Method and system for scheduling decision aggregation of an offshore hybrid platform

Abstract

The application discloses a scheduling decision aggregation method for an offshore hybrid power platform, and specifically comprises the following steps: configuring a flow-level utility function for each power service flow on each subunit, wherein the flow-level utility function is positive and a concave function in a working interval; configuring a unit-level utility function for each subunit, wherein the unit-level utility function is used for representing bus capacity occupation and fuel consumption level; constructing a platform total utility function based on the flow-level utility function and the unit-level utility function; and periodically combining power response demands accumulated between two adjacent scheduling decision update time points into a single scheduling instruction and issuing the single scheduling instruction to a corresponding execution mechanism by each sub-control unit at an optimal scheduling decision update interval period. The application only requires that the utility function satisfies positivity and concavity, can flexibly configure for heterogeneous targets such as fuel economy, load reliability and storage state of charge balance, has low online calculation complexity, and can maintain stable scheduling performance under a bus capacity limited working condition.

Description

Technical Field

[0001] This invention belongs to the technical field of ship microgrids, energy coordination and power system integrated control of offshore operation platforms, and particularly relates to a scheduling decision aggregation method and system for offshore hybrid power platforms. Background Technology

[0002] Offshore hybrid power platforms typically employ a power supply structure that combines multiple heterogeneous power units with various types of loads. The power supply side includes diesel generator sets, energy storage battery packs, waste heat recovery power generation units, and renewable energy auxiliary units, while the load side includes propulsion motors, auxiliary systems, critical safety loads, and interruptible loads. These platforms are characterized by isolated operation, high fuel costs, significant load impacts, and stringent reliability requirements. Their energy management systems need to coordinate multiple power sources and respond to multiple loads under frequently changing operating conditions.

[0003] Existing energy management methods for offshore hybrid power platforms suffer from the following shortcomings: First, rule-based scheduling methods rely on engineering experience to set switching thresholds and priorities, making it difficult to maintain optimality when operating conditions deviate from preset ranges. Furthermore, each type of platform requires parameter retuning, resulting in poor portability. Second, while scheduling methods based on model predictive control or mixed-integer linear programming can handle multivariate constraints, their online computational complexity increases non-linearly with the number of power and load units, making it difficult to complete rolling solutions on millisecond to second timescales. This is particularly problematic when bus capacity is limited or load changes abruptly, leading to solution failures. Third, most existing methods fix the optimization objective as fuel consumption or emission intensity, failing to simultaneously support heterogeneous objectives such as fuel economy, load reliability, energy storage state-of-charge balance, and voltage / frequency deviation margins. Moreover, the optimization preferences of different power and load units are difficult to configure independently. Fourth, when the platform scale expands or subunits are dynamically added or removed, existing methods typically require reconstructing the optimization problem and solving it holistically, lacking the ability to incrementally adapt to dynamic structural changes. Fifth, existing methods lack analytical performance boundaries regarding the relationship between fuel consumption and load reliability, making it difficult to theoretically guide the tuning and trade-offs of scheduling strategies.

[0004] Existing energy management methods suffer from the following shortcomings: First, rule-based scheduling methods rely on engineering experience to set switching thresholds and priorities, making it difficult to maintain optimality when operating conditions deviate from preset ranges. Furthermore, each platform requires parameter retuning, resulting in poor portability. Second, while model predictive control or mixed-integer linear programming methods can handle multivariate constraints, their online computational complexity increases non-linearly with the number of power and load units, making it difficult to complete rolling solutions within millisecond to second timescales, especially prone to failure when bus capacity is limited or load changes abruptly. Third, most existing methods fix the optimization objective as fuel consumption or emission intensity, failing to simultaneously support heterogeneous objectives such as fuel economy, load reliability, energy storage state-of-charge balance, and voltage / frequency deviation margins. Moreover, the optimization preferences of different power and load units are difficult to configure independently. Fourth, when the platform scale expands or sub-units are dynamically added or removed, existing methods typically require reconstructing the optimization problem and solving it holistically, lacking incremental adaptability to dynamic structural changes. Fifth, the lack of analytical performance boundaries between fuel consumption and load reliability makes it difficult to theoretically guide the tuning and trade-off of scheduling strategies.

[0005] Therefore, it is necessary to provide a method and system for scheduling decision aggregation of offshore hybrid power platforms that can simultaneously support configurable optimization of heterogeneous utility objectives, guarantee convergence to the global optimal solution, have low online computational complexity, support congestion-aware scheduling under bus capacity and fuel constraints, and adapt to the dynamic entry and exit of sub-units. Summary of the Invention

[0006] Purpose of the Invention: The purpose of this invention is to provide a scheduling decision aggregation method and system for offshore hybrid power platforms. This method enables the platform to simultaneously support configurable optimization of heterogeneous utility objectives, guarantee convergence to the global optimum, exhibit low online computational complexity, support congestion-aware scheduling under bus capacity and fuel constraints, and adapt to the dynamic addition and removal of sub-units.

