A source-load matching rate prediction method, system, device, medium and product for energy storage scheduling decision

Abstract

The present application relates to the technical field of power grid data prediction, and discloses a source-load matching rate prediction method, system, equipment, medium and product for energy storage scheduling decision, which constructs an initial source-load matching rate prediction model based on a deep learning algorithm, uses a prediction error function and a decision evaluation function to form a hybrid loss function, optimizes the parameters of the initial source-load matching rate prediction model based on a hybrid gradient descent learning method, and obtains a source-load matching rate prediction model, so that the source-load matching rate can be accurately predicted, and the decision accuracy under the predicted source-load matching rate is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid data prediction, and in particular to a source-load matching rate prediction method, system, device, medium and product for energy storage scheduling decision. BACKGROUND

[0002] The source-load matching rate refers to the matching degree between power output and load demand, and can reflect the dynamic relationship between the output power of the system power supply and the consumption power of the load. Due to the access of high-proportion renewable energy to the power system, the source-load matching rate fluctuation increases, and the source-load matching rate is difficult to accurately predict due to its fluctuation. The source-load matching rate prediction has a wide application in the management and scheduling of smart grids, including energy storage systems.

[0003] Traditional source-load matching rate prediction focuses on source-load matching rate prediction from the perspective of prediction range and various prediction models. In order to improve the prediction accuracy, a two-stage source-load matching rate prediction scheme is proposed, which continuously predicts the source-load matching rate peak in the first stage and the source-load matching rate in the second stage. In order to improve the prediction accuracy of neural network, the traditional source-load matching rate prediction considers energy management and control, and proposes a multi-view source-load matching rate prediction model to improve the prediction accuracy and detect the source-load matching rate peak, and uses an overall method to predict the source-load matching rate of the power grid system.

[0004] In recent years, a method of source-load matching rate interval prediction has been proposed, such as a method of predicting the source-load matching rate interval based on the Pareto optimal optimal prediction interval, which realizes the reliability and clarity of the interval, considers the dynamic source-load matching rate change of the adjacent area, and provides rolling day-ahead source-load matching rate / interval prediction and intra-day prediction through online learning method to perceive the source-load matching rate fluctuation of the adjacent area, so as to improve the prediction accuracy.

[0005] The existing optimization-based work is to formulate the uncertainty of the predicted source-load matching rate by sampling or distribution assumption, and rarely considers the reverse influence of the optimization model on the upstream prediction model. The above research shows that there is a gap between the upstream prediction and the downstream optimization model. The objective of the traditional prediction model is usually to minimize the difference between the predicted source-load matching rate and the real source-load matching rate, which is expressed by prediction error. In contrast, the energy storage system focuses on maximizing the utilization benefit from the source-load matching rate fluctuation, and the energy storage system benefit maximization can be equivalent to minimizing the deviation between the optimal actual decision under the predicted source-load matching rate and the optimal decision under the real source-load matching rate. The decision error is not necessarily consistent with the prediction error, and the decision error comes from the energy storage system scheduling model, which is difficult to be directly used for training the prediction model, which leads to inaccurate source-load matching rate prediction of the energy storage system. SUMMARY

[0006] Therefore, the application provides a source-load matching rate prediction method, system, device, medium and product for energy storage scheduling decision, which solves the technical problem of inaccurate source-load matching rate prediction of the energy storage system in the above technical background.

[0007] The first aspect of the application provides a source-load matching rate prediction method for energy storage scheduling decision, comprising:

[0008] training a historical source-load matching rate sample set based on a deep learning algorithm to construct an initial source-load matching rate prediction model;

[0009] constructing a prediction error function and a decision evaluation function, and forming a hybrid loss function according to the prediction error function and the decision evaluation function;

[0010] based on a hybrid gradient descent learning method, using the hybrid loss function to optimize the parameters of the initial source-load matching rate prediction model to obtain a source-load matching rate prediction model, which is used to predict the source-load matching rate at a future time.

