New energy power fluctuation suppression and energy storage collaborative control method

Abstract

The application relates to the technical field of new energy power generation and energy storage control, and particularly discloses a new energy power fluctuation suppression and energy storage collaborative control method, which comprises the following steps: constructing a multi-dimensional state feature set; inputting a time sequence probability graph network, and obtaining an instruction execution failure probability and a capacitor energy storage overrun risk index through a team tree propagation algorithm; comparing the risk index with a dynamic safety threshold, dynamically reconstructing a target function and a constraint condition set; based on the reconstructed target function and constraint condition, combining an ultra-short-term prediction sequence and real-time energy storage, and obtaining a power distribution instruction sequence through the introduction of a rolling optimization solution of a chaotic initialization particle swarm; and finally, the first instruction is executed, and actual response data is fed back to update the conditional probability distribution of the time sequence probability graph network; the application solves the model mismatch problem caused by the time-varying of lithium-based energy storage unit parameters, and realizes adaptive collaborative control of a hybrid energy storage system.

Description

Technical Field

[0001] This invention relates to the field of new energy power generation and energy storage control technology, specifically to a method for smoothing power fluctuations in new energy sources and coordinating control with energy storage. Background Technology

[0002] With the continuous growth of installed capacity of new energy power generation such as wind and solar power, the randomness and volatility of their output pose significant challenges to the safe and stable operation of the power grid. To mitigate power fluctuations from new energy sources, configuring energy storage systems for coordinated control has become the mainstream technical solution in the industry. Among them, hybrid energy storage systems, due to their combination of the high energy density of lithium-based energy storage units and the high power density of capacitor energy storage units, can achieve frequency-division control of fluctuations. That is, capacitor energy storage units are used to smooth high-frequency components, while lithium-based energy storage units are used to smooth low-frequency components, thereby improving the smoothing effect while optimizing the overall operating efficiency of energy storage.

[0003] The existing technology has the following shortcomings:

[0004] The time-varying nonlinear characteristics of lithium-based energy storage units under low temperature or aging conditions lead to a fundamental "sensing-execution" mismatch in model predictive control strategies based on fixed parameter models, which in turn triggers a dual crisis of systemic safety and lifespan. Summary of the Invention

[0005] The purpose of this invention is to provide a method for smoothing power fluctuations in new energy sources and coordinating control with energy storage, so as to solve the problems mentioned above.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A method for smoothing power fluctuations in new energy sources and coordinating control with energy storage includes the following steps:

[0008] S1: Collect thermodynamic characteristic parameters, health decay characteristic parameters, cycle history counts, and ultra-short-term prediction error distribution bands of new energy power generation of lithium-based energy storage units to construct a multi-dimensional state feature set characterizing the current system operating state;

[0009] S2: Input the multi-dimensional state feature set into the pre-constructed temporal probabilistic graph network, perform probabilistic reasoning through the cluster tree propagation algorithm, and calculate the probability of instruction execution failure of the lithium-based energy storage unit in the current control cycle, as well as the charge storage capacity over-limit impact risk index of the capacitor energy storage unit.

[0010] S3: Compare the instruction execution failure probability and the charge storage limit over-limit impact risk index with the preset dynamic safety threshold respectively, and dynamically select or reconstruct the objective function and corresponding constraint set of the current control cycle from multiple alternative optimization objective functions based on the comparison results.

[0011] S4: Based on the objective function and constraint set, combined with the ultra-short-term prediction sequence of new energy power generation and the real-time charge storage status of lithium-based energy storage units and capacitor energy storage units, the power allocation command sequence that satisfies multi-objective collaborative optimization is obtained by rolling the solution in the finite time domain.

[0012] S5: Issue and execute the first control command in the power allocation command sequence, and collect the actual response power and charge storage change data of the lithium-based energy storage unit after execution. Feed the actual response power and charge storage change data back to the time-series probabilistic graph network to update the conditional probability distribution of each node in the time-series probabilistic graph network.

[0013] As a further aspect of the present invention: S2 specifically includes:

[0014] The thermodynamic feature parameters and health decay feature parameters in the multidimensional state feature set are used as evidence nodes input into the temporal probabilistic graph network.

[0015] The hidden nodes in the temporal probabilistic graphical network are configured as critical thresholds characterizing the current maximum allowable charge and discharge current of the lithium-based energy storage unit.

[0016] The confidence propagation algorithm is used to update the posterior probability distribution of hidden nodes by performing confidence propagation on the temporal probabilistic graphical network.

[0017] Based on the updated posterior probability distribution of hidden nodes, the probability of instruction execution failure of lithium-based energy storage units in the current control cycle is inferred. At the same time, based on the correlation constraint between the real-time charge storage of capacitor energy storage units and hidden nodes, the charge storage limit over-limit impact risk index of capacitor energy storage units is calculated.

[0018] As a further aspect of the present invention: the calculation of the charge storage capacity over-limit impact risk index of the capacitor energy storage unit specifically includes:

[0019] Extract the power throughput boundary value of the lithium-based energy storage unit from the posterior probability distribution of the hidden node;

[0020] By comparing the power throughput boundary value with the theoretical power demand in the current control cycle, the power deficit of the lithium-based energy storage unit is determined. The power deficit represents the amount of transferred power that needs to be undertaken by the capacitor energy storage unit.

[0021] The real-time charge storage capacity of the capacitor energy storage unit is obtained, and the transferred power is used as a disturbance input to deduce the trajectory of charge storage capacity change of the capacitor energy storage unit in a future preset time domain.

[0022] The trajectory of charge storage change is compared with the preset safety limit of the storage, and the cumulative probability of the charge storage reaching the safety limit is calculated. The cumulative probability is used as the charge storage over-limit impact risk index.