[0007] Technical Solution: This invention provides a scheduling decision aggregation method for an offshore hybrid power platform. The offshore hybrid power platform includes a central energy management unit and several sub-control units corresponding to different sub-units. The sub-units include power supply units and load units. Specifically, the method involves: configuring a flow-level utility function for each power service flow on each sub-unit, wherein the flow-level utility function is positive and concave within the operating interval; configuring a unit-level utility function for each sub-unit, wherein the unit-level utility function characterizes bus capacity occupancy and fuel consumption levels; constructing a platform total utility function based on the flow-level utility function and the unit-level utility function; determining the scheduling decision update interval for each sub-unit that maximizes the platform total utility function, as the optimal scheduling decision update interval for each sub-unit; and periodically merging the accumulated power response demand between two adjacent scheduling decision update times into a single scheduling command and issuing it to the corresponding execution mechanism according to their respective optimal scheduling decision update interval.

[0008] Furthermore, the total utility function of the platform is:

[0009]

[0010] in, The number of sub-units. Subunit The number of power service streams on the network. Subunit The scheduling decision update interval, Let be the vector formed by the scheduling decision update intervals of each subunit. For unit-level utility functions, Subunit Upper The flow-level utility function of a power service stream The expected arrival rate of the power service stream.

[0011] Furthermore, the determination of the scheduling decision update interval of each sub-unit that maximizes the total utility function of the platform specifically involves: based on the first-order optimality condition of the scheduling decision update interval of each sub-unit according to the total utility function of the platform, the bisection method is used to iteratively solve the problem within a preset working interval. In each iteration, the midpoint of the current interval is taken and the boundary endpoint with the same sign as the value at the midpoint is replaced with the midpoint, until the interval length is less than the preset solution accuracy.

[0012] Furthermore, the unit-level utility function adopts the same exponential form for each sub-unit:

[0013]

[0014] in, These are the fitting coefficients, which are obtained by systematically identifying historical bus occupancy and historical fuel consumption data. Characterization subunit The number of times a scheduling decision update is initiated per unit of time. This represents the number of sub-units.

[0015] Furthermore, the optimal scheduling decision update interval for each sub-unit is implemented in a distributed manner, including:

[0016] The central energy management unit broadcasts candidate global constant vectors to each sub-control unit;

[0017] For each candidate global constant in the candidate global constant vector, each sub-control unit uses a bisection method to locally solve the local scheduling decision update interval based on the local power service flow parameters, and calculates the local response component based on the obtained local scheduling decision update interval before sending it back to the central energy management unit.

[0018] The central energy management unit accumulates the response components of the entire platform and calculates the iteration error corresponding to each candidate global constant, updates the candidate global constant vector according to the binary search method, and rebroadcasts it.

[0019] Repeat the above steps until the iteration error is less than the preset threshold. Then, take the corresponding candidate global constant as the optimal global constant and send the local optimal scheduling decision update interval of each sub-unit corresponding to the optimal global constant to each sub-control unit.

[0020] Furthermore, the local response component is:

[0021]

[0022] in, Candidate global constants, Subunit Based on candidate global constants The local scheduling decision update interval obtained from the local solution. Number the sub-units. , The total number of controlled sub-units; Subunit The number of power service streams carried on it. Subunit Upper The expected arrival rate of each power service stream, Subunit Upper The expected flow-level utility function for each power service flow.

[0023] Furthermore, the method also includes an incremental update step for the dynamic entry or exit of sub-units during operation: when a new sub-unit is entered into operation, the central energy management unit sends the current candidate global constant vector to the new sub-unit, the new sub-unit calculates its own response vector and sends it back, and the central energy management unit incrementally updates the global error corresponding to each candidate global constant based on the full platform response components; when an existing sub-unit exits operation, the central energy management unit removes the response vector of the exited sub-unit from the full platform response components and incrementally updates the global error; the candidate global constant with the smallest absolute value of the updated error is taken as the new optimal global constant.

[0024] Furthermore, the incremental update step also includes: when the deviation between the new optimal global constant and the original optimal global constant exceeds a preset broadcast threshold, the central energy management unit sends the updated local optimal scheduling decision update interval of each subunit to each subcontrol unit; when the deviation does not exceed the preset broadcast threshold, the original optimal scheduling decision update interval remains unchanged.

[0025] Furthermore, the flow level utility function is selected from at least one of the following: response delay type, non-timeout response probability type, non-timeout response quantity type, and power response and timeliness type; wherein the power response and timeliness type flow level utility function is:

[0026]

[0027] in Let x be the expected response time of the corresponding power service flow, and let x be the scheduling decision update interval.