[0011] Preferably, the prediction error function adopts a hybrid mean square error function.

[0012] Preferably, the construction process of the decision evaluation function comprises:

[0013] determining the regret degree of the decision according to the difference between the optimal decision under the true source-load matching rate and the optimal actual decision under the source-load matching rate predicted by the initial source-load matching rate prediction model;

[0014] relaxing the regret degree of the decision and determining the minimum upper bound of the regret degree of the decision to obtain a proxy regret degree;

[0015] determining a substitute regret loss function according to the proxy regret degree, and taking the substitute regret loss function as the decision evaluation function.

[0016] Preferably, forming a hybrid loss function according to the prediction error function and the decision evaluation function comprises:

[0017] weighting the prediction error function and the decision evaluation function to obtain the hybrid loss function.

[0018] Preferably, the hybrid gradient descent learning method uses the hybrid loss function to optimize the parameters of the initial source-load matching rate prediction model to obtain a source-load matching rate prediction model, comprising:

[0019] determining a hybrid gradient of the predicted source-load matching rate according to the hybrid loss function;

[0020] The initial source-load matching rate prediction model is parameter-optimized in a random gradient descent manner by using the hybrid gradient, to obtain a source-load matching rate prediction model.

[0021] Preferably, the parameter-optimized initial source-load matching rate prediction model in a random gradient descent manner by using the hybrid gradient, to obtain a source-load matching rate prediction model, comprises:

[0022] The hybrid loss value of the predicted source-load matching rate of the initial source-load matching rate prediction model is calculated according to the hybrid loss function;

[0023] The parameter gradient is obtained by back propagation from the hybrid loss value to the parameters of the initial source-load matching rate prediction model;

[0024] The parameters of the initial source-load matching rate prediction model are updated in a random gradient descent manner by using the parameter gradient;

[0025] The initial source-load matching rate prediction model is parameter-optimized by using the updated parameters, to obtain a source-load matching rate prediction model.

[0026] In a second aspect, the embodiments of the present application further provide a source-load matching rate prediction system for energy storage scheduling decision, comprising:

[0027] A prediction model construction module is configured to train a historical source-load matching rate sample set based on a deep learning algorithm, and to construct an initial source-load matching rate prediction model;

[0028] A hybrid loss construction module is configured to construct a prediction error function and a decision evaluation function, and to form a hybrid loss function according to the prediction error function and the decision evaluation function;

[0029] A prediction model optimization module is configured to parameter-optimize the initial source-load matching rate prediction model by using the hybrid loss function based on a hybrid gradient descent learning method, to obtain a source-load matching rate prediction model, which is configured to predict a source-load matching rate at a future time.

[0030] In a third aspect, the embodiments of the present application further provide an electronic device, which comprises a memory and a processor, the memory stores a computer program, and the computer program is executed by the processor to make the processor execute the steps of the source-load matching rate prediction method for energy storage scheduling decision according to the first aspect.

[0031] In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium, which stores a computer program, and the computer program is executed to implement the source-load matching rate prediction method for energy storage scheduling decision according to the first aspect.

[0032] In a fifth aspect, the embodiments of the present application further provide a computer program product, which comprises a computer program stored on a non-transitory computer-readable storage medium, and the computer program comprises program instructions, wherein when the program instructions are executed by a computer, the computer is caused to execute the source-load matching rate prediction method for energy storage scheduling decision as described in the first aspect.

