provides a distributed energy storage aggregation scheduling method considering user fatigue effect, comprising the following steps:

[0005] S1: based on the partially observable Markov decision process, a user response behavior model considering long-term response preference and short-term fatigue of users is established;

[0006] S2: an online inference strategy of user response state based on maximum likelihood estimation is established to infer unobservable states of users;

[0007] S3: based on the user response behavior model established in step S1 and the online inference strategy of user response state established in step S2, an online learning aggregation scheduling method combined with the operating state constraints of distributed energy storage units is established to realize distributed energy storage aggregation scheduling.

[0008] Further, the response behavior of the user in the user response behavior model of step S1 is composed of observable states and unobservable states , wherein, is the energy storage capacity, is the energy storage state of charge, is the stable response preference, is the fatigue degree, is the actual response probability.

[0009] Further, in the user response behavior model of step S1:

[0010] the actual response probability is jointly influenced by the stable response preference and the dynamic fatigue degree , and is described by a logistic regression model as follows:

[0011] (1)

[0012] wherein, and are coefficients quantifying the influence of and on the response probability of the user i, respectively.

[0013] Further, in the user response behavior model of step S1:

[0014] the response result of the user is represented as a Bernoulli random process:

[0015] (2)

[0016] wherein, represents that the user i complies with the dispatching instruction, represents that the user refuses to respond;

[0017] the stable response preference of the user is used to reflect the inherent and long-term willingness of the user to participate in response as an independent individual, and is determined based on historical response data of the user:

[0018] (3)

[0019] wherein, is the response result of the user; is the number of response results;

[0020] the fatigue degree of the user is used to reflect the degree of psychological fatigue accumulation of the user on continuous response events, and evolves over time, and its change process is as follows:

[0021] (4)

[0022] wherein, , , are fatigue memory factor, response elicitation strength, recovery rate respectively.

[0023] Further, the dispatch center in step S2 needs to guess the unobservable state of the user according to the user response history record, forming the guessed unobservable state so that it is closest to the user's real unobservable state ; based on the initial unobservable , according to the user response behavior model, recursively guess the next response probability of the user .

[0024] Further, the likelihood function is constructed based on the observed historical response record in step S2 as follows:

[0025] (5)

[0026] wherein, is the initial fatigue degree.

[0027] Further, the initial unobservable state of the user is inferred based on the constructed likelihood function in step S2:

[0028] (6)

[0029] The calculated is the required initial unobservable state.

[0030] Further, the scheduling process of the online learning aggregated scheduling method combined with the operating state constraints of the distributed energy storage unit in step S3 includes:

[0031] A1: Constructing distributed energy storage operating constraints;

[0032] A2: Under the distributed energy storage operating constraints, establishing the response ranking index of each user:

[0033] (7)

[0034] (8)

[0035] wherein, and are the user response ranking indexes of the discharge instruction and the charge instruction respectively; t n is the duration of participation in aggregation; is the number of historical selections; is the adjustment parameter; It is the capacity of distributed energy storage i; It is the probability of user i's response predicted by the algorithm in round t; It is the fatigue level of user i predicted by the algorithm in round t;

[0036] A3: The required aggregated target power calculated by the scheduling center Perform aggregation;

[0037] A4: Based on the distributed energy storage aggregation scheduling strategy, calculate the response ranking index for each user i:

[0038] (9)

[0039] A5: According to Users are sorted in descending order;

[0040] A6: Select m users sequentially until the following condition is met:

[0041] (10)

[0042] If the following conditions are met:

[0043] (11)

[0044] Let m = n, where n is the total number of users; where, and These represent the target charging power and target discharging power of distributed energy storage, respectively.

[0045] A7: Output the selected user group The actual polymerization power is obtained based on its response results. ;

[0046] A8: Update user response state: Update initial unobservables based on the method in step S2. Update user fatigue levels according to equations (4) and (1) respectively. and response probability Update the status information of distributed energy storage according to the operational constraints of distributed energy storage;

[0047] A9: The scheduling center calculates the target power for the next round of aggregation. Perform the next aggregation and repeat steps A4 to A8.

