Method and system for on-line identification of alumina concentration in aluminum electrolysis based on fuzzy inference

Abstract

The present disclosure provides an aluminum electrolytic alumina concentration online identification method and system based on fuzzy reasoning, which comprises: obtaining apparent alumina concentration and comprehensive discharging period through a time sequence knowledge graph, calculating the PDHC value of the current comprehensive discharging period and the previous comprehensive discharging period; inputting the PDHC value of the current comprehensive discharging period and the previous comprehensive discharging period into a preset fuzzy reasoning machine to obtain the fuzzy reasoning output variable of the alumina concentration state and the alumina concentration change trend of the current comprehensive discharging period; and performing anti-fuzzification processing on the fuzzy reasoning output variable of the alumina concentration state and the alumina concentration change trend of the current comprehensive discharging period to obtain the identification evaluation value of the alumina concentration state and the alumina concentration change trend of the current comprehensive discharging period.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of industrial semantics, and more particularly, to an aluminum electrolysis alumina concentration online identification method and system based on fuzzy reasoning. BACKGROUND

[0002] Alumina concentration is a core parameter in the aluminum electrolysis production process, directly affecting the material balance of the electrolytic cell. Alumina concentration monitoring is of great significance in reducing production energy consumption and improving product quality. However, with the large-scale process of aluminum electrolytic cells, aluminum electrolysis data presents the characteristics of massive data, dispersion, and coexistence of multiple formats, which brings challenges to the online identification of alumina concentration. Therefore, how to process the production knowledge in the aluminum electrolysis process and achieve accurate online identification of alumina concentration is the key to improving the quality and efficiency of the aluminum electrolysis industry. When the alumina concentration is in a normal state, there is an "under-raise and over-fall" relationship between the normalized cell voltage and the discharging state, i.e., when the discharging state is under-discharging, the cell voltage is in a rising state, and when the discharging state is over-discharging, the cell voltage is in a falling state, and the PDHC value is within the normal range. When the concentration is in an abnormal state allowed by the process, the "under-raise and over-fall" relationship is destroyed, and the value of PDHC also changes. Therefore, the online identification of alumina concentration can be achieved according to the value of PDHC. In the aluminum electrolysis multi-granularity time sequence knowledge graph, the value of PDHC is stored at the process semantic level.

[34]

[0003] Traditional alumina concentration identification methods mainly include mechanism model-based methods and pure data-driven methods. The mechanism model-based method extracts key parameters from process mechanism knowledge or experimental electrolytic cells to construct a model for estimating and predicting alumina concentration. In the model-based method, the key parameters are mostly fixed values, which do not match the actual situation of real-time changes in key parameters in aluminum electrolysis production and the inability to quickly and continuously measure. Therefore, the molecular dynamics and fluid dynamics models established from a microscopic perspective have a high calculation cost and are difficult to apply to industrial production. The second type of method uses data as the driving force to try to obtain knowledge from parameters such as cell voltage to estimate and predict alumina concentration. However, the above method lacks process mechanism knowledge and is not sensitive to abnormal cell conditions of the electrolytic cell, making it difficult to adapt to frequent changes in working conditions in actual production.

[0004] ​In summary, many key parameters such as aluminum output, aluminum level, electrolyte level, etc. in the existing alumina concentration identification method cannot be measured continuously and quickly, and are mostly measured offline by manual and stored in the process report. The production operation data such as cell voltage and cell current collected by industrial sensors are time series data, which are stored in a time series database. Therefore, the identification of alumina concentration involves a large amount of multi-source heterogeneous data, and it is difficult to achieve online identification. In addition, the existing method can only identify the normal state of alumina concentration, and it is difficult to identify the online identification of the abnormally low and abnormally high state of alumina concentration. The alumina concentration is too low to cause anode effect and destroy the material balance of the electrolytic cell, so the alumina concentration can be identified by predicting the anode effect. The alumina concentration is too high to reduce the current efficiency and endanger the stability of the electrolytic cell, but there are very few online detection methods for high alumina concentration in the electrolytic cell at present, which is difficult to meet the demand for comprehensive identification of the state of alumina concentration. SUMMARY

[0005] The purpose of the embodiments of the present disclosure is to provide an aluminum electrolysis alumina concentration online identification method and system based on fuzzy reasoning to improve the accuracy of online detection of high alumina concentration in the electrolytic cell.

[0006] In a first aspect, the present application provides an aluminum electrolysis alumina concentration online identification method based on fuzzy reasoning, comprising: an aluminum electrolysis alumina concentration online identification method based on fuzzy reasoning, characterized in that, comprising:

[0007] The apparent alumina concentration and the comprehensive discharging period are obtained through the time series knowledge graph, and the PDHC values of the current comprehensive discharging period and the previous comprehensive discharging period are calculated;

[0008] The PDHC values of the current comprehensive discharging period and the previous comprehensive discharging period are input into the preset fuzzy reasoning machine to obtain the fuzzy reasoning output variables of the alumina concentration state and the alumina concentration change trend of the current comprehensive discharging period;

[0009] The fuzzy reasoning output variables of the alumina concentration state and the alumina concentration change trend of the current comprehensive discharging period are subjected to inverse fuzzy processing to obtain the identification evaluation values of the alumina concentration state and the alumina concentration change trend of the current comprehensive discharging period.

[0010] Further, the preset fuzzy reasoning machine is established by the following steps:

[0011] The first threshold value, the second threshold value, the third threshold value and the fourth threshold value are selected according to the numerical range of the PDHC value in the historical data of various types of electrolytic cells under various states of alumina concentration, and a fuzzy threshold range is preset for each type of electrolytic cell; wherein the second threshold value < the first threshold value < the third threshold value < the fourth threshold value; the fuzzy threshold range includes an alumina concentration normal threshold range for representing that the PDHC value is greater than the first threshold value and less than the third threshold value, a negative small threshold range for representing that the PDHC value is less than the first threshold value and greater than the second threshold value, a negative large threshold range for representing that the PDHC value is less than the second threshold value, a positive small threshold range for representing that the PDHC value is greater than the third threshold value and less than the fourth threshold value, and a positive large threshold range for representing that the PDHC value is greater than the fourth threshold value;

[0012] Each PDHC membership function corresponding to the normal threshold range, the negative small threshold range, the negative large threshold range, the positive small threshold range and the positive large threshold range is established;

[0013] The fuzzy inference rule of the alumina concentration is established according to the PDHC value of the current comprehensive feeding period, and the fuzzy inference rule of the alumina concentration change trend includes an alumina concentration normal state for representing that the PDHC value is located in the normal threshold range, an alumina concentration high state for representing that the PDHC value is located in the negative small threshold range, an alumina concentration extremely high state for representing that the PDHC value is located in the negative large threshold range, an alumina concentration low state for representing that the PDHC value is located in the positive small threshold range, and an alumina concentration extremely low state for representing that the PDHC value is located in the positive large threshold range;

[0014] The alumina concentration membership function corresponding to the alumina concentration normal state, the alumina concentration low state, the alumina concentration high state, the alumina concentration extremely low state and the alumina concentration extremely high state is established;

[0015] The fuzzy inference rule of the alumina concentration change trend is established according to the PDHC values of two comprehensive feeding periods, and the fuzzy inference rule of the alumina concentration change trend includes a first alumina concentration change trend for representing that the alumina concentration is unchanged, a second alumina concentration change trend for representing that the alumina concentration is increased, a third alumina concentration change trend for representing that the alumina concentration is decreased, a fourth alumina concentration change trend for representing that the alumina concentration is from low to normal, and a fifth alumina concentration change trend for representing that the alumina concentration is from high to normal;

[0016] The alumina concentration change trend membership function corresponding to the first alumina concentration change trend, the second alumina concentration change trend, the third alumina concentration change trend, the fourth alumina concentration change trend and the fifth alumina concentration change trend is established.

