Resource aggregation willingness modeling and scheduling method based on markov jump system

Abstract

The application discloses a resource aggregation willingness modeling and scheduling method based on a Markov jump system, and relates to the technical field of load scheduling.The application collects multiple basic preference parameters of each user in real time, constructs a response willingness function, maps the response willingness function to a discrete state set, establishes a state transition probability matrix family in time periods based on the discrete state set, and thus can accurately depict the complex behavior characteristics of the user based on the state transition probability matrix family, determines the available resource capacity of the system at each moment based on the state transition probability matrix family, and determines the scheduling decision at each moment according to the minimum solution of a comprehensive scheduling target function, so that the scheduling method can respond to the uncertainty of the user in real time.

Description

Technical Field

[0001] This application relates to the field of load scheduling technology, and in particular to a resource aggregation intention modeling and scheduling method based on Markov jump systems. Background Technology

[0002] Against the backdrop of the digital transformation of energy systems, distributed energy resources, energy storage, and flexible user-side resources are becoming key components in enhancing the flexibility and resilience of power systems. As the proportion of renewable energy continues to increase, system volatility and uncertainty are significantly amplified. Therefore, how to achieve efficient coordination and dynamic scheduling of various types of flexible resources has become a core issue in smart grid research.

[0003] However, existing energy dispatching methods often treat user-end load as a fixed parameter, ignoring the time-varying nature and aggregation effects of user behavior, resulting in insufficient prediction accuracy and response efficiency in multi-user scenarios. For example, Chinese patent "CN120725338A A method, system, program product and electronic device for balancing the dispatching of massive air conditioning load aggregators and users based on learning optimization and differentiable game theory" proposes to achieve master-slave game equilibrium between aggregators and users through deep learning and differentiable game theory to improve solution efficiency and model accuracy. However, this method mainly focuses on the static response learning of air conditioning loads and assumes that the user group has completely rational response characteristics to stimulus signals, ignoring the differences among users and failing to accurately characterize the behavioral characteristics of different users. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to address the shortcomings of the prior art by providing a resource aggregation intention modeling and scheduling method based on Markov jump systems, which can accurately characterize the complex behavior of users.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: On the one hand, a resource aggregation intention modeling method based on Markov jump systems is provided, including:

[0006] Based on a preset interval, multiple basic preference parameters of each user are collected in real time to construct a response willingness function, which is used to reflect the scheduling willingness of each user;

[0007] Based on a preset interval, the response willingness function is mapped to a discrete state set, which is used to reflect the set of user willingness states;

[0008] Based on the discrete state set, a family of state transition probability matrices is established for different time periods. Each element in the family of state transition probability matrices represents the probability that the user's intention state changes in each time period.

[0009] The modeling method of this invention collects multiple basic preference parameters of users in real time and constructs a time-series response willingness function for each user. Therefore, this function can reflect the resource scheduling willingness of different users changing over time. Furthermore, the response willingness function is discretized into a discrete state set according to a threshold. The discrete states can represent different user willingnesses. A family of state transition probability matrices for different time periods is established based on the discrete state set. Each element in this family of matrices reflects the probability of the user's willingness state changing at different time periods. Therefore, through the modeling method of this invention, the overall willingness distribution of the system is formed, thereby achieving an accurate characterization of the complex behavioral characteristics of users.

[0010] Furthermore, the resource aggregation intention modeling method based on Markov jump systems also includes:

[0011] The maximum likelihood estimate is used to estimate the maximum likelihood probability of each element in the family of state transition probability matrices.

[0012] By introducing a forgetting factor, an updated transition probability estimate is obtained based on the maximum likelihood probability estimate and transition probability estimate of each element.

[0013] Furthermore, the resource aggregation intention modeling method based on Markov jump systems also includes:

[0014] Based on the response willingness function corresponding to the user and the weight function of each user, a group aggregation willingness function is constructed;

[0015] The group aggregation willingness function is added to the state transition probability matrix family as a driving factor.

