A control method and system of a temperature control instrument of a switch cabinet

Abstract

The application discloses a control method and system of a temperature control instrument of a switch cabinet, and relates to the technical field of intelligent electrical equipment thermal management, and comprises the following steps: constructing a thermal effect periodic change function according to obtained switch cabinet first data and a preset period length; constructing a load-thermal effect rhythm regulation and control model based on the thermal effect periodic change function and obtained current load data; generating a switch cabinet thermal effect prediction result through machine learning based on the regulation and control model; generating a temperature control response strategy through fuzzy logic reasoning according to the prediction result and in combination with real-time load state data; and adjusting the operation state of a temperature control instrument control system according to the temperature control response strategy. The application is used to solve the problem that the conventional switch cabinet temperature control instrument control method is difficult to generate a temperature control response strategy according to thermal effects.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent electrical equipment thermal management, more specifically, the present application relates to a control method and system for temperature control instrument of switch cabinet. BACKGROUND

[0002] As a key power distribution equipment in power system, the internal electrical components of switch cabinet will generate heat due to the effect of load current during long-term operation, resulting in temperature rise. Excessive temperature will accelerate the aging of insulation materials, reduce mechanical strength, cause connection site loosening and even lead to equipment failure, which seriously affects power supply reliability and equipment life. Therefore, it is crucial to effectively monitor and intelligently control the internal temperature of switch cabinet.

[0003] Traditional switch cabinet temperature control instruments mostly use start-stop control or simple PID control method based on fixed threshold. For example, when the temperature exceeds a certain set upper limit, the cooling device is started, and when the temperature falls below the lower limit, the cooling device is stopped. Although this method is simple and reliable, it has obvious limitations: first, the thermal dynamic process of switch cabinet has significant time lag and nonlinearity, and its temperature rise is not only related to the current load current, but also affected by historical load, environmental temperature and humidity, cabinet air flow organization and other factors. The fixed threshold control method is difficult to adapt to complex and variable working conditions, and is prone to control lag or frequent action; secondly, the load current often shows periodic fluctuation law (such as daily load change and seasonal electricity difference), but the traditional method cannot actively learn and use this periodic rhythm characteristics for forward-looking regulation; thirdly, the real-time load change trend and environmental boundary conditions are not deeply fused and analyzed, and the intelligentization and self-adaptation ability of the control strategy is insufficient.

[0004] For example, the invention patent with publication number CN102063056A discloses a digital temperature control instrument with intelligent setting and intelligent load adjustment function, which includes: the research on intelligent setting and intelligent load adjustment function in temperature control process in industrial control field, and provides a design of digital temperature control instrument with intelligent setting and intelligent load adjustment function. According to the correlation of multiple parameters, the user setting parameters are automatically displayed or hidden; when setting the value, the value of individual place, ten place and hundred place is intelligently determined to increase or decrease according to the time length of the key; in the state of empty machine without feeding, the pre-set temperature control is carried out to obtain the parameters of PID or fuzzy control, and in the state of feeding in actual control process, the intelligent load adjustment is carried out according to the actual control situation to obtain the corrected parameters of PID or fuzzy control in actual control process.

[0005] In the above disclosed technical solution, at least the following technical problems exist:

[0006] The traditional switch cabinet temperature control instrument control method does not consider the significant time lag and nonlinear characteristics of load change, heat accumulation and temperature rise in the switch cabinet temperature rise process, does not establish a correlation model of load and thermal effect, and the parameter correction is lagged when the load suddenly changes, so that temperature overshoot or control is not timely; secondly, the periodic rhythm characteristics accompanied by the load current of the switch cabinet are not extracted and utilized, and the parameters are adjusted only based on the real-time working condition, which cannot realize the forward-looking control and still belongs to the "passive response" mode, and it is difficult to avoid the risk of temperature sudden rise at the load peak; in addition, the fuzzy control rule and parameter correction depend on the preset logic and simple working condition feedback, do not combine the historical operation data of the switch cabinet to optimize the rule base, and lack a thermal effect prediction mechanism based on a time sequence model, so that the control strategy cannot be made in advance, and the cooling device is easy to start and stop frequently in the scene of continuous load growth, which not only affects the service life of the equipment, but also is difficult to meet the high-precision and self-adaptive temperature control requirements of the switch cabinet.

[0007] In view of the above problems, the present application provides a solution. SUMMARY

[0008] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a control method and system of a temperature control instrument of a switch cabinet, which solves the problem that the traditional switch cabinet temperature control instrument control method is difficult to generate a temperature control response strategy according to thermal effect by constructing a thermal effect periodic change function, establishing a load-thermal effect rhythm control model and fusing LSTM prediction and weighted fuzzy reasoning.

[0009] To achieve the above object, the present application provides the following technical scheme:

[0010] A control method of a temperature control instrument of a switch cabinet, characterized in that it comprises the following steps: constructing a thermal effect periodic change function according to obtained first data of the switch cabinet and presetting a period length; constructing a load-thermal effect rhythm control model based on the thermal effect periodic change function and obtained current load data; generating a thermal effect prediction result of the switch cabinet based on the control model through machine learning; generating a temperature control response strategy through fuzzy logic reasoning according to the prediction result combined with real-time load state data; and adjusting the operating state of the temperature control instrument control system according to the temperature control response strategy.

[0011] In a preferred embodiment, the period length is determined by time series analysis of the thermal effect change period.

[0012] In a preferred embodiment, the thermal effect periodic change function is constructed according to the obtained first data of the switch cabinet and the preset period length, specifically by inputting the first data of the switch cabinet into a smooth fitting algorithm, and performing data smooth fitting with the period length as the constraint condition to obtain the thermal effect periodic change function.

[0013] In a preferred embodiment, the load-thermal effect rhythm regulation model is constructed based on the thermal effect periodic change function and the current load data, specifically: based on the current load data, a load dynamic feature vector is obtained through a feature extraction algorithm; a joint feature sequence is obtained through a multi-source feature fusion algorithm according to the load dynamic feature vector and the thermal effect periodic change function; and the load-thermal effect rhythm regulation model is constructed through a time series regression algorithm based on the joint feature sequence.