[0017] Further, the step of inputting the PDHC values of the current comprehensive blanking cycle and the previous comprehensive blanking cycle into a preset fuzzy inference machine to obtain fuzzy inference output variables of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking cycle comprises:

[0018] According to the PDHC membership functions, the PDHC values of the current comprehensive blanking cycle and the previous comprehensive blanking cycle correspond to PDHC fuzzy values;

[0019] According to the PDHC fuzzy values corresponding to the PDHC values of the current comprehensive blanking cycle and the previous comprehensive blanking cycle, the fuzzy inference rules of the alumina concentration and the fuzzy inference rules of the alumina concentration change trend, the fuzzy inference output variables of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking cycle are obtained.

[0020] Further, the step of performing defuzzification processing on the fuzzy inference output variables of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking cycle to obtain recognition evaluation values of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking cycle further comprises:

[0021] According to the alumina concentration membership function and the alumina concentration change trend membership function, the fuzzy inference output variables of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking cycle are defuzzified by using the barycenter method to obtain the recognition evaluation values of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking cycle.

[0022] Further, the PDHC membership functions of the normal threshold range, the negative small threshold range, and the positive small threshold range are generalized bell-shaped membership functions; the PDHC membership function of the negative large threshold range is a Z-shaped membership function, and the PDHC membership function of the positive large threshold range is an S-shaped membership function; the alumina concentration membership function and the alumina concentration change trend membership function are triangular membership functions;

[0023] After the step of obtaining the recognition evaluation values of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking cycle, the following step is further included:

[0024] The recognition evaluation values of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking cycle are stored in a time sequence knowledge graph, and a corresponding relationship between the electrolytic cell and the alumina concentration is established according to the recognition evaluation values of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking cycle to reflect the online working condition in the aluminum electrolysis production process.

[0025] In a second aspect, the present application provides an aluminum electrolysis alumina concentration online identification system based on fuzzy reasoning, comprising:

[0026] a PDHC acquisition module, configured to acquire apparent alumina concentration and comprehensive discharging period through a timing knowledge graph, and calculate PDHC values of a current comprehensive discharging period and a previous comprehensive discharging period;

[0027] a fuzzy reasoning module, configured to input the PDHC values of the current comprehensive discharging period and the previous comprehensive discharging period into a preset fuzzy reasoning machine to obtain fuzzy reasoning output variables of alumina concentration state and alumina concentration trend of the current comprehensive discharging period;

[0028] an online identification module, configured to perform defuzzification processing on the fuzzy reasoning output variables of alumina concentration state and alumina concentration trend of the current comprehensive discharging period to obtain identification evaluation values of alumina concentration state and alumina concentration trend of the current comprehensive discharging period.

[0029] Further, the preset fuzzy reasoning machine comprises:

[0030] a PDHC membership function modeling module, configured to select first, second, third and fourth threshold values according to the numerical ranges of PDHC values in alumina concentration states of each type of electrolytic cell based on historical data of each type of electrolytic cell, and preset a fuzzy threshold range for each type of electrolytic cell; wherein the second threshold value < the first threshold value < the third threshold value < the fourth threshold value; the fuzzy threshold range comprises an alumina concentration normal threshold range for representing that the PDHC value is greater than the first threshold value and less than the third threshold value, a negative small threshold range for representing that the PDHC value is less than the first threshold value and greater than the second threshold value, a negative large threshold range for representing that the PDHC value is less than the second threshold value, a positive small threshold range for representing that the PDHC value is greater than the third threshold value and less than the fourth threshold value, and a positive large threshold range for representing that the PDHC value is greater than the fourth threshold value; and each PDHC membership function corresponding to the normal threshold range, the negative small threshold range, the negative large threshold range, the positive small threshold range and the positive large threshold range is established.

[0031] an alumina concentration membership function modeling module, configured to establish fuzzy inference rules of the alumina concentration according to the PDHC value of the current comprehensive blanking cycle, the fuzzy inference rules of the alumina concentration change trend including a normal alumina concentration state for representing that the PDHC value is within the normal threshold range, a high alumina concentration state for representing that the PDHC value is within the negative small threshold range, an extremely high alumina concentration state for representing that the PDHC value is within the negative large threshold range, a low alumina concentration state for representing that the PDHC value is within the positive small threshold range, and an extremely low alumina concentration state for representing that the PDHC value is within the positive large threshold range; and establish alumina concentration membership functions corresponding to the normal alumina concentration state, the low alumina concentration state, the high alumina concentration state, the extremely low alumina concentration state, and the extremely high alumina concentration state;

[0032] an alumina concentration change trend membership function modeling module, configured to establish fuzzy inference rules of the alumina concentration change trend according to the PDHC values of two comprehensive blanking cycles, the fuzzy inference rules of the alumina concentration change trend including a first alumina concentration change trend for representing that the alumina concentration is unchanged, a second alumina concentration change trend for representing that the alumina concentration is increased, a third alumina concentration change trend for representing that the alumina concentration is decreased, a fourth alumina concentration change trend for representing that the alumina concentration is changed from low to normal, and a fifth alumina concentration change trend for representing that the alumina concentration is changed from high to normal; and establish alumina concentration change trend membership functions corresponding to the first alumina concentration change trend, the second alumina concentration change trend, the third alumina concentration change trend, the fourth alumina concentration change trend, and the fifth alumina concentration change trend.

[0033] Further, the fuzzy inference module is specifically configured to obtain PDHC fuzzy values corresponding to the PDHC values of the current comprehensive blanking cycle and the previous comprehensive blanking cycle according to the respective PDHC membership functions; and obtain fuzzy inference output variables of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking cycle according to the PDHC fuzzy values corresponding to the PDHC values of the current comprehensive blanking cycle and the previous comprehensive blanking cycle, the fuzzy inference rules of the alumina concentration, and the fuzzy inference rules of the alumina concentration change trend.

[0034] Further, the online identification module is specifically configured to perform defuzzification processing on the fuzzy inference output variables of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking cycle by using the barycentric method according to the alumina concentration membership functions and the alumina concentration change trend membership functions, to obtain identification evaluation values of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking cycle.

[0035] Further, the PDHC membership functions of the normal threshold range, the negative small threshold range and the positive small threshold range are generalized bell-shaped membership functions; the PDHC membership function of the negative large threshold range is a Z-shaped membership function, and the PDHC membership function of the positive large threshold range is an S-shaped membership function.

[0036] The alumina concentration membership function and the alumina concentration change trend membership function are triangular membership functions.

[0037] The online alumina concentration identification method and system based on fuzzy reasoning of the aluminum electrolysis according to the present application utilizes the online alumina concentration identification method based on the multi-granularity time sequence knowledge graph, obtains the value of the PDHC through the time sequence knowledge graph to perform online identification of the alumina concentration, stores the identification result into the time sequence knowledge graph, establishes a corresponding relationship between the electrolytic cell and the alumina concentration according to the result, and reflects the online working condition in the aluminum electrolysis production process. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without any creative effort.