[0023] As a further aspect of the present invention: S3 specifically includes:

[0024] The failure probability of instruction execution and the risk index of charge storage exceeding the limit are mapped to the preset risk level division intervals to obtain the failure risk level and the risk level of exceeding the limit for the current control cycle.

[0025] Based on the failure risk level and the limit exceedance risk level, the corresponding objective function weight vector is retrieved from the pre-stored risk-strategy mapping table;

[0026] The objective function weight vector is loaded into the basic objective function to dynamically generate the objective function for the current control cycle. The basic objective function includes a grid-connected power fluctuation suppression term, a lithium-based energy storage unit lifetime loss suppression term, and a capacitor energy storage unit storage capacity safety constraint term.

[0027] Based on the failure risk level and the limit exceedance risk level, the corresponding constraint boundary parameters are retrieved from the pre-stored constraint rule base to update the power limit and storage safety limit in the constraint set.

[0028] As a further aspect of the present invention: the updated power limit and storage safety limit in the constraint set specifically include:

[0029] Input the failure risk level and the limit exceedance risk level into the pre-configured fuzzy mapping table to obtain the failure risk membership vector and the limit exceedance risk membership vector;

[0030] Using the failure risk membership vector and the limit exceedance risk membership vector as joint antecedents, and matching them with a preset fuzzy rule base, the power limit adjustment coefficient and the storage safety limit adjustment coefficient are inferred and output from the fuzzy rule base.

[0031] The power limit adjustment factor is multiplied by the power limit of the previous control cycle to obtain the updated power limit for the current control cycle.

[0032] Multiply the reserve safety limit adjustment factor by the reserve safety limit of the previous control period to obtain the updated reserve safety limit for the current control period.

[0033] Replace the original power limit parameters and storage safety limit parameters in the constraint set with the updated power limit and updated storage safety limit.

[0034] As a further aspect of the present invention: S4 specifically includes:

[0035] The objective function and constraint set are converted into numerical representation parameters that can be recognized by the solver; the numerical representation parameters, the ultra-short-term prediction sequence of new energy power generation, and the real-time charge storage status of lithium-based energy storage units and capacitor energy storage units are loaded into the rolling time domain optimization solution process.

[0036] In the rolling time-domain optimization process, a chaotic initialization particle swarm is introduced to perform iterative optimization, generating multiple candidate power allocation sequences that satisfy the constraint set.

[0037] Calculate the fitness value of each candidate power allocation sequence relative to the objective function, and select the candidate power allocation sequence with the best fitness value as the power allocation command sequence for the current control cycle;

[0038] The rolling time-domain optimization process advances by one step in each control cycle and is repeated.

[0039] As a further aspect of the present invention: in the rolling time-domain optimization process, iterative optimization is performed by introducing chaotic initialization particle swarms to generate multiple candidate power allocation sequences that satisfy the constraint set, specifically including:

[0040] A sequence of chaotic variables is generated using logistic mapping. This sequence is then mapped to the search space to obtain the initial positions of multiple particles. Each particle's initial position corresponds to a candidate power allocation sequence.

[0041] The set of constraints is introduced as a penalty term into the fitness evaluation process of each particle. The fitness value of each particle at its current position is calculated, and the optimal position of the individual and the optimal position of the group are recorded.

[0042] The velocity and spatial position of each particle are updated based on its current velocity, individual optimal position, and group optimal position, thus generating an updated particle swarm.

[0043] Determine whether the updated particle swarm satisfies the iteration termination condition. If it does, output the candidate power allocation sequence corresponding to the current optimal position of the swarm as multiple candidate power allocation sequences.

[0044] As a further aspect of the present invention: S5 specifically includes:

[0045] The actual response power of the collected lithium-based energy storage unit and the data on the change in charge storage are used as new observational evidence and injected into the corresponding evidence node in the time-series probability graph network.

[0046] Based on the likelihood between new observational evidence and the current posterior probability distribution of hidden nodes in the temporal probabilistic graphical network, calculate the conditional probability deviation between each evidence node and the hidden node.

[0047] The conditional probability table parameters between corresponding nodes in the time-series probabilistic graph network are adjusted in reverse by the conditional probability deviation value so that the adjusted conditional probability table parameters can reflect the response characteristics of the lithium-based energy storage unit under the latest operating conditions.

[0048] The adjusted temporal probabilistic graphical network is used as the input network for probabilistic inference in the next control cycle.

[0049] As a further aspect of the present invention: the calculation of the conditional probability deviation between each evidence node and the hidden node based on the likelihood between the new observation evidence and the current posterior probability distribution of the hidden nodes in the temporal probabilistic graphical network specifically includes:

[0050] Obtain the actual values ​​of each evidence node corresponding to the new observation evidence, as well as the expected values ​​of the hidden nodes in the temporal probabilistic graphical network under the current posterior probability distribution;

[0051] The deviation between the actual value of each evidence node and the expected value of the evidence node under the prior conditions is calculated to obtain the observation deviation of each evidence node.

[0052] The change in the joint probability distribution between the evidence node and the hidden node is obtained by jointly calculating the observation bias and the current posterior probability of the hidden node.

[0053] Multiply the change in the joint probability distribution by the preset update step size to obtain the conditional probability deviation value on the connection edge between the evidence node and the hidden node.