[0014] In a preferred embodiment, the switch cabinet thermal effect prediction result is generated through machine learning based on the regulation model, specifically: an LSTM input sequence is constructed based on the output of the load-thermal effect rhythm regulation model; a regular time series sample set is obtained through data standardization and time step alignment processing based on the LSTM input sequence; and the switch cabinet thermal effect prediction result is obtained through a long short-term memory network algorithm according to the time series sample set.

[0015] In a preferred embodiment, the LSTM input sequence is constructed based on the output of the load-thermal effect rhythm regulation model, specifically: a load dynamic feature vector and a thermal effect feature vector are extracted according to the output of the load-thermal effect rhythm regulation model; a fusion feature matrix is obtained through feature splicing according to the load dynamic feature vector and the thermal effect feature vector; and the LSTM input sequence is constructed through a sliding window sampling method based on the fusion feature matrix.

[0016] In a preferred embodiment, the temperature control response strategy is generated through fuzzy logic reasoning based on the prediction result and the real-time load state data, specifically: a temperature control state fuzzy set is obtained by constructing a membership function based on the prediction result and the real-time load state data; a weighted fuzzy rule base is generated according to the obtained switch cabinet historical data; a fuzzy temperature control strategy set is obtained by reasoning through the weighted fuzzy rule base based on the temperature control state fuzzy set; and the temperature control response strategy is generated through defuzzification calculation based on the fuzzy temperature control strategy set.

[0017] In a preferred embodiment, the weighted fuzzy rule base is generated according to the switch cabinet historical data, specifically: a typical operating condition mode is obtained through a clustering algorithm according to historical load and temperature data; initial fuzzy rules are extracted through an association rule mining method according to the typical operating condition mode; rule weight coefficients are calculated through a weight learning algorithm according to historical operating effect data; and the weighted fuzzy rule base is generated through integration based on the initial fuzzy rules and the weight coefficients.

[0018] In a preferred embodiment, the initial fuzzy rules are extracted from the typical operating condition mode by the association rule mining method, specifically: the features in the typical operating condition mode are discretized and fuzzy graded to form a set of operating condition feature items;Based on the set of operating condition feature items, each operating condition sample is converted into an operating condition sample record to construct an operating condition sample database;The support degree of the set of operating condition feature items is calculated in the operating condition sample database, and the frequent item set is screened out by setting a minimum support threshold;Based on the frequent item set, the association rule set is generated according to the division rule of the input item and the output item;The operating condition items in the association rule set are mapped to the corresponding fuzzy language variables to form the initial fuzzy rule set of the "if-then" structure.

[0019] A control system of a temperature control instrument of a switch cabinet, comprising: a function construction module, which constructs a thermal effect periodic change function according to obtained first data of the switch cabinet and a preset period length;A model construction module, which constructs a load-thermal effect rhythm regulation model based on the thermal effect periodic change function and obtained current load data;A thermal effect prediction module, which generates a thermal effect prediction result of the switch cabinet based on the regulation model through machine learning;A strategy generation module, which generates a temperature control response strategy through fuzzy logic reasoning according to the prediction result and in combination with real-time load state data;And a control execution module, which regulates the operating state of the temperature control instrument control system according to the temperature control response strategy.

[0020] The technical effects and advantages of the control method and system of the temperature control instrument of the switch cabinet are as follows:

[0021] The control method and system of the temperature control instrument of the switch cabinet extract thermal effect periodic characteristics from multi-source switch cabinet data and construct a periodic change function through a smoothing fitting algorithm combined with time series analysis, accurately capture the rhythmic law of thermal effects with load fluctuations, integrate load dynamic characteristics and thermal effect periodic functions through a multi-source feature fusion algorithm, construct a load-thermal effect rhythm regulation model to depict the nonlinear dynamic relationship between them, perform time series prediction on the fused feature sequence based on an LSTM algorithm, realize forward-looking trend prediction of thermal effects, break the hysteresis of the traditional "real-time response", perform fuzzy logic reasoning with the help of a weighted fuzzy rule base generated based on historical data, generate a temperature control strategy matching real-time load and environmental conditions, effectively overcome the problems of traditional methods, such as ignoring periodicity, missing correlation models, response lag and poor rule adaptability, and provide an efficient solution for intelligent electrical equipment thermal management. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 The flowchart of the control method of the temperature control instrument of the switch cabinet is as follows:

[0023] Figure 2A structure diagram of a control system of a temperature control instrument of a switch cabinet according to the present application;

[0024] Figure 3 A comparison diagram of a thermal effect periodic function fitting and a temperature prediction result;

[0025] Figure 4 A comparison diagram of temperature control effects of the present application and a traditional threshold control. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0027] It should be noted that, in this document, the relationship terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the elements defined by the statement "include" do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0028] Embodiment 1, Figure 1 A control method of a temperature control instrument of a switch cabinet is given, characterized in that it comprises the following steps:

[0029] S1, constructing a thermal effect periodic change function according to the obtained first data of the switch cabinet and a preset period length;

[0030] S2, constructing a load-thermal effect rhythm regulation model based on the thermal effect periodic change function and the obtained current load data;

[0031] S3, generating a switch cabinet thermal effect prediction result through machine learning based on the regulation model;

[0032] S4, generating a temperature control response strategy through fuzzy logic reasoning according to the prediction result and in combination with real-time load state data;

[0033] S5, dynamically regulating the operating state of the temperature control instrument control system according to the temperature control response strategy.

[0034] The application discloses a control method and system of a temperature control instrument of a switch cabinet. A thermal effect periodic characteristic is extracted from multi-source switch cabinet data by a smooth fitting algorithm combined with time series analysis, and a periodic change function is constructed, so that the rhythmic law of the thermal effect with load fluctuation is accurately captured. A load-thermal effect rhythm regulation model is constructed by integrating the load dynamic characteristic and the thermal effect periodic function through a multi-source feature fusion algorithm, so as to depict the nonlinear dynamic relationship between the load and the thermal effect. The fused feature sequence is time series predicted based on an LSTM algorithm, so as to realize a forward-looking trend prediction of the thermal effect and break the lag of the traditional "real-time response". Fuzzy logic reasoning is performed by means of a weighted fuzzy rule base generated based on historical data, so as to generate a temperature control strategy matching the real-time load and environmental conditions, and problems, such as neglecting periodicity, missing correlation model, response lag and poor rule adaptability, of a traditional method are effectively overcome, thereby providing an efficient solution for intelligent electrical equipment thermal management. The problem that a traditional switch cabinet temperature control instrument control method is difficult to generate a temperature control response strategy according to the thermal effect is effectively solved.