[0039] Figure 1 is a flowchart of the online alumina concentration identification method based on fuzzy reasoning of the aluminum electrolysis according to the embodiments of the present application.

[0040] Figure 2 is a PDHC membership function curve diagram in the online alumina concentration identification method based on fuzzy reasoning of the aluminum electrolysis according to the embodiments of the present application.

[0041] Figure 3 is a PDHC and alumina concentration change trend surface diagram in the online alumina concentration identification method based on fuzzy reasoning of the aluminum electrolysis according to the embodiments of the present application.

[0042] Figure 4 is an alumina concentration state identification result display diagram in the online alumina concentration identification method based on fuzzy reasoning of the aluminum electrolysis according to the embodiments of the present application.

[0043] Figure 5 is a structural schematic diagram of the online alumina concentration identification system based on fuzzy reasoning of the aluminum electrolysis according to the embodiments of the present application. DETAILED DESCRIPTION

[0044] The embodiments of the present application will be described in detail below with reference to the drawings.

[0045] It should be noted that the following embodiments and features in the embodiments can be combined with each other in the case of no conflict; and all other embodiments obtained by those skilled in the art based on the embodiments in the present disclosure without creative labor are within the scope of protection of the present disclosure.

[0046] It should be noted that various aspects of the embodiments described below are within the scope of the appended claims. As will be apparent, the aspects described in the embodiments can be implemented in a wide variety of forms, and that any particular structure and / or function described in the embodiments is merely illustrative. Based on the present disclosure, one of ordinary skill in the art will appreciate that one aspect described in the embodiments can be implemented independently of any other aspect and that two or more aspects described in the embodiments can be combined in various ways. For example, an apparatus can be implemented and / or a method can be practiced using any number of the aspects set forth in the embodiments. In addition, such an apparatus can be implemented and / or such a method can be practiced using other structure and / or functionality in addition to or other than one or more of the aspects set forth in the embodiments.

[0047] Figure 1 is a flowchart of a fuzzy reasoning-based on-line identification method of aluminum electrolysis alumina concentration according to an embodiment of the present disclosure.

[0048] As shown in Figure 1 the fuzzy reasoning-based on-line identification method of aluminum electrolysis alumina concentration includes:

[0049] Step 101: obtaining apparent alumina concentration and comprehensive discharging period through a time sequence knowledge graph, and calculating PDHC values of a current comprehensive discharging period and a previous comprehensive discharging period;

[0050] Step 102: inputting the PDHC values of the current comprehensive discharging period and the previous comprehensive discharging period into a preset fuzzy reasoning machine to obtain fuzzy reasoning output variables of alumina concentration state and alumina concentration change trend of the current comprehensive discharging period;

[0051] Step 103: performing defuzzification processing on the fuzzy reasoning output variables of alumina concentration state and alumina concentration change trend of the current comprehensive discharging period to obtain identification evaluation values of alumina concentration state and alumina concentration change trend of the current comprehensive discharging period.

[0052] The embodiment first acquires apparent alumina concentration PAC and comprehensive blanking cycle IFP through a time sequence knowledge graph, calculates PDHC value of the comprehensive blanking cycle, inputs the PDHC value into a fuzzy inference machine which has been set, obtains an output variable of fuzzy inference, obtains an identification evaluation value of alumina concentration state and change trend after anti-fuzzification, and finally inputs the identification result of alumina concentration represented by the evaluation value into the time sequence knowledge graph.

[0053] The alumina concentration online identification method based on fuzzy inference further has various preferred embodiments. The preset fuzzy inference machine establishes fuzzy rules through the following steps: selecting first, second, third and fourth threshold values according to the numerical range of PDHC value in each state of alumina concentration based on historical data of various types of electrolytic cells, and presetting a fuzzy threshold range for each type of electrolytic cell; wherein the second threshold value < the first threshold value < the third threshold value < the fourth threshold value; the fuzzy threshold range includes a normal threshold range of alumina concentration for representing that the PDHC value is greater than the first threshold value and less than the third threshold value, a negative small threshold range for representing that the PDHC value is less than the first threshold value and greater than the second threshold value, a negative large threshold range for representing that the PDHC value is less than the second threshold value, a positive small threshold range for representing that the PDHC value is greater than the third threshold value and less than the fourth threshold value, and a positive large threshold range for representing that the PDHC value is greater than the fourth threshold value;

[0054] Each PDHC membership function corresponding to the normal threshold range, the negative small threshold range, the negative large threshold range, the positive small threshold range and the positive large threshold range is established;

[0055] According to the PDHC value of the current comprehensive blanking cycle, the fuzzy inference rules of the alumina concentration are established, and the fuzzy inference rules of the alumina concentration change trend include a normal state of alumina concentration for representing that the PDHC value is located in the normal threshold range, a high state of alumina concentration for representing that the PDHC value is located in the negative small threshold range, an extremely high state of alumina concentration for representing that the PDHC value is located in the negative large threshold range, a low state of alumina concentration for representing that the PDHC value is located in the positive small threshold range, and an extremely low state of alumina concentration for representing that the PDHC value is located in the positive large threshold range;

[0056] The alumina concentration membership function corresponding to the normal state of alumina concentration, the low state of alumina concentration, the high state of alumina concentration, the extremely low state of alumina concentration and the extremely high state of alumina concentration is established;

[0057] According to the PDHC values of two comprehensive tapping periods, a fuzzy inference rule of the alumina concentration change trend is established, the fuzzy inference rule of the alumina concentration change trend including a first alumina concentration change trend for characterizing an invariable alumina concentration, a second alumina concentration change trend for representing an increasing alumina concentration, a third alumina concentration change trend for representing a decreasing alumina concentration, a fourth alumina concentration change trend for representing an alumina concentration changing from low to normal, and a fifth alumina concentration change trend for representing an alumina concentration changing from high to normal;

[0058] Membership functions corresponding to the first alumina concentration change trend, the second alumina concentration change trend, the third alumina concentration change trend, the fourth alumina concentration change trend, and the fifth alumina concentration change trend are established.

[0059] The above preferred embodiment specifically includes a plurality of specific optimization steps, first is the PDHC fuzzy membership function and its parameter determination, by fuzzifying the PDHC, using reasonable fuzzy rules, the PDHC is converted into a recognition evaluation index. Through fuzzification input, determination of fuzzy membership function and related parameters, the accuracy of the recognition process is further improved.

[0060] (I) Input parameter determination

[0061] In order to better identify the alumina concentration in the aluminum electrolysis process, not only the current comprehensive tapping period alumina concentration state needs to be considered, but also the alumina concentration change trend needs to be considered. If only the current PDHC value is used as the input of fuzzy reasoning, it is easy to misjudge the current electrolytic cell state, thereby providing an incorrect tapping decision. Therefore, the PDHC value of the current comprehensive tapping period and the PDHC value of the previous comprehensive tapping period are used as the input of fuzzy reasoning. Since the time range covered by one comprehensive tapping period in the used industrial data is about 2000 seconds, the PDHC of two adjacent comprehensive tapping periods can basically reflect the change trend of the alumina concentration. The fuzzy reasoning key parameter setting includes PDHC threshold determination, input parameter determination, and fuzzy membership function and its parameter determination. For the PDHC threshold, the identification of the alumina concentration is mainly realized by comparing whether the value of the PDHC exceeds the preset threshold in the positive or negative case. Due to the different models and process parameters of the electrolytic cells, the PDHC value range and threshold of each cell condition change with the electrolyte composition, which is uncertain and difficult to give an accurate threshold. In this embodiment, the approximate range of PDHC under the normal state of alumina concentration is obtained from the historical data of various types of electrolytic cells, and a fuzzy threshold is preset for each type of electrolytic cell, that is, the set threshold is a range rather than an accurate value. On this basis, the value of the PDHC is converted into a recognition evaluation index, so as to realize the online identification of the alumina concentration.