[0033] From the above technical solutions, it can be seen that the initial source-load matching rate prediction model based on the deep learning algorithm is constructed, the hybrid loss function is formed by using the prediction error function and the decision evaluation function, and the parameter optimization is performed on the initial source-load matching rate prediction model based on the hybrid gradient descent learning method to obtain the source-load matching rate prediction model, so that the source-load matching rate is accurately predicted, and the decision accuracy under the predicted source-load matching rate is improved. BRIEF DESCRIPTION OF DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0035] Figure 1 An application environment of the source-load matching rate prediction method for energy storage scheduling decision provided by the embodiments of the present application;

[0036] Figure 2 A flowchart of the source-load matching rate prediction method for energy storage scheduling decision provided by the embodiments of the present application;

[0037] Figure 3 A structural schematic diagram of the source-load matching rate prediction system for energy storage scheduling decision provided by the embodiments of the present application;

[0038] Figure 4 A structural schematic diagram of the electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0039] In order to make the person skilled in the art better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0040] The source-load matching rate prediction method for energy storage scheduling decision provided by the embodiments of the present application can be applied in the application environment as shown in Figure 1 The nodes of the energy storage system communicate with the server 102 through the network. The data storage system can store the data required to be processed by the server 102. The data storage system can be integrated on the server 102, or placed on the cloud or other network servers. The server 102 trains the historical source-load matching rate sample set based on the deep learning algorithm, constructs an initial source-load matching rate prediction model, constructs a prediction error function and a decision evaluation function, and forms a hybrid loss function according to the prediction error function and the decision evaluation function. Based on the hybrid gradient descent learning method, the hybrid loss function is used to optimize the parameters of the initial source-load matching rate prediction model to obtain a source-load matching rate prediction model, which is used to predict the source-load matching rate at a future time. The server 102 can be a stand-alone physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0041] As shown in Figure 2 The embodiments of the present application provide a source-load matching rate prediction method for energy storage scheduling decision. Taking the server 102 in Figure 1 as an example, the method includes the following steps S1 to S3. Wherein:

[0042] Step S1, training the historical source-load matching rate sample set based on the deep learning algorithm, and constructing an initial source-load matching rate prediction model.

[0043] The historical source-load matching rate sample set contains the mapping relationship between the source-load matching rate and the operation data of the energy storage system in the past period of time, and is trained based on the deep learning algorithm, with the operation data as the input quantity and the source-load matching rate as the output quantity to obtain the initial source-load matching rate prediction model.

[0044] Step S2, constructing a prediction error function and a decision evaluation function, and forming a hybrid loss function according to the prediction error function and the decision evaluation function.

[0045] In the embodiments of the present application, the prediction error function adopts a hybrid mean square error function, and other error functions such as standard deviation function can also be adopted.

[0046] For the decision evaluation function, in the proposed source-load matching rate prediction with decision awareness, the difference between the optimal actual decision under the predicted source-load matching rate and the best prediction decision under the true source-load matching rate is determined, and the low regret loss means the gradually optimal prediction machine decision.

[0047] Specifically, the process of constructing the decision evaluation function includes:

[0048] Step S201: Determine the regret degree of the decision based on the difference between the optimal decision under the actual source-load matching rate and the optimal actual decision under the source-load matching rate predicted by the initial source-load matching rate prediction model.

[0049] Among them, the regret of the decision is the difference between the benefit under the actual optimal decision and the benefit under the optimal decision:

[0050]

[0051]

[0052] In the formula, As for the degree of regret of the decision, This is the base value for the source-load matching rate. Where L is the sensitivity coefficient, G is the total system load demand, and G is the total system power output. The predicted source-load matching rate, This represents the actual source-load matching rate. , These represent the optimal decision under the actual source-load matching rate and the optimal actual decision under the predicted source-load matching rate, respectively, with T being the matrix transpose. The optimal decision uses the actual source-load matching rate as the source-load matching rate input and is the result of optimization calculations performed under the original system's optimization objective and constraints. This decision depends on the structural parameters of the selected system.

[0053] Step S202: Relax the regret of the decision and determine the minimum upper bound of the regret of the decision to obtain the proxy regret.

[0054] It should be noted that the energy storage operation model, as a typical mixed-integer programming model, has a feasible region that is a set of polyhedra. Therefore, its optimal decision may be at an extreme point of one polyhedron. Taking a two-dimensional polyhedron as an example, with the same prediction error, two different predicted source-load matching rates lead to different decision errors, and correspondingly, different degrees of regret. In the actual source-load matching rate... The decision made Matching rate with predicted source load Below The same, but with the predicted source load matching rate. Below The difference is that, in this case, the prediction error is not equal to the decision error.

