New energy power generation power prediction method and electrolytic cell load power control method

Abstract

The present specification relates to the technical field of new energy power generation, and particularly relates to a new energy power generation power prediction method and an electrolytic cell load power control method. The method comprises: acquiring input data, wherein the input data comprises first historical power data and meteorological data at a future time; predicting, according to the input data, a first probability distribution parameter of a new energy power generation device at the future time through a model; determining a probability value of the new energy power generation device at the future time according to the first probability distribution parameter; and calculating a power generation power interval of the new energy power generation device at the future time according to the probability value, wherein the power generation power interval is used to regulate a load power of an electrolytic cell at the future time, and the new energy power generation device is used to supply power to the electrolytic cell. The embodiments of the present specification can predict the power generation power interval of the new energy power generation device at the future time, thereby improving the accuracy of the load regulation of the electrolytic cell.

Description

Technical Field

[0001] This specification relates to the field of new energy power generation technology, and in particular to a new energy power generation power prediction method and an electrolytic cell load power control method. Background Technology

[0002] In recent years, utilizing new energy power generation equipment to power electrolyzers for green hydrogen production has become an important path to promote energy structure transformation. In hydrogen production systems, new energy power generation equipment such as photovoltaics and wind power can supply power to electrolyzers to drive their operation. These electrolyzers produce hydrogen through water electrolysis, which is then stored or utilized. However, due to weather conditions, new energy power generation is characterized by randomness, intermittency, and fluctuation, making it difficult to maintain stability and posing a serious challenge to the safe and economical operation of hydrogen production systems. Related technologies can predict the power output of new energy power generation equipment and provide the prediction results to the control equipment of the hydrogen production system, enabling the electrolyzer to adjust its load in advance based on the predicted power output.

[0003] However, in the aforementioned related technologies, the prediction of power generation from new energy power generation equipment is a single-point prediction. The prediction result includes the specific power generation value of the new energy power generation equipment at a future time. The power generation value cannot quantify the fluctuation of power generation and is difficult to characterize the possible range of power generation, thus hindering the regulation of electrolyzer load. Summary of the Invention

[0004] This specification provides a method for predicting the power generation capacity of new energy sources and a method for controlling the load power of electrolyzers, which are used to predict the power generation range of new energy power generation equipment in the future, so as to improve the accuracy of load regulation of electrolyzers.

[0005] This specification provides an embodiment of a method for predicting the power generation of new energy sources, including: Acquire input data, which includes first historical power data and historical meteorological data for future times; Based on the input data, the model predicts the first probability distribution parameters of the new energy power generation equipment at the future time. Based on the first probability distribution parameters, determine the probability value of the new energy power generation equipment at the future time. Based on the probability value, the power generation range of the new energy power generation equipment at the future time is calculated. The power generation range is used to regulate the load power of the electrolyzer at the future time. The new energy power generation equipment is used to supply power to the electrolyzer.

[0006] In some embodiments, the new energy power generation equipment corresponds to N discrete power points, where N>1; the step of predicting the first probability distribution parameters of the new energy power generation equipment at the future time using a model includes: The model is used to predict the N first probability distribution parameters of new energy power generation equipment at N discrete power points; Determining the probability value of the new energy power generation equipment at the future time includes: Determine the first probability mass function of the N discrete power points based on the first probability distribution parameters of the N discrete power points; Substitute each discrete power point into the corresponding first probability mass function to obtain the probability value of that discrete power point; The calculation of the power generation range of the new energy power generation equipment at the future time includes: Based on the probability values ​​of N discrete power points, calculate the power generation range of the new energy power generation equipment at the future time.

[0007] In some embodiments, the model includes an input layer, a temporal modeling layer, a convolutional layer, a transformer coding layer, a pooling and fully connected fusion layer, and N probability output heads; the step of predicting N first probability distribution parameters of new energy power generation equipment at N discrete power points through the model includes: Based on the input data, the original feature data is determined through the input layer; Based on the original feature data, the hidden state data is determined through the temporal modeling layer; Based on the hidden state data, convolutional feature data is determined through the convolutional layer. The convolutional feature data is used to represent the short-term variation characteristics of power and weather. Based on the convolutional feature data, encoded feature data is determined through the transformer coding layer. The encoded feature data is used to represent the long-term dependence of power on weather conditions. Based on the encoded feature data, the fused feature data is determined through the pooling and fully connected fusion layer; Based on the fused feature data, N first probability distribution parameters are determined through the N probability output heads.

[0008] In some embodiments, calculating the power generation range of the new energy power generation equipment at the future time includes: Determine the cumulative probability of N discrete power points based on their probability values. The upper bound of power generation is determined based on the preset upper quantile and the cumulative probability of the N discrete power points; The lower bound of power generation is determined based on the preset lower quantile and the cumulative probability of the N discrete power points; The upper and lower limits of power generation are used to form the power generation range.

[0009] In some embodiments, calculating the power generation range of the new energy power generation equipment at the future time includes: Based on the probability values ​​of N discrete power points, multiple candidate power generation intervals are determined, and the sum of the probability values ​​of each discrete power point in each candidate power generation interval is greater than or equal to the preset confidence level. Among the multiple candidate power generation ranges, the power generation range with the smallest width is selected.

[0010] In some embodiments, the model is trained in the following manner: Acquire training samples, which include feature data and label data. The feature data includes second historical power data and second historical meteorological data. The label data includes third historical power data. The third historical power data and the second historical meteorological data correspond to the same historical moment. The historical moment corresponding to the second historical power data is earlier than the historical moment corresponding to the third historical power data and the second historical meteorological data. Based on the aforementioned feature data, the second probability distribution parameters of new energy power generation equipment are predicted using a model. Loss data is determined based on label data and second probability distribution parameters; The parameters of the model are adjusted based on the loss data.