[0062] (II) Fuzzy processing of input parameters

[0063] After the input parameters for fuzzy inference are set, the input parameters need to be fuzzified. In this stage, a suitable membership function needs to be selected first.

[0064] Common fuzzy membership functions mainly include triangular membership function, trapezoidal membership function, Gaussian membership function, generalized bell-shaped membership function, S-shaped membership function, and Z-shaped membership function, etc. In this embodiment, three membership functions with relatively smooth curves are selected, i.e., generalized bell-shaped membership function, S-shaped membership function, and Z-shaped membership function. Common fuzzy methods generally divide the input parameters into central region fuzzy sets and edge region fuzzy sets, wherein the central region fuzzy sets represent that the variables are in an optimal state or a stable trend, the edge region fuzzy sets represent that the variables are in an undesirable state or have an increasing or decreasing trend, and the closer the fuzzy set is to the edge, the more extreme the state and trend are. According to the characteristics of the aluminum electrolysis production process and the possible states of the alumina concentration, the PDHC values of the input parameters are divided into fuzzy sets, and the PDHC values are divided into five fuzzy sets, i.e., normal threshold range (N), negative small threshold range (NS), negative large threshold range (NB), positive small threshold range (PS), and positive large threshold range (PB). Among them, N, NS, and PS are located in the central region, respectively representing that the PDHC values are in the normal, small, and large ranges, and NB and PB are located at the left and right ends, respectively, indicating that the PDHC values are in the extremely small or extremely large range.

[0065] Among them, since N, NS, and PS are located in the central position of all fuzzy sets, and the PDHC ranges covered by these three types of fuzzy sets (normal, small, and large) account for a small part of the overall PDHC range. Therefore, the generalized bell-shaped membership function which can be accurately adjusted is used to determine these fuzzy sets, and the form of the bell-shaped membership function is as follows.

[0066]

[0067] Among them, c i and a i are the center value positions and curve width parameters of the three fuzzy function curves, and b i is the curve slope parameter.

[0068] Since the NB fuzzy set represents that the PDHC value is in the extremely small range and is located at the left end of all fuzzy sets, the Z-shaped membership function with monotonic decreasing property is selected to determine the fuzzy set, representing the concept of "negative large", and the form of the Z-shaped membership function is as follows.

[0069]

[0070] where a i and b i determine the initial point of the function curve and the intersection point of the function and the horizontal line, respectively, and together determine the descending slope of the curve.

[0071] Since the PB fuzzy set represents the PDHC value in an extremely large region and is located at the far right end of all fuzzy sets, a S-type membership function with monotonic increasing property is selected to determine the fuzzy set, representing the "positive large" concept. The S-type membership function is shown as follows.

[0072]

[0073] where c i and d i determine the steepness of the curve and the initial horizontal translation value, respectively.

[0074] (Three) Determination of Membership Function Parameters

[0075] In the process of fuzzifying the input parameters, the parameters of the membership function can directly affect the size of each fuzzy set region and the shape of the function curve, and thus affect the results of fuzzy reasoning. Therefore, in the process of setting the parameters of the membership function, the approximate range of normal PDHC values should be determined according to the distribution of PDHC values of different electrolytic cells, and on this basis, the size of the fuzzy set and the shape of the membership function curve should be made to conform to the actual situation in industrial production as much as possible in combination with process mechanism knowledge.

[0076] Let the range of normal PDHC values be the critical point of abnormally low PDHC value be the critical point of abnormally high PDHC value be First, according to process mechanism knowledge, the membership function parameters of the three fuzzy sets located in the central region, i.e., normal, negative small, and positive small, are set.

[0077] First, the center value position c i of the membership function curve is determined. According to process mechanism knowledge, when the PDHC value is within , it indicates that the current alumina concentration is basically normal. At this time, the center point parameter c N of the membership function curve corresponding to the N fuzzy set can be calculated by and .

[0078]

[0079] where ω N is an adjustment coefficient used to correct the center point position.

[0080] When the value of PDHC is lower than and higher than , it indicates that the current alumina concentration is slightly higher than the normal value, but has not reached the case of abnormally high, and the value of PDHC at this time is set to belong to the NS fuzzy set. The center point C NS of the membership function f NS (x) can be calculated according to and :

[0081]

[0082] where ω Ns is the adjustment coefficient of the center point of f NS (x).

[0083] Similarly, when the value of PDHC is higher than and lower than , it indicates that the current alumina concentration is slightly lower than the normal value, but has not reached the case of abnormally low, and the value of PDHC at this time is set to be the PS fuzzy set. The center point C PS of the membership function f PS (x) can be calculated according to and :

[0084]

[0085] where ω PS is the adjustment coefficient of the center point of f PS (x).

[0086] When the value of PDHC is far beyond the normal range and breaks through the critical point or , it indicates that the current alumina concentration has reached the case of abnormally high or abnormally low, and the value of PDHC at this time belongs to the NB and PB fuzzy sets respectively, so the initial point a PDHC and the translation initial value d PDHC of the descending point of the membership function curves γ(x) and δ(x) corresponding to NB and PB can be determined according to the critical point or :

[0087]

[0088] where ω NB and ω PB are the adjustment coefficients of the adjustment parameters of γ(x) and δ(x) respectively.

[0089] After the above parameters are determined, the width of the bell-shaped membership function and the slope of the S-shaped and Z-shaped membership functions can be determined according to the parameters.

[0090] For the membership function corresponding to the fuzzy set of category N, the width can be determined by the center position c N of the function and the center points C NS and C PS of the adjacent two membership function curves f NS (x) and f PS (x). According to historical data, the normal value range of PDHC accounts for a small proportion in the overall range, and thus the width of the membership function corresponding to the N fuzzy set is also small. In the embodiment, the minimum value function is used to calculate the width of f N (x), and the calculation method is as follows:

[0091]

[0092] wherein, is the width correction coefficient of the curve f N (x).

[0093] For the fuzzy set corresponding to NS and PS, the maximum value function is used to calculate the width of the membership function curve, and the calculation method is as follows:

[0094]

[0095] wherein, and are the width correction coefficients of the curves f NS (x) and f PS (x), respectively.

[0096] The horizontal intersection point b PDHC and the steepness parameter c PDHC of the membership function corresponding to the fuzzy set of NB and PB are determined by the known parameters of the membership function and the width parameters of the adjacent functions. PDHC The horizontal intersection point b of γ(x) is:

[0097]

[0098] wherein, is the intersection position adjustment coefficient.

[0099] Similarly, the steepness parameter c PDHC of δ(x) is:

[0100]

[0101] wherein, The coefficient of the steepness parameter is adjusted for δ(x).