[0016] Furthermore, the resource aggregation intention modeling method based on Markov jump systems also includes:

[0017] The error is obtained based on the difference between the predicted value of each user's response willingness function and the actual observed value of each user;

[0018] The user's weight is corrected based on the error amount using an exponential decay model.

[0019] Furthermore, the resource aggregation intention modeling method based on Markov jump systems also includes:

[0020] The expression of each element in the family of state transition probability matrices is constructed using an external factor function, which is used to reflect the influence of the external environment on the user's intention state.

[0021] On the other hand, a resource aggregation intention scheduling method based on Markov jump systems is provided, including:

[0022] The power constraints of each user are obtained at preset intervals, and the power constraints are used as variables to construct a comprehensive scheduling objective function. The power constraints include adjustable power or displacement capability. The comprehensive scheduling objective function is used to reflect the resource scheduling cost of the system.

[0023] Based on a family of state transition probability matrices, the probability that each user is in each user's desired state at the preset interval is determined, wherein the family of state transition probability matrices is obtained based on the modeling method as described in any one of the first aspects.

[0024] Determine the maximum power constraint for each user under each user's desired state;

[0025] The available resource capacity of the system is determined based on the maximum power constraints of each user under each user's desired state and the probability of being in each user's desired state at the preset interval.

[0026] Based on the available resource capacity of the system and the minimum solution of the comprehensive scheduling objective function, the scheduling decision is determined at the preset interval.

[0027] The resource scheduling method of this invention periodically acquires the adjustable power or displacement capacity of each user and uses this as a variable to construct a comprehensive objective function reflecting the resource scheduling cost of the system. This function can therefore reflect the change in system resource scheduling cost as the load power of each user changes. Furthermore, by combining a family of state transition probability matrices that reflects the probability of changes in user desired states, the maximum adjustable power or displacement capacity of each user under different desired states is determined. Combined with the probability of each user being in different desired states at each time point, the available resource capacity of the system at each time point can be determined, i.e., the sum of resources that the system can call upon. Based on the available resource capacity of the system and the minimum solution of the comprehensive scheduling objective function, the scheduling decision at each time point is determined. This ensures that the scheduling decision, during execution, can adapt to the uncertainty of user responses in real time while ensuring reliable system operation.

[0028] Furthermore, the step of obtaining the power constraints of each user based on a preset time interval and constructing a comprehensive scheduling objective function using the power constraints as variables includes:

[0029] The power constraints of each user are obtained based on a preset time interval, and multiple scheduling optimization objective functions are introduced with the power constraints of each user as variables.

[0030] Based on the multiple scheduling optimization objective functions, the comprehensive scheduling objective function is constructed.

[0031] Furthermore, the resource aggregation intention scheduling method based on Markov jump systems also includes:

[0032] A robust min-max form optimization strategy is established, wherein the inner maximization layer is used to find the worst case of the comprehensive scheduling objective function in terms of resource scheduling cost, and the outer minimization layer is used to solve the minimum solution in the worst case.

[0033] Based on the optimization strategy, the minimum solution of the comprehensive scheduling objective function is obtained;

[0034] Based on the minimum solution of the comprehensive scheduling objective function, the scheduling decision is determined at the preset interval. Attached Figure Description

[0035] Figure 1 This is a flowchart illustrating a resource aggregation intention modeling method based on a Markov jump system provided in Embodiment 1 of the present invention;

[0036] Figure 2 This is the Markov jump state transition diagram provided in Embodiment 1 of the present invention;

[0037] Figure 3 This is a flowchart illustrating a resource aggregation intention scheduling method based on a Markov jump system provided in Embodiment 2 of the present invention;

[0038] Figure 4 This is a flowchart of a rolling time-domain optimization and robust scheduling solution provided in Embodiment 2 of the present invention. Detailed Implementation

[0039] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0040] Example 1:

[0041] Existing flexible resource modeling and scheduling methods mostly rely on static or centralized optimization frameworks, failing to fully consider the dynamic nature of user behavior and the differences in aggregation intentions. They suffer from a singular, static view of either the user or load side, resulting in insufficient consideration of users in resource scheduling decisions and an inability to meet the complexities of user behavior. Therefore, this application proposes a user aggregation intention modeling and resource scheduling method based on Markov jump systems to at least solve this technical problem. The embodiments of this application are described in detail below with reference to the accompanying drawings.

