Hybrid energy scheduling method, electronic device, and computer program product

Abstract

A hybrid energy scheduling method, an electronic device, and a computer program product. The method comprises: acquiring a new energy generation power hyperparameter and an electrical load power hyperparameter; respectively dividing a new energy generation power prediction layer and an electrical load power prediction layer into L-dimensional clustering subspaces, and respectively performing iterative clustering on the basis of the new energy generation power hyperparameter and the electrical load power hyperparameter to acquire a new energy generation power sequence set and an electrical load power sequence set which both have a sequence length of L; acquiring a joint probability scenario set on the basis of the new energy generation power sequence set and the electrical load power sequence set; and optimally solving a preset electricity consumption indicator on the basis of the joint probability scenario set, so as to acquire a scheduling strategy for hybrid energy scheduling.

Description

Hybrid energy scheduling method, electronic device and computer program product

[0001] Cross-reference to Related Applications

[0002] The present disclosure is based on and claims priority from Chinese patent application 2024108305675 filed on June 25, 2024, the disclosure of which is incorporated herein in its entirety by reference. TECHNICAL FIELD

[0003] Embodiments of the present disclosure relate to the field of hybrid energy scheduling, in particular, to a hybrid energy scheduling method, an electronic device and a computer program product. BACKGROUND

[0004] With the continuous transformation of global energy structure, diversified energy collaborative participation in power supply has become the main trend of future development. The collaborative participation of multiple energies in power supply not only improves the reliability and stability of energy supply, but also promotes the efficiency and sustainability of energy utilization. With the rapid development of energy storage technology and the vigorous development of new energy power supply applications, traditional diesel engines and lead-acid batteries are gradually being replaced by high-energy-density energy storage lithium batteries / sodium batteries. At the same time, clean energy such as solar and wind energy is also playing an increasingly important role in the energy supply system. Promoting the development of energy supply systems with safe and efficient participation of multiple energies means that how to manage energy, link and schedule hybrid energy power supply, and ensure the stable and reliable operation of the energy system is a problem to be solved. Therefore, hybrid energy scheduling has emerged, which is based on new energy generation prediction and load prediction, to find the best execution strategy that meets business safety, power safety, and energy storage safety constraints, so as to achieve the best economy or maximum utilization rate in the entire feasible solution set.

[0005] However, new energy generation is affected by environmental factors such as weather conditions and equipment status, and load changes are also affected by complex factors such as power demand and time distribution, so both new energy generation and load changes show obvious volatility and certain distribution dispersion; the linked scheduling of hybrid energy scheduling is a high-dimensional mathematical problem of solving uncertainty involving multiple energy participation and fine-grained strategy generation under the guarantee of safe and stable operation. Designing a solution requires a large amount of computing power to ensure the feasibility and effectiveness of the scheduling plan, and how to balance and optimize the scheduling decision and computing power consumption is a problem to be solved. Therefore, in the energy supply system with hybrid energy participation in power supply, on the one hand, the volatility of new energy output and load power needs to be considered comprehensively, and on the other hand, the balance between hybrid energy scheduling decision and computing power consumption needs to be balanced to achieve optimal scheduling decision of hybrid energy.

[0006] In the related art, no optimal scheduling scheme of mixed energy can well implement a "prediction-scheduling-adaptive" full closed-loop scheme, meet actual engineering landing scenarios, and improve deployment feasibility. SUMMARY

[0007] Embodiments of the present disclosure provide a mixed energy scheduling method, an electronic device, and a computer program product to at least solve the problem that mixed energy scheduling prediction and adaptive adjustment of scheduling strategies cannot be performed in the related art.

[0008] In one aspect of the embodiments of the present disclosure, a mixed energy scheduling method is provided, including: obtaining new energy power generation power super parameters and power load power super parameters; dividing a new energy power generation power prediction layer into an L-dimensional first clustering subspace, iteratively clustering the first clustering subspace based on the new energy power generation power super parameters, and obtaining a new energy power generation power sequence set with a sequence length of L; dividing a power load power prediction layer into an L-dimensional second clustering subspace, iteratively clustering the second clustering subspace based on the power load power super parameters, and obtaining a power load power sequence set with a sequence length of L; obtaining a joint probability scenario set according to the new energy power generation power sequence set and the power load power sequence set; and based on the joint probability scenario set, performing optimal solving of a preset power consumption index to obtain a scheduling strategy of mixed energy scheduling.

[0009] In another aspect of the embodiments of the present disclosure, a computer readable storage medium is also provided, and the computer readable storage medium stores a computer program. When the computer program is executed, the steps in any of the method embodiments described above are performed.

[0010] In yet another aspect of the embodiments of the present disclosure, an electronic device is also provided, including a memory and a processor. The memory stores a computer program, and the processor is configured to execute the computer program to perform the steps in any of the method embodiments described above.

[0011] In still another aspect of the embodiments of the present disclosure, a computer program product is also provided, including a computer program. When the computer program is executed by a processor, the steps in any of the method embodiments described above are implemented. BRIEF DESCRIPTION OF DRAWINGS

[0012] FIG. 1 is a hardware structure block diagram of a computer terminal of a mixed energy scheduling method according to an embodiment of the present disclosure;

[0013] FIG. 2 is a flowchart of a mixed energy scheduling method according to an embodiment of the present disclosure;

[0014] FIG. 3 is a structure block diagram of a mixed energy scheduling system according to an embodiment of the present disclosure;

[0015] FIG. 4 is a principle flowchart of a mixed energy scheduling method according to an embodiment of the present disclosure. Detailed Implementation

[0016] The embodiments of this disclosure will be described in detail below with reference to the accompanying drawings and examples.

[0017] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of the embodiments of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0018] In related technologies, considerations are given to the volatility of renewable energy output and electricity load: the generation of volatility involves the influence of multiple variables and factors. Currently, the main approach to optimizing this problem in the industry is to introduce uncertainty modeling, using the prediction results of renewable energy generation and load fluctuations as input into the model. The aim is to improve the accuracy of the prediction model to cope with the uncertainty of the prediction results; however, this approach of simply improving prediction accuracy cannot quantify the impact of the randomness of renewable energy generation and load changes on dispatch safety and economy, and has a fundamental methodological deficiency.

[0019] In related technologies, considerations regarding scheduling decisions and computational overhead are crucial. Hybrid energy scheduling is an optimization problem involving multiple energy sources and constraints, inherently containing uncertainty. Currently, operations research primarily uses stochastic programming or robust programming methods to solve optimization problems with uncertainty. Essentially, this involves jointly sampling a set of scenarios in a high-dimensional space to address uncertain random variables, transforming the optimization objective and constraints from the global domain to optimality and constraint satisfaction within the scenario set. However, in the field of hybrid energy scheduling, there are numerous variables with uncertainty. For example, with a 15-minute prediction granularity, there are 96 time points in a day where the predicted values ​​for renewable energy generation and load are all random variables, resulting in a very high dimensionality. Traditional sampling methods (such as Latin hypercube sampling) struggle to balance real-time performance and computational overhead when constructing scenario sets for hybrid energy scheduling. Furthermore, practical engineering applications have limited computational resources, thus posing significant challenges to the feasibility of actual deployment. In summary, the optimal scheduling of hybrid energy not only needs to consider the uncertainties of new energy power generation and load power to obtain more robust scheduling and control commands, but also needs to comprehensively consider the balance between scheduling decisions and computing power overhead to achieve a closed-loop solution of "prediction-scheduling-adaptation" to meet the actual engineering implementation scenarios and improve deployment feasibility.