[0055] Relative to the predicted source load matching rate Since the regret of decision-making is discontinuous and unmanageable, this application constructs a manageable regret loss function. , expressed as the degree of regret of the agent.

[0056] Among them, the regret of decision-making is relaxed, and the proxy regret is derived as follows:

[0057]

[0058] In the formula, Indicates the level of regret of the agent. The relaxation coefficient is... , These represent the optimal benefits under the actual source-load matching rate and the optimal benefits under the predicted source-load matching rate, respectively.

[0059] in, Optimal decision relaxation This will give the upper bound of the above expression. This applies to all... All conditions are met, and the equation can be derived:

[0060]

[0061] In fact A decreasing function. According to Definition, subgradient It can be represented as: .

[0062] yes The optimal decision under the given conditions, therefore, its gain Less than ,For example .

[0063] Less than 0, therefore, It is a monotonically decreasing function. According to the monotonically convergent theorem, It can be used Replace. With The increase, The trend of the terms can be ignored; the optimality of the maximization problem is... tending to This establishes a minimum upper bound for the degree of regret:

[0064]

[0065] The combined prediction model formula is as follows:

[0066]

[0067] in It predicts the source-load matching rate. It is input represents a prediction model mapping from an input feature vector to an output predicted source load matching rate The whole process consists of five procedures: i) data preprocessing, ii) feature engineering, iii) model selection, iv) training procedure, v) prediction target.

[0068]

[0069] The linear regression model maps the input feature vector to the predicted source load matching rate by and

[0070] The regret minimization of the prediction model can be further generalized as

[0071]

[0072] wherein represents the system model parameters, and n represents the number of scenarios.

[0073] wherein the formula for further generalization of the regret minimization of the prediction model gives the generalization form of the objective function when α takes different values, the second row of the objective function corresponds to the case of any α≥0; the third row shows that when α tends to infinity, all α can be replaced by the same parameter; the fourth row obtains the estimated value of the upper bound by setting α as 2; and the fifth row relaxes the objective function under the feasible decision value.

[0074] In step S203, the surrogate regret loss function is determined according to the agent regret, and the surrogate regret loss function is used as the decision evaluation function.

[0075] wherein given the predicted source load matching rate and the actual source load matching rate the surrogate regret loss function can be defined as

[0076]

[0077] Under the condition of given predicted source load matching rate and actual source load matching rate, the surrogate regret loss function has the following properties: regret ; is a convex function of the predicted source load matching rate .

[0078] wherein according to the definition of the surrogate regret loss function, the gradient of the surrogate regret loss function with respect to the predicted source load matching rate can be derived as

[0079]

[0080] wherein ​To replace the gradient of the predicted source-load matching rate of the regret loss function.

[0081] Wherein, the mixed mean square error and the proxy regret degree respectively pay attention to the prediction error and the decision error, and respectively view the prediction mismatch from the perspective of the prediction model and the energy storage operation model. Therefore, the embodiment of the application combines the prediction error function and the decision evaluation function to constitute a mixed loss function to train the source-load matching rate prediction model.

[0082] Wherein, the energy storage operation model is:

[0083] When the source-load matching rate is low, the ESS usually works in the charging state, and when the source-load matching rate is high, it is in the discharging state. The operation of the ESS is usually as follows:

[0084] The source-load matching rate signal maximizes its benefit to construct the objective function of the energy storage operation model, and the objective function of the energy storage operation model is:

[0085]

[0086] In the formula, P is the operating power, is the feasible region of the operating power, is the source-load matching rate at time t, is the operating power at time t.