[0042] Figure 1This is a flowchart illustrating a resource aggregation intention modeling method based on Markov jump systems provided in Embodiment 1 of the present invention, as shown below. Figure 1 As shown, the method includes:

[0043] Based on preset intervals, multiple basic preference parameters of each user are collected in real time to construct a response willingness function.

[0044] For example, the aforementioned preset interval time is the segmented time, such as collecting user parameters every 10 minutes or 15 minutes. The aforementioned basic preference parameters may include the user's own preferences for energy, such as price and comfort, as well as environmental factors (temperature, humidity), or external factors such as historical response frequency, real-time excitation signals, and time-of-use electricity prices.

[0045] In this step, time can be used as a variable to establish functions for each basic preference parameter, and these functions can be summed to obtain the above response willingness function. The constructed response willingness function can reflect the scheduling willingness of each user.

[0046] In some examples, the response willingness function can be weighted and summed, introducing different weights to the different basic preference parameters of each user, so that the resulting response willingness function can more accurately reflect the user's scheduling willingness.

[0047] Based on a preset interval, the response willingness function is mapped to a discrete set of states.

[0048] For example, the above response willingness function changes over time and can yield many different values. A preset interval is set, which can be three or four segments, or different specific values ​​can be set as needed, so that the above response willingness function falls within these preset intervals, thereby mapping it to a discrete state set. Since the values ​​in each preset interval can reflect the same user willingness, the above discrete state set can reflect a set of user willingness states. Different user willingness states can be distinguished according to the number of preset intervals.

[0049] Based on the discrete state set, a family of state transition probability matrices for different time periods is established.

[0050] For example, based on the aforementioned discrete state set, a family of state transition probability matrices for a Markov transition system is constructed using a time-segment index. This time-segment index can be based on a day and divided into morning peak, regular period, and evening peak based on different times of day, so that each time period corresponds to a different user intention state. A state transition probability matrix is ​​established for each time period, so that each element in the state transition probability matrix corresponding to each time period represents the probability of a change in the user's intention state during that time period.

[0051] In this way, the overall distribution of user intentions can be realized through this family of state transition probability matrices, enabling precise characterization of user behavior characteristics, revealing the random transition patterns of user response states, and effectively reflecting the dynamic evolution characteristics of multi-user group behavior.

[0052] The following is a specific implementation method to better understand this application:

[0053] Step 1: Collect basic preference parameters of user i at time t, including price sensitivity, comfort tolerance, historical response frequency, real-time stimulus signal, environmental factors (temperature, humidity) and time-of-use electricity price.

[0054] Step 2: Based on the variables from Step 1, construct the willingness function of user i at time t, with the expression:

[0055] (1);

[0056] in, This represents the willingness of user i to respond at time t (the larger the value, the more willing the user is to participate in scheduling).

[0057] This represents the normalized weight coefficient, reflecting the importance of different factors, and satisfies... ;

[0058] This indicates price sensitivity, that is, how sensitive users are to fluctuations in electricity prices;

[0059] This indicates comfort tolerance, which is the degree to which users tolerate disruptions to their electricity usage habits or a reduction in comfort.

[0060] This indicates the historical response frequency, representing the user's past activity level in responding to requests.

[0061] This indicates an incentive signal, such as system subsidies, rewards, or demand response incentives.

