Frequency modulation instruction prediction method and system for energy storage of coupling super capacitor of thermal power generating unit

Abstract

The invention discloses a thermal power generating unit coupling super capacitor energy storage frequency modulation instruction prediction method and system, and relates to the technical field of power system energy storage frequency modulation, and the method comprises the steps: obtaining an original frequency modulation sequence through a frequency modulation instruction of a thermal power generating unit, and dividing the original frequency modulation sequence into a training group, a proofreading group and a prediction group; training the training group through a BP neural network to generate a predicted value of the proofreading group; respectively calculating a difference error and a ratio error based on the actual value and the predicted value of the proofreading group to form a difference error sequence and a ratio error sequence; adjusting the frequency modulation instruction value of the prediction group to generate a difference error interval; adjusting the frequency modulation instruction value of the prediction group according to the ratio error sequence, and generating a ratio error interval; and combining the difference error interval with the ratio error interval to obtain an interval prediction value of the final frequency modulation instruction. The prediction precision of the frequency modulation instruction is obviously improved, and the uncertainty caused by single numerical prediction is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of power system energy storage and frequency regulation, and in particular to a method and system for predicting frequency regulation instructions for a thermal power unit coupled with supercapacitor energy storage. Background Art

[0002] As a crucial component of the power system, thermal power generation plays a crucial role in power frequency regulation. Traditional thermal power generation frequency regulation relies on coal-fired units as the primary power source for frequency regulation, a method that has long faced numerous challenges. As the proportion of renewable energy access increases, power systems are increasingly demanding higher speeds and precision in frequency regulation response. However, existing thermal power generation frequency regulation technology, limited by the characteristics of the units themselves, struggles to meet the demands for fast and precise frequency regulation. This is particularly true during long-term frequency regulation operations, which can lead to increased coal consumption, reduced reliability, and shortened equipment life. Furthermore, with the emergence of the frequency regulation ancillary services market, traditional frequency regulation methods are unable to meet the demands of modern power systems for fast response and high-efficiency energy storage devices. With the maturity of supercapacitor and lithium battery energy storage technologies, integrating them with thermal power generation frequency regulation systems has become an important research direction for improving frequency regulation performance.

[0003] Most of the existing frequency regulation energy storage systems rely on traditional numerical prediction methods, that is, frequency regulation decisions are guided by numerical sequence predictions of historical frequency regulation data. However, this prediction method has inherent shortcomings: its prediction result is a single specific value, which is often difficult to deal with the uncertainty and volatility in the frequency regulation instructions. In actual operation, the specific value prediction is easily affected by noise and environmental changes, resulting in large prediction errors, which in turn affects the accuracy and stability of the frequency regulation system. On the other hand, the current energy storage system has not fully considered how to efficiently allocate the synergistic effect of energy storage devices (such as supercapacitors and lithium batteries) in frequency regulation, resulting in the inability of energy storage devices to effectively play their due advantages. Therefore, the existing technology still has a lot of room for optimization in terms of frequency regulation response accuracy, prediction stability, and energy storage device allocation. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is that the existing thermal power unit frequency regulation technology generally has the problems of long response time, low prediction accuracy and insufficient energy storage utilization.

[0006] To solve the above technical problems, the present invention provides the following technical solution: a method for predicting frequency modulation instructions for a thermal power unit coupled with supercapacitor energy storage, comprising:

[0007] The original frequency modulation sequence is obtained through the frequency modulation instructions of the thermal power unit and divided into a training group, a calibration group and a prediction group.

[0008] The training group is trained through the BP neural network to generate the predicted value of the proofreading group.

[0009] Based on the actual values ​​and predicted values ​​of the proofreading group, the difference error and the ratio error are calculated respectively to form a difference error series and a ratio error series.

[0010] Adjust the frequency modulation command value of the prediction group to generate a difference error interval.

[0011] The frequency modulation instruction value of the prediction group is adjusted according to the ratio error sequence to generate a ratio error interval.

[0012] The difference error interval and the ratio error interval are combined to obtain the interval prediction value of the final frequency modulation instruction.