[0028] A scheduling decision aggregation system for an offshore hybrid power platform includes a central energy management unit, several sub-control units corresponding to different sub-units, and an onboard communication bus connecting the central energy management unit and each sub-control unit. Each sub-control unit includes a power supply unit and a load unit. Each sub-control unit stores the flow-level utility function and power service flow parameters for each power service flow within its sub-unit. The flow-level utility function is used to locally solve for the local scheduling decision update interval of the sub-unit based on local parameters. The central energy management unit configures unit-level utility functions characterizing bus capacity occupancy and fuel consumption levels. Each sub-control unit also periodically merges the accumulated power response demand between two adjacent scheduling decision update times into a single scheduling command and issues it to the corresponding actuator according to its optimal scheduling decision update interval. The central energy management unit determines the optimal scheduling decision update interval for each sub-unit in a distributed manner by broadcasting candidate global constant vectors to each sub-control unit, receiving local response components returned by each sub-control unit, accumulating the response components of the entire platform, and iteratively updating the candidate global constant vectors using a binary search method.

[0029] Beneficial Effects: Compared with existing energy management methods for offshore hybrid power platforms, this invention has the following beneficial effects. First, the configured utility function only needs to satisfy positivity and concavity, without restrictions on its specific form. It can be flexibly configured to address heterogeneous objectives such as fuel economy, load reliability, state-of-charge balance, voltage-frequency deviation margin, and emission intensity, without requiring algorithm redesign when switching objectives. Second, it provides independently configurable optimization granularity at three levels: power service flow level, scheduling decision update time level, and platform operating condition level, making it suitable for scenarios with heterogeneous requirements such as propulsion, auxiliary machinery, and critical safety loads. Third, the concavity condition guarantees a unique and globally optimal solution. A bisection method is used for online solution, with a single sub-unit complexity of O(log n). The convergence steps are logarithmic, enabling rolling scheduling to be completed in milliseconds to seconds. Fourth, the congestion-aware scheme models platform-level constraints as exponential unit-level utility functions, allowing the scheduling update interval to adaptively adjust with bus occupancy and fuel consumption, remaining stable even under capacity-limited or fuel-constrained conditions, and requiring only one bidirectional communication per iteration. Fifth, the incremental update mechanism allows sub-unit entry and exit to only require local response feedback and lightweight global error updates, without the need to refactor the optimization problem. Sixth, it reveals the analytical performance boundary between fuel consumption and load reliability, and based on this, introduces fine-resolution power response and timeliness indices to obtain a compromise solution that is strictly superior to the binarized objective. Attached Figure Description

[0030] Figure 1 This is a flowchart of the present invention;

[0031] Figure 2 This is a topology diagram of the energy management system for the offshore hybrid power platform of the present invention. Detailed Implementation

[0032] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0033] like Figure 1 As shown, the method steps of the present invention are as follows:

[0034] Step S1: The central energy management unit collects power service flow parameters from each sub-control unit. With the configured flow level utility function Determine the unit-level utility function Specific form and fitting coefficient ;

[0035] Step S2: The central energy management unit broadcasts candidate units to each sub-control unit. Value vector;

[0036] Step S3: Each sub-control unit uses a bisection method to solve locally based on its local parameters. and send back response components. ;

[0037] Step S4: The central energy management unit accumulates the response components of the entire platform and calculates the iteration error, then updates the candidate components using the binary search method. Value vector;

[0038] Step S5: Repeat steps S2 to S4 until the iteration error is less than the preset threshold;

[0039] Step S6: The central energy management unit will be optimized. The decision is then distributed to each sub-control unit, and each sub-control unit performs scheduling decision updates periodically according to the optimal value.

[0040] Step S7: When a sub-unit is dynamically put into or taken out of operation, refresh the optimal value according to the incremental update mechanism. Value and issue a new optimal .

[0041] The online computational complexity of the solution process on each sub-control unit is: The distributed protocol has a constant number of iteration steps, which can complete rolling scheduling within a time scale of milliseconds to seconds, meeting the real-time scheduling response requirements of offshore hybrid power platforms.

[0042] In one embodiment, the offshore hybrid power platform includes a central energy management unit and several sub-control units, each corresponding to a different type of sub-unit. The sub-units include at least two or more of the following types: main diesel generator set, auxiliary diesel generator set, energy storage battery pack, waste heat recovery power generation unit, propulsion load control unit, auxiliary machinery load control unit, and critical safety load control unit. The central energy management unit and each sub-control unit communicate bidirectionally via an internal communication bus, which includes, but is not limited to, industrial Ethernet, CAN bus, or redundant fiber optic ring network.