[0011] In some embodiments, the new energy power generation equipment corresponds to N discrete power points, where N>1; the prediction of the second probability distribution parameters of the new energy power generation equipment through a model includes: The model is used to predict the N second probability distribution parameters of new energy power generation equipment at N discrete power points; The step of determining the loss data based on the label data and the second probability distribution parameters includes: Determine the second probability mass function of the N discrete power points based on the second probability distribution parameters of the N discrete power points; Substitute each discrete power point into the corresponding second probability mass function to obtain the probability value of that discrete power point; Determine the loss data based on the loss function; The loss function includes a first loss term, a second loss term, a third loss term, a fourth loss term, and a fifth loss term; the first loss term is used to constrain the predicted probability distribution corresponding to the N predicted probability values ​​to be similar to the true probability distribution corresponding to the N true probability values; the second loss term is used to constrain the cumulative probability of the N discrete power points to be monotonically increasing; the third loss term is used to constrain the predicted probability value of the boundary power point; the fourth loss term is used to constrain the degree of difference between the predicted probability values ​​of adjacent discrete power points; and the fifth loss term is used to constrain the model weights.

[0012] This specification also provides an embodiment of an electrolytic cell load power control method, including: Receive the power generation range of new energy power generation equipment at future points in time; The load power of the electrolyzer at the future time is adjusted according to the power generation range.

[0013] In some embodiments, the power generation range includes an upper limit and a lower limit; the regulation of the load power of the electrolyzer at the future time includes: When the upper limit of the power generation is greater than the rated power of the electrolytic cell, the energy storage device of the electrolytic cell is controlled to charge. When the lower limit of the power generation is less than the minimum power of the electrolytic cell, the energy storage device of the electrolytic cell is controlled to discharge. When the upper limit of the power generation is less than or equal to the rated power of the electrolytic cell, and the lower limit of the power generation is greater than or equal to the minimum power of the electrolytic cell, the actual operating power of the electrolytic cell is controlled to be within the power generation range.

[0014] This specification also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-mentioned new energy power generation prediction method and / or the above-mentioned electrolytic cell load power control method.

[0015] This specification also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described new energy power generation prediction method and / or the above-described electrolytic cell load power control method.

[0016] This specification also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-mentioned new energy power generation prediction method and / or the above-mentioned electrolytic cell load power control method.

[0017] The technical solution of the embodiments in this specification can acquire input data, including first historical power data and meteorological data for future times. Based on the input data, a model can predict a first probability distribution parameter of the new energy power generation equipment at the future time. Based on the first probability distribution parameter, a probability value of the new energy power generation equipment at the future time can be determined. Based on the probability value, the power generation range of the new energy power generation equipment at the future time can be calculated. The power generation range is used to regulate the load power of the electrolyzer at the future time, and the new energy power generation equipment is used to supply power to the electrolyzer. Therefore, the embodiments in this specification can predict the power generation range of the new energy power generation equipment at future times. Compared to predicting a specific power generation value, predicting a power generation range can quantify the fluctuation of the power generation of the new energy power generation equipment and characterize the possible range of power generation, thus providing convenience for the regulation of the electrolyzer load. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. The drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the new energy power generation prediction method in the embodiments of this specification; Figure 2 This is a schematic diagram of the architecture of the model in the embodiments of this specification; Figure 3 This is a flowchart illustrating the electrolytic cell load power control method in the embodiments of this specification; Figure 4 This is a functional structure diagram of the new energy power generation prediction device in the embodiments of this specification; Figure 5 This is a functional structural diagram of the electrolytic cell load power control device in the embodiments of this specification. Detailed Implementation

[0020] The technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. The specific embodiments described herein are only used to explain this disclosure, and not to limit this disclosure. All other embodiments obtained by those skilled in the art based on the described embodiments of this disclosure are within the scope of protection of this disclosure. In addition, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0021] Existing methods for predicting the power output of renewable energy generation equipment are single-point predictions, providing only a deterministic power output value at a future moment. Such methods cannot quantify the fluctuation risk of power output or characterize the potential range of power output. For electrolyzer regulation, relying solely on a single power output value cannot assess the potential risk of the actual power output deviating from the predicted power output. When there is a significant deviation between the actual and predicted power output of renewable energy generation equipment, the electrolyzer may struggle to respond promptly, potentially leading to wind and solar power curtailment or damage to the electrolyzer's membrane electrodes. Furthermore, single-point predictions are insufficient to support optimal charging and discharging decisions for energy storage systems, limiting the economic efficiency and safety of the entire hydrogen production system. In addition, the time granularity of existing renewable energy generation equipment power output predictions is typically on the order of minutes, while the load regulation rate of electrolyzers can reach the order of seconds; this mismatch in time scale further exacerbates the difficulty of regulating the electrolyzer load.

[0022] Please see Figure 1 and Figure 2 This specification provides an embodiment of a method for predicting the power generation capacity of a new energy source.

[0023] The new energy power generation prediction method can be applied to power prediction equipment, and specifically includes the following steps.