[0020] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or similar computing device. Taking the case of running on a computer terminal, FIG. 1 is a hardware structure block diagram of a computer terminal of a hybrid energy scheduling method according to an embodiment of the present disclosure. As shown in FIG. 1, the computer terminal can include one or more (only one is shown in FIG. 1) processors 102 (the processor 102 can include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the computer terminal can further include a transmission device 106 for communication function and an input and output device 108. Those skilled in the art can understand that the structure shown in FIG. 1 is only schematic and does not limit the structure of the computer terminal. For example, the computer terminal can include more or fewer components than those shown in FIG. 1, or have a different configuration from that shown in FIG. 1.

[0021] The memory 104 can be used to store computer programs, for example, software programs of application software and modules, such as a computer program corresponding to the hybrid energy scheduling method in the embodiments of the present disclosure. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, that is, implements the above-mentioned method. The memory 104 can include a high-speed random access memory and can further include a non-volatile memory such as one or more magnetic storage devices, flash memories or other non-volatile solid state memories. In some examples, the memory 104 can further include a memory remotely arranged with respect to the processor 102, which can be connected to the mobile terminal through a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0022] The transmission device 106 is used to receive or send data via a network. Specific examples of the above-mentioned network can include a wireless network provided by a communication provider of the computer terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (Radio Frequency, RF) module, which is used to communicate with the Internet in a wireless manner.

[0023] The embodiments of the present disclosure provide a hybrid energy scheduling method, and FIG. 2 is a flow chart of the hybrid energy scheduling method according to an embodiment of the present disclosure. As shown in FIG. 2, the flow includes the following steps:

[0024] In step S202, new energy power generation super parameters and power load power super parameters are obtained.

[0025] In an example embodiment, the new energy power generation parameter and the power load parameter are obtained by adjusting the hybrid energy scheduling parameter through the super parameter optimization value network.

[0026] In the embodiments of the present disclosure, the super parameter is adjusted adaptively through the super parameter optimization value network. The key super parameters in the tth scheduling are adaptively corrected according to the feedback learning of the t-1th scheduling execution effect, and the key super parameters at least include the new energy power generation parameter m and the power load parameter n in the embodiments of the present disclosure. Wherein, t, m and n are positive integers.

[0027] In an example embodiment, the super parameter optimization value network is adjusted through the advantage ratio parameter, wherein the advantage ratio parameter is the difference between the first ratio and the second ratio, the first ratio is the ratio of the actual income of the hybrid energy scheduling at any moment to the actual operation time at the moment, the second ratio is the ratio of the income of the scene set at the moment to the operation time of the scene set at the moment, and the scene set is constructed based on the actual power load and the new energy power generation.

[0028] In the embodiments of the present disclosure, the super parameter optimization value network is adjusted through the advantage ratio parameter AdvantageRatio(t-1), wherein the advantage ratio parameter is the difference between the first ratio ActualValue(t-1) / ActualCost(t-1) and the second ratio BaseValue(t-1) / BaseCost(t-1), the first ratio ActualValue(t-1) / ActualCost(t-1) is the ratio of the actual income ActualValue(t-1) of the hybrid energy scheduling at any moment to the actual operation time ActualCost(t-1) at the moment, the second ratio BaseValue(t-1) / BaseCost(t-1) is the ratio of the income BaseValue(t-1) of the scene set at the moment to the operation time BaseCost(t-1) of the scene set at the moment, and the scene set is constructed based on the actual power load and the new energy power generation.

[0029] In step S204, the new energy power generation prediction layer is divided into a first clustering subspace of L dimensions, the first clustering subspace is iteratively clustered based on the new energy power generation parameter, and a new energy power generation sequence set with a sequence length of L is obtained; and the power load prediction layer is divided into a second clustering subspace of L dimensions, the second clustering subspace is iteratively clustered based on the power load parameter, and a power load sequence set with a sequence length of L is obtained.

[0030] In the embodiments of the present disclosure, the new energy power generation power prediction layer and the electricity load power prediction layer are respectively divided into L-dimensional clustering subspaces, and iterative clustering is performed based on the new energy power generation power hyperparameters and the electricity load power hyperparameters to obtain a new energy power generation power sequence set and an electricity load power sequence set with a sequence length of L.

[0031] In one example embodiment, each first clustering subspace corresponds to different predicted values of new energy power generation power at a time point and probability distribution values corresponding to the different predicted values; and each second clustering subspace corresponds to different predicted values of electricity load power at a time point and probability distribution values corresponding to the different predicted values.

[0032] In the embodiments of the present disclosure, different dimensions correspond to different time points, and at the same time point, i.e., the same dimension, the first clustering subspaces respectively correspond to predicted values of new energy power generation power, and each predicted value corresponds to a probability distribution value. The second clustering subspaces of the same dimension respectively correspond to predicted values of electricity load power, and each predicted value corresponds to a probability distribution value.

[0033] In one example embodiment, the first clustering subspaces are iteratively clustered based on new energy power generation power hyperparameters to obtain a new energy power generation power sequence set with a sequence length of L, including: iteratively clustering the first clustering subspaces of the first dimension to the Lth dimension in turn to obtain the new energy power generation power sequence set with the sequence length of L.

[0034] In the embodiments of the present disclosure, the new energy power generation power hyperparameters are represented as m, and the predicted values of all new energy power generation power corresponding to the first clustering subspaces are iteratively clustered to obtain m clustering centroids. In the embodiments of the present disclosure, the probability distributions corresponding to the predicted values of new energy power generation power corresponding to the first clustering subspaces are normalized, and the normalization method is to assign the probability values of other centroids except the m clustering centroids to the clustering centroids corresponding to the m predicted values by Euclidean distance or sequence edit distance.

[0035] In the embodiments of the present disclosure, in the process of probability normalization of iterative dimension increasing clustering, sequences in smaller clusters are assigned to larger clusters, and the probability percentage of each cluster is redistributed. By performing the above operation of redistributing the probability, the problem that the probabilities of various clusters are very small after the dimension is increased to a certain degree, resulting in unstable final clustering effect, is avoided.

[0036] In an example embodiment, the first clustering subspace of the first dimension to the Lth dimension is sequentially iteratively clustered respectively to obtain a new energy power sequence set with a sequence length of L, including: clustering the first clustering subspace of the first dimension, and performing probability normalization on the predicted value of the new energy power corresponding to the first dimension to obtain m predicted values of the new energy power corresponding to the first dimension as a first new energy power sequence set, where m is equal to the value of the new energy power hyperparameter, and m is a positive integer; cross-combining the first new energy power sequence set and m predicted values of the corresponding new energy power of the second dimension, and performing probability normalization on the predicted value of the corresponding new energy power of the second dimension to obtain m predicted value sequences of the corresponding new energy power of the second dimension as a second new energy power sequence set; and repeatedly cross-combining m predicted values of the corresponding new energy power of the third dimension to the Lth dimension in sequence until a new energy power sequence set with a sequence length of L is obtained.