[0043] The theoretical basis of the method includes the construction and simplification of the platform's total utility function, the optimality condition and proof of global optimality, the correction of the total utility function under the congestion-aware scheme and the proof of the diagonal dominance of the Hessian matrix, and the complete derivation of the analytical performance boundary between fuel consumption and load reliability.

[0044] Construction and simplification of the platform's total utility function: Suppose that the offshore hybrid power ship platform includes Each controlled subunit is numbered... Identification; each sub-unit configuration A power service stream, the power service stream being numbered Identifier. Sub-unit. The scheduling decision update interval is The vector formed by the scheduling decision update interval of each sub-unit is denoted as Let the parameters characterizing the platform-level operating conditions be denoted as . The parameters are used to characterize bus capacity occupancy, fuel consumption levels, or other platform-level states. (Subunit) Upper The instantaneous flow level utility function of each power service flow is: Subunit The unit-level utility function is Assume that the power scheduling task has elapsed since the decision time. Arrive in seconds The probability of one is The response time limit is set to a value of The probability density is Then the platform's total utility function The original form is:

[0045]

[0046] To facilitate the solution, the following two assumptions are introduced. First, the flow-level utility function is independent of absolute time and depends only on the relative time from the generation of the power scheduling task to its inclusion in the scheduling update, i.e.:

[0047]

[0048] Secondly, the power arrival process is smooth within the scheduling decision update interval, i.e. Define the expected arrival rate of the power service flow. for:

[0049]

[0050] Substitute into the platform's total utility function:

[0051]

[0052] make ,have to:

[0053]

[0054] The combined integral of the response time distribution and relative duration is defined as the expected flow level utility:

[0055]

[0056] We obtain a compact form of the platform's total utility function:

[0057]

[0058] Optimality condition and proof of global optimality: The total utility function of the platform with respect to any sub-unit Scheduling decision update interval Taking the partial derivative, we can obtain the Jacobian component. :

[0059]

[0060] Due to the positivity and concavity of the configured utility function, i.e. , , , For the equation to have a solution, one of the two first-order terms must be negative:

[0061]

[0062] Equivalent land:

[0063]

[0064] Taking the partial derivative of the Jacobian matrix further, we obtain the diagonal elements of the Hessian matrix as follows:

[0065]

[0066] The off-diagonal elements of the Hessian matrix are:

[0067]

[0068] Sign analysis is performed on the three terms of the diagonal elements respectively: In the first term , The product is negative; in the second term , The product is negative; the third term is negative according to the aforementioned first-order term sign constraint. Therefore, the diagonal elements are always negative. Combined with the fact that the off-diagonal elements are always zero, the Hessian matrix is ​​a negative definite matrix. Thus, it can be concluded that when the optimality equation has a solution, this solution is unique and corresponds to the global maximum value of the platform's total utility function.

[0069] Total utility function correction under the congestion-aware scheme: In the congestion-aware scheme, the unit-level utility function depends on the platform-wide scheduling decision update interval vector. The flow-level utility function depends only on the current sub-unit. The platform's total utility function is rewritten as follows:

[0070]

[0071] To ensure fairness among all sub-units, all sub-units adopt the same form of unit-level utility function. Jacobian components are rewritten as:

[0072]

[0073] Equivalent rewrite as:

[0074]

[0075] Proof of diagonal dominance of the Hessian matrix under the congestion-aware scheme: Under the congestion-aware scheme, the cell-level utility function adopts the same exponential form for each sub-cell. ,in The fitting coefficients are obtained by least-squares identification of historical bus occupancy and historical fuel consumption data. The second-order mixed partial derivative of the Hessian matrix is ​​calculated for the platform's total utility function under the congestion-aware scheme.

[0076]

[0077] The sum of the absolute values ​​of the off-diagonal elements in any row is:

[0078]

[0079] The absolute values ​​of the diagonal elements are:

[0080]

[0081] The diagonal dominance condition is:

[0082]

[0083] Simplifying, we get:

[0084]

[0085] Since all diagonal elements are negative and the Hessian matrix is ​​strictly diagonally dominant, the Hessian matrix is ​​a negative definite matrix. The optimal solution is unique and corresponds to the global maximum value of the platform's total utility function.

[0086] Analytical performance boundary between fuel consumption and load reliability: A complete derivation process is given for the existence of an analytical performance boundary between fuel consumption and load reliability. The derivation supports the assertion about the analytical performance boundary in the aforementioned invention and serves as the theoretical basis for subsequent selection of flow-level utility functions.