[0102] After parameter setting, the PDHC membership function parameter summary table is shown in Table 1, and the PDHC membership function curve is shown in Figure 1. Figure 2

[0103] Table 1

[0104]

[0105] The identification of alumina concentration not only considers the current state of alumina concentration in the comprehensive feeding period, but also considers the trend of alumina concentration. Good trend can not only indicate that the current electrolytic cell is in a stable state, but also can improve the current state of alumina concentration under certain conditions, so that the alumina concentration develops from high or low to normal. Therefore, the alumina concentration identification fuzzy inference rule needs to be set according to the PDHC values of two adjacent comprehensive feeding periods.

[0106] According to the characteristics of the aluminum electrolysis process, the current state of alumina concentration is divided into five categories: normal state of alumina concentration (Z), high state of alumina concentration (H), abnormally high state of alumina concentration (AH), low state of alumina concentration (L), and abnormally low state of alumina concentration (AL).

[0107] The trend of alumina concentration can be divided into five categories: the state of alumina concentration is basically unchanged (S), the state of alumina concentration changes from high to normal (HN), the state of alumina concentration changes from low to normal (LN), the state of alumina concentration decreases (TL), and the state of alumina concentration increases (TH).

[0108] There is a corresponding relationship between the PDHC value in the current comprehensive feeding period and the state of alumina concentration. According to the corresponding relationship, the fuzzy inference rule is as follows:

[0109] IF PDHC1=N THEN P=Z

[0110] IF PDHC1=NS THEN P=H

[0111] IF PDHC1=PS THEN P=L

[0112] ​IF PDHC1 = NB THEN P = AH IF PDHC1 = PB THEN P = AL

[0113] wherein PDHC1 is the value of the current comprehensive tapping cycle PDHC, and P is the current comprehensive tapping cycle alumina concentration state.

[0114] The alumina concentration trend T is inferred from the PDHC value PDHC2 of the previous comprehensive tapping cycle and the PDHC value PDHC1 of the current comprehensive tapping cycle. If the PDHC2 value of the previous tapping cycle and the PDHC1 value of the current comprehensive tapping cycle are in the same fuzzy set, i.e. the alumina concentration state of the previous comprehensive tapping cycle and the alumina concentration state of the current comprehensive tapping cycle are basically the same, it indicates that the alumina concentration state is basically unchanged. The inference rule for this case is:

[0115] IF (PDHC1 = N and PDHC2 = N) or (PDHC1 = NS and PDHC2 = NS) or (PDHC1 = PS and PDHC2 = PS) or (PDHC1 = NB and PDHC2 = NB) or (PDHC1 = PB and PDHC2 = PB), THEN T = S

[0116] When the PDHC value returns to the normal range from a smaller value (NS) or an abnormally small value (NB) beyond the normal range, it indicates that the alumina concentration is from high to normal. The inference rule for this case is:

[0117] IF (PDHC2 = NS and PDHC1 = N) or (PDHC2 = NB and PDHC1 = N), THEN T = HN

[0118] Among them, the weight of PDHC2 = NB and PDHC1 = N is smaller, because in the actual aluminum electrolysis production process, it is difficult for the alumina concentration to adjust from abnormally high to normal state within one comprehensive tapping cycle.

[0119] Similarly, when the PDHC value returns to the normal range from a larger value (PS) or an abnormally large value (PB) beyond the normal range, it indicates that the alumina concentration is from low to normal. The inference rule for this case is:

[0120] IF (PDHC2 = PS and PDHC1 = N) or (PDHC2 = PB and PDHC1 = N), THEN T = LN

[0121] Similarly, in the case of In this case, PDHC2=PB and PDHC1=N have smaller weights.

[0122] The PDHC value of the current comprehensive blanking cycle and the PDHC value of the current comprehensive blanking cycle do not belong to the same fuzzy set, and the fuzzy set in which the PDHC value of the current comprehensive blanking cycle is located is on the right side of the membership function curve given in FIG. 1. This indicates that the alumina concentration is reduced. The reasoning rule for this case is: Figure 2

[0123] IF (PDHC2=NB and PDHC1=NS) or (PDHC2=NB and PDHC1=PS) or (PDHC2=N and PDHC1=PS) or (PDHC2=N and PDHC1=PB) or (PDHC2=PS and PDHC1=PB) or (PDHC2=NS and PDHC1=PS) or (PDHC2=NS and PDHC1=PB), THEN T=TL

[0124] Similarly, when the PDHC2 value and the PDHC1 value do not belong to the same fuzzy set, and the fuzzy set in which the PDHC1 is located is on the left side of the membership function curve given in FIG. 1, it indicates that the alumina concentration is increased. The reasoning rule for this case is: Figure 2

[0125] IF (PDHC2=PB and PDHC1=PS) or (PDHC2=PB and PDHC1=NS) or (PDHC2=N and PDHC1=NS) or (PDHC2=N and PDHC1=NB) or (PDHC2=NS and PDHC1=NB) or (PDHC2=PS and PDHC1=NS) or (PDHC2=PS and PDHC1=NB), THEN T=TH

[0126] The above-mentioned five fuzzy reasoning rules are summarized in the form of a table, and the result is shown in Table 2. Among them, S indicates that the alumina concentration is basically unchanged, TH indicates that the alumina concentration is increased, TL indicates that the alumina concentration is reduced, and HN and LN respectively indicate that the alumina concentration is from low to normal and from high to normal.

[0127] Table 2: Alumina concentration identification rule based on PDHC

[0128]

[0129] ​​Since in the actual aluminum electrolysis process, the alumina concentration of the previous comprehensive discharging cycle is not abnormally high (abnormally low), and the alumina concentration of the next comprehensive discharging cycle is not abnormally low (abnormally high). Therefore, when PDHC2=PB and PDHC1=NB or PDHC2=NB and PDHC1=PB, the fuzzy inference machine will not output the change trend of the alumina concentration.

[0130] After obtaining the identification variable through the fuzzy inference method, the identification variable needs to be converted into a real variable through the defuzzification method. Before defuzzification, the value range of the identification variable needs to be set, and five types of membership functions corresponding to the identification result of the current alumina concentration state and the change trend of the alumina concentration are set on the value range. In this embodiment, the identification value range of the current alumina concentration state and the change trend of the alumina concentration is determined as [-3, 3]. For the current alumina concentration state (P), the identification value is close to 0, indicating that the current alumina concentration is in a normal state, and the larger the identification value is and the closer to 3, the higher the alumina concentration is; the smaller the identification value is and the closer to -3, the lower the alumina concentration is. For the change trend of the alumina concentration (Z), when the identification value is close to 0, it indicates that the alumina concentration is basically unchanged, the larger the identification value is and the closer to 3, the higher the alumina concentration is; the smaller the identification value is and the closer to -3, the lower the alumina concentration is. In this embodiment, the triangular membership function is used to establish the membership functions of the current state and the change trend of the alumina concentration, and the form of the triangular membership function is:

[0131]

[0132] wherein a i and c i are the horizontal coordinates when σ i is 0, and b i is the horizontal coordinate when σ i takes the maximum value.

[0133] According to the characteristics of the aluminum electrolysis production data, the parameters of each membership function are set, and the parameters are shown in Table 3.

[0134] Table 3 Membership function parameters of the alumina concentration state and the change trend

[0135]

[0136] After the membership function of the fuzzy system output value is determined, according to the fuzzy inference result, the online identification semantic value of the alumina concentration is converted into an accurate numerical value through the defuzzification method. In this embodiment, the gravity method is used for defuzzification, and the gravity of the identification result fuzzy set is used as the result after defuzzification, and the calculation formula is:

[0137]

[0138] wherein v0 is the exact value after de-fuzzification, v k is the identified variable, σ i represents the membership function of the identification result.