[0037] In the embodiment of the present disclosure, the first clustering subspace of the first dimension is iteratively clustered first, and then probability normalization is performed, and the sum of the probability values of the m clustering centroids obtained thereafter is 100%. Thereafter, the m predicted values of the corresponding new energy power of the second dimension are cross-combined with the iterative clustering result of the first dimension, and then probability normalization is performed on the predicted value of the corresponding new energy power of the second dimension, where the probability normalization manner is the same as above, and then m predicted value sequences of the corresponding new energy power of the second dimension are obtained. The process is repeatedly cycled thereafter until a new energy power sequence set with a sequence length of L is obtained.

[0038] In an example embodiment, the second clustering subspace is iteratively clustered based on the electricity load power hyperparameter to obtain an electricity load power sequence set with a sequence length of L, including: iteratively clustering the second clustering subspace of the first dimension to the Lth dimension respectively to obtain an electricity load power sequence set with a sequence length of L.

[0039] In the embodiment of the present disclosure, the electricity load power hyperparameter is represented as n, the predicted values of all electricity load powers corresponding to the first clustering subspace are iteratively clustered to obtain n clustering centroids, and in the embodiment of the present disclosure, the probability distribution corresponding to the predicted value of the electricity load power corresponding to the first clustering subspace is probability normalized, where the probability normalization manner is to assign the probability values of the centroids other than the n clustering centroids to the n predicted values corresponding to the clustering centroids through Euclidean distance or sequence edit distance.

[0040] In an example embodiment, the second clustering subspaces of the first dimension to the Lth dimension are sequentially iteratively clustered respectively to obtain the power consumption load power sequence set with the sequence length of L, including: clustering the second clustering subspace of the first dimension, and performing probability normalization on the predicted value of the power consumption load power corresponding to the first dimension to obtain n predicted values of the power consumption load power corresponding to the first dimension as the first power consumption load power sequence set, wherein n is equal to the value of the power consumption load power hyperparameter, and n is a positive integer; cross-combining the first power consumption load power sequence set and n predicted values of the power consumption load power corresponding to the second dimension, and performing probability normalization on the predicted value of the power consumption load power corresponding to the second dimension to obtain n predicted value sequences of the power consumption load power corresponding to the second dimension as the second power consumption load power sequence set; and repeatedly cross-combining n predicted values of the power consumption load power corresponding to the third dimension to the Lth dimension in sequence until the power consumption load power sequence set with the sequence length of L is obtained.

[0041] In the embodiment of the present disclosure, the second clustering subspace of the first dimension is iteratively clustered first, and then probability normalization is performed, and the sum of the probability values of the n clustering centroids obtained thereafter is 100%. Thereafter, the n predicted values of the power consumption load power corresponding to the second dimension are cross-combined with the iterative clustering result of the first dimension, and then probability normalization is performed on the predicted value of the power consumption load power corresponding to the second dimension, wherein the probability normalization manner is the same as above, and n predicted value sequences of the power consumption load power corresponding to the second dimension are further obtained. The process is repeatedly cycled thereafter until the power consumption load power sequence set with the sequence length of L is obtained.

[0042] In an example embodiment, the probability normalization on the predicted value of the new energy power generation power or the power consumption load power includes: distributing the probability values of the clustering centroids other than the clustering centroids corresponding to the m predicted values of the new energy power generation power to the clustering centroids corresponding to the m predicted values according to the Euclidean distance or the sequence edit distance; and distributing the probability values of the clustering centroids other than the clustering centroids corresponding to the n predicted values of the power consumption load power to the clustering centroids corresponding to the n predicted values according to the Euclidean distance or the sequence edit distance.

[0043] In step S206, the joint probability scenario set is obtained according to the new energy power generation power sequence set and the power consumption load power sequence set.

[0044] In an example embodiment, the joint probability scenario set is obtained according to the new energy power generation power sequence set and the power consumption load power sequence set, including: cross-combining the new energy power generation power sequence set and the power consumption load power sequence set to obtain the joint probability scenario set.

[0045] In the embodiment of the present disclosure, all possible scenarios are obtained by the cross method, and the corresponding probability distribution is obtained, and a joint probability maximum probability scenario set S is obtained.

[0046] In step S208, based on the joint probability scenario set, the optimal solution of the preset power consumption index is performed to obtain the scheduling strategy of the hybrid energy scheduling.

[0047] In the embodiment of the present disclosure, the state of charge (SOC) of the energy storage cell is taken as the optimal solution target; and the following two energy storage operations are not performed between any adjacent decision points unless the power grid is powered off: charging first and discharging later, or discharging first and charging later. The solution space of the scheduling strategy is searched by using an intelligent optimization algorithm, and the most suitable control strategy is selected: the control strategy can meet the standby power safety constraint and the energy storage safety constraint under all scenarios in the scenario set S, and can achieve the optimal economic index or the maximum consumption index under the probability weighting of all scenarios in the scenario set S.

[0048] In the embodiment of the present disclosure, the constraints of the optimal solution of the preset power consumption index can include the standby power safety constraint and the energy storage safety constraint. In the actual implementation process, other types of constraints can also be established according to the actual situation, which is not limited here.

[0049] In one example embodiment, based on the joint probability scenario set, the standby power safety constraint and / or the energy storage safety constraint are taken as the constraint function, the optimal solution of the preset power consumption index is performed, and the state of charge (SOC) of the energy storage cell at different times is obtained as the scheduling strategy of the hybrid energy scheduling.

[0050] Through the embodiment of the present disclosure, a hybrid energy scheduling method is provided. The new energy power generation power super parameter and the power consumption load power super parameter are obtained; the new energy power generation power prediction layer is divided into an L-dimensional first clustering subspace, the first clustering subspace is iteratively clustered based on the new energy power generation power super parameter, and a new energy power generation power sequence set with a sequence length of L is obtained; the power consumption load power prediction layer is divided into an L-dimensional second clustering subspace, the second clustering subspace is iteratively clustered based on the power consumption load power super parameter, and a power consumption load power sequence set with a sequence length of L is obtained; the joint probability scenario set is obtained according to the new energy power generation power sequence set and the power consumption load power sequence set; the optimal solution of the preset power consumption index is performed based on the joint probability scenario set, and the scheduling strategy of the hybrid energy scheduling is obtained. The problem that the hybrid energy scheduling prediction and adaptive adjustment of the scheduling strategy cannot be performed in the related art is solved, the hybrid energy scheduling prediction and adaptive adjustment of the scheduling strategy are achieved, and the effect of improving the feasibility of the hybrid energy scheduling is achieved.

[0051] Those skilled in the art can clearly understand, through the description of the foregoing embodiments, that the method according to the foregoing embodiments can be implemented by means of software on a general hardware platform as necessary, and of course can also be implemented by hardware, but in many cases the former is a better implementation. Based on such understanding, the technical solutions of the present disclosure can be embodied in the form of a software product in essence or in the form of a part that contributes to the prior art, which is stored in a storage medium (such as a ROM / RAM, a magnetic disk, or an optical disk) and includes a number of instructions for causing an end device (which can be a mobile phone, a computer, a server, or a network device) to perform the method described in each embodiment of the present disclosure.

[0052] In the embodiments of the present disclosure, a hybrid energy scheduling device is also provided, which is used to implement the above-mentioned embodiments and preferred embodiments, and will not be described again. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, implementation of hardware or a combination of software and hardware is also possible and contemplated.