[0035] S1, constructing a thermal effect periodic change function according to the obtained switch cabinet first data and a preset periodic length.

[0036] In this embodiment, the switch cabinet first data includes temperature data, humidity data and environmental parameters; and the current load data includes a loop real-time current value, a current change rate, a rated current and a load duration.

[0037] In this embodiment, the periodic length is determined by extracting the thermal effect change period through time series analysis.

[0038] In this embodiment, the thermal effect periodic change function is constructed according to the obtained switch cabinet first data and the preset periodic length, specifically, the switch cabinet first data is input into a smooth fitting algorithm, and data smooth fitting is performed with the periodic length as a constraint condition, so as to obtain the thermal effect periodic change function.

[0039] The periodic length of the thermal effect change is extracted from the switch cabinet first data, so as to provide a constraint basis for subsequent fitting.

[0040] The periodicity of the data is analyzed by calculating the autocorrelation coefficient, that is, the similarity degree of the data at the lag time point and the original data is compared for any time lag.

[0041] The lag time with a higher correlation coefficient is screened out, and the most significant periodic length is found through cluster analysis, so as to be determined as the main period.

[0042] The maximum difference of the temperature in one main period is calculated, and a certain fault tolerance range is reserved as the amplitude constraint for subsequent fitting; and the offset of the time relative to the start point of the period is calculated as the reference of the phase of the period.

[0043] Smooth fitting algorithm parameter configuration sets algorithm parameters according to periodic characteristics, balancing data fitting accuracy and curve smoothness:

[0044] Smooth window size is determined according to the length of the main period, usually 1 / 8 of the main period, to ensure that the window can cover the local fluctuation characteristics in the period.

[0045] Different positions in the window are given different weights-the data in the center position has the highest weight, and gradually decreases to both sides, and the weight change conforms to Gaussian distribution.

[0046] The regularization parameter is determined by cross-validation. Specifically, the data is divided into a training set (80%) and a validation set (20%), and multiple parameter values are tested to select the parameter that minimizes the validation set fitting error.

[0047] The formula of the smooth fitting algorithm is:

[0048]

[0049] Among them, is the target function value of the thermal effect period, is the thermal effect period smoothing window size, determined according to the length of the thermal effect main period, is the index of the data point in the window, is the Gaussian weight of the th data point in the window, is the actual value of the th original data point in the window, is the smooth fitting value of the center position of the window, is the thermal effect period regularization parameter, is the smooth fitting value of the th position in the window, is the smooth fitting value of the th position in the window, is the smooth fitting value of the th position in the window.

[0050] The periodic change function of the thermal effect is expressed in the form of a sine function to quantify the periodic change pattern of the thermal effect, and the formula is as follows:

[0051]

[0052] Among them, is the periodic characteristic function of the thermal effect change; is the amplitude of the periodic characteristic of the thermal effect change; is the angular frequency; is the phase shift; is the periodic characteristic mean term of the thermal effect change, is a sinusoidal function.

[0053] S2, constructing a load-thermal effect rhythm regulation model based on the thermal effect periodic change function and the obtained current load data.

[0054] In this embodiment, the load-thermal effect rhythm regulation model is constructed based on the thermal effect periodic change function and the current load data, specifically:

[0055] Based on the current load data, a load dynamic feature vector is obtained through a feature extraction algorithm;

[0056] According to the load dynamic feature vector and the thermal effect periodic change function, a joint feature sequence is obtained through a multi-source feature fusion algorithm;

[0057] Based on the joint feature sequence, a load-thermal effect rhythm regulation model is constructed through a time series regression algorithm.

[0058] The load dynamic feature vector is extracted. Based on the current load data, a structured feature vector is converted through a feature extraction algorithm. Specifically, statistical features and dynamic features of the current are calculated from the time domain, and spectrum features of the load fluctuation may also be extracted through frequency domain analysis, finally forming a high-dimensional vector that can fully reflect the real-time state and trend of the load.

[0059] The feature extraction algorithm refers to converting the original time series signal into a structured feature vector based on the current load data through multi-dimensional analysis method, specifically including: extracting mean, variance, maximum value, peak factor, change rate, and proportion of continuous high load in the time domain, which are used to describe the current load level and transient fluctuation; identifying load mutation, impact peak and its duration at the event level, reflecting the process characteristics of the load; calculating power spectral density and harmonic distribution through Fourier transform in the frequency domain, capturing periodic fluctuation and random jitter features; after normalization and standardization, a high-dimensional load dynamic feature vector is formed, which can fully reflect the load intensity and dynamic trend.

[0060] Multi-source feature fusion is realized. The above load dynamic feature vector and the thermal effect periodic change function obtained in step S1 are jointly processed: the thermal effect periodic function is discretized into time series features, and then the two types of features are integrated into a joint feature sequence through a multi-source feature fusion algorithm. For example, if the load feature vector at a certain time is [500A, 0.2A / s, 1 time, 80%], and the thermal effect periodic function output at the same time is [35℃, in the rising period], the fusion will form a sequence containing "load-thermal effect" coordination information, which reflects the current load intensity and its position in the thermal effect period, thereby capturing the dynamic correlation between the two.

[0061] The multi-source feature fusion algorithm refers to joint processing of the load dynamic feature vector and the output of the thermal effect periodic change function to form a joint feature sequence, specifically including: first, aligning the two types of data based on timestamps to ensure that fusion is achieved at the same time; second, discretizing the thermal effect periodic change function into structured features such as temperature value, periodic phase, and periodic paragraph; then, concatenating the above periodic features and load features into an extended vector, and explicitly expressing typical coupling relationships such as high load and periodic peak, low load and periodic valley through the construction of cross features; finally, organizing the extended feature vector into a joint feature sequence in chronological order to provide input basis for time series modeling.