[0048] Furthermore, the distributed energy storage operation constraints in step A1 are expressed as follows:

[0049] (12)

[0050] in: an upper limit of the aggregated power of the distributed energy storage; a SoC of the distributed energy storage at time t; and respectively, a charging efficiency and a charging efficiency of the whole; a capacity of the distributed energy storage as a whole; and respectively, an upper limit of the SoC and a lower limit of the SoC of the whole distributed energy storage; and respectively, a total charging power and a total discharging power of the currently aggregated distributed energy storage.

[0051] The application also provides a distributed energy storage aggregation scheduling system considering user fatigue effects, comprising:

[0052] A user response behavior model establishment module, which establishes a user response behavior model considering long-term response preferences and short-term fatigue of users based on a partially observable Markov decision process;

[0053] An unobservable state inference module, which establishes an online inference strategy of user response states based on maximum likelihood estimation to infer unobservable states of users;

[0054] An aggregation scheduling execution module, which establishes an online learning aggregation scheduling method combining with operating state constraints of distributed energy storage units based on the user response behavior model and the online inference strategy of user response states to realize distributed energy storage aggregation scheduling.

[0055] Advantages: Compared with the prior art, the application can simultaneously consider user response uncertainty and dynamic fatigue effects in aggregation scheduling, and form an improved CUCB-MLE strategy in combination with state of charge (SoC) of energy storage units, thereby guaranteeing feasibility and reliability of scheduling results. By introducing a partially observable Markov decision process (POMDP), the hidden states of users can be dynamically tracked and accurately inferred. Compared with existing methods that simplify fatigue effects as static parameters, the application has significant advantages in terms of accuracy of user behavior modeling and long-term adaptability. Compared with traditional scheduling strategies based on MAB or simple feedback mechanisms, the application has better performance in terms of target power tracking accuracy, stability of aggregation strategy, and economy of system operation. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 is a flowchart of the method of the application;

[0057] Figure 2 is a comparison chart of aggregated powers of distributed energy storage of different algorithms;

[0058] Figure 3 is a relative power regulation mismatch chart. DETAILED DESCRIPTION

[0059] The present application will be further clarified by the following examples, which should be considered as illustrating rather than limiting the scope of the application. Modifications of the application, in addition to those described, will be apparent to those with ordinary skill in the art, having the benefit of the attention drawn to the principles of the application by the description and examples given.

[0060] Example 1

[0061] As shown in the figure, the embodiment provides a distributed energy storage aggregation scheduling method considering user fatigue effect, including the following steps: Figure 1

[0062] S1: based on the partially observable Markov decision process, a user response behavior model considering user long-term response preference and short-term fatigue is established;

[0063] The response behavior of the user in the user response behavior model is composed of observable state and unobservable state , wherein, is the energy storage capacity, is the energy storage state of charge, is the stable response preference, is the fatigue degree, is the actual response probability.

[0064] In the user response behavior model:

[0065] The actual response probability is jointly influenced by the stable response preference and the dynamic fatigue degree , and a logistic regression model is used to describe:

[0066] (1)

[0067] wherein, and are coefficients quantifying the influence of and on the response probability of the user i.

[0068] In the user response behavior model:

[0069] The response result of the user is a Bernoulli random process:

[0070] (2)

[0071] wherein, represents that the user i follows the scheduling instruction, represents that the user refuses to respond;

[0072] ​The response reward of the user under the charging and discharging instructions is:

[0073] (3)

[0074] wherein, and are the response rewards of charging and discharging, respectively;

[0075] Stable response preference of the user for reflecting the inherent and long-term participation response intention of the user as an independent individual. The preference is a relatively fixed static quantity, which is determined based on the historical response data of the user:

[0076] (4)

[0077] wherein, is the response result of the user;

[0078] Fatigue degree of the user for reflecting the accumulation degree of the psychological fatigue of the user to the continuous response event, which evolves over time, and the change process is as follows:

[0079] (5)

[0080] wherein, , , are the fatigue memory factor, the response excitation intensity, and the recovery rate, respectively; the transfer mechanism makes the fatigue state have accumulation and recovery, and can reflect the psychological reaction characteristics of the user to the continuous dispatch.