[0053] The hybrid energy scheduling device provided by the embodiments of the present disclosure can include: an acquisition module configured to acquire new energy power generation super parameters and power load power super parameters. A clustering module is configured to divide a new energy power generation prediction layer into an L-dimensional first clustering subspace, iteratively cluster the first clustering subspace based on the new energy power generation super parameters, and obtain a new energy power generation sequence set with a sequence length of L; divide a power load power prediction layer into an L-dimensional second clustering subspace, iteratively cluster the second clustering subspace based on the power load power super parameters, and obtain a power load power sequence set with a sequence length of L. A combination module is configured to obtain a joint probability scenario set according to the new energy power generation sequence set and the power load power sequence set. A scheduling module is configured to perform optimal solving of a preset power consumption index based on the joint probability scenario set, and obtain a scheduling strategy of hybrid energy scheduling.

[0054] In actual implementation, the naming and function division of the modules of the hybrid energy scheduling device described above are not specifically limited, as long as the steps in the hybrid energy scheduling method embodiments described above can be implemented.

[0055] It should be noted that each of the above modules can be implemented by software or hardware. For the latter, the following implementation methods can be used, but are not limited thereto: all the above modules are located in the same processor; or the above modules are located in different processors in any combination.

[0056] The embodiment of the present disclosure further provides a computer readable storage medium, which stores a computer program, and the computer program is configured to execute the steps in any of the method embodiments.

[0057] In an example embodiment, the computer readable storage medium can include, but is not limited to, a U disk, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store computer programs.

[0058] The embodiment of the present disclosure further provides an electronic device, which comprises a memory and a processor, the memory stores a computer program, and the processor is configured to execute the computer program to perform the steps in any of the method embodiments.

[0059] In an example embodiment, the electronic device can further comprise a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0060] The embodiment of the present disclosure further provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the steps in any of the method embodiments.

[0061] The embodiment of the present disclosure provides a computer program product, which comprises a non-volatile computer readable storage medium, the non-volatile computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps in the method of various embodiments of the present disclosure.

[0062] The specific examples in the embodiment can refer to the examples described in the above embodiments and example embodiments, and the embodiment will not be described here.

[0063] Obviously, those skilled in the art should understand that the modules or steps of the present disclosure described above can be realized by general computing devices, which can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and they can be realized by program codes executable by computing devices, so that they can be stored in storage devices and executed by computing devices, and in some cases, the steps shown or described can be executed in different order, or they can be manufactured into individual integrated circuit modules, or multiple modules or steps can be manufactured into a single integrated circuit module. Therefore, the present disclosure is not limited to any specific hardware and software combination.

[0064] In order for those skilled in the art to better understand the technical solutions of the embodiments of the present disclosure, the following is described in conjunction with different embodiments.

[0065] Embodiment one

[0066] FIG. 3 is a structural block diagram of a hybrid energy scheduling system according to an embodiment of the present disclosure, as shown in FIG. 3, including a new energy power generation predictor, a load predictor, an adaptive hyperparameter adjuster, a scenario set generator, and a robust solver. The hybrid energy scheduling system provided by the embodiment of the present disclosure is applied in association with a power supply control system including new energy power generation equipment and energy storage equipment.

[0067] FIG. 4 is a principle flow chart of a hybrid energy scheduling method according to an embodiment of the present disclosure, as shown in FIG. 4, the hybrid energy scheduling method is implemented based on the hybrid energy scheduling system described above, and includes the following steps:

[0068] Step S401, adaptive hyperparameter adjustment is performed by the hyperparameter tuning value network.

[0069] According to the feedback learning of the t-1th scheduling execution effect, the key hyperparameters in the tth scheduling are adaptively corrected. In the embodiment of the present disclosure, the key hyperparameters at least include a new energy power generation power hyperparameter m and a power load power hyperparameter n.

[0070] In the embodiment of the present disclosure, the hyperparameter tuning value network is used to adjust the hybrid energy scheduling hyperparameters to obtain the new energy power generation power hyperparameters and the power load power hyperparameters. Before the hyperparameter adjustment, the hyperparameter tuning value network is designed and adjusted.

[0071] By defining the state variables, the state of the hyperparameter tuning value network is designed. In the embodiment of the present disclosure, the state variable definition is shown in Table 1.

[0072] Table 1: Example of state variable definition table

[0073] In the embodiment of the present disclosure, the action of the hyperparameter tuning value network is defined and designed as shown in Table 2.

[0074] Table 2: Example of Action definition table

[0075] In the embodiment of the present disclosure, the definition of the environment feedback Reward of the hyperparameter tuning value network is shown in Table 3.

[0076] Table 3: Example of environment feedback Reward definition table

[0077] In the embodiments of the present disclosure, as shown in Table 3, the hyperparameter tuning value network is feedback adjusted by an advantage ratio parameter AdvantageRatio(t-1), wherein the advantage ratio parameter is the difference between a first ratio ActualValue(t-1) / ActualCost(t-1) and a second ratio BaseValue(t-1) / BaseCost(t-1), the first ratio ActualValue(t-1) / ActualCost(t-1) is the ratio of the actual revenue ActualValue(t-1) of the hybrid energy scheduling at any moment to the actual operation time consumption ActualCost(t-1) at the moment, the second ratio BaseValue(t-1) / BaseCost(t-1) is the ratio of the revenue BaseValue(t-1) of the scenario set at the moment to the operation time consumption BaseCost(t-1) of the scenario set at the moment, and the scenario set is constructed based on the actual power load and the new energy power generation.

[0078] According to the definition of the above state variable, action and environment feedback Reward, adaptive feedback learning is performed at the t-th moment. The process is as follows:

[0079] A value learning sample valuerow(t) is constructed, the independent variable of which is [State(t-1),Action(t-1)], and the dependent variable of which is Reward(t-1), and the hyperparameter tuning value network sp_value_nn is updated based on the value learning sample valuerow(t) by using a stochastic gradient descent (SGD) algorithm.

[0080] For the state variable State(t), all selectable actions are traversed to construct an inference sample input sp_value_nn.

[0081] The action a* that returns the maximum Reward(a*) is selected, and the key hyperparameter items of the t-th hybrid energy scheduling, i.e., the new energy power generation hyperparameter m and the power load hyperparameter n, are set.

[0082] In step S402, scenario clustering is performed to obtain a joint probability scenario set.

[0083] The scenario clustering is performed by the scenario generator in the hybrid energy scheduling system to obtain the joint probability scenario set S=f1(α,β,m,n).

[0084] Wherein, the new energy power generation power prediction layer outputs a probability distribution sequence α of length L, that is, a new energy power generation power sequence set α; the power consumption load power prediction layer outputs a probability distribution sequence β of length L, that is, a power consumption load power sequence set β; through the step-by-step iterative clustering scene construction method, first, the maximum joint probability new energy power generation power sequence A containing m pieces and the maximum joint probability power consumption load power sequence B containing n pieces are extracted from α and β respectively, and the scene set S containing m*n maximum joint probabilities in the future K time periods is obtained through cross combination, that is, the joint probability scene set S.