[0062] A time series regression model is constructed. The joint feature sequence is used as training data to model using a time series regression algorithm: the input of the model is the joint feature sequence in a certain period of time, and the output is the actual value of the thermal effect in the corresponding period; the model parameters are optimized through iterative training to learn the nonlinear mapping relationship between load dynamic change and thermal effect periodic law. The final load-thermal effect rhythm regulation model can output thermal effect correlation results matching the periodic law based on load feature dynamics, providing core correlation basis for subsequent thermal effect prediction.

[0063] The time series regression algorithm refers to the process of establishing a time series prediction model using the relationship between the joint feature sequence and the actual thermal effect response, specifically including: using a sliding window to intercept a fixed length joint feature sequence as input, and using measured temperature, temperature rise rate or thermal effect index as output target; a nonlinear mapping relationship between input and output is established through an autoregressive model, a state space regression model or a regression method based on machine learning; in the training process, the parameters are optimized through iteration to minimize the error between the predicted value and the actual thermal effect, so that the model can accurately learn the coupling features between load dynamics and thermal effect periodic law; the final time series regression model can output thermal effect prediction values matching the working conditions when new load and periodic features are input, providing core support for load-thermal effect rhythm regulation.

[0064] S3, based on the regulation model, generating switch cabinet thermal effect prediction results through machine learning.

[0065] In this embodiment, the switch cabinet thermal effect prediction results are generated through machine learning based on the regulation model, specifically:

[0066] Based on the output of the load-thermal effect rhythm regulation model, an LSTM input sequence is constructed;

[0067] Based on the LSTM input sequence, a standardized time series sample set is obtained through data standardization and time step alignment processing;

[0068] According to the time series sample set, a switch cabinet thermal effect prediction result is obtained through a long short-term memory network algorithm.

[0069] The LSTM input sequence is constructed. Based on the output of the load-thermal effect rhythm regulation model, two types of core features are first extracted from the output: one is a load dynamic feature vector, and the other is a thermal effect feature vector. Then, the two types of vectors are fused into a fusion feature matrix containing “load-thermal effect” collaborative information through feature splicing. Then, the input sequence of the LSTM is generated by using the sliding window sampling method: the window size is set, and the window is slid in time steps, and the feature matrix in each window constitutes an input sample, aiming to capture the short-term time sequence dependence of the thermal effect on the load.

[0070] Secondly, data standardization and time step alignment processing. Because the dimensions of different features in the fusion feature matrix are quite different, standardization is needed to eliminate the dimension influence - Z-score standardization or Min-Max standardization is commonly used to avoid the dominance of a feature due to its large value range. At the same time, due to the possible slight differences in sensor collection frequency, linear interpolation or nearest neighbor interpolation is used to unify the time step length, ensuring that the features at each time point in the input sequence are strictly aligned. Finally, a regular time series sample set is formed, each sample containing “input sequence” and “target output”.

[0071] Finally, the prediction result is generated through the LSTM algorithm. Based on the regular sample set, the LSTM network is constructed: the input layer dimension is consistent with the dimension of the fusion feature; the hidden layer is set to 2-3 layers, each layer contains 32-128 memory cells, which selectively retain key time sequence information through the gating mechanism; the output layer is one neuron, corresponding to the thermal effect prediction value at a future time. During the training process, the mean square error is used as the loss function, and the network parameters are iteratively adjusted through the Adam optimizer, while 80% of the samples are divided into the training set and 20% into the validation set to prevent overfitting through the early stopping method. After the model training is completed, the new fusion feature sequence is input, and the prediction result of the switch cabinet thermal effect is output, providing a forward-looking basis for the subsequent temperature control strategy generation.

[0072] In this embodiment, the LSTM input sequence is constructed based on the output of the load-thermal effect rhythm regulation model, specifically:

[0073] According to the output of the load-thermal effect rhythm regulation model, the load dynamic feature vector and the thermal effect feature vector are extracted;

[0074] According to the load dynamic feature vector and the thermal effect feature vector, a fusion feature matrix is obtained through feature splicing;

[0075] Based on the fusion feature matrix, an LSTM input sequence is constructed through a sliding window sampling method.

[0076] When constructing the LSTM input sequence based on the load-thermal effect rhythm regulation model output, the logic of feature extraction, feature fusion, and sequence generation needs to be gradually advanced to ensure that the input sequence can accurately carry the time sequence correlation information of load and thermal effect.

[0077] Extract the load dynamic feature vector and the thermal effect feature vector. The output of the load-thermal effect rhythm regulation model already contains the correlation analysis results of load and thermal effect, and the two types of core features need to be separated from it: on the one hand, the load dynamic feature vector needs to cover the real-time state and change trend of the load, specifically including real-time current value, current change rate, rated current proportion, load duration, etc. to fully characterize the dynamic characteristics of the load; on the other hand, the thermal effect feature vector is based on the thermal effect periodic change function, including the reference thermal effect of the switch cabinet, the thermal effect cycle phase, the humidity correction coefficient, etc. to accurately reflect the periodicity of the thermal effect and the environmental influencing factors.

[0078] The fusion feature matrix is obtained by feature splicing. Since the load dynamic feature vector and the thermal effect feature vector correspond to the same time dimension, the two types of vectors need to be strictly aligned according to the timestamp to ensure that the load features at a certain time match the thermal effect features at the same period; then the load dynamic feature vector and the thermal effect feature vector at the same time are combined into a multi-dimensional feature row vector by column splicing; and then all the feature row vectors are arranged in chronological order to form a fusion feature matrix, where each row of the matrix corresponds to the "load-thermal effect" collaborative features at a time point, and each column corresponds to a type of feature, realizing the deep integration of load and thermal effect multi-source features.