[0024] Step 11, obtain input data, which includes first historical power data and historical meteorological data for future times; Step 12: Based on the input data, predict the first probability distribution parameters of the new energy power generation equipment at future times using the model; Step 13: Determine the probability value of the new energy power generation equipment at future times based on the first probability distribution parameters; Step 14: Calculate the power generation range of the new energy power generation equipment in the future based on the probability value. The power generation range is used to regulate the load power of the electrolytic cell in the future. The new energy power generation equipment is used to supply power to the electrolytic cell.

[0025] The technical solution of the embodiments in this specification can acquire input data, including first historical power data and meteorological data for future times. Based on the input data, a model can predict a first probability distribution parameter of the new energy power generation equipment at the future time. Based on the first probability distribution parameter, a probability value of the new energy power generation equipment at the future time can be determined. Based on the probability value, the power generation range of the new energy power generation equipment at the future time can be calculated. The power generation range is used to regulate the load power of the electrolyzer at the future time, and the new energy power generation equipment is used to supply power to the electrolyzer. Therefore, the embodiments in this specification can predict the power generation range of the new energy power generation equipment at future times. Compared to predicting a specific power generation value, predicting a power generation range can quantify the fluctuation of the power generation of the new energy power generation equipment and characterize the possible range of power generation, thus providing convenience for the regulation of the electrolyzer load.

[0026] In some embodiments, new energy power generation equipment refers to equipment that generates electricity using renewable natural resources. New energy power generation equipment may include photovoltaic power generation equipment, wind power generation equipment, etc. In a hydrogen production system, new energy power generation can supply power to an electrolyzer. The electrolyzer may include a hydrogen production electrolyzer. The electrolyzer produces hydrogen by electrolyzing water and then stores or utilizes it. Limited by meteorological conditions, new energy power generation is characterized by randomness, intermittency, and fluctuation, making it difficult to maintain stability. Therefore, it is necessary to predict the power output of new energy power generation in order to adaptively adjust the electrolyzer load so that the power generation output matches the electrolyzer load.

[0027] In some embodiments, the input data may include first historical power data of the renewable energy power generation equipment at one or more historical moments. The first historical power data may include the actual power generation of the renewable energy power generation equipment at those historical moments. For example, the input data may include multiple first historical power data points within a time window preceding the current moment. The length of the time window may be 1 day, 1 week, etc. The end point of the time window is the current moment. The input data may also include meteorological data for one or more future moments. The future moments may be the predicted moments for power generation. Meteorological data may include wind speed, irradiance, temperature, etc.

[0028] Optionally, the input data may also include first historical meteorological data for one or more historical moments.

[0029] Optionally, the input data may also include one or more historical moments and one or more future moments.

[0030] For example, the input data may include meteorological data at time t0, first historical power data at time t1, and first historical power data at time t2. Time t0 is a future time, while times t1 and t2 are historical times. Times t1 and t2 are earlier than time t0. Using this input data, the power generation range of the new energy power generation equipment at time t0 can be predicted.

[0031] In some embodiments, the input data can be feature-engineered to obtain a feature data sequence. The feature data sequence includes several feature data arranged in chronological order. Each feature data corresponds to a time point, which can be a historical time point or a future time point. When a feature data corresponds to a historical time point, it can be obtained from one or more of the following: first historical power data for that historical time point, meteorological data for that historical time point, or the historical time point itself. When a feature data corresponds to a future time point, it can be obtained from one or more of the following: meteorological data for that future time point, or the future time point itself. Each feature data point can include one or more of the following: power generation characteristics, meteorological characteristics, and time characteristics. The power generation characteristics can include actual power generation; the meteorological characteristics can include wind speed, irradiance, temperature, etc.; and the time characteristics are obtained by encoding the time point data.

[0032] For example, the time features include hourly encoding features, weekday encoding features, month encoding features, etc. The hourly encoding features may include... , etc., where h represents the hour information of the time data. The weekday encoding feature may include , And so on, where w represents the weekday information of the time data. The month encoding feature may include , , w represents the month information of the time data.

[0033] In some embodiments, the installed capacity of the new energy power generation equipment can be obtained; the installed capacity of the new energy power generation equipment can be discretized according to the power resolution to obtain N discrete power points. N>1. N is a positive integer. The power resolution can be a preset value. Among the N discrete power points, the difference between adjacent discrete power points is the power resolution. Each discrete power point can be less than or equal to the installed capacity, specifically understood as one power generation capacity of the new energy power generation equipment.

[0034] For example, if the installed capacity of new energy power generation equipment is Cap=300MW and the power resolution is 5MW, then the installed capacity Cap=300MW can be discretized according to the power resolution, resulting in 61 discrete power points {0, 5, 10, ..., 300}.

[0035] In some embodiments, the feature data sequence can be input into a trained model to obtain the model's output data. The output data may include the first probability distribution parameters of the new energy power generation equipment at M future time points. M ≥ 1. M is a positive integer. The M future time points are consecutive. The time interval between adjacent future time points can be 1 minute, 2 minutes, 5 minutes, etc.

[0036] The first probability distribution parameters for each future time point include N first probability distribution parameters for the new energy power generation equipment at N discrete power points. Each first probability distribution parameter corresponds to one discrete power point. The first probability distribution parameters are used to represent probability distributions. For example, the first probability distribution parameters include the mean and variance, and are used to represent a Gaussian distribution.