[0013] As a preferred solution of the frequency modulation instruction prediction method for a thermal power unit coupled with supercapacitor energy storage according to the present invention, the original frequency modulation sequence includes, assuming that the length of the original frequency modulation sequence is N, the formula is expressed as:

[0014] [x1,x2,x3,…,x N ]

[0015] The proofreading group is the first 90% of the original FM sequence, which is used to train the neural network model. The formula is:

[0016] [x1,x2,x3,…,x 0.9N ]

[0017] The prediction group is the last 10% of the original FM sequence and is used to generate the proportional error and the difference error. The formula is:

[0018] [x 0.9N+1 ,x 0.9N+2 ,…,x N ]

[0019] The prediction group is the unknown 10% of the original FM sequence in the future, and the formula is:

[0020] [x N+1 ,x N+2 ,...,x 1.1N ]

[0021] Where N represents the length of the original FM sequence, N>300. i Indicates the value of the i-th frequency modulation instruction.

[0022] As a preferred solution of the frequency modulation instruction prediction method of a thermal power unit coupled with supercapacitor energy storage described in the present invention, generating the prediction value of the proofreading group includes inputting the training group into a BP neural network for training to obtain the prediction value of the proofreading group and the value of the prediction group.

[0023] The predicted value of the proofreading group is expressed as:

[0024] [x′ 0.9N+1 ,x′ 0.9N+2 ,...,x′ N ]

[0025] Where N represents the length of the original FM sequence, N>300. i Indicates the value of the i-th frequency modulation instruction.

[0026] As a preferred solution of the frequency modulation instruction prediction method of the thermal power unit coupled with supercapacitor energy storage described in the present invention, the actual value and the predicted value of the calibration group will respectively generate a difference error sequence and a ratio error sequence.

[0027] The difference error sequence reflects the absolute difference between the actual value and the predicted value of the proofreading group, and the formula is expressed as:

[0028] [c1,c2,c3,…,c 0.1N ]c i =x i -x i

[0029] Among them, c i represents the i-th difference error, x i represents the actual value of the proofreading group, x′ i represents the predicted value of the proofreading group, c i =x i -x′ i Represents the absolute difference between the actual value and the predicted value. N represents the length of the original FM sequence, N>300.

[0030] The ratio error series reflects the relative error between the actual value and the predicted value of the proofreading group, and the formula is expressed as:

[0031]

[0032] Where N represents the length of the original FM sequence, N>300. i represents the i-th ratio error, x i represents the actual value of the proofreading group, x′ i represents the predicted value of the proofreading group, Indicates the relative error between the actual value and the predicted value.

[0033] As a preferred solution of the frequency modulation instruction prediction method of a thermal power unit coupled with supercapacitor energy storage described in the present invention, the difference error interval includes, after adjustment of the difference error sequence, the value of the prediction group changes from a numerical value to a range interval.

[0034] The formula is:

[0035]

[0036] Among them, x N+1 ,x N+2 ,…,x 1.1N Represents the original frequency modulation command value in the prediction group. β1,β2,…,β 1.1N Indicates the error range of the FM command value. N represents the length of the original FM sequence, N>300. exp represents the exponential function, tanh represents the activation function, and q represents the number of errors greater than 0 in the error sequence. i Represents the i-th element in the difference error sequence, (c1+c2+c3+...+c 0.1N ) / 0.1N represents the average value of the difference error, c max represents c1+c2+c3+…+c 0.1N The maximum value in .

[0037] As a preferred solution of the frequency modulation instruction prediction method of the thermal power unit coupled with supercapacitor energy storage according to the present invention, wherein: the ratio error interval includes, and the value of the prediction group after the ratio error sequence adjustment is expressed as:

[0038] [(x n+1 *k1,x n+1 *k1'),(x n+2 *k2,x n+2 *k2'),(x n+3 *k3,x n+3 *k3'),…,(x*k 1.1N ,x 1.1N *k 1.1N )]

[0039] Among them, x N+1 ,x N+2 ,…,x 1.1N k1, k2, k3, ..., k represents the original frequency modulation command value in the prediction group. 1.1N represents the adjustment coefficient in the ratio error sequence, k′1, k′2, k′3, …, k′ 1.1N Indicates that k1, k2, k3, ..., k 1.1N The corresponding upper limit coefficient. N represents the length of the original FM sequence, N>300.