[0087] Assuming that the execution fuel consumption corresponding to each scheduling decision update is approximately constant, the unit-level utility function is equivalent to... The form is given by:

[0088]

[0089] The flow-level utility function is chosen to be of the non-timeout response quantity type, i.e.:

[0090]

[0091]

[0092] Substituting the aforementioned form into the optimality condition, for each sub-unit... After simplification and rearrangement, the optimal scheduling decision update interval can be obtained as follows:

[0093]

[0094] The optimal solution can be determined from the equation. The following conditions must be met: Under this value, the cumulative probability of timeout for all power service flows on this sub-unit is suppressed to the boundary state, that is, the cumulative timeout quality approaches zero.

[0095] Furthermore, construct the expression for the total number of successful response loads per unit of fuel consumption:

[0096]

[0097] Expand the expression:

[0098]

[0099] Substituting the aforementioned optimality condition into the equation, and using the definition of expected response time:

[0100]

[0101] Simplifying, we get:

[0102]

[0103] Therefore, the total number of successful response loads per unit of fuel consumption depends only on the expected response time of each power service flow. With expected arrival rate The scheduling decision update interval of each subunit The specific value is irrelevant.

[0104] The analytical conclusions indicate that, under the implementation condition where the optimization objective is solely the probability of load response without timeout and fuel consumption is proportional to the number of scheduling decision updates, there is an upper bound on the total number of successfully responded loads achievable per unit of fuel consumption. This upper bound is jointly determined by the expected response time and expected arrival rate of each power service flow, and cannot be further improved by adjusting the scheduling decision update interval. In other words, under these conditions, it is impossible to further reduce fuel consumption without unequally worsening load reliability, nor is it possible to further improve load reliability without unequally increasing fuel consumption. Based on these analytical conclusions, in scenarios involving a trade-off between fuel consumption and load reliability, a power response time-based flow-level utility function should be used instead of a flow-level utility function based on the number of timeout responses.

[0105] Sub-unit identification and power service flow configuration: The central energy management unit identifies all controlled sub-units and assigns a unique number to each sub-unit. Each subunit configuration Each power service flow is a collection of power scheduling tasks with the same type of response demand and the same priority. In this embodiment, the configuration examples of sub-unit types and power service flows are as follows: the main diesel generator set is configured with a propulsion drive scheduling flow and an auxiliary power supply scheduling flow; the energy storage battery pack is configured with a dynamic frequency support flow, a voltage support flow, and a peak shaving and valley filling scheduling flow; the propulsion load control unit is configured with a propulsion given scheduling flow and a propulsion power reduction response flow; the auxiliary load control unit is configured with an auxiliary start / stop scheduling flow and an auxiliary power adjustment flow; and the critical safety load control unit is configured with a safety margin monitoring flow.

[0106] The specific construction of the flow-level utility function: the flow-level utility function It is positive and concave within the operating range. This embodiment provides four optional specific construction forms, which can be selected by those skilled in the art according to the specific power service flow response requirements, or other forms that satisfy the positive and concave conditions can be selected.

[0107] Response delay type: Response delay is a negative indicator; its negative value is taken as the utility. Let the power scheduling task be generated after a relative time elapsed. Then during the scheduling decision update interval The following flow level utility is:

[0108]

[0109] This utility function is independent of the response time distribution, therefore regarding The integral of is always 1. Its maximum value is at . The utility function, obtained at [location], reflects the asymptotic objective of minimizing response latency. This utility function is applicable to power service flows that are highly sensitive to response latency, such as energy storage dynamic frequency support flows.

[0110] Probabilistic non-timeout response: When the power scheduling task waits for too long, it may exceed the response time limit and cause a decrease in the load safety margin. The utility function represents the probability of a non-timeout response.

[0111]

[0112] Quantity of non-timeout responses: The number of loads that did not time out in a single scheduling decision update can be used as utility, which is equivalent to multiplying the probability of non-timeout responses by the scheduling decision update interval.

[0113]

[0114] The utility function expression is always non-negative, and its second derivative is always non-positive, satisfying the concavity condition.

[0115] Power response timeliness type: To obtain finer resolution compared to binary timeout and non-timeout determination, this embodiment introduces a power response timeliness type utility function, which characterizes the remaining time of the power scheduling task before its response deadline:

[0116]

[0117] in This is the expected value for the response time of the power service flow. When The utility function reaches its maximum value when the scheduling decision update interval is short or long. The utility function provides a trade-off between two extremes: when the scheduling decision update interval is short or long. When the time interval is short, the accumulated power response demand within a single scheduling update is less, resulting in lower aggregation benefits; when Over a longer period, accumulated demand increases, but response timeliness decreases.

[0118] As demonstrated by the analytical performance boundary derivation, under the implementation condition where the only optimization objective is the probability of load response without timeout, there exists an upper bound between fuel consumption and load reliability that cannot be improved by adjusting the scheduling decision update interval. The power response and timeliness-based utility function provides fine resolution at a linear scale, and compared to binarization, it can more precisely characterize the load response margin. Therefore, in scenarios involving trade-offs between fuel consumption and load reliability, this type of utility function should be preferred, enabling the present invention to obtain a compromise solution that is strictly superior to the binarization objective.