[0037] For each of the M future times, N first probability quality functions can be determined based on the N discrete power points at that future time and their corresponding N first probability distribution parameters. Each first probability quality function corresponds to a discrete power point. For example, each first probability distribution parameter includes the mean and variance, and a Gaussian distribution function can be determined as the first probability quality function based on the mean and variance. Thus, each discrete power point can correspond to a first probability quality function. Substituting each discrete power point into the corresponding first probability quality function yields the probability value of that discrete power point. The probability value of the discrete power point represents the probability that the power generation of the new energy power generation equipment is at that discrete power point. Therefore, for each of the M future times, N probability values ​​of discrete power points can be obtained. The i-th discrete power point and its corresponding probability value can be represented as {P}. i ,p i}. P i p represents the discrete power point. i This represents the probability value, where 1 ≤ i ≤ N.

[0038] For each of the M future times, the power generation range for that future time can be calculated based on the N discrete power points and their corresponding N probability values. Thus, M power generation ranges for the M future times can be obtained. Each power generation range corresponds to one future time. Each power generation range may include an upper bound and a lower bound. The upper bound and lower bound together form the power generation range.

[0039] In some embodiments, please refer to Figure 2The model is a machine learning model. It comprises an input layer, a temporal modeling layer, a convolutional layer, a transformer encoding layer, a pooling and fully connected fusion layer, and probability output heads, connected sequentially. The number of probability output heads in the model can be determined based on the installed capacity and power resolution of the new energy power generation equipment. The number of probability output heads in the model is equal to the number of discrete power points of the new energy power generation equipment. Specifically, the number of probability output heads is N. Each of the N probability output heads corresponds to a discrete power point and is used to output the first probability distribution parameters of that discrete power point. Specifically, based on the input data, the input layer can determine the original feature data; based on the original feature data, the temporal modeling layer can determine the hidden state data; based on the hidden state data, the convolutional layer can determine the convolutional feature data; based on the convolutional feature data, the transformer encoding layer can determine the encoded feature data; based on the encoded feature data, the pooling and fully connected fusion layer can determine the fused feature data; and based on the fused feature data, the N first probability distribution parameters of the N discrete power points can be determined through the N probability output heads.

[0040] The input layer maps the feature data sequence to the original feature data. For example, the original feature data can be represented as H0, where H0 = Linear(X), X represents the feature data, and Linear represents the input layer. The temporal modeling layer can include bidirectional GRU or bidirectional LSTM, etc. The temporal modeling layer extracts the temporal dependency between power data and meteorological data based on the original feature data. For example, H1 = BiGRU(H0), where H1 represents the hidden state data output by the temporal modeling layer, representing the temporal dependency between power data and meteorological data. The convolutional layer extracts short-term fluctuation features, such as power ramp-up, rapid decline, and local meteorological disturbances. For example, H2 = Conv(H1, kernel), where H2 represents the convolutional feature data output by the convolutional layer, representing the short-term variation features of power and meteorology. Multiple convolutional kernel sizes can be set in parallel within the convolutional layer and then concatenated to capture local changes at different time scales. The transformer encoding layer captures long-distance temporal dependencies, such as the impact of weather change trends on future power distribution. For example, H3 = TransformerEncoder(H2), where H3 represents the encoded feature data output by the transformer encoding layer, which is used to represent the long-term dependence of power on weather conditions. The pooling and fully connected fusion layer is used to perform attention pooling on the time dimension or to extract the hidden vector at the last moment to obtain the overall hidden vector, which is then fused through the fully connected layer. For example, h = AttentionPooling(H3), where h represents the overall hidden vector, and z = ReLU(Linear(h)), where z represents the fused feature data. Each probability output head can output the mean and variance. The mean μ = Linear(z), and the variance σ = Softplus(Linear(z)).

[0041] In some embodiments, upper and lower quantiles can be pre-configured. Therefore, for each of the M future times, the cumulative probability of the N discrete power points can be determined based on the N discrete power points and their corresponding N probability values ​​at that future time; an upper bound for power generation can be determined based on the upper quantile and the cumulative probability of the N discrete power points; and a lower bound for power generation can be determined based on the lower quantile and the cumulative probability of the N discrete power points. The upper and lower bounds for power generation are used to form the power generation range for that future time.

[0042] The i-th discrete power point and its corresponding probability value can be represented as {P} i ,p i}. P i p represents the discrete power point. i This represents the probability value, where 1 ≤ i ≤ N. It can be calculated using a formula. . . This represents the cumulative probability of the k-th power point. Used to indicate that the power generation capacity of new energy power generation equipment does not exceed The probability of.

[0043] The upper quantile represents the cumulative probability threshold corresponding to the upper bound of power generation, for example, it could be 0.95. The lower quantile represents the cumulative probability threshold corresponding to the lower bound of power generation, for example, it could be 0.05. The minimum quantile can be found among N discrete power points. Make The lower limit of power generation is .Right now, We can find the smallest one. Make Upper limit of power generation .Right now, . Indicates the lower quantile. Indicates the upper quantile. . This represents a range of power generation capacity. For example, N discrete power points can be represented as... The probability values ​​of the N discrete power points can be expressed as: The lower quantile is 0.05, and the upper quantile is 0.95. The cumulative probability of the N discrete power points can then be expressed as F = [0.1, 0.3, 0.7, 0.9, 1.0]. Therefore, the lower bound of the power generation is discrete power point 0 among the N discrete power points, and the upper bound of the power generation is discrete power point 20 among the N discrete power points.

[0044] In some embodiments, a confidence level can be pre-configured. To this end, for each of the M future times, multiple candidate power generation intervals can be determined based on N discrete power points and their corresponding N probability values ​​at that future time. Each candidate power generation interval may include one or more consecutive discrete power points. The number of discrete power points contained in different candidate power generation intervals may be the same or different. The sum of the probability values ​​of each discrete power point within each candidate power generation interval is greater than or equal to the confidence level. The power generation interval with the smallest width can be selected from the multiple candidate power generation intervals. The width of the power generation interval can be represented by the difference between the upper and lower bounds of the power generation.