[0087] Wherein, the operating constraints of the energy storage operation model include:

[0088] The feasible region of is subject to a set of operation and technical restrictions as follows:

[0089]

[0090] In the formula, represents that the ESS is charged from the power grid, represents that the ESS discharges to the power grid

[0091]

[0092]

[0093]

[0094] In the formula, , respectively are the stored energy in the ESS at time t-1 and time t. Ensure that the energy in the ESS at time t is is located in the allowable value, , represents the minimum value and the maximum value of the battery system capacity.

[0095]

[0096]

[0097]

[0098]

[0099]

[0100] To prevent simultaneous charging and discharging, the big-M method is used to charge the ESS in the embodiments of the application. M is a large positive number. and is a binary variable indicating the charging and discharging state, where 1 indicates the state and 0 indicates the opposite. and is the maximum value of the charging and discharging power. The entire ESS scheduling model is expressed as a mixed integer linear programming.

[0101] Specifically, the prediction error function and the decision evaluation function are weighted to obtain a hybrid loss function. The hybrid loss function is represented as:

[0102]

[0103] In the formula, is the hybrid loss function, is a weighting coefficient, indicating the importance of minimizing the prediction error, is the total time.

[0104] Step S3, based on the hybrid gradient descent learning method, the initial source load matching rate prediction model is optimized by using the hybrid loss function to obtain the source load matching rate prediction model, and the source load matching rate prediction model is used to predict the source load matching rate at the future time.

[0105] Specifically, the hybrid gradient descent learning method in step S3 is used to optimize the initial source load matching rate prediction model by using the hybrid loss function to obtain the source load matching rate prediction model, including:

[0106] Step S301, determining the hybrid gradient of the predicted source load matching rate according to the hybrid loss function.

[0107] The hybrid gradient is:

[0108]

[0109] In the formula, is the hybrid gradient.

[0110] The mixed gradient of the mixed loss function can be calculated explicitly by the above formula. In order to improve the calculation efficiency, the automatic differentiation algorithm in the Autograd tool in Pytorch is used to implicitly calculate the gradient of MSE. Autograd is a reverse automatic differentiation system for calculating derivatives, which is the core of Pytorch. It records the graph of all operations performed on the gradient-enabled tensor and creates a loop-free graph called dynamic computation graph. The gradient is calculated by tracing the computation from root to leaf and multiplying each gradient by the chain rule. The automatic differentiation algorithm of Autograd has high calculation efficiency due to its core logic in c++, and can implicitly calculate the gradient of MSE, which is more efficient than the explicit calculation of the above formula. On the contrary, the gradient of the agent regret is explicitly calculated by solving the mixed integer programming problem.

[0111] In step S302, the initial source-load matching rate prediction model is parameter-optimized in a random gradient descent manner using the mixed gradient, and a source-load matching rate prediction model is obtained.

[0112] In some embodiments, the parameter optimization of the initial source-load matching rate prediction model in a random gradient descent manner using the mixed gradient in step S302 to obtain the source-load matching rate prediction model comprises:

[0113] In step S3021, the mixed loss value of the predicted source-load matching rate of the initial source-load matching rate prediction model is calculated according to the mixed loss function.

[0114] In step S3022, the parameter gradient is obtained by backpropagation of the mixed loss value to the parameters of the initial source-load matching rate prediction model.

[0115] In step S3023, the parameter gradient is used to update the parameters of the initial source-load matching rate prediction model in a random gradient descent manner.

[0116] In step S3024, the initial source-load matching rate prediction model is parameter-optimized using the updated parameters, and a source-load matching rate prediction model is obtained.

[0117] It should be noted that the mixed gradient proposed in the embodiments of the present application indicates that the prediction model learning reduces the decision error of the prediction model. The prediction error is measured by mean squared error (MSE). Based on MSE and agent regret, a mixed loss function is designed, a mixed stochastic gradient descent (SGD) learning method for updating the parameters of the prediction model is proposed, and the prediction and decision accuracy is improved.

[0118] Exemplarily, in order to solve the two different gradient calculation methods, the prediction model training process under the mixed gradient is divided into three steps:

[0119] 1) Calculate the gradient of the hybrid mean square error loss function of the predicted source load matching rate, backpropagate the gradient of the hybrid mean square error loss function to the parameters of the initial source load matching rate prediction model, and reserve the gradient for subsequent parameter gradient update.