[0139] After the fuzzification, fuzzy rule reasoning and de-fuzzification, the relationship between the PDHC value and the current alumina concentration state and alumina concentration trend can be obtained. Since the alumina concentration state is only related to the current PDHC1 value of the discharging state, this embodiment only gives the surface graph of the PDHC1 value and the alumina concentration trend, as shown in FIG. 6. For the experimental results, the evaluation indexes used in this embodiment mainly include: accuracy, precision, recall and F1-score, etc. Among them, accuracy is the most common evaluation index, which calculates the ratio of the number of correctly classified samples to the total number of samples, but this index is less applicable to class- imbalanced data sets. The experimental data used for alumina concentration identification is real-time industrial data obtained from industrial sensors, most of which are in a normal state. Therefore, accuracy cannot comprehensively evaluate the classification performance, and is only used as a reference in this embodiment. Figure 3 Precision represents the proportion of all samples classified as positive classes that are actually positive classes, and its focus is on the accuracy of the positive classes of the classifier, i.e. the reliability of the classifier when classifying positive classes.

[0140] Recall represents the proportion of the number of samples correctly classified as positive classes among all samples that are actually positive classes, and its focus is on the coverage of all true positive classes of the classifier, i.e. the integrity of the classifier when classifying all samples that are actually positive classes.

[0141] F1-score is the harmonic mean of precision and recall, which considers the balance between precision and recall, and can better evaluate the overall performance of the classifier.

[0142] The calculation methods of the four evaluation indexes are as follows:

[0143]

[0144]

[0145] Among them, T and F in TP, TN, FP and FN represent whether the classification result matches the actual value, and P and N represent whether the classification result is a positive class or a negative class. For example, in a multi-classification task, there are A, B and C three types in total, and for A, the positive class is A and the negative class is B and C.

[0146] ​The embodiment adopts an alumina concentration fuzzy recognition method based on a multi-granularity timing knowledge graph, and uses PDHC data for online recognition of alumina concentration. The PDHC data used in the experiment is derived from the multi-granularity knowledge graph, and contains 8935 pieces of PDHC values and corresponding comprehensive discharge periods of the No. 5201 electrolytic cell from August 2021 to June 2022. The above data is used as the input of the fuzzy inference machine, and the fuzzy inference method is used for alumina concentration state recognition. The final recognition result is shown in Figure 4

[0147] The results of alumina concentration fuzzy recognition are represented by a confusion matrix. Among them, AL, L, Z, H and AH are alumina concentration state labels, respectively representing alumina concentration abnormally low, low, normal, high and abnormally high. The horizontal coordinate of the confusion matrix represents the actual value label, and the vertical coordinate represents the classification result label. The elements in each cell represent the classification proportion, and the elements on the main diagonal represent the proportion of correct classification of each category to the actual sample, that is, the recall rate of the category. The evaluation index values of each alumina concentration state are shown in Table 4.

[0148] Table 4: Evaluation index value summary of each alumina concentration state

[0149]

[0150] According to the experimental results, the following conclusions can be drawn:

[0151] (1) Overall, the alumina concentration fuzzy recognition method can accurately identify the alumina concentration state. The classification accuracy Accuracy of the method can be calculated by the total number of correct classifications in Table 3 and the total number of actual samples, which is about 0.995, indicating that the method has high accuracy in identifying alumina concentration state.

[0152] (2) For samples with alumina concentration state as "normal", the recognition effect of the method is excellent. Since the proportion of samples with alumina concentration state as "normal" is very large, the precision and recall rate need to be considered to ensure that the classification method can avoid misclassification of other types of samples as "normal" as much as possible while capturing all truly "normal" samples. As can be seen from Table 3, the precision and recall rate of alumina concentration state as "normal" are both 0.999, indicating that the method has high classification reliability and integrity when facing samples with alumina concentration as "normal", and can achieve comprehensive coverage of "normal" samples.

[0153] ​(3) For other individual state samples, the recognition effect of the method is good. For the "abnormal" samples with alumina concentration in the "abnormally low", "low", "high" and "abnormally high" states, the precision, recall rate and F1 score of the method are at a high level, indicating that the method has good recognition performance in the face of samples with alumina concentration in the "abnormal" state.

[0154] (4) The method basically does not misidentify the samples with alumina concentration in the high state as low, and vice versa. In the aluminum electrolysis process, when the alumina concentration is in the low or high state, different control strategies are needed to adjust the alumina concentration to the normal state. If the high state is misidentified as low, it will lead to the use of the wrong control strategy, causing the alumina concentration to be out of control. Figure 4 As shown in the confusion matrix, the proportion of samples misidentified by the method from the "low" state to the "high" state is 0.001, and the proportion of samples misidentified from the "high" state to the "low" state is 0.004, which indicates that the method can accurately identify samples with alumina concentration in the low or high state, thereby providing correct decision support for aluminum electrolysis production discharging.

[0155] Table 5 Comparison between different methods

[0156]

[0157] The PDHC and IFP used in the method proposed in this embodiment can be obtained online, and through fuzzy reasoning, online identification of the normal and abnormal states of the alumina concentration can be realized. The identification time required for each comprehensive discharging period is about one discharging interval, which basically meets the needs of online identification of the alumina concentration. The method proposed in this embodiment is superior to existing methods in terms of identification range, real-time performance, accuracy and other dimensions. Subsequent optimization of the method is needed to reduce the time required for each identification and improve the time performance of the method.

[0158] Figure 5 is a structural diagram of an aluminum electrolysis alumina concentration online identification system based on fuzzy reasoning according to an embodiment of the present disclosure. Figures 1-4 The embodiment shown in FIG. 1 can be used to explain the present embodiment. As shown in FIG. 1, an aluminum electrolysis alumina concentration online identification system based on fuzzy reasoning includes: Figure 5 The aluminum electrolysis alumina concentration online identification system based on fuzzy reasoning includes:

[0159] The PDHC acquisition module 501 is configured to acquire the apparent alumina concentration and the comprehensive discharging period through the time sequence knowledge graph, and calculate the PDHC value of the current comprehensive discharging period and the previous comprehensive discharging period.

[0160] The fuzzy inference module 502 is configured to input the PDHC value of the current comprehensive blanking cycle and a previous comprehensive blanking cycle into a preset fuzzy inference machine to obtain fuzzy inference output variables of an alumina concentration state and an alumina concentration change trend of the current comprehensive blanking cycle.

[0161] The online identification module 503 is configured to perform defuzzification processing on the fuzzy inference output variables of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking cycle to obtain identification evaluation values of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking cycle.