[0045] Confidence level is used to represent the probability that a power generation range includes the actual power generation of renewable energy power generation equipment at a future time. For example, a 90% confidence level means that the power generation range covers the actual power generation at a future time with a 90% probability.

[0046] In some embodiments, one or more training samples may be obtained. The training samples may include feature data and label data. The feature data may include second historical power data and second historical meteorological data, and the label data may include third historical power data. The second historical power data is the actual power generation of the new energy power generation equipment at a historical time; the second historical meteorological data is the actual meteorological data at a historical time; and the third historical power data is the actual power generation of the new energy power generation equipment at a historical time. The third historical power data and the second historical meteorological data correspond to the same historical time. The historical time corresponding to the second historical power data is earlier than the historical time corresponding to the third historical power data and the second historical meteorological data. Thus, a training sample is constructed with the actual power generation at a historical time (second historical power data) and the meteorological data at a predicted time (second historical meteorological data) as input, and the actual power generation at the predicted time (third historical power data) as the label.

[0047] For each feature data, the model can predict the second probability distribution parameters of new energy power generation equipment; the loss data can be determined based on the corresponding label data and the second probability distribution parameters; and the parameters of the model can be adjusted based on the loss data.

[0048] For example, the model includes, in sequence, an input layer, a temporal modeling layer, a convolutional layer, a transformer encoding layer, a pooling and fully connected fusion layer, and N probability output heads. Based on the feature data, the input layer determines the original feature data; the temporal modeling layer determines the hidden state data; the convolutional layer determines the convolutional feature data, which represents the short-term variation characteristics of power and weather; the transformer encoding layer determines the encoded feature data, which represents the long-term dependency between power and weather; the pooling and fully connected fusion layer determines the fused feature data; and the N probability output heads determine N second probability distribution parameters. These N second probability distribution parameters correspond to N discrete power points. Based on these parameters, a second probability quality function for each discrete power point can be determined. Substituting each discrete power point into the corresponding second probability quality function yields its predicted probability value. Thus, for each feature data point, predicted probability values ​​for N discrete power points can be obtained.

[0049] For each tag data point, the distance between the tag data point and N discrete power points can be calculated. The true probability value of the tag data point among the N discrete power points can then be determined based on these distances. For example, it can be checked among the N discrete power points to see if a target discrete power point equal to the tag data point exists. If it does, the true probability value of the target discrete power point is set to 1, and the true probability values ​​of all other discrete power points among the N discrete power points are set to 0. If it does not exist, one or more first candidate discrete power points greater than the tag data point can be searched among the N discrete power points. The smallest first candidate discrete power point can be selected as the first adjacent discrete power point. Similarly, one or more second candidate discrete power points smaller than the tag data point can be searched among the N discrete power points. The largest second candidate discrete power point can be selected as the second adjacent discrete power point. A first difference between the first adjacent discrete power point and the tag data can be calculated; a second difference between the second adjacent discrete power point and the tag data can be calculated; and a third difference between the second adjacent discrete power point and the first adjacent discrete power point can be calculated. Dividing the first difference by the third difference yields the true probability value of the first adjacent discrete power point; similarly, dividing the second difference by the third difference yields the true probability value of the second adjacent discrete power point. The true probability values ​​of all discrete power points among the N discrete power points, except for the first and second adjacent discrete power points, can be set to 0. Therefore, for each tag data point, the true probability values ​​of N discrete power points can be obtained.

[0050] The loss function may include a first loss term, a second loss term, a third loss term, a fourth loss term, and a fifth loss term; the first loss term is used to constrain the predicted probability distribution corresponding to the N predicted probability values ​​to be similar to the true probability distribution corresponding to the N true probability values; the second loss term is used to constrain the cumulative probability of the N discrete power points to be monotonically increasing; the third loss term is used to constrain the predicted probability value of the boundary power point; the fourth loss term is used to constrain the degree of difference between the predicted probability values ​​of adjacent discrete power points; and the fifth loss term is used to constrain the weights of the model.

[0051] For example, L total =λ1·L NLL +λ2·L monotonic +λ3·L boundary +λ4·L smooth +λ5·L reg Among them, L NLL This represents the first loss item. L monotonic This represents the second loss term. L boundary This represents the third loss item. L smooth This represents the fourth loss item. Lreg λ represents the fifth loss term. λ1, λ2, λ3, λ4, and λ5 represent the weights of each loss term.

[0052] The first loss term is used to calculate the negative log-likelihood loss. For example, . This represents the negative log-likelihood loss. Q represents the true probability vector true The i-th element in Q true Includes the true probability values ​​of N discrete power points. Represents the prediction probability vector P pred The i-th element in P pred Including the predicted probability values ​​of N discrete power points. and They correspond to the same discrete power point. ε represents the minimum value.

[0053] The second loss term is used to calculate the monotonicity loss. For example, . This indicates monotonicity loss. . Indicates the first Predicted probability values ​​for discrete power points. This represents the cumulative probability of the i-th discrete power point.