[0079] Based on the fusion feature matrix, the LSTM input sequence is constructed by sliding window sampling. Considering that LSTM needs to learn the rules through historical time sequence information, reasonable sliding window parameters need to be set: the window size is determined according to the data acquisition frequency and prediction requirements, and the sliding step is usually consistent with the data acquisition period. Then slide the window along the time axis, and after each slide, the fusion features of 12 time points contained in the window form an LSTM input sample; by continuously sliding the window, the entire fusion feature matrix is converted into multiple continuous input samples, forming the input sequence required by LSTM, which can effectively capture the short-term time sequence dependence between load changes and thermal effect evolution, laying a foundation for subsequent LSTM model learning and prediction rules.

[0080] In the prediction process, first, the thermal effect prediction value at the future time based on the regulation model output is predicted, as follows:

[0081]

[0082] where, the first predicted temperature of the switchgear at a future time instant; the switchgear number monitored by the sensor in the switchgear; the load-thermal effect rhythm regulation model; the multi-modal sensor data input vector at the time instant; the prediction time window span.

[0083] A continuous prediction curve is drawn according to the predicted value, and the temperature change curve reflects the temperature evolution trend of each region in the prediction period (such as 2 hours, 4 hours or 24 hours in the future), which is used to depict the possible overheating risk or abnormal change of the region.

[0084] S4, according to the prediction result, combining real-time load state data, a temperature control response strategy is generated by fuzzy logic reasoning.

[0085] In this embodiment, the temperature control response strategy is generated by fuzzy logic reasoning according to the prediction result and real-time load state data, specifically:

[0086] Based on the prediction result and real-time load state data, a membership function is constructed to obtain a temperature control state fuzzy set;

[0087] According to the obtained switchgear historical data, a weighted fuzzy rule base is generated;

[0088] Based on the temperature control state fuzzy set, a fuzzy temperature control strategy set is obtained by reasoning through the weighted fuzzy rule base;

[0089] Based on the fuzzy temperature control strategy set, a temperature control response strategy is generated by defuzzification calculation.

[0090] Fuzzy processing of input variables. The fuzzy interface converts precise numerical input into a fuzzy set to adapt to the "linguistic reasoning" characteristics of fuzzy logic.

[0091] Scenario naming of language variables. Each input variable is defined with a language variable that fits the temperature control scenario, such as "real-time load rate" corresponding to "load intensity" and "predicted temperature deviation" corresponding to "temperature deviation degree", so that the fuzzy result is more consistent with human language description habits of temperature control scenarios.

[0092] Reasonable division of fuzzy subsets. "Load intensity" is divided into "low load", "medium load" and "high load".

[0093] "Temperature deviation degree" (derived from prediction result) is divided into "negative deviation" (predicted temperature lower than target), "zero deviation" (predicted temperature close to target) and "positive deviation" (predicted temperature higher than target).

[0094] In the selection of membership function type, the triangular function is adopted in the temperature control scenario. It is defined by three parameters: "left endpoint (a), vertex (b), and right endpoint (c)," which can intuitively reflect the degree to which the precise value belongs to a certain fuzzy subset.

[0095] The formula for the membership function of the triangle is:

[0096]

[0097] in, Fuzzy subset of real-time data from switchgear Enter the exact value. For the type of fuzzy subset, Fuzzy subset of real-time data from switchgear membership degree Let be the left endpoint of the trigonometric function. The vertex of the trigonometric function, is the right endpoint of the trigonometric function.

[0098] The fuzzy set of input variables can be represented as:

[0099]

[0100] in, For a fuzzy set of input variables, Fuzzy subset of real-time data from switchgear Enter the exact value. Fuzzy subset of real-time data from switchgear membership degree For the type of fuzzy subset, The general type of fuzzy subsets.

[0101] Generation of a weighted fuzzy rule base. The rule base construction requires combining historical data and introducing a weighting mechanism: First, based on historical load and temperature data, typical operating conditions are divided using clustering algorithms such as K-means to reduce rule redundancy; second, for each operating condition, initial fuzzy rules are extracted using the Apriori algorithm; third, based on historical operating performance data, a weight learning algorithm is used to assign weight coefficients to each rule; finally, these are integrated to form a weighted fuzzy rule base, where rules with higher weights have higher priority in inference.

[0102] Then comes fuzzy inference based on a rule base. Using the fuzzy set of input variables as a basis, triggering conditions are matched in a weighted rule base; the strength of each triggering rule is calculated; subsequently, the outputs of all triggering rules are aggregated and superimposed according to their strength to obtain a set of fuzzy temperature control strategies, reflecting the applicability of different strategies.

[0103] In this embodiment, the de-obfuscation calculation is specifically:

[0104] The fuzzy temperature control strategy set obtained based on the fuzzy inference is received, and the fuzzy temperature control strategy set is in the form of a membership function to represent the applicability of different temperature control action intensities.

[0105] The barycentric coordinates of the fuzzy temperature control strategy set are calculated by using the barycentric method, and the abscissa value of the barycentric coordinates is mapped to an accurate temperature control instruction initial value.

[0106] According to the physical characteristics of the switch cabinet cooling equipment, the accurate temperature control instruction initial value is standardized and discretized to generate a specific temperature control response strategy that can be directly issued.

[0107] In this embodiment, the weighted fuzzy rule base is generated according to the historical data of the switch cabinet, specifically:

[0108] According to the historical load and temperature data, a typical operating condition mode is obtained by a clustering algorithm.

[0109] According to the typical operating condition mode, an initial fuzzy rule is extracted by an association rule mining method.

[0110] According to the historical operating effect data, the weight coefficients of each rule are calculated by a weight learning algorithm.

[0111] Based on the initial fuzzy rule and the weight coefficient, a weighted fuzzy rule base is generated by integration.

[0112] Based on the historical load and temperature data, a typical operating condition mode is extracted by a clustering algorithm. The core of clustering is to classify similar operating states in a large amount of historical data into one category, reduce redundancy, and highlight key modes. The load dimension and temperature dimension that can reflect the operating state in the historical data are selected, and a K-means clustering algorithm is used for grouping: the elbow method is used to determine the optimal clustering number, and the algorithm will cluster the data points into several clusters according to the feature similarity, and each cluster represents a typical operating condition. These typical operating conditions strip the data noise and provide a clear scene basis for subsequent rule extraction.

[0113] According to the typical operating condition mode, an initial fuzzy rule is extracted by an association rule mining method.