[0081] S2: Establish an online inference strategy of the user response state based on maximum likelihood estimation, and infer the unobservable state of the user;

[0082] Since the unobservable state cannot be directly obtained, the dispatch center needs to guess the unobservable state of the user according to the historical record of the user response, form the guessed unobservable state so that it is maximally close to the real unobservable state of the user ; based on the initial unobservable , the response probability of the user in the next step can be recursively guessed according to the formula (1) and the formula (5) of the user response behavior model .

[0083] The likelihood function is constructed based on the observed historical response record as follows:

[0084] (6)

[0085] wherein, is the initial fatigue degree;

[0086] The goal of MLE is to identify a set of model parameters that maximizes the likelihood function. To facilitate computation and improve numerical stability, the estimation problem is transformed into a minimization problem of the negative log-likelihood function, which is mathematically equivalent to the original maximization problem;

[0087] Therefore, based on the constructed likelihood function, the initial unobservable state of the user is inferred:

[0088] (7)

[0089] The inferred is the required initial unobservable state.

[0090] S3: Based on the user response behavior model established in step S1 and the online inference strategy established in step S2, an online learning aggregated scheduling method combined with the operating state constraints of the distributed energy storage units is established to realize the aggregated scheduling of the distributed energy storage;

[0091] The scheduling process of the online learning aggregated scheduling method combined with the operating state constraints of the distributed energy storage units includes:

[0092] A1: Construct the operating constraints of the distributed energy storage;

[0093] (8)

[0094] Wherein: is the upper limit of the aggregated power of the distributed energy storage; is the SoC of the entire distributed energy storage at time t; and are the charging efficiency and the charging efficiency of the entire distributed energy storage, respectively; is the capacity of the entire distributed energy storage; and are the upper limit of the SoC and the lower limit of the SoC of the entire distributed energy storage, respectively; and represent the total charging power and the total discharging power of the current aggregated distributed energy storage, respectively;

[0095] A2: Under the operating constraints of the distributed energy storage, establish the response ranking index of each user:

[0096] (9)

[0097] (10)

[0098] Wherein, and are the user response ranking indexes of the discharging instruction and the charging instruction, respectively; t nis the duration of participation in aggregation; is the number of times of historical selection; is the adjustment parameter, balancing exploration and utilization; is the capacity of distributed energy storage i; is the estimated response probability of user i in the tth round of algorithm; is the estimated fatigue level of user i in the tth round of algorithm. This makes the scheduling strategy take into account the user's response willingness and fatigue level while further embodying the perception and adaptability of the operating state of the energy storage unit;

[0099] A3: The scheduling center calculates the required aggregation target power , executes aggregation;

[0100] A4: Based on the distributed energy storage aggregation scheduling strategy, calculate the response ranking index of each user i:

[0101] (11)

[0102] A5: According to , arrange the users in descending order;

[0103] A6: Select m users in order until the following conditions are met:

[0104] (12)

[0105] If the following conditions are met:

[0106] (13)

[0107] Let m=n, n is the total number of users; wherein, and represent the distributed energy storage charging target power and discharging target power respectively;

[0108] A7: Output the selected user group , based on the response result to get the actual aggregation power ;

[0109] (14)

[0110] A8: Update the user response state: update the initial unobservable based on the method in step S2; update the user fatigue level and the response probability according to formula (5) and formula (1) respectively; update the state information of the distributed energy storage according to formula (8);

[0111] A9: The scheduling center calculates the next round of aggregation target power The next aggregation is performed, and steps A4-A8 are repeated.