[0062] Step 3, Mapped to a discrete state set according to a preset interval ,in, These represent the user's willingness status as high, medium, low, and no response, respectively.

[0063] Step 4: Construct a time period index based on the discrete state set. Family of transition probability matrices under (morning rush hour, off-peak hours, evening rush hour) The details are as follows:

[0064] (2);

[0065] in, Indicates the first State transition probability matrix for each time period (morning peak, off-peak, evening peak);

[0066] element Indicates the first During a given time period, the user's status... Jump to status The conditional probability, Indicates the status number.

[0067] It should be noted that the state transition probability matrix should satisfy probability normalization, and the sum of the transition probabilities of each row should be 1. Figure 2 This is the Markov transition diagram provided in Embodiment 1 of the present invention, such as... Figure 2 As shown in the figure, 1, 2, 3, and 4 represent the four user intention states mentioned above. At each moment, the user intention state has a different probability of changing to other user intention states or remaining unchanged, and the sum of these probabilities should be 1.

[0068] In some implementations, the representation of each element in the aforementioned family of state transition probability matrices can be constructed as an external factor function.

[0069] For example, external factors such as the user's historical response frequency, incentive intensity, weather conditions, and real-time electricity price are introduced. For each user, an external factor function with time as the variable is established, so that the state evolution is adaptive and can dynamically reflect the influence of the external environment on the user's intentions.

[0070] For example, the elements of the transition probability matrix The external factor functional representation is introduced as follows:

[0071] (3);

[0072] in, This indicates that at time t, the user starts from the user's desired state. Jump to status The probability of; Represents the regression coefficient, used to characterize the effect of different transfer factors on probability; Indicates environmental factors (such as temperature, humidity, and weather conditions); This indicates the real-time electricity price.

[0073] In this implementation, the probabilistic properties are guaranteed by constructing an exponential normalization form, thereby enabling the comprehensive perception and dynamic adjustment of the transition probability to external factors, such as historical responses, incentives, weather, and electricity prices.

[0074] In summary, this application embodiment collects multiple basic preference parameters of users in real time and constructs a time-series response willingness function for each user. Therefore, this function can reflect the resource scheduling willingness of different users changing over time. Furthermore, the response willingness function is discretized into a discrete state set according to a threshold. The discrete states can represent different user willingnesses. A family of state transition probability matrices for different time periods is established based on the discrete state set. Each element in this family of matrices reflects the probability of changes in the user's willingness state at different time periods. Then, through the modeling method proposed in this application embodiment, the overall willingness distribution of the system is formed, thereby achieving an accurate characterization of the complex behavioral characteristics of users.

[0075] In some cases, the method proposed in the embodiments of this application may further include:

[0076] The maximum likelihood estimate is used to estimate the maximum likelihood probability of each element in the family of state transition probability matrices.

[0077] By introducing a forgetting factor, an updated transition probability estimate is obtained based on the maximum likelihood probability estimate and transition probability estimate of each element.

[0078] For example, the original frequencies are estimated using maximum likelihood estimation, as shown below:

[0079] (4);

[0080] in, This represents the maximum likelihood probability estimate from state i to j;

[0081] This represents the number of observed transitions from user's desired state i to j;

[0082] This represents the total number of occurrences of user's desired state i.

[0083] Introducing smooth updates with a forgetting factor to improve time-series response capability:

[0084] (5);

[0085] in, This represents the updated transition probability estimate; This represents the estimated transition probability before the update; This represents the maximum likelihood probability estimate based on the current observations; It is a forgetting factor.

[0086] The method proposed in this application involves statistically analyzing the historical transition counts of users between different states, using the maximum likelihood estimation method to correct the transition probability in real time, and employing online adaptive estimation and exponential smoothing updates for matrix family elements. This ensures that the model is responsive to both sudden changes and long-term trends, while avoiding drastic parameter fluctuations caused by a single abnormal observation. This allows the model to continuously correct prediction biases and achieve a closed-loop iteration of user intention modeling and scheduling optimization.