[0085] In the embodiments of the present disclosure, the joint probability scene set S is constructed based on the method of clustering as the dimension increases:

[0086] (1) In the k-1 dimensional subspace, d cluster centers and their probability distributions are obtained through clustering convergence;

[0087] In the embodiments of the present disclosure, the new energy power generation power prediction layer is divided into a first clustering subspace of L dimensions, the first clustering subspace is iteratively clustered based on new energy power generation power hyperparameters, and a new energy power generation power sequence set with a sequence length of L is obtained.

[0088] Through clustering of the first clustering subspace of the first dimension, m predicted values of the new energy power generation power corresponding to the first dimension are obtained as a first new energy power generation power sequence set. Wherein, the first dimension corresponds to the k-1 dimension above, and d corresponds to m.

[0089] (2) Cross generation of d*V(k) k-1 dimensional subspace centers in the extension set U(k) of the k dimension (assuming that the k dimension has V(k) density peak values);

[0090] In the embodiments of the present disclosure, the first new energy power generation power sequence set and the m predicted values of the new energy power generation power corresponding to the second dimension are cross combined, m predicted value sequences of the new energy power generation power corresponding to the second dimension are obtained as a second new energy power generation power sequence set.

[0091] Wherein, d is the d cluster centers of the first new energy power generation power sequence set, the second dimension is the k dimension, the k dimension has V(k) density peak values, and d*V(k) is the m predicted value sequences of the new energy power generation power corresponding to the second dimension obtained through cross combination, as the number of spatial centers of the second new energy power generation power sequence set U(k).

[0092] (3) Selecting the d centers with the maximum joint distribution probability in the above extension set U(k) as the cluster representatives of the k dimensional subspace;

[0093] (4) Through the Euclidean distance or sequence edit distance, other new centers are merged into the cluster where the first d new centers are located.

[0094] (5) Calculate the normalized probability distribution of each cluster in the k-th dimensional subspace, jump to step (1) until the completion of the entire L-dimensional expansion.

[0095] In the embodiments of the present disclosure, the predicted value of the corresponding new energy power of the second dimension is normalized in probability, that is, an expansion set U(k) is obtained, the d centroids with the maximum joint distribution probability are taken as the centroids in the final second power consumption load sequence set, and the probability values corresponding to the other centroids except the d centroids are distributed to the d centroids through Euclidean distance or sequence edit distance to complete the probability normalization.

[0096] Then, the n predicted values of the corresponding power consumption load of the third dimension to the L-th dimension are repeatedly cross combined in sequence until the power consumption load sequence set with a sequence length of L is obtained.

[0097] It should be noted that in the process of constructing the joint probability scenario set S based on the method of edge dimension increasing and clustering, the predicted value of the new energy power is introduced as edge dimension increasing and clustering, so d corresponds to m and is equal. In the actual implementation process, when the power consumption load is edge dimension increasing and clustering, d corresponds to n and is equal.

[0098] Therefore, each time the dimension increasing and converging operation only needs to be operated on the d*|V(k)| expanded centroid set, and the total operation time complexity is O(d*|V(k)|*L), which greatly reduces the operation pressure; otherwise, based on the ordinary clustering of first dimension increasing, sampling, and then clustering, the operation time complexity is O(d*V(k)^L).

[0099] Next, the process of constructing the joint probability scenario set S based on the above edge dimension increasing and clustering method will be introduced in combination with specific formula process.

[0100] (1) Denote the new energy power sequence set α and the power consumption load sequence set β with a length of L, respectively.

[0101] wherein, wherein v i represents the predicted value of the i-th time point, p i represents the sequence value probability of the i-th time point, represents the j-th possible value of the predicted value of the i-th time point, represents the value probability of the j-th possible value of the predicted value of the i-th time point, wherein i=1,...,L.

[0102] Taking the new energy power sequence as an example, denote the i-th sequence clustering set as S iWhen i = 1, S1 is the value and probability distribution of the first prediction point, i.e.

[0103] a. The first sequence cluster set S1 and the value and probability distribution of the next prediction point i+1th prediction point are clustered to generate all possible sequences S2 of length 2, i.e.

[0104]

[0105] For the selected joint probability topm sequences in S2 (i.e. the first m probability values arranged in descending order of probability value), a joint sampling sequence set is obtained, and for the unselected sequences, their probabilities are assigned to the sequences in the joint sampling sequence set by calculating the Euclidean distance (or sequence edit distance) to achieve probability normalization, updating S2, i.e.

[0106] b. The second sequence cluster set S2 and the value and probability distribution of the next prediction point i+1th prediction point are clustered to generate all possible sequences S3 of length 3, i.e.

[0107] For the selected joint probability topm sequences in S3, a joint sampling sequence set is obtained, and for the unselected sequences, their probabilities are assigned to the sequences in the joint sampling sequence set by calculating the Euclidean distance (or sequence edit distance) to achieve probability normalization, updating S3, i.e.

[0108] c. Repeat the above steps layer by layer until i+1 = L, to generate a new energy power sequence set A of length L containing the maximum joint probability topm, i.e.

[0109] Similarly, the power consumption load sequence set B containing the maximum joint probability topn (i.e. the first n probability values arranged in descending order of probability value) can be obtained, i.e.

[0110]

[0111] d. Through the crossover method, all possible scenarios and their corresponding probability distributions are obtained, and the joint probability maximum probability scenario set S is obtained, i.e.

[0112] Step S403, based on the joint probability scenario set, robustly solve the optimization target.

[0113] ​​In the embodiments of the present disclosure, based on the joint probability scenario set, optimal solution of preset power consumption indicators is performed to obtain a scheduling strategy of hybrid energy scheduling.

[0114] The scheduling strategy θ is defined as a sequence with a length of L, each element of the sequence represents a State of Charge (SOC) target of a corresponding decision time point; and the following two energy storage operations are not performed between any adjacent decision points unless the power grid is powered off: charging first and discharging later, or discharging first and charging later.

[0115] The solution space of the scheduling strategy θ is searched by using an intelligent optimization algorithm, and the most suitable control strategy is selected: the control strategy can meet the standby power safety μ and the energy storage safety constraint under all scenarios of the scenario set S , and can achieve the optimal probability-weighted economic indicator or the maximum consumption indicator under all scenarios of the scenario set S.

[0116] Embodiment Two

[0117] In the present embodiment, under the hybrid energy scenario (the hybrid energy mentioned in the embodiment includes the power grid, photovoltaic and energy storage), the economic scheduling with the optimal power consumption cost is taken as the scheduling strategy.

[0118] In the present embodiment, the 100th hybrid energy scheduling and scheduling-prediction dynamic perception feedback optimization are performed.

[0119] (1) Adaptive hyperparameter adjustment is performed.

[0120] In the i=99th prediction-scheduling decision, the hyperparameters are m=2 and n=3, adaptive feedback adjustment is performed based on the 99th prediction-scheduling result, and the scenario set containing m*n=4 joint probability maximums generated in the 99th time is recorded as S(99), that is:

[0121] S(99)={ Scene1:{Solar1:(0,0,...,350,300,...,0,0),Load1:(200,200...,300,320,...,200)},

[0122] Scene2:{Solar1:(0,0,...,350,300,...,0,0),Load2:(200,220...,300,320,...,200)},

[0123] Scene3:{Solar1:(0,0,...,350,300,...,0,0),Load3:(200,220...,300,320,...,200)},

[0124] Scene4:{Solar1:(0,0,...,350,300,...,0,0),Load4:(200,220...,300,320,...,200)}}.