[0054] The beneficial effects of this invention are:

[0055] (1) This invention introduces a time-series probabilistic graphical network to probabilistically describe the power throughput capability of lithium-based energy storage units and uses actual response data to update network parameters online, overcoming the problem of model parameter drift caused by low temperature or aging. By calculating the probability of instruction execution failure and the risk of capacitor energy storage exceeding limits in real time, the objective function and constraint boundary can be dynamically reconstructed, ensuring that the control strategy always matches the actual physical state of the energy storage unit, effectively avoiding instruction failure and equipment overload caused by model mismatch.

[0056] (2) This invention introduces a chaotic initialization particle swarm optimization algorithm in the rolling optimization solution, which can quickly search for the optimal power allocation sequence under multiple constraints. By incorporating the lithium-based energy storage unit lifetime loss term and the capacitor energy storage unit storage capacity safety constraint term into the objective function, and dynamically adjusting the weights according to the risk level, it achieves the goal of smoothing out fluctuations while prioritizing the protection of lithium batteries from deep charge and discharge shocks, preventing supercapacitors from frequently exceeding limits due to bearing low-frequency energy, and improving the operational safety and cycle life of the hybrid energy storage system under extreme conditions. Attached Figure Description

[0057] The invention will now be further described with reference to the accompanying drawings.

[0058] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Please see Figure 1 As shown, this invention is a method for smoothing power fluctuations in new energy sources and coordinating control with energy storage, comprising the following steps:

[0061] S1: Collect thermodynamic characteristic parameters, health decay characteristic parameters, cycle history counts, and ultra-short-term prediction error distribution bands of new energy power generation of lithium-based energy storage units to construct a multi-dimensional state feature set characterizing the current system operating state;

[0062] S2: Input the multi-dimensional state feature set into the pre-constructed temporal probabilistic graph network, perform probabilistic reasoning through the cluster tree propagation algorithm, and calculate the probability of instruction execution failure of the lithium-based energy storage unit in the current control cycle, as well as the charge storage capacity over-limit impact risk index of the capacitor energy storage unit.

[0063] S3: Compare the instruction execution failure probability and the charge storage limit over-limit impact risk index with the preset dynamic safety threshold respectively, and dynamically select or reconstruct the objective function and corresponding constraint set of the current control cycle from multiple alternative optimization objective functions based on the comparison results.

[0064] S4: Based on the objective function and constraint set, combined with the ultra-short-term prediction sequence of new energy power generation and the real-time charge storage status of lithium-based energy storage units and capacitor energy storage units, the power allocation command sequence that satisfies multi-objective collaborative optimization is obtained by rolling the solution in the finite time domain.

[0065] S5: Issue and execute the first control command in the power allocation command sequence, and collect the actual response power and charge storage change data of the lithium-based energy storage unit after execution. Feed the actual response power and charge storage change data back to the time-series probabilistic graph network to update the conditional probability distribution of each node in the time-series probabilistic graph network.

[0066] In S1, thermodynamic characteristic parameters, health degradation characteristics, cycle history counts, and ultra-short-term prediction error distribution bands of new energy power generation from lithium-based energy storage units are collected to construct a multi-dimensional state feature set characterizing the current system operating state, specifically including:

[0067] First, basic data is collected and preprocessed. A thermocouple temperature sensor array installed on the surface of the lithium-based energy storage unit cell or inside the module collects the temperature value of the lithium-based energy storage unit under current operating conditions in real time at a preset sampling frequency (e.g., once per second), serving as a thermodynamic characteristic parameter characterizing its thermodynamic state. Simultaneously, using an interface communicating with the lithium-based energy storage unit management system, the current health degradation characteristic of the unit is periodically read. This health degradation characteristic is specifically a percentage value of health calculated by converting the cumulative charge / discharge ampere-hours to the rated capacity, reflecting the current aging degree of the lithium-based energy storage unit.

[0068] Subsequently, by reading the cycle count register inside the lithium-based energy storage unit management system, the cumulative number of complete charge-discharge cycles completed by the unit since its commissioning is obtained, and this number is used as the cycle history count. On the other hand, ultra-short-term forecast data from the new energy power generation forecast system is obtained. This data contains a sequence of predicted power values ​​for a preset time period in the future (e.g., the next 15 minutes to 4 hours). At the same time, historical forecast data and actual power generation data for the same period are read. By statistically analyzing the deviation distribution between the predicted and actual values, the ultra-short-term forecast error distribution band at the current moment is calculated. This error distribution band is specifically represented by the standard deviation of the forecast error or the width of the confidence interval corresponding to the forecast time.

[0069] Finally, the collected thermodynamic characteristic parameters, health decay characteristic parameters, cycle history counts, and ultra-short-term prediction error distribution bands are timestamped and cleaned to remove abnormal jump points and invalid data, forming a set of multi-dimensional state characteristics that can comprehensively reflect the current system operating conditions, which serves as input data for subsequent control steps.

[0070] In S2, a multi-dimensional state feature set is input into a pre-constructed temporal probabilistic graph network. Probabilistic inference is performed using a cluster tree propagation algorithm to calculate the probability of instruction execution failure of the lithium-based energy storage unit within the current control cycle, as well as the charge storage capacity over-limit impact risk index of the capacitor energy storage unit. Specifically, this includes:

[0071] This step first constructs a probabilistic graphical network to describe the temporal dependencies between variables. This network is built based on a historical offline dataset containing the operating parameters of lithium-based energy storage units at different temperatures and aging stages, along with their corresponding maximum allowable charge / discharge current values. By statistically analyzing the frequency of simultaneous occurrences of each variable in the historical data, a conditional probability table is calculated and defined for each node in the network. This conditional probability table defines the probability of a child node taking each possible value given that the parent node takes a specific value.