[0054] The third loss term is used to calculate the boundary constraint loss. This third loss term ensures that the predicted probability values ​​of the discrete power points at the boundary conform to physical meaning. For example, . This represents the boundary constraint loss. The boundary discrete power points include the minimum and maximum discrete power points among the N discrete power points. This represents the predicted probability value of the minimum discrete power point. This represents the predicted probability value of the maximum discrete power point. Among the N discrete power points, the minimum discrete power point can be, for example, 0, and the maximum discrete power point can be, for example, the installed capacity. This represents the expected probability value of the minimum discrete power point. This represents the expected probability value of the point with the maximum discrete power. and This can be a preset value. For example, if it is believed that the probability of a new energy power generation device equaling the boundary discrete power point should not be abnormally high under normal circumstances, then this can be set to... and Set to a smaller value.

[0055] The fourth loss term is used to calculate the smoothness loss. This fourth loss term is used to avoid drastic jumps in the predicted probability values ​​between adjacent power points. The larger the value of the fourth loss term, the more pronounced the jumps in the predicted probability values ​​between adjacent discrete power points. For example, . This indicates the loss of smoothness.

[0056] The fifth loss term is used to calculate the regularization loss. This fifth loss term is used to suppress model overfitting. For example, . This represents the regularization loss. In the model, the first The weight matrix of the layer, This represents the sum of squares of all elements in the weight matrix. Layers may include any one or more of the following: input layer, temporal modeling layer, convolutional layer, transformer encoding layer, pooling and fully connected fusion layer, and N probability output heads.

[0057] The parameters of the model can be adjusted using methods such as gradient descent based on the loss data to achieve model training.

[0058] In some embodiments, M power generation ranges for M future times can also be sent to the control equipment of the electrolyzer, so that the control equipment can control the load power of the electrolyzer at the M future times.

[0059] Please see Figure 3 This specification also provides an embodiment of an electrolytic cell load power control method, which can be applied to power control equipment and may specifically include the following steps.

[0060] Step 31: Receive the power generation range of the new energy power generation equipment at a future time. Step 32: Adjust the load power of the electrolyzer at the future time according to the power generation range.

[0061] In some embodiments, the power prediction device can predict the power generation range of the new energy power generation equipment at M future time points; and can send the power generation range of the new energy power generation equipment at M future time points to the power control device. The power control device can receive the power generation range. The process of the power prediction device predicting the power generation range can be found in the previous embodiments.

[0062] In some embodiments, the load power of the electrolytic cell can be the actual operating power or the actual power consumed by the electrolytic cell.

[0063] The power generation range at each future time point can include an upper limit and a lower limit. These upper and lower limits can serve as safety boundaries for electrolyzer load regulation. Upon reaching a future time point, the upper and lower limits can be obtained based on the power generation range at that future time point. When the upper limit is greater than the rated power of the electrolyzer, the energy storage device of the electrolyzer can be controlled to charge. When the lower limit is less than the minimum power of the electrolyzer, the energy storage device can be controlled to discharge. When the upper limit is less than or equal to the rated power of the electrolyzer, and the lower limit is greater than or equal to the minimum power of the electrolyzer, the actual operating power of the electrolyzer can be controlled to remain within the power generation range.

[0064] When the upper limit of power generation exceeds the rated power of the electrolyzer, it indicates that the output of the new energy power generation equipment may exceed the electrolyzer's absorption capacity. In this case, the energy storage device can be charged to absorb the excess power. Furthermore, the electrolyzer load can be limited to not exceed its rated power. For example, the power difference between the upper limit of power generation and the rated power of the electrolyzer can be calculated, and the energy storage device can be controlled to absorb this power difference, allowing the electrolyzer to operate at its rated power. When the lower limit of power generation is less than the minimum stable operating power of the electrolyzer, it indicates that the output of the new energy power generation equipment may be insufficient to maintain stable operation of the electrolyzer. In this case, the energy storage device can be discharged to make up the power shortfall. If the energy storage cannot make up the shortfall, the electrolyzer load can be reduced in advance or a safe load reduction strategy can be implemented to prevent the electrolyzer from falling out of its stable operating range. For example, the power difference between the lower limit of power generation and the minimum power of the electrolyzer can be calculated, and the energy storage device can be controlled to provide the power difference to the electrolyzer, allowing the electrolyzer to operate at its minimum power. When the power generation range is within the range formed by the electrolyzer's rated power and minimum power, the actual operating power of the electrolyzer can be controlled to remain within the power generation range. The aforementioned energy storage devices can be, for example, batteries or supercapacitors, used to absorb electrical energy when the renewable energy power generation equipment has excess output and to supplement electrical energy when its output is insufficient. The energy storage devices are connected to both the renewable energy power generation equipment and the electrolyzer. Therefore, the load regulation of the electrolyzer no longer relies on a single power forecast value, but can make decisions based on the range of possible future power generation, thereby better addressing the uncertainty and fluctuation risks of renewable energy equipment output and improving the safety and economy of the hydrogen production system.

[0065] Optionally, when the upper limit of power generation is less than or equal to the rated power of the electrolyzer and the lower limit of power generation is greater than or equal to the minimum power of the electrolyzer, the power generation range variation trend can be determined based on the power generation range of M future times. Based on the power generation range variation trend, the target power generation can be selected within the power generation range of the future times, and the actual operating power of the electrolyzer can be controlled to be equal to the target power generation.