[0125] Scenario 3: {Solar2: (0, 0,..., 350, 300,..., 0, 0), Load1: (200, 200,..., 320, 320,..., 200)}

[0126] Scenario 4: {Solar2: (0, 0,..., 350, 300,..., 0, 0), Load2: (200, 200,..., 320, 320,..., 200)}

[0127] Probability: {Scenario 1: 0.1, Scenario 2: 0.7,

[0128] Scenario 3: 0.1, Scenario 4: 0.1}

[0129] }

[0130] }

[0131] A value sample valuerow(100) is constructed, with independent variables of [State(99), Action(99)] and dependent variable of Reward(99). Based on valuerow(100), the hyperparameter tuning value network sp_value_nn is updated using the SGD method.

[0132] For different Actions, inference sample inputs sp_value_nn are constructed, and the efficiency ratio AdvantageRatio(99) is calculated under actual scenarios and the coarsest granularity scenario set (i.e., the scenario with the highest probability, which is scenario 2 in this embodiment). The corresponding Reward is obtained by subtracting the efficiency ratio calculated last time. Table 4 is a comparison table of Action and Advantage Ratio Reward of Example 2.

[0133] Table 4 Comparison table of Action and Advantage Ratio Reward of Example 2

[0134] Referring to Table 4, the action a* = 1001 that returns the maximum Reward(a*) is selected, i.e., the key hyperparameters m and n are updated to the current value +1 and the current value -1, respectively, to obtain the maximum Reward value. The key hyperparameters are updated, i.e., m = 3 and n = 2.

[0135] (2) Perform scenario clustering f1(α, β, 3, 2) (with a sampling granularity of 15 minutes).

[0136] wherein the new energy power generation power prediction layer outputs a probability distribution sequence a of length 96: a = {0: {v: [0], p: [1]},..., 40: {v: [220, 240, 300], p: [0.2, 0.6, 0.2]},..., 95: {v: [0], p: [1]}}

[0137] The power consumption load power prediction layer outputs a probability distribution sequence β of length 96: β = {0: {v: [200, 210], p: [0.8, 0.2]},..., 40: {v: [300, 310, 320], p: [0.1, 0.1, 0.8]},..., 95: {v: [200, 205], p: [0.9, 0.1]}}

[0138] a. By step-by-step iterative clustering method, m = 3 power generation power sequences A of length 96 with maximum joint probability in a are obtained respectively:

[0139] A = {sequence: {Solar1: (0,..., 300,..., 0),

[0140] Solar2: (0,..., 240,..., 0),

[0141] Solar3: (0,..., 280,..., 0)},

[0142] Probability: {Solar1: 0.6,

[0143] Solar2: 0.2,

[0144] Solar3: 0.2}

[0145] n = 1 load power sequence B of length 96 with maximum joint probability in β:

[0146] B = {sequence: {Load1: (200,..., 300,..., 200),

[0147] Load2: (200,..., 320,..., 200)},

[0148] Probability: {Load1: 0.4,

[0149] Load2: 0.6}

[0150] b. By cross method, 6 scenario sets S with maximum joint probability in 24 hours of the future day are obtained:

[0151] S = {scenario: {scenario1: {Solar1: (0,..., 300,..., 0),

[0152] Load1:(200,...,300,...,200)},

[0153] Scenario 2: {Solar1:(0,...,300,...,0),

[0154] Load2:(200,...,320,...,200)},

[0155] Scenario 3: {Solar2:(0,...,240,...,0),

[0156] Load1:(200,...,300,...,200)},

[0157] Scenario 4: {Solar2:(0,...,240,...,0),

[0158] Load2:(200,...,320,...,200)},

[0159] Scenario 5: {Solar3:(0,...,240,...,0),

[0160] Load1:(200,...,300,...,200)},

[0161] Scenario 6: {Solar3:(0,...,240,...,0),

[0162] Load2:(200,...,320,...,200)},

[0163] Probabilities: {Scenario 1: 0.24,

[0164] Scenario 2: 0.08,

[0165] Scenario 3: 0.08,

[0166] Scenario 4: 0.36,

[0167] Scenario 5: 0.12,

[0168] Scenario 6: 0.12}

[0169] (3) Define the policy θ as a sequence of length 96, θ * The scheduling policy (control policy) to be solved.

[0170] Define the objective function of the optimal electricity cost as:

[0171] Where, Cost *represents the electricity cost generated during the dispatch, F1 represents the objective function of the economic index optimization, S represents the scenario set obtained in 1), which contains 6 scenarios, S[i].p represents the weight of the i-th scenario, P elec (t) represents the grid power at time t, Price(t) represents the grid price at time t.

[0172] The strategy θ = {OPC1, OPC2,..., OPC 96} is defined as a sequence of length 96, each element of the sequence represents the SOC target of the energy storage at the corresponding decision time; the most suitable control strategy θ * = {OPC1 * , OPC2 * ,..., OPC 96 *} is obtained by using an intelligent optimization algorithm.

[0173] θ * On the one hand, it can meet the constraint conditions under all scenarios in the scenario set S, including:

[0174] a. Backup power safety constraint μ, that is, to ensure that the power balance meets the normal operation of the equipment, and the energy storage reserves enough power to ensure the backup time: P elec = P load -P solar -P bat

[0175] Where, P elec represents the grid power, P load represents the load power, P solar represents the solar power, and P bat represents the energy storage battery power; SOC min ≤ SOC ≤ SOC max

[0176] Where, SOC represents the energy storage SOC at each decision point, SOC min represents the minimum SOC threshold to ensure backup power safety, and SOC max represents the maximum SOC threshold of the energy storage;

[0177] b. Energy storage safety constraint φ, that is, to ensure that the energy storage charging is safe during the dispatch process: P bat = AH*(SOC(t)-SOC(t-1))*BatteryU / Δt 0 ≤ OPC ≤ ChargeMaxRatio

[0178] Wherein, AH represents the equipped energy storage battery capacity, SOC(t) represents the corresponding energy storage SOC value at decision point t, BatteryU represents the energy storage working voltage, Δt represents the decision sampling interval, which is 15 min by default, OPC(t) represents the corresponding charging rate at decision point t, and ChargeMaxRatio represents the maximum charging rate of the energy storage health.

[0179] On the other hand, the weighted economic indicators of all scenarios in the scenario set S can be optimized, that is:

[0180] Wherein, F1 represents the objective function of the optimal economic indicator, S represents the scenario set obtained in 1), which contains 6 scenarios, P elec (t) represents the city power at t time point, and Price(t) represents the city power price at t time point.

[0181] (4) Let i=i+1, repeat the above steps (1), (2) and (3), and carry out continuous prediction-scheduling-feedback dynamic coordination.

[0182] Example three

[0183] In this embodiment, the photovoltaic consumption optimal consumption scheduling scenario is taken as an example to carry out the 100th hybrid energy scheduling and scheduling-prediction dynamic sensing feedback optimization.

[0184] (1) Adaptive super parameter adjustment.