[0114] In this embodiment, the initial fuzzy rule is extracted from the typical operating condition mode by the association rule mining method, specifically:

[0115] The features in the typical operating condition mode are discretized and fuzzy graded to form a set of operating condition feature items.

[0116] Based on the set of operating condition feature items, each operating condition sample is converted into an operating condition sample record to construct an operating condition sample database.

[0117] The support degree of the working condition characteristic item set is calculated in the working condition sample database, and a minimum support degree threshold is set to screen out a frequent item set;

[0118] Based on the frequent item set, an association rule set is generated according to the division rule of input items and output items;

[0119] The working condition items in the association rule set are mapped to corresponding fuzzy language variables to form an initial fuzzy rule set of "if-then" structure.

[0120] Preferably, the feature data in the typical operating condition mode is preprocessed first, and the feature data includes temperature data, current data, temperature rise rate, humidity data and environmental parameters. The preprocessing step includes removing abnormal data and completing missing data to ensure the integrity and accuracy of the input data. Further, the above-mentioned feature data is segmented, the continuous numerical value is divided into several intervals, and the fuzzy language variables are assigned to the intervals according to the operating characteristics, for example, the temperature is divided into low temperature, medium temperature and high temperature, the current is divided into light load, rated and overload, and the temperature rise rate is divided into decline, stable and rise. In this way, a working condition characteristic item set is formed to provide a basis for subsequent fuzzy rule establishment.

[0121] Further, based on the working condition characteristic item set, each operating condition sample is converted into a working condition sample record, and each sample contains several working condition items. By collecting a large number of working condition samples and converting them, a working condition sample database is constructed. The working condition sample database can comprehensively reflect the characteristic combination relationship of the switchgear under different working conditions.

[0122] Preferably, in the working condition sample database, the frequency of the occurrence of the working condition characteristic item set is counted, the support degree is calculated, and a minimum support degree threshold is set to screen out a frequent item set. The screening of the frequent item set can ensure that the rules generated subsequently have certain representativeness and reliability, and avoid the interference of low-frequency events on the effectiveness of the rules.

[0123] Further, based on the frequent item set, a candidate association rule is generated according to the division mode of input items and output items. For example, the input item can be the combination of high load current and temperature rise, and the output item can be temperature rise. Subsequently, the confidence, lift and correlation indexes of the candidate rule are calculated, and compared with the preset threshold to obtain an effective association rule set. At the same time, the effective rules are subjected to redundancy elimination and conflict detection, and the repeated or contradictory rules are deleted to ensure the simplicity and accuracy of the rule set.

[0124] Preferably, the working condition items in the reduced association rule set are mapped to corresponding fuzzy language variables, and combined with their membership function relationship to form an initial fuzzy rule set. For example, if the current is high and the temperature rise rate is rising, the temperature is high. The initial fuzzy rule set provides a direct basis for subsequent establishment of a weighted fuzzy rule base.

[0125] It should be noted that the purpose of the association rule mining is to discover the potential association relationship between the working condition characteristics and the effective temperature control strategy by analyzing the characteristics of the typical operating condition mode. Specifically, the association rule mining can not only reveal the coupling law between the operating parameters such as current, temperature, humidity, etc., but also deduce the corresponding temperature control strategy mode on this basis, thereby providing data support for the automatic generation of fuzzy rules. Compared with the traditional method of relying on artificial experience to set rules, this method can extract a rule set with stronger universality and higher adaptability in a larger scale of data samples, ensuring the matching and foresight of the temperature control strategy and the actual operating condition.

[0126] The weight coefficients of each rule are calculated based on historical operating effect data through a weight learning algorithm. The role of the weight is to quantify the actual effectiveness of the rule, so that the rules with better performance have higher priority in reasoning. The historical operating effect data includes key indicators after the execution of each rule in the past, such as temperature reduction efficiency, energy consumption cost, and equipment wear. The weight learning algorithm is used to calculate the weight: first, set the weight for each effect indicator, then calculate the comprehensive effect score of each rule, and finally normalize the score to obtain the rule weight coefficient. The higher the weight, the better the comprehensive performance of the rule in historical application.

[0127] Finally, the initial fuzzy rules and the weight coefficients are integrated to generate a weighted fuzzy rule base. Each initial rule is bound with its corresponding weight coefficient to form a structured entry of "rule content + weight value", such as "Rule 1: If the load is heavy and the temperature is high, the strategy is fan full speed + heat sink on" and "Rule 2: If the load is medium and the temperature fluctuation is medium, the strategy is fan medium speed + 10 minutes monitoring". At the same time, a rule indexing mechanism is established to ensure quick matching of related rules during reasoning. The final weighted fuzzy rule base not only contains the decision logic extracted from historical data, but also reflects the actual effect difference of the rules through the weight, providing a decision basis for subsequent fuzzy reasoning that takes into account experience and efficiency.

[0128] S5, according to the temperature control response strategy, regulating the operating state of the temperature control instrument control system.

[0129] In this embodiment, according to the temperature control response strategy, the operating state of the temperature control instrument control system is regulated, specifically:

[0130] Based on the temperature control response strategy, specific control instructions and target parameters are generated through a pre-set analysis protocol;

[0131] According to the control instruction and the target parameter, each execution module in the temperature control instrument control system is regulated.

[0132] The specific control instruction and the target parameter are generated through the preset analysis protocol, and specifically:

[0133] The regulation and control requirements and the target indicators in the temperature control response strategy are input to the protocol analysis module;

[0134] According to the execution module corresponding to the strategy, the module address is determined, and the corresponding instruction frame header is generated;

[0135] The instruction code is determined based on the regulation and control type, such as heat dissipation control, heating control, or sensor sampling control;

[0136] The target indicator values in the temperature control response strategy are mapped to the corresponding parameter fields;

[0137] The parameter range of the instruction content is verified, and a complete frame structure is generated;

[0138] A check field is generated through CRC check to ensure the integrity of the instruction in the transmission process;

[0139] Finally, the control instruction conforming to the analysis protocol format is formed, and the target parameter is issued to the corresponding execution module.