[0119] The specific construction and fitting coefficient identification of the unit-level utility function: the unit-level utility function It is a positive and concave function within the working interval. This implementation provides two specific construction forms.

[0120] In implementations that do not consider platform-level coupling constraints, the unit-level utility function is a constant or depends only on the scheduling decision update interval of the sub-unit. The unit-level utility function can be used to increase the relative importance of specific sub-units. For example, setting the unit-level utility function to be related to the energy storage state of charge can give energy storage sub-units with low state of charge higher priority in scheduling optimization.

[0121] Congestion awareness index form: In an implementation considering bus capacity and fuel consumption constraints, the unit-level utility function depends on the scheduling decision update interval vector of all sub-units across the entire platform. Take the exponential form:

[0122]

[0123] in The fitting coefficients are obtained by identifying historical bus occupancy and fuel consumption data using the least squares method. Characterization subunit The number of times a scheduling decision update is initiated per unit time. The unit-level utility function... The first-order partial derivative is:

[0124]

[0125] The second diagonal partial derivative is:

[0126]

[0127] The second-order mixed partial derivative is:

[0128]

[0129] Historical power scheduling task arrival data are collected for each power service flow to estimate the expected arrival rate. With response time probability density The parameters are identified from historical operating data and sent to the local storage of the corresponding sub-control unit.

[0130] To construct reasonable candidates The initial search interval for the value vector is determined first, and the scheduling decision update interval for each sub-unit is also determined. The operating range. When the flow-level utility function is selected as a power response and time-sensitivity type. ,in Subunit Upper The expected value of the response time of a power service flow. Substituting into the equivalent equation, the necessary condition for the existence of a solution can be obtained from the fact that the right side of the equation is always non-positive:

[0131]

[0132] Right now:

[0133]

[0134] The lower bound Subunit The expected response time of all power service flows is the average value weighted by arrival rate, and its physical meaning is the minimum feasible scheduling decision update interval of the sub-unit.

[0135] The right side of the diagonal dominance condition is in all Take the minimum feasible value The maximum value is reached at that time. This is combined with the aforementioned lower bound. We can obtain the lower bound of the solution and the upper bound in the worst case:

[0136]

[0137] Candidate Value vector initialization and broadcasting: Central energy management unit based on the number of sub-units with fitting coefficients And refer to the upper bound to estimate the working upper bound of the global constant. And set a solution accuracy threshold. .

[0138] Central Energy Management Unit Constructs Initial Candidates Value vector:

[0139]

[0140] The central energy management unit transmits the candidate data via the ship's internal communication bus. value vector The message is broadcast to all sub-control units at once.

[0141] Derivation of the local solution equation: The equivalent equation is further rewritten as:

[0142]

[0143] Substituting the exponential congestion-aware utility function, we get:

[0144]

[0145] because For sub-unit numbering Irrelevant global constants, given candidate The equations described after the value are reduced to single-variable equations that depend only on local parameters and can be solved independently by each sub-control unit.

[0146] Bisection method online solution: The local solution equation is in the working interval It only crosses the zero axis once. According to the mean value theorem, the left side of the equations on both sides of the zero point is the candidate... The differences in values ​​have opposite signs. This implementation uses the bisection method for online solution:

[0147] Step 1: Determine the initial working range ,in The upper bound of the preset scheduling decision update interval is determined according to the upper bound; the left side of the equation at both endpoints and the candidate... The sign of the difference in values;

[0148] Step 2, take the midpoint of the interval Calculate the left side of the equation at the midpoint and the candidate The difference in values;

[0149] Step 3: Replace the endpoints with the same sign as the difference at the midpoint with... This results in a new working interval, with the interval length reduced to half of the original length.

[0150] Step four, repeat steps two and three until... The solution accuracy is less than the preset solution accuracy.

[0151] Each sub-control unit for candidates For each candidate value in the value vector, the binary search method is performed to obtain the corresponding local value. The number of iterations in the solution process is:

[0152]

[0153] The online computational complexity of a single sub-unit is The computational complexity of the entire platform online is .

[0154] Calculation and feedback of local response components: Each sub-control unit calculates and transmits the local response components based on the obtained local response components. For candidates For each candidate value in the value vector, calculate the local response component:

[0155]

[0156] Each sub-control unit transmits the local response vector, composed of the local response components, back to the central energy management unit via the ship's internal communication bus.