[0072] During real-time control, the thermodynamic characteristic parameters and health decay characteristic parameters from the multidimensional state feature set are first input into the time-series probabilistic graphical network as observed evidence nodes. Simultaneously, a hidden node in the network that is not directly observed is configured as a critical threshold characterizing the maximum allowable charge / discharge current of the lithium-based energy storage unit under the current operating conditions. This critical threshold is a variable to be solved through probabilistic reasoning.

[0073] Subsequently, the clique tree propagation algorithm is executed on the temporal probabilistic graphical network to propagate the confidence level. The execution process of the clique tree propagation algorithm is as follows: First, the original probabilistic graphical network is transformed into a tree-like clique tree, where each node contains a set of variables; then, bidirectional message passing occurs between adjacent nodes in the clique tree, with each message being a numerical function of the variable combination; after multiple iterations, when the belief states of all nodes stabilize, the joint probability distribution of the variables within each clique node becomes the updated posterior probability distribution. Through this algorithm, the posterior probability distribution of the hidden node (i.e., the node representing the maximum allowable charge / discharge current threshold) under the current observational evidence is finally updated. This distribution, in the form of probabilities corresponding to different current values, provides a probabilistic description of the power throughput capability of the lithium-based energy storage unit under the current operating conditions.

[0074] Based on the posterior probability distribution of the aforementioned hidden nodes, the charge storage capacity over-limit impact risk index of the capacitor energy storage unit is further calculated. The specific execution process is as follows:

[0075] First, extract the range of values ​​with the highest probability density from the updated posterior probability distribution of the hidden nodes. Then, determine the upper limit of this range as the power throughput boundary value of the lithium-based energy storage unit at the current moment. This boundary value represents the maximum power that the lithium-based energy storage unit has a high probability of safely outputting under the current operating conditions.

[0076] Then, the theoretical power demand calculated to smooth out power fluctuations in new energy sources within the current control cycle is obtained. This theoretical power demand is the total power expected to be borne by both lithium-based energy storage units and capacitor energy storage units after comprehensively considering the grid-connected power smoothing target. The power throughput boundary value is compared with the theoretical power demand, and the portion of the theoretical power demand exceeding the power throughput boundary value is calculated. This excess portion is identified as the power deficit of the lithium-based energy storage units. This power deficit represents the amount of power transferred that must be additionally borne by the capacitor energy storage units due to the limited capacity of the lithium-based energy storage units.

[0077] Next, the real-time charge storage of the capacitor energy storage unit is obtained through the communication interface. This real-time charge storage is expressed as a percentage, representing the proportion of the capacitor energy storage unit's current remaining energy to its rated capacity. Using the calculated transfer power as a disturbance input, and combining it with the rated capacity parameters of the capacitor energy storage unit, the trajectory of charge storage changes within a predetermined future time domain is deduced through integral calculations. Specifically, the future time domain is divided into multiple consecutive time segments. Within each time segment, the increase or decrease in charge storage is calculated cumulatively based on the sign (charging or discharging) and magnitude of the transfer power, thus obtaining a time-varying charge storage sequence.

[0078] Finally, the derived charge storage change trajectory is compared with the preset storage safety limit. This storage safety limit includes two boundaries: an excessively high storage safety limit and an excessively low storage safety limit. The number of time segments in the trajectory where the charge storage value exceeds the excessively high storage safety limit or falls below the excessively low storage safety limit is counted. The ratio of the number of time segments exceeding the boundary to the total number of derived time segments is calculated as the cumulative probability of the charge storage reaching the safety limit. This cumulative probability value is used as the final output charge storage limit exceedance impact risk index. This index reflects the risk that, within the current control cycle, the transfer of power from the lithium-based energy storage unit to the capacitor energy storage unit due to insufficient capacity may cause the capacitor energy storage unit to operate beyond its safe range.

[0079] In S3, the probability of instruction execution failure and the risk index of charge storage exceeding the limit are compared with preset dynamic safety thresholds. Based on the comparison results, the objective function of the current control cycle and the corresponding set of constraints are dynamically selected or reconstructed from multiple alternative optimization objective functions. Specifically, this includes:

[0080] This step first maps the calculated command execution failure probability and the charge storage over-limit impact risk index into a hierarchical system. Several risk level division intervals are pre-defined; for example, the command execution failure probability is divided into low-risk, medium-risk, and high-risk intervals based on its numerical value. Similarly, corresponding level division intervals are set for the charge storage over-limit impact risk index. By determining the numerical intervals into which the two risk indices fall, the failure risk level and over-limit risk level corresponding to the current control cycle are determined. This risk level is represented by discrete values ​​or labels, such as Level 1, Level 2, or Level 3.

[0081] Then, based on the determined failure risk level and limit exceedance risk level, the risk-policy mapping table pre-stored within the controller is queried. This mapping table is constructed as follows: through offline simulation or experimental calibration, for each combination of failure risk level and limit exceedance risk level, a set of objective function weight vectors that optimizes the overall system performance is pre-defined. This weight vector contains three weight coefficients, corresponding to the grid-connected power fluctuation suppression term, the lithium-based energy storage unit lifetime loss suppression term, and the capacitor energy storage unit storage capacity safety constraint term in the basic objective function. During the query, the failure risk level and limit exceedance risk level of the current control cycle are used as a joint index to retrieve the specific values ​​of the three corresponding weight coefficients from the mapping table.

[0082] Next, the objective function is dynamically generated. The three weighting coefficients obtained from the query are multiplied by the three preset sub-terms in the basic objective function, and the three results are summed to form the objective function actually used in the current control cycle. Specifically, the grid-connected power fluctuation suppression term is the absolute value of the deviation between the actual power at the grid connection point and the smoothing target power; the lithium-based energy storage unit lifetime loss suppression term is the absolute value of the charging and discharging power of the lithium-based energy storage unit; and the capacitor energy storage unit storage safety constraint term is the absolute value of the deviation between the real-time charge storage of the capacitor energy storage unit and the preset storage target value. By adjusting the weighting coefficients, the degree of emphasis on different control objectives is dynamically changed.