[0087] It is understandable that energy systems, such as power systems, serve numerous users. These users' loads, electricity consumption levels, and importance to the entire system are not the same. Furthermore, each user's load may change over time, thus altering their importance within the system. Related technologies simply treat all users as a single entity and view each user statically, making it difficult to characterize the changing aggregation intentions and irrational decision-making characteristics of users in dynamic environments. Therefore, the method proposed in this application may further include:

[0088] Based on the response willingness function corresponding to the user and the weight function of each user, a group aggregation willingness function is constructed by weighting the weight function equally.

[0089] The group aggregation willingness function is added to the state transition probability matrix family as a driving factor.

[0090] For example, a weighted function with time as the variable is established for each user, and a group aggregated willingness function is constructed by weighting the corresponding response willingness function of each user in an equal manner, such as:

[0091] (6);

[0092] in, This represents the weighted group's willingness to aggregate; This represents the weight of user i at time t, reflecting the reliability of the data or the degree of trust in the user.

[0093] The constructed group aggregation willingness function is used as the driving factor of the state transition probability matrix family, so that the dynamic changes of users' group aggregation willingness can be reflected from the model constructed in the embodiments of this application.

[0094] In specific implementation, the constructed group aggregation willingness function can be combined with the external factor function involved in the above formula (3) to realize this application.

[0095] In some examples, the above method also includes:

[0096] The error is calculated based on the difference between the predicted value of each user's response willingness function and the actual observed value of each user.

[0097] The user's weight is corrected based on the error amount using an exponential decay model.

[0098] For example, after the system schedule is executed, the user's willingness to respond may change. For each user, the actual value observed by the system is compared with the predicted value obtained through the willingness-to-respond function. The difference between the two is used to obtain the error. Based on this error, the weighting function is corrected using an exponential decay model, for example:

[0099] (7);

[0100] in, This represents the weight of user i at the next time step; This represents the current weight of user i; This represents the prediction error, which is the difference between user i's prediction intention and the actual observation. This represents the sensitivity coefficient, used to adjust the degree of influence of errors on weights.

[0101] This implementation method takes into account user heterogeneity during the aggregation process and reduces the impact of abnormal or unreliable data on the transition matrix and scheduling decisions through weight correction.

[0102] In some examples, a preset threshold can also be set. If the error exceeds the threshold, the system will determine that the user data is abnormal and reduce its weight to a lower level to avoid it from causing a deviation in the overall willingness level.

[0103] It should be noted that the updated weights for all users are normalized to ensure... This is to maintain the numerical stability of the weighted aggregation results.

[0104] It should be noted that the process of adjusting the weighted users can be regarded as a cyclical process, continuously updating the aggregation intention and weight distribution at each moment, thereby forming an adaptive user aggregation mechanism and realizing the dual sensitivity of group intention to dynamic behavior and data reliability.

[0105] In some examples, the above-mentioned group aggregation willingness function can be established in an equal-weighted manner when all users in the system belong to the same category, as shown below:

[0106] (8);

[0107] The group aggregation willingness function determined in this way can reduce the system's computational load. Furthermore, it can also... That is, the group aggregation willingness function is used as the above response willingness function to map the above discrete state set. That is, the average willingness is used as the individual willingness, which can further reduce the system's computational load and thus improve the system's scheduling efficiency.

[0108] In some examples, the group aggregation willingness function of formula (8) can be used as the driving factor of the above state transition probability matrix. Compared with formula (6), this can reduce the amount of system computation and thus improve the system scheduling efficiency.

[0109] Example 2:

[0110] To address the technical shortcomings of existing technologies where optimization objectives and decision-making mechanisms primarily serve the power supply side and fail to consider the feedback impact of user-side intentions on scheduling results, this application proposes a resource scheduling method based on Markov transition systems to at least solve the aforementioned technical problems. Figure 3 This is a flowchart illustrating a resource aggregation intention scheduling method based on a Markov jump system provided in Embodiment 2 of the present invention, as shown below. Figure 3 As shown, the method includes:

[0111] The power constraints of each user are obtained at preset intervals, and the power constraints are used as variables to construct a comprehensive scheduling objective function.