[0185] In the i=99th prediction-scheduling decision, the super parameter is m=2 and n=3, and adaptive feedback adjustment is carried out based on the 99th prediction-scheduling result, and the scenario set containing m*n=6 joint probability maximum generated in the 99th time is recorded as S(99), that is:

[0186] S(99)={ Scene1: {Solar1:(0,0,...,350,300,...,0,0), Load1:(200,200...,300,320,...,200)},

[0187] Scene2: {Solar1:(0,0,...,350,300,...,0,0), Load2:(200,220...,300,320,...,200)},

[0188] Scene3: {Solar1:(0,0,...,350,300,...,0,0), Load3:(200,220...,300,320,...,200)},

[0189] Scene4: {Solar1:(0,0,...,350,300,...,0,0), Load4:(200,220...,300,320,...,200)},

[0190] Scenario 3: {Solar1: (0, 0,..., 350, 300,..., 0, 0), Load3: (200, 200,..., 320, 320,..., 200)}

[0191] Scenario 4: {Solar2: (0, 0,..., 350, 300,..., 0, 0), Load1: (200, 200,..., 320, 320,..., 200)}

[0192] Scenario 5: {Solar2: (0, 0,..., 350, 300,..., 0, 0), Load2: (200, 200,..., 340, 320,..., 200)}

[0193] Scenario 6: {Solar2: (0, 0,..., 340, 300,..., 0, 0), Load3: (200, 200,..., 320, 300,..., 200)}

[0194] }

[0195] Probabilities:

[0196] Scenario 1: 0.5, Scenario 2: 0.1, Scenario 3: 0.1,

[0197] Scenario 4: 0.1, Scenario 5: 0.1, Scenario 6: 0.1}

[0198] }

[0199] A value sample valuerow(100) is constructed, with independent variables of <State(99), Action(99)]> and dependent variable of Reward(99), and the hyperparameter tuning value network sp_value_nn is updated based on valuerow(100) using the SGD method.

[0200] The inference sample input sp_value_nn is constructed by traversing different Actions, and the efficiency ratio AdvantageRatio(99) is calculated under the actual scenario and the coarsest granularity scenario set (i.e. the scenario with the largest probability, which is scenario 1 in this embodiment), and the corresponding Reward is obtained by subtracting the efficiency ratio obtained in the last calculation from the efficiency ratio. Table 5 is a comparison table of Action and Advantage Ratio Reward in Example Three.

[0201] Table 5 Comparison table of Action and Advantage Ratio Reward in Example Three

[0202] The sp_value_nn selects the action a* = 0001 with the maximum Reward(a*), and the key hyperparameters m remain unchanged, and n is updated to the current value -1 to obtain the maximum Reward value. Update the key hyperparameters, i.e. m = 2 and n = 2.

[0203] (2) Perform scene clustering f1(a, b, 2, 2) with a sampling granularity of 15 minutes.

[0204] where a and b are respectively:

[0205] The new energy power generation prediction layer outputs a probability distribution sequence a of length 96: a = {0: {v: [0], p: [1]},..., 40: {v: [220, 240, 300], p: [0.2, 0.6, 0.2]},..., 95: {v: [0], p: [1]}

[0206] The power load prediction layer outputs a probability distribution sequence b of length 96: b = {0: {v: [200, 210], p: [0.8, 0.2]},..., 40: {v: [300, 310, 320], p: [0.1, 0.1, 0.8]},..., 95: {v: [200, 205], p: [0.9, 0.1]}}

[0207] Through step-by-step iterative clustering, m = 2 sequences of length 96 of the maximum joint probability in a are obtained:

[0208] A = {sequence: {Solar1: (0,..., 300,..., 0),

[0209] Solar2: (0,..., 240,..., 0)},

[0210] Probability: {Solar1: 0.8,

[0211] Solar2: 0.2}

[0212] n = 2 sequences of length 96 of the maximum joint probability in b are obtained:

[0213] B = {sequence: {Load1: (200,..., 300,..., 200),

[0214] Load2: (200,..., 320,..., 200)},

[0215] Probability: {Load1: 0.4,

[0216] Load2: 0.6}

[0217] The four scenario sets S with the maximum joint probability in the 24-hour period of the next day are obtained by the cross method:

[0218] S = {Scenario: {Scenario1: {Solar1: (0,..., 300,..., 0),

[0219] Load1: (200,..., 300,..., 200)},

[0220] Scenario2: {Solar1: (0,..., 300,..., 0),

[0221] Load2: (200,..., 320,..., 200)},

[0222] Scenario3: {Solar2: (0,..., 240,..., 0),

[0223] Load1: (200,..., 300,..., 200)},

[0224] Scenario4: {Solar2: (0,..., 240,..., 0),

[0225] Load2: (200,..., 320,..., 200)}}.

[0226] Probability: {Scenario1: 0.32,

[0227] Scenario2: 0.48,

[0228] Scenario3: 0.08,

[0229] Scenario4: 0.12}

[0230] (3) Robust optimization is performed to obtain θ* = f2(S, μ, φ):

[0231] The objective function of optimal consumption is defined as:

[0232] where Consume * represents the generated photovoltaic consumption during the scheduling period, F2 represents the objective function of optimal photovoltaic consumption, S[i] represents the ith scenario of the scenario set S obtained in 1), S[i].p represents the weight of the ith scenario, S[i].solar represents the photovoltaic power sequence of the ith scenario, S[i].load represents the load power sequence of the ith scenario, soc_begin represents the energy storage SOC value at the time point at the beginning of the scheduling period, and OPCD represents the energy priority configuration at different time points.

[0233] Define the strategy θ = {OPCD1, OPCD2,..., OPCD96} as a sequence of length 96, each element of the sequence represents the energy storage SOC target at the corresponding decision time; use intelligent optimization algorithm to obtain the most appropriate control strategy θ 96 = {OPCD1 * , OPCD2 * ,..., OPCD * 96 *}. Where OPCD represents the energy priority strategy corresponding to the sampling point, taking values of 0 / 1 / 2 / 3 respectively.

[0234] θ * On the one hand, it can meet the constraint conditions of all scenarios in the scenario set S, including:

[0235] a. Backup power safety constraint μ, that is, ensure that the energy storage retains enough power to guarantee the backup time at each sampling point: SOC min ≤ SOC ≤ SOC max

[0236] Where SOC represents the energy storage SOC at each decision point, SOC min represents the minimum threshold of SOC to ensure backup power safety, and SOC max represents the maximum threshold of the energy storage SOC;

[0237] b. Energy storage safety constraint φ, including energy storage SOC change constraint and charge / discharge current constraint:

[0238] The change of SOC within the sampling interval is: SOC temp (t, S[i]) = SOC(t-1, S[i]) + I(t, S[i].GF, S[i].FH)*Δt / AH

[0239] Then the SOC at the next time that meets the change constraint is:

[0240] SOC(t, S[i]) =

[0241] The charge / discharge current constraint is:

[0242] ​Wherein, SOC(t, S[i]) represents the SOC value of the i-th scene of the scene set S at time t, the energy storage SOC value at the decision point t, OPCD(t) represents the energy priority policy at time t, BatteryU1 represents the energy storage operating voltage under the priority of photovoltaic > energy storage > commercial power, BatteryU2 represents the energy storage operating voltage under the priority of energy storage > photovoltaic > commercial power, MaxBatteryCP represents the maximum charging power of the energy storage, MaxBatteryDP represents the maximum discharging power of the energy storage, and AH represents the equipped energy storage battery capacity.