[0112] Based on the above, the method of the application first establishes a user response behavior model based on a partially observable Markov decision process, taking into account long-term response preferences and short-term dynamic fatigue modeling; then, an online inference strategy of user response state based on maximum likelihood estimation is constructed to realize dynamic estimation and belief state recursion of unobservable user states, thereby improving the accuracy of user behavior characterization; on this basis, an online learning aggregation scheduling method combined with the operating state constraints of distributed energy storage units is proposed.

[0113] Embodiment 2:

[0114] Based on the method provided in Embodiment 1, this embodiment provides a distributed energy storage aggregation scheduling system considering user fatigue effects, comprising:

[0115] A user response behavior model establishment module establishes a user response behavior model considering long-term response preferences and short-term fatigue of users based on a partially observable Markov decision process;

[0116] An unobservable state inference module establishes an online inference strategy of user response state based on maximum likelihood estimation to infer unobservable states of users.

[0117] An aggregation scheduling execution module establishes an online learning aggregation scheduling method combined with operating state constraints of distributed energy storage units based on the user response behavior model and the online inference strategy of user response state to realize distributed energy storage aggregation scheduling.

[0118] Embodiment 3:

[0119] In order to verify the effectiveness and effect of the method of the application, the following experiments and analyses are performed in this embodiment:

[0120] This embodiment compares and analyzes the power system scheduling results under four scenarios:

[0121] The four scenarios are as follows: scenario 1: without considering the participation of distributed energy storage in power system scheduling, scenario 2: aggregating distributed energy storage participating in power system scheduling based on CUCB strategy, scenario 3: aggregating distributed energy storage participating in power system scheduling based on CUCB-avg strategy according to user historical response, and scenario 4: aggregating distributed energy storage participating in power system scheduling based on the strategy proposed in the application, considering the influence of fatigue degree and SoC of distributed energy storage on the basis of learning user historical response.

[0122] As Figure 2As shown, from the target tracking effect, the strategy proposed by the application shows the best performance, the matching degree of the power curve and the target power curve is the highest, and the average relative power mismatch rate is only 4.63%, which is significantly better than CUCB and CUCB-avg.

[0123] As Figure 3 shown, the CUCB-avg strategy improves the power deviation rate by 69.9% compared with the reference CUCB, and the strategy proposed by the application further optimizes the power deviation rate to 4.63% on this basis, while the user fatigue is reduced by 11.6%. This result shows that the method of the application can not only guarantee the power adjustment accuracy in dealing with the challenge brought by the non-stationary customer fatigue environment, but also effectively alleviate the user fatigue problem.