[0112] It should be noted that user power constraints can be determined from two aspects: adjustable power, i.e., the range of user power variation, and displacement capability, i.e., phase angle displacement capability. Both of these can reflect the range of load variation on the user side. The timing of acquiring user power constraint data can be consistent with the timing of collecting user basic preference parameters mentioned above. The above-mentioned comprehensive scheduling objective function is used to reflect the resource scheduling cost of the system. This step can clearly express the scheduling pressure on the system caused by changes in user load through functional form.

[0113] In some implementations, the above-mentioned acquisition of each user's power constraints based on preset time intervals, and the construction of a comprehensive scheduling objective function using the power constraints as variables, may include:

[0114] The power constraints of each user are obtained based on a preset time interval, and multiple scheduling optimization objective functions are introduced with the power constraints of each user as variables.

[0115] Based on the multiple scheduling optimization objective functions, the comprehensive scheduling objective function is constructed.

[0116] For example, scheduling costs can encompass multiple aspects, such as economic, security, or user-side costs. A specific implementation can be as follows:

[0117] (9);

[0118] in, This represents the overall scheduling objective function described above; The aforementioned power constraints are represented as scheduling variables; This represents the weighting coefficient, indicating the trade-off ratio between different optimization objectives; This represents an economic cost function, such as electricity purchase cost or equipment start-up and shutdown cost. This represents the comfort loss function, indicating the impact of scheduling on user comfort; These indicate system operational safety indicators, such as voltage deviation and reserve margin.

[0119] This formula allows us to calculate how the system scheduling cost will be reflected when user load changes.

[0120] Based on the family of state transition probability matrices, the probability that each user is in each user's desired state at the preset interval is determined.

[0121] For example, the probability of each user being in their desired state at a preset interval can be calculated from the matrix group of the aforementioned formula (2). Combining this with the aforementioned formula, the expression can be:

[0122] (10);

[0123] in, It should belong to the aforementioned discrete state set.

[0124] Determine the maximum power constraint for each user under each user's desired state.

[0125] For example, the maximum power constraint mentioned above represents the physical upper limit of a user's desired state. This physical upper limit is usually determined by the inherent parameters of the device. Examples include the rated power range, adjustable speed, and temperature range of loads such as air conditioners, water heaters, and electric vehicles, or the charging and discharging power upper limit of energy storage devices.

[0126] Through the above steps, a model capable of accurately characterizing complex behaviors of multiple users is introduced. The embodiments of this application ensure that the final scheduling decision can adapt to the uncertainty of user responses.

[0127] The available resource capacity of the system is determined based on the maximum power constraints of each user under each user's desired state and the probability of being in each user's desired state at each preset interval.

[0128] For example, the available resource capacity of the system at a certain moment can be determined by summing the results based on the aforementioned maximum power constraint.

[0129] (11);

[0130] in, This indicates the flexible resource capacity available to the system at any given time.

[0131] Based on the available resource capacity of the system and the minimum solution of the comprehensive scheduling objective function, the scheduling decision is determined at the preset interval.

[0132] For example, the available resource capacity of the system represents the range of resources that the system can adjust at a certain moment, and scheduling decisions cannot exceed this range. Therefore, calculating the minimum solution of the above-mentioned comprehensive scheduling objective function within the range of the system's available resource capacity, this solution can be used as the adjustment capacity for scheduling decisions. This can be achieved by establishing... Perform rolling optimization to obtain the optimal scheduling decision at each time step. This expression represents the calculation... time Within the time domain The minimum solution can be found, and the time domain span can be determined according to specific requirements.