[0140] According to the control instruction and the target parameter, each execution module is regulated, and multi-module collaborative action and dynamic feedback are realized. The core execution modules of the temperature control instrument control system include a heat dissipation module, a heating module, and a monitoring module. After receiving the corresponding instructions, each module completes the action through a drive circuit and control logic. If multiple modules are involved, the main control unit will issue instructions according to the preset time sequence, and simultaneously receive feedback data from the monitoring module. If it is found that the actual temperature reduction effect does not meet the expectation, the strategy will be reanalyzed to generate supplementary instructions until the temperature stabilizes in the target range, forming a closed-loop regulation and control of instruction execution, data feedback, and dynamic adjustment, ensuring that the temperature control system adapts to the real-time changes of the load and thermal effect.

[0141] In the closed-loop regulation and control process, the fuzzy rule weight is continuously optimized based on the temperature control execution feedback result, and the weight update formula is as follows:

[0142]

[0143] wherein, is the updated temperature control execution fuzzy rule weight coefficient; is the weight of the th temperature control execution fuzzy rule at time ; is the adaptive learning rate; an adjustment factor calculated from a feedback error function.

[0144] Example 2, to verify the applicability and effectiveness of the method of the present application, a reduced example system of a provincial power grid is selected as the research object, and the method described in Example 1 is used for analysis.

[0145] 1) Example scenario description

[0146] The reduced example system contains 10 main nodes and 15 transmission lines. The operation scenarios cover three typical load intervals: low valley (0-6h), flat section (6-18h), and peak (18-24h). The collected characteristic data includes: node current, voltage amplitude, cabinet temperature and humidity. Table 1 shows some examples of operating parameters:

[0147] Table 1

[0148]

[0149] The system data fully reflects the law of load and temperature changes with the daily cycle, providing a typical test scenario for verifying the prediction of thermal effects and the generation of temperature control strategies.

[0150] 2) Model construction and thermal effect prediction

[0151] Based on Step S1 of Example 1, the thermal effect cycle characteristics are extracted from the node temperature and humidity data and environmental parameters. In this example, the cycle amplitude is set to 5°C, the reference temperature is 30°C, and the phase shift is 0.3 rad. The main cycle is determined using autocorrelation analysis, and the periodic variation function is obtained through the smooth smoothing fitting algorithm.

[0152] Based on Step S2 of Example 1, based on the node current, voltage, and cabinet temperature data in Table 2, the load dynamic characteristic vector is extracted, and it is combined with the thermal effect cycle function to form a joint feature sequence. Through time series regression algorithm, the load-thermal effect rhythm regulation model is constructed, which can dynamically predict the future thermal effect.

[0153] Based on Step S3 of Example 1, the fusion feature matrix is constructed and the LSTM input sequence is generated to predict the future 24-hour cabinet temperature. Table 2 shows the prediction results:

[0154] Table 2

[0155]

[0156] Error analysis shows that the maximum deviation between the predicted value and the measured value is not more than 0.3°C, verifying the high precision and forward-looking nature of the model.

[0157] Figure 3The comparison of the thermal effect cycle function fitting and the temperature prediction results is shown in the figure. It can be seen from the figure that the prediction value is highly consistent with the measured value, and the error is small, which shows that the model can accurately capture the periodic change trend of the temperature of the switch cabinet, and has good prediction performance.

[0158] 3) Temperature control response strategy generation and verification

[0159] According to the predicted temperature and real-time load data, a temperature control state fuzzy set (low load / medium load / high load, temperature deviation degree, etc.) is constructed, and a weighted fuzzy rule base is called for fuzzy reasoning. The specific temperature control response strategy is generated by the barycentric method.

[0160] Table 3

[0161]

[0162] The strategy is sent to the temperature control instrument execution module to monitor the temperature change in the cabinet, and the results show that:

[0163] In the load peak (18h) stage, through the fan full speed + auxiliary cooling fin strategy, the temperature in the cabinet is stabilized at the target value of 36±0.5℃; in the low load stage, the temperature control system automatically reduces the fan speed, realizing the energy saving and temperature control target; in the temperature control closed loop, the rule weight is dynamically updated according to the feedback error, and the strategy adaptability is optimized.

[0164] Figure 4 The temperature control effect of the intelligent temperature control method proposed in the present application and the traditional fixed threshold control method in actual operation is compared. It can be seen from the figure that the temperature control difference of the two methods: the present application method can smoothly control, the temperature fluctuation is small; the traditional method fluctuates greatly, and there is a risk of overheating. Through comparison, the advantages of the present application are highlighted: higher control accuracy, lower energy consumption, fewer device action times, thereby prolonging the service life of the equipment and improving the system reliability.

[0165] In order to quantify the advantages of the present application, the present application method and the traditional threshold control method (temperature > 70℃ start strong cooling: fan 100%, cooling fin 100%; temperature < 65℃ close) are compared under the same example. The performance comparison of the present application and the traditional method is shown in Table 4.

[0166] Table 4

[0167]

[0168] Control effect: The method of the application can stably control the cabinet temperature below 68.5 DEG C during the whole peak period, without overshoot. While the traditional method continuously rises the temperature under the action of load inertia and triggers the 70 DEG C threshold multiple times, resulting in the highest temperature reaching 72.5 DEG C, which has the risk of overheating.

[0169] Economy and equipment life: The method of the application avoids frequent start-stop and long-time operation of the actuator at 100% power due to early and smooth intervention, effectively prolonging the equipment life.

[0170] Control quality: The temperature fluctuation standard deviation under the control of the application is 1.2 DEG C, which is much lower than the 3.5 DEG C of the traditional method, and the control process is more stable.

[0171] Example 3, Figure 2 The control system of the temperature control instrument of the switch cabinet is given, which comprises a function construction module, a model construction module, a thermal effect prediction module, a strategy generation module and a control execution module, and there is a connection between the modules:

[0172] The function construction module constructs a thermal effect periodic change function according to the obtained switch cabinet first data and a preset period length;

[0173] The model construction module constructs a load-thermal effect rhythm regulation model based on the thermal effect periodic change function and the obtained current load data;

[0174] The thermal effect prediction module generates a switch cabinet thermal effect prediction result based on the regulation model through machine learning;

[0175] The strategy generation module generates a temperature control response strategy through fuzzy logic reasoning according to the prediction result, combined with the current load level and environmental boundary conditions;

[0176] The control execution module dynamically regulates the running state of the temperature control instrument control system according to the temperature control response strategy.