[0040] 0 <k i <1,1 <k′ i .

[0041] k i Expressed as:

[0042]

[0043] [α1,α2,α3,…,α 0.1N ] is represented as:

[0044]

[0045] Among them, α1, α2, α3, α4 represent intermediate variables or adjustment coefficients. b1, b2, b3, …, b 0.1N Represents the individual error values ​​in the ratio error sequence, b max represents the maximum value in the ratio error sequence, represents the exponential part of the Gaussian distribution, Represents the Gaussian function. N represents the length of the original FM sequence, where N>300.

[0046] k′1 is expressed as:

[0047]

[0048] Among them, k′ i Indicates the adjustment coefficient of the upper limit of the ratio error range, M1, M2, M3, ..., M 0.1N represents the weight coefficient, Swish represents the activation function, and Mish represents the activation function. N represents the length of the original FM sequence, N>300. b1,b2,b3,…,b 0.1N Represents the individual error values ​​in the ratio error sequence. N represents the length of the original sequence, M i Indicates the adjustment coefficient, (b1+b2+…+b i-1 ) i-1 Represents the (i-1)th power of the cumulative sum of all errors before the current ratio error, (b i+1 +b i+2 +…+b 0.1N ) i Represents the i-th power of the cumulative sum of all errors after the current ratio error.

[0049] As a preferred solution of the frequency modulation instruction prediction method for a thermal power unit coupled with supercapacitor energy storage described in the present invention, the interval prediction value for obtaining the final frequency modulation instruction includes, for each value in the prediction group, giving an interval for the difference error interval and the ratio error interval respectively, and the final interval prediction value is the intersection of these two intervals.

[0050] A frequency modulation instruction prediction system for a thermal power unit coupled with supercapacitor energy storage, wherein:

[0051] The frequency modulation instruction prediction module processes the frequency modulation instructions of the thermal power unit and generates a frequency modulation instruction sequence. The prediction module is responsible for training and predicting the input frequency modulation sequence.

[0052] The error calculation module calculates the proportional error and the difference error, and generates the corresponding error sequences. By making error adjustments to the prediction group results, a probabilistic prediction range is formed.

[0053] The probability range prediction module combines the prediction results of the proportional error and the difference error to obtain the intersection of the prediction intervals, ultimately forming a more accurate prediction range.

[0054] The coupled energy storage control module uses the predicted frequency modulation command results to control the power distribution between the supercapacitor and the lithium battery, and optimizes the adjustment strategy of the capacitor and battery during the frequency modulation process.

[0055] A computer device includes a memory and a processor, wherein the memory stores a computer program, and wherein the processor implements the steps of any one of the methods of the present invention when executing the computer program.

[0056] A computer-readable storage medium stores a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of any one of the methods of the present invention.

[0057] Beneficial effects of the present invention: The frequency modulation instruction prediction method for a thermal power unit coupled with supercapacitor energy storage provided by the present invention solves the problem of large errors in traditional numerical predictions by introducing a BP neural network to train the frequency modulation instruction sequence and combining the probabilistic range prediction of the difference error and the ratio error. This method significantly improves the prediction accuracy of the frequency modulation instruction and avoids the uncertainty brought about by a single numerical prediction. At the same time, the present invention optimizes the power distribution of the energy storage device by coupling the energy storage system of the supercapacitor and the lithium battery, effectively improving the response speed of the system and the energy storage utilization rate. This technology not only improves the stability of the frequency modulation of the thermal power unit, but also significantly reduces the coal consumption and maintenance costs of the unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0059] Figure 1 This is an overall flow chart of a method for predicting frequency modulation instructions for a thermal power unit coupled with supercapacitor energy storage, provided as a first embodiment of the present invention. DETAILED DESCRIPTION

[0060] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0061] Example 1, reference Figure 1 , as one embodiment of the present invention, provides a method for predicting frequency modulation instructions for a thermal power unit coupled with supercapacitor energy storage, comprising:

[0062] S1: Obtain the original frequency modulation sequence through the frequency modulation instructions of the thermal power unit and divide it into training group, proofreading group and prediction group.