[0133] In some examples, only one execution is performed at each time step or for each scheduling decision publication. The optimal scheduling strategy is determined, and the rest are used as prediction references. The state probability and available capacity are then updated over time to achieve continuous tracking and real-time adaptation to the dynamic intentions of users.

[0134] In this embodiment, the adjustable power or displacement capacity of each user is acquired periodically, and this data is used as variables to construct a comprehensive objective function reflecting the system's resource scheduling cost. This function reflects the change in system resource scheduling cost as the load power of each user changes. Furthermore, by combining a family of state transition probability matrices that reflects the probability of changes in users' desired states, the maximum adjustable power or displacement capacity of each user under different desired states is determined. Combined with the probability of each user being in different desired states at each time point, the available resource capacity of the system at each time point can be determined, i.e., the sum of resources that the system can call upon. Based on the available resource capacity of the system and the minimum solution of the comprehensive scheduling objective function, the scheduling decision at each time point is determined. This ensures that the scheduling decision, during execution, can adapt to the uncertainty of user responses in real time while ensuring reliable system operation.

[0135] In some implementations, since the prediction error of the state transition probability matrix introduces uncertainty, this application considers introducing a robust optimization mechanism during the rolling optimization process. Therefore, the above resource scheduling method further includes:

[0136] A robust min-max form optimization strategy is established, wherein the inner maximization layer is used to find the worst case of the comprehensive scheduling objective function in terms of resource scheduling cost, and the outer minimization layer is used to solve the minimum solution in the worst case.

[0137] Based on the optimization strategy, the minimum solution of the comprehensive scheduling objective function is obtained;

[0138] For example, by introducing a set of uncertainties, the scheduling problem is extended to a robust min-max form:

[0139] (12);

[0140] in, Belongs to the set of uncertainty. It belongs to the estimation matrix.

[0141] In this expression, the inner maximization represents the worst-case assessment of system cost or risk under all perturbations of the allowed transition matrix, while the outer minimization represents the minimum solution of the aforementioned integrated scheduling objective function. For ease of solution, It can be defined by statistical confidence intervals or historical fluctuation limits, and in practice, dual relaxation or contextual approximation is used to transform robust problems into solvable convex or decomposable problems.

[0142] Based on the minimum solution of the comprehensive scheduling objective function, the scheduling decision is determined at the preset interval.

[0143] For example, the result after robust optimization can be used as the scheduling decision, which can still ensure the feasibility and near-optimality of scheduling in the worst case, thereby improving the stability of the system in complex environments.

[0144] The two embodiments above provide a modeling method and a resource scheduling method. The modeling method introduces the maximum likelihood method to iteratively optimize the transition probabilities of each element in the state transition probability matrix family, allowing for real-time correction of the transition probabilities. The scheduling method introduces a robust min-max form to combat the uncertainty caused by matrix prediction errors. Combining these two methods enables the state transition probability matrix to more accurately characterize complex user behaviors. Figure 4 This is a flowchart of a rolling time-domain optimization and robust scheduling solution provided in Embodiment 2 of the present invention, as follows: Figure 4As shown, this process includes constructing the aforementioned state transition probability matrix and user intention state model based on the acquired initialization parameters, and then establishing a cyclical update system of rolling prediction—robust modeling—solving the optimization problem—and executing the update. Specifically, the maximum likelihood method is used to perform rolling prediction of the transition probabilities of elements in the state transition probability matrix. By introducing an uncertainty set, a robust mini-maximum form optimization strategy is established. The scheduling optimization decision is obtained based on the solution of the comprehensive scheduling objective function, and the "rolling prediction—robust modeling—solving the optimization problem" process is iteratively executed, thereby significantly improving the stability and optimality of the scheduling results.