[0066] For example, the changing trend can include a downward trend and an upward trend. A downward trend indicates that the power generation range over M future time points is shifting towards lower power; for example, the upper and / or lower bounds of the power generation range over the M power generation ranges show a decreasing trend. An upward trend indicates that the power generation range over M future time points is shifting towards higher power; for example, the upper and / or lower bounds of the power generation range over the M power generation ranges show an increasing trend. Under a downward trend, a lower power value within the power generation range can be selected as the target power generation to reduce the electrolyzer load in advance and reserve energy storage discharge capacity, thereby matching the subsequent power decrease trend. Under an upward trend, a higher power value within the power generation range can be selected as the target power generation to increase the electrolyzer load in advance and reserve energy storage charging capacity to absorb prediction deviations. For example, the median value of the power generation range can be obtained; the power generation range can be divided into a first sub-range and a second sub-range based on the median value. The first sub-range is the range formed by the lower bound of the power generation and the median value, and the second sub-range is the range formed by the upper bound of the power generation and the median value. For example, This indicates the range of power generation capacity. () represents the intermediate value. The first subinterval can be represented as The second subinterval can be represented as In a downward trend, the target power generation can be selected within the first sub-interval, resulting in a smaller target power generation. In an upward trend, the target power generation can be selected within the second sub-interval, resulting in a larger target power generation.

[0067] Furthermore, under a downward trend, a linear fit can be performed on the upper or lower bounds of the power generation intervals of the M power generation intervals to obtain a first slope. This first slope is used to quantitatively characterize the downward trend of the power generation intervals at the M future times. Based on the first slope, a target power generation can be selected within a first sub-interval. For example, the first slope can be normalized to the [0,1] interval to obtain a first weight; this can be achieved through a formula... Select the target power generation capacity within the first sub-interval. Indicates the first weight. This represents the target power generation capacity selected within the first sub-interval. For example, it can be expressed using the formula... k represents the first slope. Therefore, the faster the downward trend, the smaller the target power generation selected within the first sub-interval, and the closer the target power generation is to... .

[0068] Accordingly, under an upward trend, a linear fit can be performed on the upper or lower bounds of the power generation intervals of the M power generation intervals to obtain a second slope. This second slope is used to quantitatively characterize the upward trend of the power generation intervals at the M future times. Based on the second slope, a target power generation can be selected within a second sub-interval. For example, the second slope can be normalized to the [0,1] interval to obtain a second weight; this can be achieved through a formula... Select the target power generation capacity within the second sub-interval. Indicates the second weight. This represents the target power generation selected within the second sub-interval. The method for normalizing the second slope is similar to that for the first slope, and the two can be explained in comparison. Therefore, the faster the upward trend, the larger the target power generation selected within the second sub-interval, and the closer the selected target power generation is to... .

[0069] Please see Figure 4 This specification provides a new energy power generation prediction device, which can be applied to power prediction equipment and may include the following units.

[0070] Acquisition unit 41 is used to acquire input data, which includes first historical power data and future meteorological data; Prediction unit 42 is used to predict the first probability distribution parameters of new energy power generation equipment at future times based on the input data and through a model. The determining unit 43 is used to determine the probability value of the new energy power generation equipment at a future time based on the first probability distribution parameters. The calculation unit 44 is used to calculate the power generation range of the new energy power generation equipment in the future based on the probability value. The power generation range is used to adjust the load power of the electrolyzer in the future. The new energy power generation equipment is used to supply power to the electrolyzer.

[0071] Please see Figure 5 This specification provides an embodiment of an electrolytic cell load power control device. The control device can be applied to power control equipment and may include the following units.

[0072] The receiving unit 51 is used to receive the power generation range of the new energy power generation equipment at a future time. The control unit 52 is used to control the load power of the electrolyzer at the future time according to the power generation range.

[0073] This specification also provides a hydrogen production system. The hydrogen production system may include a new energy power generation device, a power prediction device, a power control device, and an electrolyzer. The new energy power generation device supplies power to the electrolyzer. The power prediction device predicts the power output of the new energy power generation. Specifically, the new energy power generation device acquires input data, including first historical power data and future meteorological data; based on the input data, it predicts a first probability distribution parameter of the new energy power generation device at a future time using a model; based on the first probability distribution parameter, it determines a probability value of the new energy power generation device at a future time; and based on the probability value, it calculates the power generation range of the new energy power generation device at a future time. The power prediction device can send the power generation range of the new energy power generation device at a future time to the power control device. The power control device can receive the power generation range. The power control device can control the power consumption of the electrolyzer based on the power generation range of the new energy power generation device at a future time. For example, the power control device can adjust the load power of the electrolyzer at a future time based on the power generation range.

[0074] This specification also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-mentioned new energy power generation prediction method and / or the above-mentioned electrolytic cell load power control method.

[0075] This specification also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described new energy power generation prediction method and / or the above-described electrolytic cell load power control method.

[0076] This specification also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-mentioned new energy power generation prediction method and / or the above-mentioned electrolytic cell load power control method.

[0077] Those skilled in the art will understand that this specification can be provided as a method, system, or computer program product. Therefore, this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware. Furthermore, this specification may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0078] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments thereof. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. The computer may be a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0079] The functional units in the embodiments of this specification can be integrated into one processing unit, or each functional unit can exist physically separately, or two or more functional units can be integrated into one processing unit.

[0080] Those skilled in the art will understand that the descriptions of the various embodiments in this specification have different focuses, and parts not described in detail in a certain embodiment can be referred to in the relevant descriptions of other embodiments. Furthermore, it is understood that those skilled in the art, after reading this specification, can conceive of any combination of some or all of the embodiments listed in this specification without creative effort, and such combinations are also within the scope of disclosure and protection of this specification.

[0081] Although this specification has been described through embodiments, those skilled in the art will understand that the above embodiments are merely illustrative of the core ideas of this specification. Those skilled in the art will appreciate that many variations and modifications are possible with this specification. It is intended that the appended claims encompass these variations and modifications without departing from the spirit of this specification.