[0083] Subsequently, the boundary parameters in the constraint set are updated based on the failure risk level and the limit exceedance risk level. The specific update process is implemented through fuzzy logic reasoning, and the steps are as follows:

[0084] First, the failure risk level and the limit violation risk level are input into a pre-configured fuzzy mapping table. This mapping table defines the membership degree of different fuzzy sets corresponding to each risk level. For example, for a failure risk level, it may correspond to three fuzzy sets: "small," "medium," and "large," each with a membership function. By looking up the table, the input discrete risk levels are converted into failure risk membership vectors and limit violation risk membership vectors, each consisting of several membership values. Each membership value is a numerical value between zero and one, representing the degree to which the risk level belongs to the corresponding fuzzy set.

[0085] Then, the failure risk membership vector and the over-limit risk membership vector are merged as a joint antecedent and input into a preset fuzzy rule base for matching. This fuzzy rule base consists of multiple logical rules in the form of "if...then...", where the antecedent of each rule is a combination of failure risk membership and over-limit risk membership, and the consequent is a fuzzy set of the corresponding output power limit adjustment coefficient and storage safety limit adjustment coefficient. By calculating the excitation intensity of each rule and performing weighted synthesis and defuzzification operations (e.g., using the centroid method) on the consequents of all rules, two specific numerical outputs are finally obtained: the power limit adjustment coefficient and the storage safety limit adjustment coefficient, both of which are real numbers between zero and one.

[0086] Next, the power limit adjustment coefficient obtained through reasoning is multiplied by the power limit used in the previous control cycle to calculate the updated power limit for the current control cycle; similarly, the storage safety limit adjustment coefficient obtained through reasoning is multiplied by the storage safety limit used in the previous control cycle to calculate the updated storage safety limit for the current control cycle. The power limit and storage safety limit for the previous control cycle are initially given by system preset values.

[0087] Finally, the original power limit parameters in the constraint set are replaced with the updated power limit values ​​obtained from the calculation, and the original storage safety limit parameters in the constraint set are replaced with the updated storage safety limit values, forming the constraint set actually used in the current control cycle. This constraint set will serve as the boundary basis for rolling optimization in subsequent steps.

[0088] In S4, based on the objective function and constraint set, and combined with the ultra-short-term prediction sequence of new energy power generation and the real-time charge storage status of lithium-based energy storage units and capacitor energy storage units, a rolling solution is performed in the finite time domain to obtain a power allocation command sequence that satisfies multi-objective collaborative optimization, specifically including:

[0089] This step, based on the reconstructed objective function and the updated constraint set, combines the ultra-short-term prediction sequence of new energy power generation with the real-time charge storage status of lithium-based energy storage units and capacitor energy storage units. Within a finite time domain, it generates the power allocation command sequence for the current control cycle through rolling optimization. The specific execution process is as follows:

[0090] First, the determined objective function and constraint set are converted into numerical representation parameters. These numerical representation parameters include the specific values ​​of each weight coefficient in the objective function, the specific values ​​of the power limits and reserve safety limits in the constraint set, as well as the boundary values ​​of each variable's range. The converted parameters are stored in a data format that the solver can recognize.

[0091] Then, the aforementioned numerical characterization parameters, the ultra-short-term prediction sequence of new energy power generation (which includes predicted power values ​​for N future time points, where N is the prediction time domain length), and the real-time charge storage status of lithium-based energy storage units and capacitor energy storage units obtained through the communication interface are all loaded into the rolling time domain optimization solution process. This rolling time domain optimization solution process is executed once in each control cycle, and each solution covers a future time domain of a preset length, but only outputs the control command at the current moment.

[0092] In the rolling time-domain optimization process, a chaotic initialization particle swarm is introduced for iterative optimization to generate multiple candidate power allocation sequences that satisfy the constraint set. The specific implementation of this process is as follows:

[0093] The first step is to generate a sequence of chaotic variables using the logistic mapping. The iterative calculation formula for the logistic mapping is as follows:

[0094] ;

[0095] in, Indicates the first Chaotic variables generated in the next iteration. The value ranges from 0 to 1; To control parameters, set The value of is 4 to ensure that the system enters a completely chaotic state; The initial value is set to a random number between 0 and 1, and not equal to 0.5. Through iterative calculation using the above formula, a sequence of chaotic variables with a length equal to the preset population size multiplied by the variable dimension is generated.

[0096] The second step involves transforming the generated chaotic variable sequence into the actual search space using a linear mapping to obtain the initial positions of multiple particles. The search space is defined by the value range of each decision variable, which includes the charging and discharging power values ​​of the lithium-based energy storage unit and the capacitor energy storage unit at each future time. The initial position of each particle corresponds to a candidate power allocation sequence, which contains the power command values ​​of the lithium-based energy storage unit and the capacitor energy storage unit at each time point within a preset time domain.

[0097] The third step involves incorporating the defined set of constraints as a penalty term into the fitness evaluation process for each particle. For each particle's current position corresponding to a candidate power allocation sequence, its original fitness value under the objective function is first calculated. Then, it is checked whether all decision variables in the sequence satisfy the power limit constraint and the reserve safety limit constraint. For sequences that violate the constraints, a penalty term is added to their original fitness value. The greater the degree of violation, the larger the penalty term, thereby reducing the overall fitness value of the sequence. After calculating the overall fitness value for each particle, the best position searched so far for each particle is recorded as the individual optimal position, and the best position searched so far for the entire particle swarm is recorded as the swarm optimal position.