[0145] The aforementioned scheduling optimization and robust optimization models, based on the modeling and scheduling methods of this application, form an integrated modeling and scheduling framework encompassing "user intention modeling—state transition mechanism—scheduling optimization—robust solution—parameter update—execution decision." This application incorporates user-side behavioral uncertainties into the scheduling system in a progressive manner, ultimately achieving a user-centric flexible resource optimization scheduling method. This approach can at least solve the integrated modeling and scheduling problem in microgrid scenarios involving multiple types of distributed energy sources and multiple users.

[0146] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for modeling resource aggregation intentions based on Markov jump systems, characterized in that, include: Based on a preset interval, multiple basic preference parameters of each user are collected in real time to construct a response willingness function, which is used to reflect the scheduling willingness of each user; Based on a preset interval, the response willingness function is mapped to a discrete state set, which is used to reflect the set of user willingness states; Based on the discrete state set, a family of state transition probability matrices is established for different time periods. Each element in the family of state transition probability matrices represents the probability that the user's intention state changes in each time period.

2. The resource aggregation intention modeling method based on Markov jump systems according to claim 1, characterized in that, Also includes: The maximum likelihood estimate is used to estimate the maximum likelihood probability of each element in the family of state transition probability matrices. By introducing a forgetting factor, an updated transition probability estimate is obtained based on the maximum likelihood probability estimate and transition probability estimate of each element.

3. The resource aggregation intention modeling method based on Markov jump systems according to claim 1, characterized in that, Also includes: Based on the response willingness function corresponding to the user and the weight function of each user, a group aggregation willingness function is constructed; The group aggregation willingness function is added to the state transition probability matrix family as a driving factor.

4. The resource aggregation intention modeling method based on Markov jump systems according to claim 3, characterized in that, Also includes: The error is obtained based on the difference between the predicted value of each user's response willingness function and the actual observed value of each user; The user's weight is corrected based on the error amount using an exponential decay model.

5. The resource aggregation intention modeling method based on Markov jump systems according to any one of claims 1 to 4, characterized in that, Also includes: The expression of each element in the family of state transition probability matrices is constructed using an external factor function, which is used to reflect the influence of the external environment on the user's intention state.

6. A resource aggregation intention scheduling method based on Markov jump systems, characterized in that, include: The power constraints of each user are obtained at preset intervals, and the power constraints are used as variables to construct a comprehensive scheduling objective function. The power constraints include adjustable power or displacement capability. The comprehensive scheduling objective function is used to reflect the resource scheduling cost of the system. Based on the family of state transition probability matrices, the probability that each user is in each user's desired state at the preset interval is determined. The family of state transition probability matrices is obtained based on the resource aggregation intention modeling method based on Markov jump systems as described in any one of claims 1 to 5. Determine the maximum power constraint for each user under each user's desired state; The available resource capacity of the system is determined based on the maximum power constraints of each user under each user's desired state and the probability of being in each user's desired state at the preset interval. Based on the available resource capacity of the system and the minimum solution of the comprehensive scheduling objective function, the scheduling decision is determined at the preset interval.

7. The resource aggregation intention scheduling method based on Markov jump systems according to claim 6, characterized in that, The step of obtaining the power constraints of each user based on a preset time interval and constructing the comprehensive scheduling objective function using the power constraints as variables includes: The power constraints of each user are obtained based on a preset time interval, and multiple scheduling optimization objective functions are introduced with the power constraints of each user as variables. Based on the multiple scheduling optimization objective functions, the comprehensive scheduling objective function is constructed.

8. The resource aggregation intention scheduling method based on Markov jump systems according to claim 6 or 7, characterized in that, Also includes: A robust min-max form optimization strategy is established, wherein the inner maximization layer is used to find the worst case of the comprehensive scheduling objective function in terms of resource scheduling cost, and the outer minimization layer is used to solve the minimum solution in the worst case. Based on the optimization strategy, the minimum solution of the comprehensive scheduling objective function is obtained; Based on the minimum solution of the comprehensive scheduling objective function, the scheduling decision is determined at the preset interval.

Images