[0120] 2) Calculate the gradient of the proxy regret of the predicted source load matching rate, and backpropagate the gradient to the same parameters of the initial source load matching rate prediction model;

[0121] 3) Update the parameters of the initial source load matching rate prediction model according to the cumulative gradient, and the hybrid gradient update is realized by twice backpropagation and once update.

[0122] It is worth noting that the random gradient descent learning method for energy storage system scheduling proposed in the embodiments of the application can be applied to simple linear prediction models and complex deep learning models.

[0123] In order to distinguish from the previous research, the innovation of the application lies in three aspects.

[0124] 1) The method of the application uses prediction and decision errors to train the source load matching rate prediction model for energy storage scheduling, while the existing research only considers the prediction error.

[0125] 2) In the context of deep learning, the application proposes a processable proxy regret to train a multi-layer neural network called residual neural network (ResNet).

[0126] 3) The application proposes a hybrid loss design and learning method to improve the accuracy of energy storage operation decision and improve the utilization efficiency, which is first applied to energy storage scheduling prediction.

[0127] The embodiments of the application use prediction error and decision error to learn the parameters of the prediction model, and propose a decision-centered source load matching rate prediction method for energy storage scheduling. Compared with the previous prediction method, the decision-centered method proposed in the embodiments of the application focuses on the influence of prediction error on the downstream optimization model, and improves the decision accuracy under the predicted source load matching rate.

[0128] The embodiments of the application propose a processable alternative regret to measure the decision error between the actual decision under the predicted source load matching rate and the decision under the true source load matching rate. Further derivation of its gradient with the predicted source load matching rate is the focus of this paper. The decision regret under the predicted source load matching rate is discontinuous and difficult to process. Therefore, this paper relaxes the feasible region of the original problem, derives an upper bound of the processable regret to measure the decision error, and uses the proxy regret to derive the gradient of the proxy regret of the predicted source load matching rate, and passes it from the downstream optimization model to the upstream prediction model to reduce decision errors.

[0129] The embodiment of the application designs a hybrid loss function to measure the prediction and decision errors based on the weighted sum of the prediction error and the proxy regret. Then a hybrid stochastic gradient descent learning method for energy storage scheduling is proposed to train the prediction model to improve the prediction and decision accuracy based on the hybrid loss. The case analysis shows that the decision-centered method proposed in the application can more accurately capture and predict the trend of the source-load matching rate, thereby improving the decision accuracy and energy utilization efficiency.

[0130] It should be noted that, in the embodiment of the application, an initial source-load matching rate prediction model based on a deep learning algorithm is constructed, a hybrid loss function is formed by using a prediction error function and a decision evaluation function, and the initial source-load matching rate prediction model is parameter optimized based on a hybrid gradient descent learning method to obtain a source-load matching rate prediction model, so that the source-load matching rate is accurately predicted, and the decision accuracy under the predicted source-load matching rate is improved.

[0131] Based on the same inventive concept, the embodiment of the application also provides a source-load matching rate prediction system for energy storage scheduling decision.

[0132] The implementation scheme for solving the problem provided by the system is similar to the implementation scheme described in the above method, so the specific limitations in one or more source-load matching rate prediction system embodiments for energy storage scheduling decision provided below can be referred to the limitations of the source-load matching rate prediction method for energy storage scheduling decision in the above, which will not be described here.

[0133] As shown in Figure 3 The embodiment of the application also provides a source-load matching rate prediction system for energy storage scheduling decision, which comprises:

[0134] The prediction model construction module 100 is configured to train a historical source-load matching rate sample set based on a deep learning algorithm, and construct an initial source-load matching rate prediction model.

[0135] The hybrid loss construction module 200 is configured to construct a prediction error function and a decision evaluation function, and form a hybrid loss function according to the prediction error function and the decision evaluation function.