[0157] The central energy management unit receives the local response vectors returned by all sub-control units and performs selection on candidate units. Each candidate value in the value vector The global response is obtained by summing the results using the following formula:

[0158]

[0159] The central energy management unit calculates the iteration error for each candidate value:

[0160]

[0161] Central energy management unit determines midpoint candidate values Iteration error The symbol will be the candidate value vector and The boundary endpoints on the same side of the symbol are replaced with This reduces the length of the candidate search interval to half of its original length, and the midpoint of the new interval is taken as the new... .

[0162] The central energy management unit compares the minimum absolute value of the current iteration error. Compared with the preset solution accuracy threshold If the absolute value of the minimum iteration error is greater than the solution accuracy threshold, then the updated candidate... The value vector is rebroadcast to each sub-control unit, and the process returns to step S3; if the absolute value of the minimum iteration error is less than or equal to the solution accuracy threshold, convergence is determined, and the candidate corresponding to the minimum iteration error is selected. The value is taken as the final optimal global constant. .

[0163] The number of convergence steps of the distributed solution protocol is , and the number of sub-units Irrelevant This indicates rounding up to the nearest integer.

[0164] The central energy management unit is based on the final optimal global constant. The corresponding local optimal scheduling decision update interval is retrieved from the local solution already obtained by the sub-control unit in step S3. and the The distributed protocol is transmitted to the corresponding sub-control unit via the ship's internal communication bus. The distributed protocol runs once during the initialization phase, and after that, each sub-control unit operates according to its own... The scheduling decision is updated periodically; if the platform's operating conditions change significantly, the protocol is restarted.

[0165] Each sub-control unit updates its interval according to the local optimal scheduling decision. The scheduling decision update action is performed periodically. The scheduling decision update action includes: accumulating all arriving power response demands between two adjacent scheduling decision update times, and merging the accumulated power response demands into a single scheduling instruction and issuing it to the corresponding execution mechanism at the scheduling decision update time.

[0166] Central Energy Management Unit Maintenance Candidate value vector Historical response tables for each subunit. The central energy management unit monitors subunit activation and deactivation events, including but not limited to diesel generator set paralleling / de-parallelization, energy storage module switching, and load tiering activation / de-activation.

[0167] When a new sub-unit is detected to be operational, the central energy management unit issues the current candidate list to the new sub-unit. value vector When it is detected that an existing sub-unit has exited the runtime, the historical response vector of that exited sub-unit is removed from the response table.

[0168] When a new subunit is put into operation, the central energy management unit updates the global error using the following formula:

[0169]

[0170] When a subunit exits operation, the central energy management unit updates the global error using the following formula:

[0171]

[0172] Candidate with the smallest absolute error after update The value is the new optimal global constant. The central energy management unit determines whether the deviation between the new optimal global constant and the original optimal global constant exceeds a preset broadcast threshold; if it exceeds the broadcast threshold, the new local optimal scheduling decision update interval is then applied. The update is sent to all controlled sub-units; if the broadcast threshold is not exceeded, the original optimal scheduling decision update interval remains unchanged to reduce communication overhead. Since both the local and global equations have unique solutions, the incremental update is guaranteed to converge by the mean value theorem.

[0173] Those skilled in the art should understand that the specific utility function form, parameter identification method, and communication bus type given in this embodiment are merely examples. Without departing from the inventive concept, the utility function can be replaced with any other function satisfying the positive and concave conditions; the fitting coefficients... The identification method can be replaced by any system identification method such as maximum likelihood estimation, ridge regression, or online recursive least squares; the communication bus can be replaced by any shipboard communication medium that supports bidirectional communication. All such substitutions fall within the protection scope of this invention.

Claims

1. A scheduling decision aggregation method for offshore hybrid power platforms, characterized in that, The offshore hybrid power platform includes a central energy management unit and several sub-control units corresponding to different sub-units. Each sub-unit includes a power supply unit and a load unit. The method specifically involves: configuring a flow-level utility function for each power service flow on each sub-unit, where the flow-level utility function is positive and concave within the operating range; configuring a unit-level utility function for each sub-unit, where the unit-level utility function characterizes bus capacity occupancy and fuel consumption levels; constructing a total platform utility function based on the flow-level and unit-level utility functions; determining the optimal scheduling decision update interval for each sub-unit that maximizes the total platform utility function; and periodically merging the accumulated power response demand between two adjacent scheduling decision update times into a single scheduling command and issuing it to the corresponding execution mechanism according to their respective optimal scheduling decision update intervals.

2. The scheduling decision aggregation method for a marine hybrid power platform according to claim 1, characterized in that, The overall utility function of the platform is: ； in, Number of sub-units Subunit The number of power service streams on the network. Subunit The scheduling decision update interval, Let be the vector formed by the scheduling decision update intervals of each subunit. For unit-level utility functions, Subunit Upper The flow-level utility function of a power service stream. The expected arrival rate of the power service stream.