[0098] The fourth step involves updating the velocity and spatial position of each particle based on its current velocity, individual optimal position, and swarm optimal position using the standard particle swarm optimization (PSO) formulas. This generates an updated particle swarm. The inertia weight coefficient in the velocity update formula is set to decrease linearly with the number of iterations to balance global and local search capabilities; the learning factor coefficient is set to a fixed value to control the degree to which particles learn towards their individual and swarm optimal positions.

[0099] The fifth step is to determine whether the updated particle swarm meets the preset iteration termination conditions. These conditions include reaching the maximum number of iterations or the optimal position of the swarm showing no significant improvement in consecutive iterations. If the termination conditions are met, the candidate power allocation sequence corresponding to the current optimal position of the swarm is output as one of the multiple candidate power allocation sequences obtained in this optimization solution; otherwise, the process returns to step three to continue iterating.

[0100] After obtaining all candidate power allocation sequences, the fitness value of each candidate power allocation sequence relative to a defined objective function is calculated. The fitness value is calculated as follows: for each candidate sequence, it is substituted into the objective function expression to obtain the corresponding objective function value. Since the objective function includes grid-connected power fluctuation suppression terms, lithium-based energy storage unit lifetime loss suppression terms, and capacitor energy storage unit storage capacity safety constraint terms, and all terms are weighted by weighting coefficients, a smaller objective function value indicates better overall performance of the sequence.

[0101] The candidate power allocation sequence with the best fitness value (i.e., the smallest objective function value) among all candidate power allocation sequences is selected as the power allocation command sequence for the current control cycle. This command sequence contains power command values ​​for several future control cycles starting from the current time.

[0102] Finally, after each control cycle, the entire rolling time-domain optimization solution process is shifted forward by one step. Specifically, the current time is shifted backward by one control cycle, the latest ultra-short-term prediction sequence of new energy power generation and the latest real-time charge storage status of lithium-based energy storage units and capacitor energy storage units are re-acquired, and steps one to four above are repeated to achieve rolling optimization of the entire control process.

[0103] In S5, the first control command in the power allocation command sequence is issued and executed. After execution, the actual response power and charge storage changes of the lithium-based energy storage unit are collected. This data is then fed back to the time-series probabilistic graph network to update the conditional probability distribution of each node in the network. Specifically, this includes:

[0104] First, the first control command in the power distribution command sequence is issued to the actuators of the lithium-based energy storage unit and the capacitor energy storage unit, instructing them to charge or discharge according to the command. After the command is executed, the actual response power value of the lithium-based energy storage unit within the control cycle is collected by the current and voltage sensors in the lithium-based energy storage unit management system. Simultaneously, the charge storage value before and after the command execution is obtained by reading the charge storage register in the lithium-based energy storage unit management system, and the difference between the two is calculated as the charge storage change data within the control cycle. The collected actual response power and charge storage change data are used as new observation evidence and injected into the evidence node corresponding to the state of the lithium-based energy storage unit in the time-series probabilistic graph network. This evidence node is the node corresponding to the input thermodynamic characteristic parameter and the health decay characteristic.

[0105] Subsequently, based on the likelihood between the new observational evidence and the current posterior probability distribution of the hidden nodes in the temporal probabilistic graphical network, the conditional probability deviation between each evidence node and the hidden nodes is calculated. The specific implementation of this process is as follows:

[0106] The first step is to obtain the actual values ​​of each evidence node corresponding to the new observation evidence. These actual values ​​are the collected actual response power values ​​and charge storage change data. Simultaneously, the expected values ​​of the hidden nodes in the time-series probabilistic graphical network under the current posterior probability distribution are obtained. These expected values ​​are calculated by weighted averaging of the posterior probability distributions of the hidden nodes. Specifically, each possible value of the hidden node is multiplied by its corresponding posterior probability, and then all products are summed.

[0107] The second step involves calculating the deviation between the actual value and the expected value of each evidence node under prior conditions, thus obtaining the observation bias for each evidence node. The expected value of an evidence node under prior conditions refers to the expected value calculated solely based on the probability distribution of its parent nodes in the temporal probabilistic graphical network before introducing any new observational evidence. For each evidence node, the absolute value of the difference between its actual value and its expected value under prior conditions is taken as the observation bias for that evidence node.

[0108] The third step involves jointly calculating the observation bias and the current posterior probability of the hidden node to obtain the change in the joint probability distribution between the evidence node and the hidden node. Specifically, for each possible value of the hidden node, its corresponding probability value under the current posterior probability distribution is multiplied by the observation bias of the current evidence node, resulting in a set of product values. These product values ​​are then normalized by dividing each product value by the sum of all product values, yielding a set of normalized values. This set of values ​​represents the change in the joint probability distribution between the evidence node and the hidden node, reflecting the magnitude of the change in the degree of correlation between them under the current observation bias.

[0109] The fourth step involves multiplying the change in the joint probability distribution by a preset update step size to obtain the conditional probability deviation value on the connection edge between the evidence node and the hidden node. The preset update step size is a positive number less than 1, for example, set to 0.05, to control the magnitude of each update and avoid drastic fluctuations in network parameters due to noise from a single observation. During multiplication, each value in the change in the joint probability distribution is multiplied by the update step size to obtain a new set of values. This set of values ​​represents the conditional probability deviation value on the connection edge between the evidence node and the hidden node, used to characterize the adjustment magnitude of the conditional probability parameters caused by the introduction of new observation evidence.