[0136] The prediction model optimization module 300 is configured to parameter optimize the initial source-load matching rate prediction model based on a hybrid gradient descent learning method and by using the hybrid loss function, to obtain a source-load matching rate prediction model, and the source-load matching rate prediction model is used to predict the source-load matching rate at a future time.

[0137] In some embodiments, the prediction error function adopts a hybrid mean square error function.

[0138] In some embodiments, the construction process of the decision evaluation function includes:

[0139] The regret degree of the decision is determined according to the gap between the optimal decision under the real source-load matching rate and the optimal actual decision under the source-load matching rate predicted by the initial source-load matching rate prediction model;

[0140] The regret degree of the decision is relaxed, and the minimum upper bound of the regret degree of the decision is determined to obtain a proxy regret degree;

[0141] The proxy regret degree is determined to determine a substitute regret loss function, and the substitute regret loss function is used as the decision evaluation function.

[0142] In some embodiments, the hybrid loss function is composed of the prediction error function and the decision evaluation function, including:

[0143] The prediction error function and the decision evaluation function are weighted to obtain the hybrid loss function.

[0144] In some embodiments, the prediction model optimization module 300 is configured to determine a hybrid gradient of the predicted source-load matching rate according to the hybrid loss function; and perform parameter optimization on the initial source-load matching rate prediction model in a random gradient descent manner by using the hybrid gradient to obtain the source-load matching rate prediction model.

[0145] The parameter optimization on the initial source-load matching rate prediction model in a random gradient descent manner by using the hybrid gradient to obtain the source-load matching rate prediction model includes:

[0146] The hybrid loss value of the predicted source-load matching rate of the initial source-load matching rate prediction model is calculated according to the hybrid loss function;

[0147] The parameter gradient is obtained by performing back propagation to the parameters of the initial source-load matching rate prediction model according to the hybrid loss value;

[0148] The parameters of the initial source-load matching rate prediction model are updated in a random gradient descent manner by using the parameter gradient;

[0149] The initial source-load matching rate prediction model is parameter optimized by using the updated parameters to obtain the source-load matching rate prediction model.

[0150] As shown in Figure 4 The embodiments of the present application also provide an electronic device, which includes a memory 20 and a processor 30, the memory 20 stores a computer program, and the computer program is executed by the processor 30 to make the processor 30 execute the steps of the source-load matching rate prediction method for energy storage scheduling decision in any of the above embodiments.

[0151] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed to implement the source-load matching rate prediction method for energy storage scheduling decision in any of the foregoing embodiments.

[0152] The embodiment of the present application further provides a computer program product, which comprises a computer program stored on a non-transitory computer readable storage medium, and the computer program comprises program instructions, wherein when the program instructions are executed by a computer, the computer is caused to execute the source-load matching rate prediction method for energy storage scheduling decision in any of the foregoing embodiments.

[0153] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the system, the electronic device, the computer storage medium and the computer program product described above can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0154] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification and claims of the present application and the foregoing drawings are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device comprising a series of steps or units does not have to be limited to the clearly listed steps or units, but can include other steps or units that are not clearly listed or inherent to the process, method, product or device.

[0155] In several embodiments provided by the present application, it can be understood that each block in the flowchart or block diagram can represent a module, a program segment or a part of code, and the module, the program segment or the part of code include one or more executable instructions for implementing a specified logic function. It should also be noted that, in some alternative implementation manners, the functions annotated in the blocks can also occur in an order different from that annotated in the drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and they can also be executed in reverse order in some cases, depending on the functions involved.

[0156] In several embodiments provided by the present application, it can be understood that the disclosed system, electronic device, computer storage medium, computer program product and method can be implemented in other manners. For example, the above-described device embodiments are merely schematic, and the division of units is merely a logical function division, and there can be another division manner in actual implementation, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or in other forms.

[0157] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., may be located in one place, or may be distributed to multiple network units. Part or all of the units may be selected according to actual needs to achieve the purpose of the embodiment.