3. The scheduling decision aggregation method for a marine hybrid power platform according to claim 1, characterized in that, The specific method for determining the scheduling decision update interval of each sub-unit that maximizes the total utility function of the platform is as follows: based on the first-order optimality condition of the scheduling decision update interval of each sub-unit according to the total utility function of the platform, the bisection method is used to iteratively solve the problem within the preset working interval. In each iteration, the midpoint of the current interval is taken and the boundary endpoint with the same sign as the value at the midpoint is replaced with the midpoint, until the interval length is less than the preset solution accuracy.

4. The scheduling decision aggregation method for a marine hybrid power platform according to claim 1, characterized in that, The unit-level utility function adopts the same exponential form for each sub-unit: ； in, These are the fitting coefficients, which are obtained by systematically identifying historical bus occupancy and historical fuel consumption data. Characterization subunit The number of times a scheduling decision update is initiated per unit of time. This represents the number of sub-units.

5. The scheduling decision aggregation method for a marine hybrid power platform according to claim 1, characterized in that, The optimal scheduling decision update interval for each sub-unit is implemented in a distributed manner, including: The central energy management unit broadcasts candidate global constant vectors to each sub-control unit; For each candidate global constant in the candidate global constant vector, each sub-control unit uses a bisection method to locally solve the local scheduling decision update interval based on the local power service flow parameters, and calculates the local response component based on the obtained local scheduling decision update interval before sending it back to the central energy management unit. The central energy management unit accumulates the response components of the entire platform and calculates the iteration error corresponding to each candidate global constant, updates the candidate global constant vector according to the binary search method, and rebroadcasts it. Repeat the above steps until the iteration error is less than the preset threshold. Then, take the corresponding candidate global constant as the optimal global constant and send the local optimal scheduling decision update interval of each sub-unit corresponding to the optimal global constant to each sub-control unit.

6. The scheduling decision aggregation method for a marine hybrid power platform according to claim 5, characterized in that, The local response component is: ； in, Candidate global constants, Subunit Based on candidate global constants The local scheduling decision update interval obtained from the local solution; Number the sub-units. , The total number of controlled sub-units; Subunit The number of power service streams carried on it. Subunit Upper The expected arrival rate of each power service stream, Subunit Upper The expected flow-level utility function for each power service flow.

7. The scheduling decision aggregation method for a marine hybrid power platform according to claim 1, characterized in that, The method further includes an incremental update step for the dynamic entry or exit of sub-units during operation: when a new sub-unit is entered into operation, the central energy management unit sends the current candidate global constant vector to the new sub-unit, the new sub-unit calculates its own response vector and sends it back, and the central energy management unit incrementally updates the global error corresponding to each candidate global constant based on the full platform response components; when an existing sub-unit exits operation, the central energy management unit removes the response vector of the exited sub-unit from the full platform response components and incrementally updates the global error; the candidate global constant with the smallest absolute value of the updated error is taken as the new optimal global constant.

8. The scheduling decision aggregation method for a marine hybrid power platform according to claim 7, characterized in that, The incremental update step further includes: when the deviation between the new optimal global constant and the original optimal global constant exceeds a preset broadcast threshold, the central energy management unit sends the updated local optimal scheduling decision update interval of each sub-unit to each sub-control unit; when the deviation does not exceed the preset broadcast threshold, the original optimal scheduling decision update interval remains unchanged.

9. The scheduling decision aggregation method for a marine hybrid power platform according to claim 1, characterized in that, The flow level utility function is selected from at least one of the following: response delay type, non-timeout response probability type, non-timeout response quantity type, and power response and timeliness type; wherein the power response and timeliness type flow level utility function is: ； in Let x be the expected response time of the corresponding power service flow, and let x be the scheduling decision update interval.

10. A scheduling decision aggregation system for an offshore hybrid power platform, used to implement the method as described in claim 1, characterized in that, The system includes a central energy management unit, several sub-control units corresponding to different sub-units, and an internal communication bus connecting the central energy management unit and each sub-control unit. Each sub-unit includes a power supply unit and a load unit. Each sub-control unit stores the flow-level utility function and power service flow parameters for each power service flow within its sub-unit. The flow-level utility function is used to locally solve for the local scheduling decision update interval of the sub-unit based on local parameters. The central energy management unit configures unit-level utility functions characterizing bus capacity occupancy and fuel consumption levels. Each sub-control unit also periodically merges the accumulated power response demand between two adjacent scheduling decision update times into a single scheduling command and issues it to the corresponding actuator according to its optimal scheduling decision update interval. The central energy management unit determines the optimal scheduling decision update interval for each sub-unit in a distributed manner by broadcasting candidate global constant vectors to each sub-control unit, receiving local response components returned by each sub-control unit, accumulating the full platform response components, and iteratively updating the candidate global constant vectors using a binary search method.

Images