[0063] The original FM sequence includes, assuming the length of the original FM sequence is N, the formula is expressed as:

[0064] [x1,x2,x3,…,x N ]

[0065] The proofreading group is the first 90% of the original FM sequence, which is used to train the neural network model. The formula is:

[0066] [x1,x2,x3,…,x 0.9N ]

[0067] The prediction group is the last 10% of the original FM sequence and is used to generate the proportional error and the difference error. The formula is:

[0068] [x 0.9N+1 ,x 0.9N+2 ,…,x N ]

[0069] The prediction group is the unknown 10% of the original FM sequence in the future, and the formula is:

[0070] [x N+1 ,x N+2 ,...,x 1.1N ]

[0071] Where N represents the length of the original FM sequence, N>300. i Indicates the value of the i-th frequency modulation instruction.

[0072] Furthermore, by serializing the frequency regulation instructions of thermal power units and dividing the data into training, proofreading, and prediction groups, it is helpful to establish a reasonable model training and verification mechanism to ensure that the prediction algorithm can not only learn from historical data but also verify its generalization performance on new data.

[0073] S2: Train the training group through BP neural network to generate prediction values ​​for the proofreading group.

[0074] Generating the predicted value of the proofreading group includes inputting the training group into the BP neural network for training to obtain the predicted value of the proofreading group and the value of the prediction group.

[0075] The predicted value of the proofreading group is expressed as:

[0076] [x′ 0.9N+1 ,x′ 0.9N+2 ,...,x′ N ]

[0077] Where N represents the length of the original FM sequence, N>300. i Indicates the value of the i-th frequency modulation instruction.

[0078] Furthermore, a BP neural network was used to train the training data, leveraging its powerful nonlinear mapping capabilities to effectively capture the complex frequency modulation instruction change patterns and generate prediction values, which provided an accurate basis for subsequent error analysis.

[0079] S3: Based on the actual values ​​and predicted values ​​of the proofreading group, the difference error and the ratio error are calculated respectively to form a difference error sequence and a ratio error sequence.

[0080] The actual and predicted values ​​of the calibration group will produce difference error series and ratio error series respectively.

[0081] The difference error sequence reflects the absolute difference between the actual value and the predicted value of the proofreading group, and the formula is expressed as:

[0082] [c1,c2,c3,…,c 0.1N ]c i =x i -x i

[0083] Among them, c i represents the i-th difference error, x i represents the actual value of the proofreading group, x′ i represents the predicted value of the proofreading group, c i =x i -x′ i Represents the absolute difference between the actual value and the predicted value. N represents the length of the original FM sequence, N>300.

[0084] The ratio error series reflects the relative error between the actual value and the predicted value of the proofreading group, and the formula is expressed as:

[0085]

[0086] Where N represents the length of the original FM sequence, N>300. i represents the i-th ratio error, x i represents the actual value of the proofreading group, x′ i represents the predicted value of the proofreading group, Indicates the relative error between the actual value and the predicted value.

[0087] Furthermore, by calculating the difference error and the ratio error, we can measure the accuracy of the prediction from both absolute and relative perspectives. This dual error analysis method can fully reflect the performance of the model at different scales, ensuring more reliable prediction results.

[0088] S4: Adjust the frequency modulation instruction value of the prediction group to generate a difference error interval.

[0089] The difference error interval includes that after the difference error sequence is adjusted, the value of the prediction group is changed from a numerical value to a range interval.

[0090] The formula is:

[0091]

[0092] Among them, x N+1 ,x N+2 ,…,x 1.1N Represents the original frequency modulation command value in the prediction group. β1,β2,…,β 1.1N Indicates the error range of the FM command value. N represents the length of the original FM sequence, N>300. exp represents the exponential function, tanh represents the activation function, and q represents the number of errors greater than 0 in the error sequence. i Represents the i-th element in the difference error sequence, (c1+c2+c3+...+c 0.1N ) / 0.1N represents the average value of the difference error, c max represents c1+c2+c3+…+c 0.1N The maximum value in .

[0093] Furthermore, the frequency modulation command values ​​of the prediction group are adjusted using the difference error sequence to generate an error interval. This error interval reflects the uncertainty range of the prediction value, providing a more flexible reference for frequency modulation decisions and reducing decision errors caused by prediction errors.