[0162] Preferably, the preset fuzzy inference machine comprises:

[0163] The PDHC membership function modeling module is configured to select a first threshold value, a second threshold value, a third threshold value and a fourth threshold value according to a numerical range of the PDHC value in each state of the alumina concentration based on historical data of each type of electrolytic cell, and preset a fuzzy threshold range for each type of electrolytic cell; the second threshold value < the first threshold value < the third threshold value < the fourth threshold value; the fuzzy threshold range comprises an alumina concentration normal threshold range for representing that the PDHC value is greater than the first threshold value and less than the third threshold value, a negative small threshold range for representing that the PDHC value is less than the first threshold value and greater than the second threshold value, a negative large threshold range for representing that the PDHC value is less than the second threshold value, a positive small threshold range for representing that the PDHC value is greater than the third threshold value and less than the fourth threshold value, and a positive large threshold range for representing that the PDHC value is greater than the fourth threshold value;

[0164] The alumina concentration membership function modeling module is configured to establish a fuzzy inference rule of the alumina concentration according to the PDHC value of the current comprehensive blanking cycle; the fuzzy inference rule of the alumina concentration change trend comprises an alumina concentration normal state for representing that the PDHC value is located in the normal threshold range, an alumina concentration high state for representing that the PDHC value is located in the negative small threshold range, an alumina concentration extremely high state for representing that the PDHC value is located in the negative large threshold range, an alumina concentration low state for representing that the PDHC value is located in the positive small threshold range, and an alumina concentration extremely low state for representing that the PDHC value is located in the positive large threshold range; and the alumina concentration membership function corresponding to the alumina concentration normal state, the alumina concentration low state, the alumina concentration high state, the alumina concentration extremely low state and the alumina concentration extremely high state is established.

[0165] The alumina concentration change trend membership function modeling module is configured to establish fuzzy inference rules of the alumina concentration change trend according to the PDHC values of two comprehensive blanking periods, the fuzzy inference rules of the alumina concentration change trend including a first alumina concentration change trend for representing an unchanged alumina concentration, a second alumina concentration change trend for representing an increased alumina concentration, a third alumina concentration change trend for representing a decreased alumina concentration, a fourth alumina concentration change trend for representing a low-to-normal alumina concentration, and a fifth alumina concentration change trend for representing a high-to-normal alumina concentration; and establish alumina concentration change trend membership functions corresponding to the first alumina concentration change trend, the second alumina concentration change trend, the third alumina concentration change trend, the fourth alumina concentration change trend, and the fifth alumina concentration change trend.

[0166] Preferably, the fuzzy inference module 502 is specifically configured to obtain PDHC fuzzy values corresponding to the PDHC values of the current comprehensive blanking period and the previous comprehensive blanking period according to the respective PDHC membership functions; and obtain fuzzy inference output variables of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking period according to the PDHC fuzzy values corresponding to the PDHC values of the current comprehensive blanking period and the previous comprehensive blanking period, the fuzzy inference rules of the alumina concentration, and the fuzzy inference rules of the alumina concentration change trend.

[0167] The online identification module 503 is specifically configured to perform defuzzification processing on the fuzzy inference output variables of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking period by using the barycentric method according to the alumina concentration membership function and the alumina concentration change trend membership function, to obtain identification evaluation values of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking period.

[0168] The embodiment utilizes the alumina concentration online identification method based on the multi-granularity time sequence knowledge graph, performs online identification of the alumina concentration by obtaining the PDHC values through the time sequence knowledge graph, stores the identification results in the time sequence knowledge graph, establishes a corresponding relationship between the electrolytic cell and the alumina concentration according to the results, and reflects the online working conditions in the aluminum electrolysis production process.

[0169] It should be noted that not all steps and modules in the above processes and system structure diagrams are necessary, and some steps or modules can be omitted according to actual needs. The execution order of each step is not fixed and can be adjusted as needed. The system structure described in each of the above embodiments can be a physical structure or a logical structure, that is, some modules can be implemented by the same physical entity, or some modules can be implemented by multiple physical entities, or can be implemented by some components in multiple independent devices together.

[0170] In each of the above embodiments, a hardware unit can be implemented mechanically or electrically. For example, a hardware unit can include permanent dedicated circuitry or logic (such as a dedicated processor, FPGA, or ASIC) to complete the corresponding operation. The hardware unit can also include programmable logic or circuitry (such as a general-purpose processor or other programmable processor), which can be temporarily set by software to complete the corresponding operation. The specific implementation method (mechanical method, or dedicated permanent circuit, or temporarily set circuit) can be determined based on cost and time considerations.

[0171] The above has been disclosed and described in detail through the drawings and preferred embodiments, but the present application is not limited to these disclosed embodiments, and those skilled in the art can know that the code review means in the above different embodiments can be combined to obtain more embodiments of the present application, and these embodiments are also within the protection scope of the present application.

Claims

1. A method for on-line identification of alumina concentration in aluminum electrolysis based on fuzzy inference, characterized by, The method comprises the following steps: Obtaining the apparent alumina concentration and the comprehensive blanking cycle through the time sequence knowledge graph, calculating the PDHC value of the current comprehensive blanking cycle and the previous comprehensive blanking cycle; Inputting the PDHC value of the current comprehensive blanking cycle and the previous comprehensive blanking cycle into a preset fuzzy inference machine to obtain the fuzzy inference output variable of the alumina concentration state and the alumina concentration trend of the current comprehensive blanking cycle; Performing defuzzification processing on the fuzzy inference output variable of the alumina concentration state and the alumina concentration trend of the current comprehensive blanking cycle to obtain the recognition evaluation value of the alumina concentration state and the alumina concentration trend of the current comprehensive blanking cycle, and the preset fuzzy inference machine is established by the following steps: According to the historical data of various types of electrolytic cells, the numerical range of the PDHC value under each state of the alumina concentration is obtained to select the first threshold value, the second threshold value, the third threshold value and the fourth threshold value, and a fuzzy threshold range is preset for each type of electrolytic cell; wherein the second threshold value < the first threshold value < the third threshold value < the fourth threshold value; the fuzzy threshold range includes an alumina concentration normal threshold range for representing that the PDHC value is greater than the first threshold value and less than the third threshold value, a negative small threshold range for representing that the PDHC value is less than the first threshold value and greater than the second threshold value, a negative large threshold range for representing that the PDHC value is less than the second threshold value, a positive small threshold range for representing that the PDHC value is greater than the third threshold value and less than the fourth threshold value, and a positive large threshold range for representing that the PDHC value is greater than the fourth threshold value; Establishing each PDHC membership function corresponding to the normal threshold range, the negative small threshold range, the negative large threshold range, the positive small threshold range and the positive large threshold range; According to the PDHC value of the current comprehensive blanking cycle, the fuzzy inference rule of the alumina concentration is established, and the fuzzy inference rule of the alumina concentration trend includes an alumina concentration normal state for representing that the PDHC value is located in the normal threshold range, an alumina concentration high state for representing that the PDHC value is located in the negative small threshold range, an alumina concentration extremely high state for representing that the PDHC value is located in the negative large threshold range, an alumina concentration low state for representing that the PDHC value is located in the positive small threshold range, and an alumina concentration extremely low state for representing that the PDHC value is located in the positive large threshold range; Establishing the alumina concentration membership function corresponding to the alumina concentration normal state, the alumina concentration low state, the alumina concentration high state, the alumina concentration extremely low state and the alumina concentration extremely high state. According to the PDHC values of two comprehensive discharging periods, a fuzzy inference rule of the alumina concentration change trend is established, and the fuzzy inference rule of the alumina concentration change trend includes a first alumina concentration change trend for characterizing an invariable alumina concentration, a second alumina concentration change trend for representing an increasing alumina concentration, a third alumina concentration change trend for representing a decreasing alumina concentration, a fourth alumina concentration change trend for representing a low-to-normal alumina concentration, and a fifth alumina concentration change trend for representing a high-to-normal alumina concentration; Membership functions corresponding to the first alumina concentration change trend, the second alumina concentration change trend, the third alumina concentration change trend, the fourth alumina concentration change trend, and the fifth alumina concentration change trend are established.