[0177] The above formulas are all dimensionless numerical calculations, and the formulas are obtained by software simulation of a large amount of data to obtain a formula of the most recent real situation, and the preset parameters in the formula are set by the person skilled in the art according to the actual situation.

[0178] The above embodiments can be realized all or partially by software, hardware, firmware or any combination thereof. When realized by software, the above embodiments can be realized in the form of a computer program product all or partially.

[0179] Those skilled in the art can understand that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0180] In addition, each functional module in each embodiment of the present application can be integrated in one processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0181] The above is merely specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any skilled in the art within the technical scope disclosed in the present application can easily think of changes or replacements, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0182] Finally: the above is only the preferred embodiment of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A control method of a temperature control instrument of a switchgear, characterized by, The method comprises the following steps: According to the obtained switch cabinet first data, a thermal effect periodic change function is constructed with a preset period length, specifically, the switch cabinet first data is input into a smooth fitting algorithm, and data smoothing fitting is performed with the period length as a constraint condition to obtain the thermal effect periodic change function, wherein the period length is determined by time series analysis to extract the period of thermal effect change; Based on the thermal effect periodic change function and the obtained current load data, a load-thermal effect rhythm regulation model is constructed, specifically: based on the current load data, a load dynamic feature vector is obtained through a feature extraction algorithm; according to the load dynamic feature vector and the thermal effect periodic change function, a joint feature sequence is obtained through a multi-source feature fusion algorithm; based on the joint feature sequence, a load-thermal effect rhythm regulation model is constructed through a time series regression algorithm; Based on the regulation model, a switch cabinet thermal effect prediction result is generated through machine learning; According to the prediction result, a temperature control response strategy is generated through fuzzy logic reasoning combined with real-time load state data; According to the temperature control response strategy, the operating state of the temperature control instrument control system is regulated.

2. The control method of the temperature control instrument of the switchboard according to claim 1, characterized in that, The switch cabinet thermal effect prediction result is generated through machine learning based on the regulation model, specifically: Based on the output of the load-thermal effect rhythm regulation model, an LSTM input sequence is constructed; Based on the LSTM input sequence, a regular time series sample set is obtained through data standardization and time step alignment processing; According to the time series sample set, a switch cabinet thermal effect prediction result is obtained through a long short-term memory network algorithm.

3. The control method of the temperature control instrument of the switchboard according to claim 2, characterized in that, Based on the output of the load-thermal effect rhythm regulation model, an LSTM input sequence is constructed, specifically: According to the output of the load-thermal effect rhythm regulation model, a load dynamic feature vector and a thermal effect feature vector are extracted; According to the load dynamic feature vector and the thermal effect feature vector, a fusion feature matrix is obtained through feature splicing; Based on the fusion feature matrix, an LSTM input sequence is constructed through a sliding window sampling method.

4. The control method of the temperature control instrument of the switchboard according to claim 3, characterized by, According to the prediction result, a temperature control response strategy is generated through fuzzy logic reasoning combined with real-time load state data, specifically: Based on the prediction result and the real-time load state data, a membership function is constructed to obtain a temperature control state fuzzy set; According to the obtained switch cabinet historical data, a weighted fuzzy rule base is generated; Based on the temperature control state fuzzy set, inference is performed through the weighted fuzzy rule base to obtain a fuzzy temperature control strategy set; Based on the fuzzy temperature control strategy set, a temperature control response strategy is generated through defuzzification calculation.

5. The control method of the temperature control instrument of the switchboard according to claim 4, characterized in that, According to the obtained switch cabinet historical data, a weighted fuzzy rule base is generated, specifically: According to the historical load and temperature data, a typical operating condition mode is obtained through a clustering algorithm; According to the typical operating condition mode, initial fuzzy rules are extracted through an association rule mining method; According to the historical operating effect data, weight learning algorithm is used to calculate the weight coefficients of each rule; Based on the initial fuzzy rules and the weight coefficients, a weighted fuzzy rule base is generated through integration.

6. The control method of the temperature control instrument of the switchboard according to claim 5, characterized in that, According to the typical operating condition mode, initial fuzzy rules are extracted through an association rule mining method, specifically: Discretization and fuzzy classification of the features in the typical operating condition mode are performed to form a set of condition feature items; Based on the set of condition feature items, each operating condition sample is converted into a condition sample record to construct a condition sample database; Support degrees of the set of condition feature items are calculated in the condition sample database, and a minimum support threshold is set to screen out a frequent item set; Based on the frequent item set, an association rule set is generated according to the division rules of input items and output items; Condition items in the association rule set are mapped to corresponding fuzzy language variables to form an initial fuzzy rule set of "if-then" structure.

7. A system for controlling a temperature control instrument using the switchgear according to any one of claims 1 to 6, characterized in that, It comprises: A function construction module constructs a thermal effect periodic change function according to the obtained first data of the switch cabinet and a preset period length, specifically, the first data of the switch cabinet is input into a smooth fitting algorithm, and data smoothing fitting is performed with the period length as a constraint condition to obtain the thermal effect periodic change function, and the period length is determined by time series analysis to extract the period of thermal effect change; A model construction module constructs a load-thermal effect rhythm regulation model based on the thermal effect periodic change function and the obtained current load data, specifically: based on the current load data, a load dynamic feature vector is obtained through a feature extraction algorithm; according to the load dynamic feature vector and the thermal effect periodic change function, a joint feature sequence is obtained through a multi-source feature fusion algorithm; based on the joint feature sequence, a load-thermal effect rhythm regulation model is constructed through a time series regression algorithm; A thermal effect prediction module generates a switch cabinet thermal effect prediction result through machine learning based on the regulation model; A strategy generation module generates a temperature control response strategy through fuzzy logic reasoning according to the prediction result and in combination with real-time load state data; A control execution module regulates the operating state of a temperature control instrument control system according to the temperature control response strategy.

Images