[0094] S5: Adjust the frequency modulation instruction value of the prediction group according to the ratio error sequence to generate a ratio error interval.

[0095] The ratio error interval includes the values ​​of the forecast group after the ratio error series adjustment as follows:

[0096] [(x n+1*k1,x n+1 *k1'),(x n+2 *k2,x n+2 *k2'),(x n+3 *k3,x n+3 *k3'),…,(x*k 1.1N ,x 1.1N *k 1.1N )]

[0097] Among them, x N+1 ,x N+2 ,…,x 1.1N k1, k2, k3, ..., k represents the original frequency modulation command value in the prediction group. 1.1N represents the adjustment coefficient in the ratio error sequence, k′1, k′2, k′3, …, k′ 1.1N Indicates that k1, k2, k3, ..., k 1.1N The corresponding upper limit coefficient. N represents the length of the original FM sequence, N>300.

[0098] 0 <k i <1,1 <k′ i .

[0099] k i Expressed as:

[0100]

[0101] [α1,α2,α3,…,α 0.1N ] is represented as:

[0102]

[0103] Among them, α1, α2, α3, α4 represent intermediate variables or adjustment coefficients. b1, b2, b3, …, b 0.1N Represents the individual error values ​​in the ratio error sequence, b max represents the maximum value in the ratio error sequence, represents the exponential part of the Gaussian distribution, Represents the Gaussian function. N represents the length of the original FM sequence, where N>300.

[0104] k′1 is expressed as:

[0105]

[0106] Among them, k i Indicates the adjustment coefficient of the upper limit of the ratio error range, M1, M2, M3, ..., M 0.1Nrepresents the weight coefficient, Swish represents the activation function, and Mish represents the activation function. N represents the length of the original FM sequence, N>300. b1,b2,b3,…,b 0.1N Represents the individual error values ​​in the ratio error sequence. N represents the length of the original sequence, M i Indicates the adjustment coefficient, (b1+b2+…+b i-1 ) i-1 Represents the (i-1)th power of the cumulative sum of all errors before the current ratio error, (b i+1 +b i+2 +…+b 0.1N ) i Represents the i-th power of the cumulative sum of all errors after the current ratio error.

[0107] Furthermore, the prediction group is adjusted using the ratio error to further improve the accuracy of the prediction interval. The ratio error can effectively deal with the deviation of the predicted value on the scale, ensuring that the prediction interval is consistent with the actual situation over a larger range.

[0108] S6: Combine the difference error interval and the ratio error interval to obtain the interval prediction value of the final frequency modulation instruction.

[0109] The method of obtaining the interval prediction value of the final frequency modulation instruction includes, for each value in the prediction group, giving an interval of the difference error interval and the ratio error interval respectively, and the final interval prediction value is the intersection of these two intervals.

[0110] The value of the ratio error prediction group becomes [(x n+1 *k1,x n+1 *k1'),(x n+2 *k2,x n+2 *k2'),(x n+3 *k3,x n+3 *k3'),…,(x*k 1.1N ,x 1.1N *k 1.1N )]. The value of the difference error prediction group becomes (x N+1 -β1,x N+1 +β1),(x N+2 -β2,x N+2 +β2),…,(x 1.1N -β 1.1N ,x 1.1N +β 1.1N ).

[0111] Now we need to couple the results of these two methods.

[0112] The specific method is to take the intersection of these two results. For example, x N+1 =1,βi =0.1,β i =0.15, then the difference error group prediction for x N+1 The interval division is [0.91.15], and k i =0.95,k i =1.2 then the ratio error prediction group x N+1 The interval division is [0.951.20], so the final probability distribution interval is [0.91.15]∩[0.951.20]=[0.951.15]. The remaining x n+2 ...and so on. Following this method, we get the interval distribution of the final predicted value.

[0113] Furthermore, by combining the difference error interval with the ratio error interval, a more accurate interval prediction value is obtained. The significance of this design is to make full use of different types of error information to further improve the stability and accuracy of the prediction results, thereby providing a reliable basis for the frequency regulation control of thermal power units.