2. The fuzzy inference based on-line identification method of alumina concentration in aluminum electrolysis according to claim 1, characterized in that, The step of inputting the PDHC values of the current comprehensive discharging period and the previous comprehensive discharging period into a preset fuzzy inference machine to obtain fuzzy inference output variables of the alumina concentration state and the alumina concentration change trend of the current comprehensive discharging period includes: According to the PDHC membership functions, PDHC fuzzy degree values corresponding to the PDHC values of the current comprehensive discharging period and the previous comprehensive discharging period are obtained; According to the PDHC fuzzy degree values corresponding to the PDHC values of the current comprehensive discharging period and the previous comprehensive discharging period, the fuzzy inference rule of the alumina concentration, and the fuzzy inference rule of the alumina concentration change trend, fuzzy inference output variables of the alumina concentration state and the alumina concentration change trend of the current comprehensive discharging period are obtained.

3. The fuzzy-reasoning-based on-line identification method of aluminum electrolytic alumina concentration according to claim 2, characterized in that, The step of performing defuzzification processing on the fuzzy inference output variables of the alumina concentration state and the alumina concentration change trend of the current comprehensive discharging period to obtain recognition evaluation values of the alumina concentration state and the alumina concentration change trend of the current comprehensive discharging period further includes: According to the alumina concentration membership function and the alumina concentration change trend membership function, the fuzzy inference output variables of the alumina concentration state and the alumina concentration change trend of the current comprehensive discharging period are defuzzified by using the barycentric method to obtain the recognition evaluation values of the alumina concentration state and the alumina concentration change trend of the current comprehensive discharging period.

4. The fuzzy-reasoning-based on-line identification method of aluminum electrolysis alumina concentration according to claim 3, characterized in that, The PDHC membership functions of the normal threshold range, the negative small threshold range, and the positive small threshold range are generalized bell-shaped membership functions; the PDHC membership function of the negative large threshold range is a Z-shaped membership function, and the PDHC membership function of the positive large threshold range is an S-shaped membership function; the alumina concentration membership function and the alumina concentration change trend membership function are triangular membership functions; The step of obtaining the recognition evaluation values of the alumina concentration state and the alumina concentration change trend of the current comprehensive discharging period further includes: The recognition evaluation value of the alumina concentration state and alumina concentration change trend of the current comprehensive blanking cycle is stored in the time sequence knowledge graph, and a corresponding relationship between the electrolytic cell and the alumina concentration is established according to the recognition evaluation value of the alumina concentration state and alumina concentration change trend of the current comprehensive blanking cycle, so as to reflect the online working condition in the aluminum electrolysis production process.

5. A fuzzy inference based on-line identification system for alumina concentration in aluminum electrolysis, characterized by, Comprise: The PDHC acquisition module is used for acquiring the apparent alumina concentration and the comprehensive blanking cycle through the time sequence knowledge graph, and calculating the PDHC value of the current comprehensive blanking cycle and the previous comprehensive blanking cycle; The fuzzy reasoning module is used for inputting the PDHC value of the current comprehensive blanking cycle and the previous comprehensive blanking cycle into the preset fuzzy reasoning machine to obtain the fuzzy reasoning output variable of the alumina concentration state and alumina concentration change trend of the current comprehensive blanking cycle; The online recognition module is used for anti-fuzzification processing of the fuzzy reasoning output variable of the alumina concentration state and alumina concentration change trend of the current comprehensive blanking cycle to obtain the recognition evaluation value of the alumina concentration state and alumina concentration change trend of the current comprehensive blanking cycle, The preset fuzzy reasoning machine comprises: The PDHC membership function modeling module is used for selecting the first threshold value, the second threshold value, the third threshold value and the fourth threshold value according to the numerical range of the PDHC value in each state of the alumina concentration, and pre-setting the fuzzy threshold range for each type of electrolytic cell; wherein the second threshold value < the first threshold value < the third threshold value < the fourth threshold value; the fuzzy threshold range comprises an alumina concentration normal threshold range for representing that the PDHC value is greater than the first threshold value and less than the third threshold value, a negative small threshold range for representing that the PDHC value is less than the first threshold value and greater than the second threshold value, a negative large threshold range for representing that the PDHC value is less than the second threshold value, a positive small threshold range for representing that the PDHC value is greater than the third threshold value and less than the fourth threshold value, and a positive large threshold range for representing that the PDHC value is greater than the fourth threshold value; each PDHC membership function corresponding to the normal threshold range, the negative small threshold range, the negative large threshold range, the positive small threshold range and the positive large threshold range is established; The alumina concentration membership function modeling module is used for establishing the fuzzy reasoning rule of the alumina concentration according to the PDHC value of the current comprehensive blanking cycle, and the fuzzy reasoning rule of the alumina concentration change trend comprises an alumina concentration normal state for representing that the PDHC value is located in the normal threshold range, an alumina concentration high state for representing that the PDHC value is located in the negative small threshold range, an alumina concentration extremely high state for representing that the PDHC value is located in the negative large threshold range, an alumina concentration low state for representing that the PDHC value is located in the positive small threshold range, and an alumina concentration extremely low state for representing that the PDHC value is located in the positive large threshold range; The alumina concentration change trend membership function modeling module is configured to establish fuzzy inference rules of the alumina concentration change trend according to the PDHC values of two comprehensive blanking cycles, the fuzzy inference rules of the alumina concentration change trend including a first alumina concentration change trend for representing no change of the alumina concentration, a second alumina concentration change trend for representing an increase of the alumina concentration, a third alumina concentration change trend for representing a decrease of the alumina concentration, a fourth alumina concentration change trend for representing a change of the alumina concentration from low to normal, and a fifth alumina concentration change trend for representing a change of the alumina concentration from high to normal; and establish alumina concentration change trend membership functions corresponding to the first alumina concentration change trend, the second alumina concentration change trend, the third alumina concentration change trend, the fourth alumina concentration change trend, and the fifth alumina concentration change trend.

6. The fuzzy-reasoning-based on-line identification system for alumina concentration in aluminum electrolysis according to claim 5, wherein The fuzzy inference module is specifically configured to obtain PDHC fuzzy values corresponding to the PDHC values of the current comprehensive blanking cycle and the previous comprehensive blanking cycle according to the PDHC membership functions; and obtain fuzzy inference output variables of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking cycle according to the PDHC fuzzy values of the current comprehensive blanking cycle and the previous comprehensive blanking cycle, the fuzzy inference rules of the alumina concentration, and the fuzzy inference rules of the alumina concentration change trend.

7. The fuzzy-reasoning-based on-line identification system for alumina concentration in aluminum electrolysis according to claim 6, characterized in that, The online identification module is specifically configured to perform defuzzification processing on the fuzzy inference output variables of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking cycle by using the gravity center method according to the alumina concentration membership function and the alumina concentration change trend membership function, to obtain identification evaluation values of the alumina concentration state and the alumina concentration change trend of the current comprehensive blanking cycle.

8. The fuzzy-reasoning-based on-line identification system for alumina concentration in aluminum electrolysis according to claim 7, characterized in that, The PDHC membership functions of the normal threshold range, the negative small threshold range, and the positive small threshold range are generalized bell-shaped membership functions; the PDHC membership function of the negative large threshold range is a Z-shaped membership function, and the PDHC membership function of the positive large threshold range is an S-shaped membership function. The alumina concentration membership function and the alumina concentration change trend membership function are triangular membership functions.

Images