[0110] Then, using the conditional probability deviation values ​​calculated above, the conditional probability table parameters between corresponding nodes in the time-series probabilistic graph network are adjusted in reverse. The adjustment method is as follows: for each connection edge between an evidence node and a hidden node, the corresponding parameter value in the original conditional probability table is added to or subtracted from the calculated conditional probability deviation value to form the updated conditional probability table parameters. After adjustment, the new conditional probability table parameters can more accurately reflect the response characteristics of lithium-based energy storage units under the latest operating conditions, such as the correspondence between actual response power and theoretical commands in low-temperature environments.

[0111] Finally, the adjusted temporal probabilistic graphical network is used as the input network for probabilistic inference in the next control cycle. At the beginning of the next control cycle, step two will re-perform risk probabilistic inference based on this updated temporal probabilistic graphical network, thereby achieving online adaptive updating of the temporal probabilistic graphical network parameters, enabling it to continuously track the dynamic characteristic changes of lithium-based energy storage units caused by temperature variations or aging processes.

[0112] The working principle of this invention is as follows: First, thermodynamic characteristic parameters, health decay characteristic parameters, cycle history counts, and ultra-short-term prediction error distribution bands of new energy power generation of lithium-based energy storage units are collected to construct a multi-dimensional state feature set. Then, the multi-dimensional state feature set is input into a pre-constructed time-series probabilistic graph network, and probabilistic inference is performed through a cluster tree propagation algorithm to calculate the command execution failure probability of the lithium-based energy storage unit in the current control cycle and the charge storage limit over-limit impact risk index of the capacitor energy storage unit. Next, the command execution failure probability and the charge storage limit over-limit impact risk index are compared with preset dynamic safety thresholds, and based on the comparison results, the objective function and corresponding constraint set of the current control cycle are dynamically selected or reconstructed from multiple alternative optimization objective functions. Based on the objective function and the constraint set, combined with the ultra-short-term prediction sequence of new energy power generation and the real-time charge storage status of lithium-based energy storage units and capacitor energy storage units, a rolling optimization solution that satisfies multi-objective collaborative optimization is obtained within the finite time domain by introducing chaotic initialization particle swarm optimization. Finally, the first control command in the power allocation command sequence is issued and executed, and the actual response power and charge storage change data of the lithium-based energy storage unit after execution are collected. The actual response power and charge storage change data are fed back to the time-series probabilistic graph network to update the conditional probability distribution of each node in the time-series probabilistic graph network. The updated time-series probabilistic graph network is used as the input for probabilistic inference in the next control cycle, thereby realizing closed-loop adaptive collaborative control.

[0113] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for coordinating new energy power fluctuation mitigation with energy storage control, characterized in that, Includes the following steps: S1: Collect thermodynamic characteristic parameters, health decay characteristic parameters, cycle history counts, and ultra-short-term prediction error distribution bands of new energy power generation of lithium-based energy storage units to construct a multi-dimensional state feature set characterizing the current system operating state; S2: Input the multi-dimensional state feature set into the pre-constructed temporal probabilistic graph network, perform probabilistic reasoning through the cluster tree propagation algorithm, and calculate the probability of instruction execution failure of the lithium-based energy storage unit in the current control cycle, as well as the charge storage capacity over-limit impact risk index of the capacitor energy storage unit. S3: Compare the instruction execution failure probability and the charge storage limit over-limit impact risk index with the preset dynamic safety threshold respectively, and dynamically select or reconstruct the objective function and corresponding constraint set of the current control cycle from multiple alternative optimization objective functions based on the comparison results. S4: Based on the objective function and constraint set, combined with the ultra-short-term prediction sequence of new energy power generation and the real-time charge storage status of lithium-based energy storage units and capacitor energy storage units, the power allocation command sequence that satisfies multi-objective collaborative optimization is obtained by rolling the solution in the finite time domain. S5: Issue and execute the first control command in the power allocation command sequence, and collect the actual response power and charge storage change data of the lithium-based energy storage unit after execution. Feed the actual response power and charge storage change data back to the time-series probabilistic graph network to update the conditional probability distribution of each node in the time-series probabilistic graph network.

2. The method for coordinating new energy power fluctuation mitigation and energy storage control according to claim 1, characterized in that, S2 specifically includes: The thermodynamic feature parameters and health decay feature parameters in the multidimensional state feature set are used as evidence nodes input into the temporal probabilistic graph network. The hidden nodes in the temporal probabilistic graphical network are configured as critical thresholds characterizing the current maximum allowable charge and discharge current of the lithium-based energy storage unit. The confidence propagation algorithm is used to update the posterior probability distribution of hidden nodes by performing confidence propagation on the temporal probabilistic graphical network. Based on the updated posterior probability distribution of hidden nodes, the probability of instruction execution failure of lithium-based energy storage units in the current control cycle is inferred. At the same time, based on the correlation constraint between the real-time charge storage of capacitor energy storage units and hidden nodes, the charge storage limit over-limit impact risk index of capacitor energy storage units is calculated.

3. The method for coordinating new energy power fluctuation mitigation and energy storage control according to claim 2, characterized in that, The calculated charge storage capacity over-limit impact risk index of the capacitor energy storage unit specifically includes: Extract the power throughput boundary value of the lithium-based energy storage unit from the posterior probability distribution of the hidden node; By comparing the power throughput boundary value with the theoretical power demand in the current control cycle, the power deficit of the lithium-based energy storage unit is determined. The power deficit represents the amount of transferred power that needs to be undertaken by the capacitor energy storage unit. The real-time charge storage capacity of the capacitor energy storage unit is obtained, and the transferred power is used as a disturbance input to deduce the trajectory of charge storage capacity change of the capacitor energy storage unit in a future preset time domain. The trajectory of charge storage change is compared with the preset safety limit of the storage, and the cumulative probability of the charge storage reaching the safety limit is calculated. The cumulative probability is used as the charge storage over-limit impact risk index.