Claims

1. A method for predicting the power generation capacity of new energy sources, characterized in that, include: Acquire input data, which includes first historical power data and meteorological data for future times; Based on the input data, the model predicts the first probability distribution parameters of the new energy power generation equipment at the future time. Based on the first probability distribution parameters, determine the probability value of the new energy power generation equipment at the future time. Based on the probability value, the power generation range of the new energy power generation equipment at the future time is calculated. The power generation range is used to regulate the load power of the electrolyzer at the future time. The new energy power generation equipment is used to supply power to the electrolyzer. The new energy power generation equipment corresponds to N discrete power points, where N>1; the model includes an input layer, a temporal modeling layer, a convolutional layer, a converter coding layer, a pooling and fully connected fusion layer, and N probability output heads; The method of predicting the first probability distribution parameters of new energy power generation equipment at the future time using a model includes: Based on the input data, the original feature data is determined through the input layer. Based on the original feature data, the hidden state data is determined through the temporal modeling layer. Based on the hidden state data, the convolutional feature data is determined through the convolutional layer. The convolutional feature data is used to represent the short-term variation characteristics of power and weather. Based on the convolutional feature data, the encoded feature data is determined through the transformer coding layer. The encoded feature data is used to represent the long-term dependence of power and weather. Based on the encoded feature data, the fused feature data is determined through the pooling and fully connected fusion layer. Based on the fused feature data, N first probability distribution parameters are determined through the N probability output heads. Determining the probability value of the new energy power generation equipment at the future time includes: Based on the first probability distribution parameters of N discrete power points, determine the first probability mass function of N discrete power points, and substitute each discrete power point into the corresponding first probability mass function to obtain the probability value of that discrete power point. The calculation of the power generation range of the new energy power generation equipment at the future time includes: Based on the probability values ​​of N discrete power points, calculate the power generation range of the new energy power generation equipment at the future time.

2. The method according to claim 1, characterized in that, The calculation of the power generation range of the new energy power generation equipment at the future time includes: Determine the cumulative probability of N discrete power points based on their probability values. The upper bound of power generation is determined based on the preset upper quantile and the cumulative probability of the N discrete power points; The lower bound of power generation is determined based on the preset lower quantile and the cumulative probability of the N discrete power points; The upper and lower limits of power generation are used to form the power generation range.

3. The method according to claim 1, characterized in that, The calculation of the power generation range of the new energy power generation equipment at the future time includes: Based on the probability values ​​of N discrete power points, multiple candidate power generation intervals are determined, and the sum of the probability values ​​of each discrete power point in each candidate power generation interval is greater than or equal to the preset confidence level. Among the multiple candidate power generation ranges, the power generation range with the smallest width is selected.

4. The method according to claim 1, characterized in that, The model was trained in the following way: Acquire training samples, which include feature data and label data. The feature data includes second historical power data and second historical meteorological data. The label data includes third historical power data. The third historical power data and the second historical meteorological data correspond to the same historical moment. The historical moment corresponding to the second historical power data is earlier than the historical moment corresponding to the third historical power data and the second historical meteorological data. Based on the aforementioned feature data, the second probability distribution parameters of new energy power generation equipment are predicted using a model. Loss data is determined based on label data and second probability distribution parameters; The parameters of the model are adjusted based on the loss data.

5. The method according to claim 4, characterized in that, The new energy power generation equipment corresponds to N discrete power points, where N>1; the second probability distribution parameters of the new energy power generation equipment predicted by the model include: The model is used to predict the N second probability distribution parameters of new energy power generation equipment at N discrete power points; The step of determining the loss data based on the label data and the second probability distribution parameters includes: Determine the second probability mass function of the N discrete power points based on the second probability distribution parameters of the N discrete power points; Substitute each discrete power point into the corresponding second probability mass function to obtain the predicted probability value of that discrete power point. Calculate the distance between each tag data point and N discrete power points; The true probability value of the labeled data at N discrete power points is determined based on the distance. Determine the loss data based on the loss function; The loss function includes a first loss term, a second loss term, a third loss term, a fourth loss term, and a fifth loss term; the first loss term is used to constrain the predicted probability distribution corresponding to the N predicted probability values ​​to be similar to the true probability distribution corresponding to the N true probability values; the second loss term is used to constrain the cumulative probability of the N discrete power points to be monotonically increasing; the third loss term is used to constrain the predicted probability value of the boundary power point; the fourth loss term is used to constrain the degree of difference between the predicted probability values ​​of adjacent discrete power points; and the fifth loss term is used to constrain the model weights.

6. A method for controlling the load power of an electrolytic cell, characterized in that, include: Receive the power generation range of the new energy power generation equipment at a future time, wherein the power generation range is obtained by the method according to any one of claims 1-5; The load power of the electrolyzer at the future time is adjusted according to the power generation range.

7. The method according to claim 6, characterized in that, The power generation range includes an upper limit and a lower limit; the load power of the electrolyzer at the future time includes: When the upper limit of the power generation is greater than the rated power of the electrolytic cell, the energy storage device of the electrolytic cell is controlled to charge. When the lower limit of the power generation is less than the minimum power of the electrolytic cell, the energy storage device of the electrolytic cell is controlled to discharge. When the upper limit of the power generation is less than or equal to the rated power of the electrolytic cell, and the lower limit of the power generation is greater than or equal to the minimum power of the electrolytic cell, the actual operating power of the electrolytic cell is controlled to be within the power generation range.

8. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method of any one of claims 1-7.

Images