[0114] Example 2 is an embodiment of the present invention, which provides a frequency modulation instruction prediction system for a thermal power unit coupled with supercapacitor energy storage, including: a frequency modulation instruction prediction module, an error calculation module, a probability range prediction module, and a coupled energy storage control module.

[0115] The frequency modulation instruction prediction module is used to generate a frequency modulation instruction sequence by processing the frequency modulation instructions of the thermal power unit. The prediction module is responsible for training and predicting the input frequency modulation sequence.

[0116] The error calculation module is used to calculate the proportional error and the difference error, and generate the corresponding error sequences. By making error adjustments to the prediction group results, a probabilistic prediction range is formed.

[0117] The probability range prediction module is used to obtain the intersection of the prediction intervals by combining the prediction results of the proportional error and the difference error, and finally form a more accurate prediction range.

[0118] The coupled energy storage control module is used to use the predicted frequency modulation command results to control the power distribution between the supercapacitor and the lithium battery, and optimize the adjustment strategy of the capacitor and battery during the frequency modulation process.

[0119] Example 3, an embodiment of the present invention, is different from the previous two embodiments in that:

[0120] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0121] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0122] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0123] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0124] In Example 4, the experiment simulates a frequency modulation response system of a thermal power unit. The system includes a frequency modulation control module of the thermal power unit, energy storage devices such as supercapacitors and lithium batteries, and a BP neural network module for data processing and prediction.

[0125] At the outset of the experiment, historical frequency modulation command data was extracted from the thermal power unit dispatching system to form an original frequency modulation sequence. This frequency modulation sequence was preprocessed to remove outliers and noise data and then divided into a training set, a calibration set, and a prediction set. The training set contained 90% of the original frequency modulation data and was used to train the neural network model; the calibration set consisted of 10% of the historical data and was used to correct and generate error sequences; and the prediction set contained future frequency modulation command data, used to generate the final frequency modulation command interval prediction value.

[0126] During the data processing phase, a BP neural network is used to train the training set. The neural network consists of a three-layer, fully connected network. The input layer receives the frequency modulation instruction data sequence, the hidden layer contains 20 neurons, and the output layer generates the predicted values ​​for the proofreading set. By comparing the actual values ​​of the proofreading set with the predicted values, the difference error and ratio error series are calculated, respectively. The difference error series represents the absolute deviation between the predicted and actual values, while the ratio error series measures the relative deviation of the predicted values ​​from the actual values.

[0127] Next, the frequency modulation command values ​​for the prediction group are adjusted. First, the prediction group is adjusted based on the difference error sequence to generate a difference error interval. The prediction range for each frequency modulation command value is determined by the error sequence. Then, the prediction group is further adjusted based on the ratio error sequence to generate a ratio error interval. Finally, the difference error interval and the ratio error interval are combined to obtain the final frequency modulation command interval prediction value. This process effectively reduces the prediction error and improves the accuracy and efficiency of the frequency modulation response. The experimental data is shown in Table 1.

[0128] Table 1 Experimental data table

[0129]

[0130] The experimental data demonstrates that the proposed frequency modulation instruction prediction method has significant advantages. Both the difference error and the ratio error between the predicted values ​​generated by the BP neural network and the actual values ​​of the proofreading group remain within a relatively small range, demonstrating high prediction accuracy. Furthermore, the final prediction interval effectively covers the actual values ​​of the proofreading group, demonstrating that combining the difference error interval and the ratio error interval prediction method can significantly reduce the error bias in traditional numerical prediction.

[0131] Compared to existing technologies, traditional numerical prediction methods typically directly output a specific predicted value. However, due to the volatility of frequency modulation commands in power systems, a single-value prediction often fails to accurately reflect actual demand, easily causing delayed or excessive frequency modulation responses. The method of the present invention produces a prediction result as an interval value. This probabilistic range prediction can better reflect the uncertainty in frequency modulation commands, reduce the error range of frequency modulation responses, and thus improve the stability and response speed of the power system.

[0132] The data in the table further demonstrates the effectiveness of the present invention in reducing errors. For example, the difference error of frequency modulation command 1 is -4 MW, and the ratio error is 3.92%. After interval prediction, the actual value of 102 MW falls within the final predicted range of 98.5-103.5 MW. This effect has been verified in multiple tests of frequency modulation commands, demonstrating that the frequency modulation command prediction method of the present invention can provide stable and reliable prediction results under different operating conditions.