4. The method for coordinating new energy power fluctuation mitigation and energy storage control according to claim 1, characterized in that, S3 specifically includes: The failure probability of instruction execution and the risk index of charge storage exceeding the limit are mapped to the preset risk level division intervals to obtain the failure risk level and the risk level of exceeding the limit for the current control cycle. Based on the failure risk level and the limit exceedance risk level, the corresponding objective function weight vector is retrieved from the pre-stored risk-strategy mapping table; The objective function weight vector is loaded into the basic objective function to dynamically generate the objective function for the current control cycle. The basic objective function includes a grid-connected power fluctuation suppression term, a lithium-based energy storage unit lifetime loss suppression term, and a capacitor energy storage unit storage capacity safety constraint term. Based on the failure risk level and the limit exceedance risk level, the corresponding constraint boundary parameters are retrieved from the pre-stored constraint rule base to update the power limit and storage safety limit in the constraint set.

5. The method for coordinating new energy power fluctuation mitigation and energy storage control according to claim 4, characterized in that, The updated set of power limits and reserve safety limits specifically includes: Input the failure risk level and the limit exceedance risk level into the pre-configured fuzzy mapping table to obtain the failure risk membership vector and the limit exceedance risk membership vector; Using the failure risk membership vector and the limit exceedance risk membership vector as joint antecedents, and matching them with a preset fuzzy rule base, the power limit adjustment coefficient and the storage safety limit adjustment coefficient are inferred and output from the fuzzy rule base. The power limit adjustment factor is multiplied by the power limit of the previous control cycle to obtain the updated power limit for the current control cycle. Multiply the reserve safety limit adjustment factor by the reserve safety limit of the previous control period to obtain the updated reserve safety limit for the current control period. Replace the original power limit parameters and storage safety limit parameters in the constraint set with the updated power limit and updated storage safety limit.

6. The method for coordinating new energy power fluctuation mitigation and energy storage control according to claim 1, characterized in that, S4 specifically includes: The objective function and constraint set are converted into numerical representation parameters that can be recognized by the solver; the numerical representation parameters, the ultra-short-term prediction sequence of new energy power generation, and the real-time charge storage status of lithium-based energy storage units and capacitor energy storage units are loaded into the rolling time domain optimization solution process. In the rolling time-domain optimization process, a chaotic initialization particle swarm is introduced to perform iterative optimization, generating multiple candidate power allocation sequences that satisfy the constraint set. Calculate the fitness value of each candidate power allocation sequence relative to the objective function, and select the candidate power allocation sequence with the best fitness value as the power allocation command sequence for the current control cycle; The rolling time-domain optimization process advances by one step in each control cycle and is repeated.

7. The method for coordinating new energy power fluctuation mitigation and energy storage control according to claim 6, characterized in that, In the rolling time-domain optimization process, iterative optimization is performed by introducing chaotic initialization of the particle swarm to generate multiple candidate power allocation sequences that satisfy the constraint set, specifically including: A sequence of chaotic variables is generated using logistic mapping. This sequence is then mapped to the search space to obtain the initial positions of multiple particles. Each particle's initial position corresponds to a candidate power allocation sequence. The set of constraints is introduced as a penalty term into the fitness evaluation process of each particle. The fitness value of each particle at its current position is calculated, and the optimal position of the individual and the optimal position of the group are recorded. The velocity and spatial position of each particle are updated based on its current velocity, individual optimal position, and group optimal position, thus generating an updated particle swarm. Determine whether the updated particle swarm satisfies the iteration termination condition. If it does, output the candidate power allocation sequence corresponding to the current optimal position of the swarm as multiple candidate power allocation sequences.

8. The method for coordinating new energy power fluctuation mitigation and energy storage control according to claim 1, characterized in that, S5 specifically includes: The actual response power of the collected lithium-based energy storage unit and the data on the change in charge storage are used as new observational evidence and injected into the corresponding evidence node in the time-series probability graph network. Based on the likelihood between new observational evidence and the current posterior probability distribution of hidden nodes in the temporal probabilistic graphical network, calculate the conditional probability deviation between each evidence node and the hidden node. The conditional probability table parameters between corresponding nodes in the time-series probabilistic graph network are adjusted in reverse by the conditional probability deviation value so that the adjusted conditional probability table parameters can reflect the response characteristics of the lithium-based energy storage unit under the latest operating conditions. The adjusted temporal probabilistic graphical network is used as the input network for probabilistic inference in the next control cycle.

9. The method for coordinating new energy power fluctuation mitigation and energy storage control according to claim 8, characterized in that, The calculation of the conditional probability deviation between each evidence node and the hidden nodes is based on the likelihood between the new observation evidence and the current posterior probability distribution of the hidden nodes in the temporal probabilistic graphical network. Specifically, this includes: Obtain the actual values ​​of each evidence node corresponding to the new observation evidence, as well as the expected values ​​of the hidden nodes in the temporal probabilistic graphical network under the current posterior probability distribution; The deviation between the actual value of each evidence node and the expected value of the evidence node under the prior conditions is calculated to obtain the observation deviation of each evidence node. The change in the joint probability distribution between the evidence node and the hidden node is obtained by jointly calculating the observation bias and the current posterior probability of the hidden node. Multiply the change in the joint probability distribution by the preset update step size to obtain the conditional probability deviation value on the connection edge between the evidence node and the hidden node.

Images