[0158] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0159] If the integrated unit is realized in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the whole or part of the technical solutions that make essential contributions to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for executing all or part of the steps of the method described in each embodiment of the present application by a computer device (which can be a personal computer, a server, or a network device, etc.). The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (English full name: Read-Only Memory, English abbreviation: ROM), a random access memory (English full name: Random Access Memory, English abbreviation: RAM), a magnetic disk or an optical disk, and various program code storage media.

[0160] The above embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A source-load matching rate prediction method for energy storage scheduling decision, characterized in that, The method comprises the steps of: training a historical source-load matching rate sample set based on a deep learning algorithm to construct an initial source-load matching rate prediction model; constructing a prediction error function and a decision evaluation function, and forming a hybrid loss function according to the prediction error function and the decision evaluation function, comprising: weighting the prediction error function and the decision evaluation function to obtain the hybrid loss function; the prediction error function adopts a hybrid mean square error function; the construction process of the decision evaluation function comprises: determining the regret degree of the decision according to the difference between the optimal decision under the true source-load matching rate and the optimal actual decision of the source-load matching rate predicted by the initial source-load matching rate prediction model; relaxing the regret degree of the decision and determining the minimum upper bound of the regret degree of the decision to obtain a proxy regret degree; determining a substitute regret loss function according to the proxy regret degree, and taking the substitute regret loss function as the decision evaluation function; based on a hybrid gradient descent learning method, the hybrid loss function is used to optimize the parameters of the initial source-load matching rate prediction model to obtain a source-load matching rate prediction model, comprising: determining a hybrid gradient of a predicted source-load matching rate according to the hybrid loss function; using the hybrid gradient in a random gradient descent manner to optimize the parameters of the initial source-load matching rate prediction model to obtain a source-load matching rate prediction model; the source-load matching rate prediction model is used to predict the source-load matching rate at a future time.

2. The method of claim 1, wherein, the optimization of the initial source-load matching rate prediction model to obtain a source-load matching rate prediction model using the hybrid gradient in a random gradient descent manner, comprising: calculating the hybrid loss value of the predicted source-load matching rate of the initial source-load matching rate prediction model according to the hybrid loss function; propagating the hybrid loss value to the parameters of the initial source-load matching rate prediction model in a reverse direction to obtain a parameter gradient; updating the parameters of the initial source-load matching rate prediction model in a random gradient descent manner using the parameter gradient; optimizing the parameters of the initial source-load matching rate prediction model using the updated parameters to obtain a source-load matching rate prediction model.

3. A source-load matching rate prediction system for energy storage scheduling decision based on the source-load matching rate prediction method for energy storage scheduling decision according to claim 1 or 2, characterized in that, The method comprises the steps of: a prediction model construction module is configured to train a historical source-load matching rate sample set based on a deep learning algorithm to construct an initial source-load matching rate prediction model; a hybrid loss construction module is configured to construct a prediction error function and a decision evaluation function, and form a hybrid loss function according to the prediction error function and the decision evaluation function; a prediction model optimization module is configured to optimize the parameters of the initial source-load matching rate prediction model based on a hybrid gradient descent learning method using the hybrid loss function to obtain a source-load matching rate prediction model, and the source-load matching rate prediction model is used to predict the source-load matching rate at a future time.

4. An electronic device, comprising: The electronic device comprises a memory and a processor, the memory stores a computer program, and the computer program is executed by the processor to make the processor execute the steps of the source-load matching rate prediction method for energy storage scheduling decision according to any one of claims 1-2.

5. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed to implement the source-load matching rate prediction method for energy storage scheduling decision as claimed in any one of claims 1-2.

6. A computer program product, characterised in that, The computer program product comprises a computer program stored on a non-transitory computer-readable storage medium, and the computer program comprises program instructions, wherein when the program instructions are executed by a computer, the computer is caused to execute the source-load matching rate prediction method for energy storage scheduling decision as claimed in any one of claims 1-2.

Images