[0133] Furthermore, the mean absolute percentage error (MAPE) measures the error between the frequency modulation command and the actual response. Experimental data shows that the MAPE of the present invention is significantly lower than that of traditional BP prediction methods. While the MAPE of traditional BP prediction methods is approximately 10-17%, the present invention, through a range prediction method, controls the MAPE to between 1.5-6%, demonstrating that the present invention significantly improves the accuracy and stability of frequency modulation command prediction. This further validates the feasibility and advantages of this method in practical applications.

[0134] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting frequency modulation instructions for a thermal power unit coupled with supercapacitor energy storage, characterized in that: include: The original frequency modulation sequence is obtained through the frequency modulation instructions of the thermal power unit and divided into a training group, a calibration group and a prediction group; The training group is trained through BP neural network to generate the predicted value of the proofreading group; Based on the actual values ​​and predicted values ​​of the proofreading group, the difference error and the ratio error are calculated respectively to form a difference error series and a ratio error series; Adjust the frequency modulation command value of the prediction group to generate a difference error interval; Adjust the frequency modulation instruction value of the prediction group according to the ratio error sequence to generate a ratio error interval; The difference error interval and the ratio error interval are combined to obtain the interval prediction value of the final frequency modulation instruction.

2. The method for predicting frequency modulation instructions for a thermal power unit coupled with supercapacitor energy storage according to claim 1, characterized in that: The original frequency modulation sequence includes, assuming that the length of the original frequency modulation sequence is N, the formula is expressed as: [x1,x2,x3,…,x N ] The proofreading group is the first 90% of the original FM sequence, which is used to train the neural network model. The formula is: [x1,x2,x3,…,x 0.9N ] The prediction group is the last 10% of the original frequency modulation sequence, which is used to generate the proportional error and the difference error; the formula is expressed as: [x 0.9N+1 ,x 0.9N+2 ,…,x N ] The prediction group is the unknown 10% of the original FM sequence in the future, and the formula is: [x N+1 ,x N+2 ,...,x 1.1N ] Where N represents the length of the original FM sequence, N>300; x i Indicates the value of the i-th frequency modulation instruction.

3. The method for predicting frequency modulation instructions for a thermal power unit coupled with supercapacitor energy storage according to claim 2, characterized in that: Generating the predicted value of the proofreading group includes inputting the training group into the BP neural network for training to obtain the predicted value of the proofreading group and the value of the prediction group; The predicted value of the proofreading group is expressed as: [x′ 0.9N+1 ,x′ 0.9N+2 ,...,x′ N ] Where N represents the length of the original FM sequence, N>300; x i Indicates the value of the i-th frequency modulation instruction.

4. The method for predicting frequency modulation instructions for a thermal power unit coupled with supercapacitor energy storage according to claim 3, characterized in that: The actual and predicted values ​​of the calibration group will produce difference error series and ratio error series respectively; The difference error sequence reflects the absolute difference between the actual value and the predicted value of the proofreading group, and the formula is expressed as: [c1,c2,c3,…,c 0.1N ]c i =x i -x i Among them, c i represents the i-th difference error, x i represents the actual value of the proofreading group, x′ i represents the predicted value of the proofreading group, c i =x i -x′ i Represents the absolute difference between the actual value and the predicted value; N represents the length of the original FM sequence, N>300; The ratio error series reflects the relative error between the actual value and the predicted value of the proofreading group, and the formula is expressed as: Where N represents the length of the original FM sequence, N>300; b i represents the i-th ratio error, x i represents the actual value of the proofreading group, x′ i represents the predicted value of the proofreading group, Indicates the relative error between the actual value and the predicted value.