Claims

1. A distributed energy storage aggregation scheduling method considering user fatigue effect, characterized in that, Includes the following steps: S1: Based on a partially observable Markov decision process, establish a user response behavior model that considers users' long-term response preferences and short-term fatigue. S2: Establish an online inference strategy for user response states based on maximum likelihood estimation to infer unobservable user states; S3: Based on the user response behavior model established in step S1 and the online inference strategy for user response status established in step S2, establish an online learning aggregation scheduling method that combines the operating status constraints of distributed energy storage units to realize distributed energy storage aggregation scheduling; The response behavior of the user in the user response behavior model of step S1 is composed of observable states and unobservable states together, wherein is the energy storage capacity, is the energy storage state of charge, is the stable response preference, is the fatigue degree, is the actual response probability; The scheduling center in step S2 needs to guess the unobservable state of the user according to the history of user responses to form the guessed unobservable state , so as to be as close as possible to the real unobservable state of the user ; based on the initial unobservable state , recursively guess the response probability of the user in the next step according to the user response behavior model ; The scheduling process of the online learning aggregation scheduling method combining the operating state constraints of distributed energy storage units in step S3 includes: A1: Constructing operational constraints for distributed energy storage; A2: Under the constraints of distributed energy storage operation, establish response ranking indicators for each user: （7）； （8）； in, and These are the user response ranking metrics for discharge and charge commands, respectively; t n It refers to the duration of participation in the aggregation; It is the number of times history has been chosen; It refers to adjusting parameters; It is the capacity of distributed energy storage i; It is the response probability of user i predicted by the algorithm in round t; It is the fatigue level of user i predicted by the algorithm in round t; A3: The required aggregated target power calculated by the scheduling center Perform aggregation; A4: Based on the distributed energy storage aggregation scheduling strategy, calculate the response ranking index for each user i: （9）； A5: According to Users are sorted in descending order; A6: Select m users sequentially until the following condition is met: （10）； If the following conditions are met: （11）； Let m = n, where n is the total number of users; where, and These represent the target charging power and target discharging power of distributed energy storage, respectively. A7: Output the selected user group The actual polymerization power is obtained based on its response results. ; A8: Update user response state: Update initial unobservables based on the method in step S2. Update user fatigue level and response probability Update the status information of distributed energy storage according to the operational constraints of distributed energy storage; A9: The scheduling center calculates the target power for the next round of aggregation. Perform the next aggregation and repeat steps A4 to A8.

2. The distributed energy storage aggregation and scheduling method considering user fatigue effect according to claim 1, characterized in that, In the user response behavior model of step S1: Actual response probability Subject to its stable response preference and dynamic fatigue The combined effects are described using a logistic regression model: （1）； in, and Quantitative and The coefficient that influences the response probability of user i.

3. A distributed energy storage aggregation and scheduling method considering user fatigue effects according to claim 2, characterized in that, In the user response behavior model of step S1: The user's response is represented by a Bernoulli random process: （2）； in, This indicates that user i follows the scheduling instructions. This indicates a refusal to respond; User's stable response preference This is used to reflect a user's inherent and long-term willingness to participate and respond as an independent individual, and is determined based on the user's historical response data: （3）； in, The user's response result; The number of response results; User fatigue This reflects the degree of psychological fatigue accumulated by users in response to continuous events, and its evolution over time is as follows: （4）； in, , , These are fatigue memory factor, response excitation intensity, and recovery rate, respectively.

4. A distributed energy storage aggregation and scheduling method considering user fatigue effect according to claim 3, characterized in that, In step S2, the likelihood function is constructed based on the observed historical response records as follows: （5）； in, This represents the initial fatigue level.

5. A distributed energy storage aggregation and scheduling method considering user fatigue effect according to claim 4, characterized in that, In step S2, the user's initial unobservable state is inferred based on the constructed likelihood function: （6）； The calculated This is the desired initial unobservable state.

6. A distributed energy storage aggregation and scheduling method considering user fatigue effect according to claim 5, characterized in that, The distributed energy storage operation constraints in step A1 are expressed as follows: （12）； in: This represents the upper limit of the aggregated power of distributed energy storage. The overall SoC of the distributed energy storage system at time t; and These are the overall charging efficiency and the charging efficiency, respectively. This refers to the overall capacity of distributed energy storage; and These represent the upper and lower limits of the overall distributed energy storage SoC (System-on-Chips). and These represent the total charging power and discharging power of the currently aggregated distributed energy storage, respectively.

7. A distributed energy storage aggregation and scheduling system considering user fatigue effects, characterized in that, For implementing the method of claim 1, the system comprises: The user response behavior model building module, based on a partially observable Markov decision process, establishes a user response behavior model that considers users' long-term response preferences and short-term fatigue. The unobservable state inference module establishes an online inference strategy for user response state based on maximum likelihood estimation to infer the user's unobservable state. The aggregation scheduling execution module establishes an online learning aggregation scheduling method that combines the user response behavior model and the online inference strategy of user response status with the constraints of the operating status of distributed energy storage units, thereby realizing the aggregation scheduling of distributed energy storage.

Images