[0243] On the other hand, the probability-weighted economic indicators of all scenes in the scene set S can be optimized, that is: Consume * = argmax F2(S, θ * )

[0244] (4) Let i = i + 1, and repeat the above steps (1), (2), and (3) to perform continuous prediction-scheduling-feedback dynamic coordination.

[0245] In summary, the embodiment of the present disclosure provides a hybrid energy scheduling method, which clusters the photovoltaic power generation prediction sequence and the load prediction sequence based on the iterative strategy of ascending dimension and converging centroid, constructs m*n future scenes with maximum joint probability, converts the uncertainty problem into a deterministic optimization problem for solving, reduces the investment of computing resources, meets the requirements of engineering practice, and is convenient for actual engineering deployment. The value learning is introduced, the computing power investment of the scene set construction is dynamically adjusted according to the decision cost-effectiveness ratio, the adaptive balance of scheduling effect and computing power investment is achieved, the prediction-scheduling-feedback integrated dynamic coordination is realized, and the dynamic compromise between the effectiveness of hybrid energy scheduling decision and resource feasibility is realized, which is convenient for landing deployment on platforms of different computing power levels.

[0246] The hybrid energy scheduling method provided by the embodiment of the present disclosure can be applied to a hybrid energy scheduling system containing new energy power generation equipment and energy storage equipment, and can perform scheduling work containing economic scheduling and consumption scheduling targets; and can be used in the fields of communication site hybrid energy scheduling, virtual power plant and other energy applications.

[0247] The above only describes the preferred embodiments of the present disclosure and is not intended to limit the present disclosure. For those skilled in the art, the present disclosure can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the principles of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A hybrid energy scheduling method, comprising: obtaining new energy power generation power super parameters and power load power super parameters; dividing a new energy power generation power prediction layer into L-dimensional first clustering subspaces, iteratively clustering the first clustering subspaces based on the new energy power generation power super parameters, and obtaining a new energy power generation power sequence set with a sequence length of L; dividing a power load power prediction layer into L-dimensional second clustering subspaces, iteratively clustering the second clustering subspaces based on the power load power super parameters, and obtaining a power load power sequence set with a sequence length of L; obtaining a joint probability scenario set according to the new energy power generation power sequence set and the power load power sequence set; based on the joint probability scenario set, performing optimal solving of a preset power consumption index, and obtaining a scheduling strategy of the hybrid energy scheduling.

2. The method of claim 1, wherein, The obtaining of the new energy power generation power super parameters and the power load power super parameters comprises: adjusting the hybrid energy scheduling super parameters by a super parameter tuning value network to obtain the new energy power generation power super parameters and the power load power super parameters.

3. The method of claim 2, wherein: the super parameter tuning value network is feedback adjusted by an odds ratio parameter, wherein the odds ratio parameter is a difference between a first ratio and a second ratio, the first ratio is a ratio of an actual income of the hybrid energy scheduling at any moment to an actual operation time consumption at the moment, and the second ratio is a ratio of an income of a scenario set at the moment to an operation time consumption of the scenario set at the moment, the scenario set being constructed based on actual power load power and new energy power generation power.

4. The method of claim 1, wherein: each of the first clustering subspaces corresponds to different predicted values of new energy power generation power at a time point and probability distribution values corresponding to different predicted values; each of the second clustering subspaces corresponds to different predicted values of power load power at a time point and probability distribution values corresponding to different predicted values.

5. The method of claim 4, wherein, The iteratively clustering of the first clustering subspaces based on the new energy power generation power super parameters to obtain the new energy power generation power sequence set with the sequence length of L comprises: iteratively clustering the first clustering subspaces of the first dimension to the Lth dimension in sequence to obtain the new energy power generation power sequence set with the sequence length of L.

6. The method of claim 5, wherein, The iteratively clustering of the first clustering subspaces of the first dimension to the Lth dimension in sequence to obtain the new energy power generation power sequence set with the sequence length of L comprises: clustering the first clustering subspaces of the first dimension, and performing probability normalization on predicted values of new energy power generation power corresponding to the first dimension to obtain m predicted values of the new energy power generation power corresponding to the first dimension as a first new energy power generation power sequence set, wherein m is equal to a value of the new energy power generation power super parameters, and m is a positive integer. cross-combining the m predicted values of the new energy power generation corresponding to the second dimension, and performing probability normalization on the predicted values of the new energy power generation corresponding to the second dimension, to obtain m predicted value sequences of the new energy power generation corresponding to the second dimension as a second new energy power generation sequence set; repeating the cross-combining on the m predicted values of the new energy power generation corresponding to the third dimension to the Lth dimension in sequence until the new energy power generation sequence set with the sequence length L is obtained.

7. The method of claim 4, wherein, The iterative clustering of the second clustering subspace based on the electricity load power super parameter to obtain a sequence length L electricity load power sequence set comprises: iteratively clustering the second clustering subspace of the first dimension to the Lth dimension in sequence to obtain a sequence length L electricity load power sequence set.

8. The method of claim 7, wherein, The iterative clustering of the second clustering subspace of the first dimension to the Lth dimension in sequence to obtain a sequence length L electricity load power sequence set comprises: clustering the second clustering subspace of the first dimension, and performing probability normalization on the predicted values of the electricity load power corresponding to the first dimension, to obtain n predicted values of the electricity load power corresponding to the first dimension as a first electricity load power sequence set, wherein n is equal to the value of the electricity load power super parameter, and n is a positive integer; cross-combining the n predicted values of the electricity load power corresponding to the second dimension with the first electricity load power sequence set, and performing probability normalization on the predicted values of the electricity load power corresponding to the second dimension, to obtain n predicted value sequences of the electricity load power corresponding to the second dimension as a second electricity load power sequence set; repeating the cross-combining on the n predicted values of the electricity load power corresponding to the third dimension to the Lth dimension in sequence until the sequence length L electricity load power sequence set is obtained.

9. The method of any of claims 6 or 8, wherein, The probability normalization on the predicted values of the new energy power generation or the electricity load power comprises: assigning, according to Euclidean distance or sequence edit distance, the probability values of the clustering centroids of the new energy power generation except the clustering centroids corresponding to the m predicted values to the clustering centroids corresponding to the m predicted values; assigning, according to Euclidean distance or sequence edit distance, the probability values of the clustering centroids of the electricity load power except the clustering centroids corresponding to the n predicted values to the clustering centroids corresponding to the n predicted values.

10. The method of claim 1, wherein, The obtaining of the joint probability scenario set according to the new energy power generation sequence set and the electricity load power sequence set comprises: cross-combining the new energy power generation sequence set and the electricity load power sequence set to obtain a joint probability scenario set.

11. The method of claim 1, wherein, Based on the joint probability scenario set, the preset power consumption index is optimally solved with the power safety constraint and / or the energy storage safety constraint as a constraint function, so as to obtain an energy storage cell power SOC of the mixed energy scheduling at different time as the scheduling strategy.

12. A computer-readable storage medium having stored therein a computer program, wherein, The computer program, when executed by a processor, implements the method recited in any one of claims 1 to 11.

13. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the method recited in any one of claims 1 to 11 when executing the computer program.

14. A computer program product comprising a computer program, wherein the computer program, when executed by a processor, implements the method recited in any one of claims 1 to 11.

Images