5. The method for predicting frequency modulation instructions for a thermal power unit coupled with supercapacitor energy storage according to claim 4, characterized in that: The difference error interval includes that after the difference error sequence is adjusted, the value of the prediction group changes from a numerical value to a range interval; The formula is: (x N+1 -β1,x N+1 +β1),(x N+2 -β2,x N+2 +β2),…,(x 1.1N -b 1.1N ,x 1.1N +b 1.1N ) Among them, x N+1 ,x N+2 ,…,x 1.1N represents the original frequency modulation command value in the prediction group; β1,β2,…,β 1.1N Indicates the error range of the FM command value; N represents the length of the original FM sequence, N>300; exp represents the exponential function, tanh represents the activation function, and q represents the number of errors greater than 0 in the error sequence; c i Represents the i-th element in the difference error sequence, (c1+c2+c3+...+c 0.1N ) / 0.1N represents the average value of the difference error, c max represents c1+c2+c3+…+c 0.1N The maximum value in .

6. The method for predicting frequency modulation instructions for a thermal power unit coupled with supercapacitor energy storage according to claim 5, characterized in that: The ratio error interval includes the value of the prediction group after the ratio error sequence adjustment is expressed as: [(x n+1 *k1,x n+1 *k1’),(x n+2 *k2,x n+2 *k2’),(x n+3 *k3,x n+3 *k3’),…,(x*k 1.1N ,x 1.1N *k 1.1N )] Among them, 0 <k i <1,1 <k′ i , x N+1 ,x N+2 ,…,x 1.1N represents the original frequency modulation command value in the prediction group; k1, k2, k3, ..., k 1.1N represents the adjustment coefficient in the ratio error sequence, k′1, k′2, k′3, …, k′ 1.1N Indicates that k1, k2, k3, ..., k 1.1N The corresponding upper limit coefficient; N represents the length of the original FM sequence, N>300; k i Expressed as: [a1,a2,a3,…,a 0.1N ] shown as: Among them, α1, α2, α3, α4 represent intermediate variables or adjustment coefficients; b1, b2, b3, ..., b 0.1N Represents the individual error values ​​in the ratio error sequence, b max represents the maximum value in the ratio error sequence, represents the exponential part of the Gaussian distribution, represents the Gaussian function; N represents the length of the original FM sequence, N>300; k′1 is expressed as: Among them, k′ i Indicates the adjustment coefficient of the upper limit of the ratio error range, M1, M2, M3, ..., M 0.1N represents the weight coefficient, Swish represents the activation function, Mish represents the activation function; N represents the length of the original FM sequence, N>300; b1, b2, b3,…, b 0.1N Represents the error values ​​in the ratio error sequence; N represents the length of the original sequence, M i Indicates the adjustment coefficient, (b1+b2+…+b i-1 ) i-1 Represents the (i-1)th power of the cumulative sum of all errors before the current ratio error, (b i+1 +b i+2 +…+b 0.1N ) i Represents the i-th power of the cumulative sum of all errors after the current ratio error.

7. The method for predicting frequency modulation instructions for a thermal power unit coupled with supercapacitor energy storage according to claim 6, characterized in that: The interval prediction value of the final frequency modulation instruction is obtained, for each value in the prediction group, a difference error interval and a ratio error interval are respectively given an interval, and the final interval prediction value is the intersection of these two intervals.

8. A frequency modulation instruction prediction system for a thermal power unit coupled with supercapacitor energy storage using the method according to any one of claims 1 to 7, characterized in that: Including frequency modulation instruction prediction module, error calculation module, probability range prediction module, coupled energy storage control module; The frequency modulation instruction prediction module is used to generate a frequency modulation instruction sequence by performing data processing on the frequency modulation instructions of the thermal power generation unit; The prediction module is responsible for training and predicting the input FM sequence; The error calculation module is used to calculate the proportional error and the difference error, and generate corresponding error sequences respectively; by performing error adjustment on the results of the prediction group, a probabilistic prediction range is formed; The probability range prediction module is used to obtain the intersection of the prediction intervals by combining the prediction results of the proportional error and the difference error, and ultimately form a more accurate prediction range; The coupled energy storage control module is used to use the predicted frequency modulation instruction result to control the power distribution between the supercapacitor and the lithium battery, and optimize the adjustment strategy of the capacitor and the battery during the frequency modulation process.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for predicting frequency modulation instructions of a thermal power unit coupled with supercapacitor energy storage according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting frequency modulation instructions of a thermal power unit coupled with supercapacitor energy storage according to any one of claims 1 to 7 are implemented.

Images